Whamcloud - gitweb
03c8d4d8d65dd2f8e7dc909ef013e112f517d99a
[fs/lustre-release.git] / lustre / llite / llite_lib.c
1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2  * vim:expandtab:shiftwidth=8:tabstop=8:
3  *
4  * Lustre Light Super operations
5  *
6  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
7  *
8  *   This file is part of Lustre, http://www.lustre.org.
9  *
10  *   Lustre is free software; you can redistribute it and/or
11  *   modify it under the terms of version 2 of the GNU General Public
12  *   License as published by the Free Software Foundation.
13  *
14  *   Lustre is distributed in the hope that it will be useful,
15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  *   GNU General Public License for more details.
18  *
19  *   You should have received a copy of the GNU General Public License
20  *   along with Lustre; if not, write to the Free Software
21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22  */
23
24 #define DEBUG_SUBSYSTEM S_LLITE
25
26 #include <linux/module.h>
27 #include <linux/types.h>
28 #include <linux/random.h>
29 #include <linux/version.h>
30
31 #include <linux/lustre_lite.h>
32 #include <linux/lustre_ha.h>
33 #include <linux/lustre_dlm.h>
34 #include <linux/lprocfs_status.h>
35 #include <linux/lustre_acl.h>
36 #include "llite_internal.h"
37
38 kmem_cache_t *ll_file_data_slab;
39 kmem_cache_t *ll_intent_slab;
40
41 extern struct address_space_operations ll_aops;
42 extern struct address_space_operations ll_dir_aops;
43
44 #ifndef log2
45 #define log2(n) ffz(~(n))
46 #endif
47
48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
49 {
50         struct ll_sb_info *sbi = NULL;
51         class_uuid_t uuid;
52         ENTRY;
53
54         OBD_ALLOC(sbi, sizeof(*sbi));
55         if (!sbi)
56                 RETURN(NULL);
57
58         spin_lock_init(&sbi->ll_lock);
59         INIT_LIST_HEAD(&sbi->ll_pglist);
60         sbi->ll_pglist_gen = 0;
61         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
62                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
63         else
64                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
65         INIT_LIST_HEAD(&sbi->ll_conn_chain);
66         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
67         INIT_LIST_HEAD(&sbi->ll_mnt_list);
68         
69         sema_init(&sbi->ll_gns_sem, 1);
70         spin_lock_init(&sbi->ll_gns_lock);
71         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
72         init_waitqueue_head(&sbi->ll_gns_waitq);
73         init_completion(&sbi->ll_gns_mount_finished);
74
75         /* this later may be reset via /proc/fs/... */
76         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
77         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
78         
79         /* this later may be reset via /proc/fs/... */
80         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
81                strlen("/usr/sbin/gns_upcall"));
82         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
83
84         /* default values, may be changed via /proc/fs/... */
85         sbi->ll_gns_state = LL_GNS_IDLE;
86         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
87         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
88
89         sbi->ll_gns_timer.data = (unsigned long)sbi;
90         sbi->ll_gns_timer.function = ll_gns_timer_callback;
91         init_timer(&sbi->ll_gns_timer);
92
93         ll_set_sbi(sb, sbi);
94
95         generate_random_uuid(uuid);
96         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
97         RETURN(sbi);
98 }
99
100 void lustre_free_sbi(struct super_block *sb)
101 {
102         struct ll_sb_info *sbi = ll_s2sbi(sb);
103         ENTRY;
104
105         if (sbi != NULL) {
106                 list_del(&sbi->ll_gns_sbi_head);
107                 del_timer(&sbi->ll_gns_timer);
108                 OBD_FREE(sbi, sizeof(*sbi));
109         }
110         ll_set_sbi(sb, NULL);
111         EXIT;
112 }
113
114 int lustre_init_dt_desc(struct ll_sb_info *sbi)
115 {
116         int valsize, rc;
117         ENTRY;
118         
119         valsize = sizeof(sbi->ll_dt_desc);
120         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
121         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
122                           "lovdesc", &valsize, &sbi->ll_dt_desc);
123         RETURN(rc);
124 }
125
126 extern struct dentry_operations ll_d_ops;
127
128 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
129                              char *security, __u32 *nllu, int async)
130 {
131         struct ll_sb_info *sbi = ll_s2sbi(sb);
132         struct ptlrpc_request *request = NULL;
133         struct lustre_handle dt_conn = {0, };
134         struct lustre_handle md_conn = {0, };
135         struct inode *root = NULL;
136         struct obd_device *obd;
137         struct obd_statfs osfs;
138         struct lustre_md md;
139         kdev_t devno;
140         int err;
141         ENTRY;
142
143         obd = class_name2obd(lmv);
144         if (!obd) {
145                 CERROR("MDC %s: not setup or attached\n", lmv);
146                 RETURN(-EINVAL);
147         }
148         obd_set_info(obd->obd_self_export, strlen("async"), "async",
149                      sizeof(async), &async);
150
151         if (security == NULL)
152                 security = "null";
153
154         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
155                            strlen(security), security);
156         if (err) {
157                 CERROR("LMV %s: failed to set security %s, err %d\n",
158                         lmv, security, err);
159                 RETURN(err);
160         }
161
162         err = obd_set_info(obd->obd_self_export, strlen("nllu"), "nllu",
163                            sizeof(__u32) * 2, nllu);
164         if (err) {
165                 CERROR("LMV %s: failed to set NLLU, err %d\n",
166                         lmv, err);
167                 RETURN(err);
168         }
169
170         if (proc_lustre_fs_root) {
171                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
172                                                   lov, lmv);
173                 if (err < 0)
174                         CERROR("could not register mount in /proc/lustre");
175         }
176
177         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
178         if (err == -EBUSY) {
179                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
180                        " client is not a part.  Please wait for recovery to "
181                        "complete, abort, or time out.\n", lmv);
182                 GOTO(out, err);
183         } else if (err) {
184                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
185                 GOTO(out, err);
186         }
187         sbi->ll_md_exp = class_conn2export(&md_conn);
188         err = obd_statfs(obd, &osfs, jiffies - HZ);
189         if (err)
190                 GOTO(out_lmv, err);
191
192         if (!osfs.os_bsize) {
193                 CERROR("Invalid block size is detected.");
194                 GOTO(out_lmv, err);
195         }
196
197         sb->s_magic = LL_SUPER_MAGIC;
198         sb->s_blocksize = osfs.os_bsize;
199         sb->s_blocksize_bits = log2(osfs.os_bsize);
200         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
201        
202         devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid, 
203                              strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
204
205         sb->s_dev = devno;
206
207         obd = class_name2obd(lov);
208         if (!obd) {
209                 CERROR("OSC %s: not setup or attached\n", lov);
210                 GOTO(out_lmv, err);
211         }
212         obd_set_info(obd->obd_self_export, strlen("async"), "async",
213                      sizeof(async), &async);
214
215         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
216         if (err == -EBUSY) {
217                 CERROR("An OST (lov %s) is performing recovery, of which this"
218                        " client is not a part.  Please wait for recovery to "
219                        "complete, abort, or time out.\n", lov);
220                 GOTO(out, err);
221         } else if (err) {
222                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
223                 GOTO(out_lmv, err);
224         }
225         sbi->ll_dt_exp = class_conn2export(&dt_conn);
226
227         err = lustre_init_dt_desc(sbi);
228         if (err == 0) {
229                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
230                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
231                                  sbi->ll_dt_desc.ld_tgt_count *
232                                  sizeof(struct llog_cookie));
233         }
234         
235         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
236         if (err) {
237                 CERROR("cannot mds_connect: rc = %d\n", err);
238                 GOTO(out_lov, err);
239         }
240         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
241
242         sb->s_op = &lustre_super_operations;
243
244         /* make root inode
245          * XXX: move this to after cbd setup? */
246         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
247                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
248                          0, &request);
249         if (err) {
250                 CERROR("md_getattr failed for root: rc = %d\n", err);
251                 GOTO(out_lov, err);
252         }
253
254         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0, 
255                                 sbi->ll_dt_exp, &md);
256         if (err) {
257                 CERROR("failed to understand root inode md: rc = %d\n", err);
258                 ptlrpc_req_finished(request);
259                 GOTO(out_lov, err);
260         }
261
262         LASSERT(id_ino(&sbi->ll_rootid) != 0);
263         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
264
265         ptlrpc_req_finished(request);
266
267         if (root == NULL || is_bad_inode(root)) {
268                 CERROR("lustre_lite: bad iget4 for root\n");
269                 GOTO(out_root, err = -EBADF);
270         }
271
272         err = ll_close_thread_start(&sbi->ll_lcq);
273         if (err) {
274                 CERROR("cannot start close thread: rc %d\n", err);
275                 GOTO(out_root, err);
276         }
277
278         ll_gns_add_timer(sbi);
279
280         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
281            backing dev info assigned to inode mapping is used for
282            determining maximal readahead. */
283 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
284     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
285         /* bug 2805 - set VM readahead to zero */
286         vm_max_readahead = vm_min_readahead = 0;
287 #endif
288
289         sb->s_root = d_alloc_root(root);
290         sb->s_root->d_op = &ll_d_ops;
291
292 #ifdef S_PDIROPS
293         CWARN("Enabling PDIROPS\n");
294         sb->s_flags |= S_PDIROPS;
295 #endif
296
297         RETURN(err);
298 out_root:
299         if (root)
300                 iput(root);
301 out_lov:
302         obd_disconnect(sbi->ll_dt_exp, 0);
303 out_lmv:
304         obd_disconnect(sbi->ll_md_exp, 0);
305 out:
306         lprocfs_unregister_mountpoint(sbi);
307         return err;
308 }
309
310 void lustre_common_put_super(struct super_block *sb)
311 {
312         struct ll_sb_info *sbi = ll_s2sbi(sb);
313         struct hlist_node *tmp, *next;
314         ENTRY;
315
316         ll_gns_del_timer(sbi);
317         ll_close_thread_shutdown(sbi->ll_lcq);
318
319         list_del(&sbi->ll_conn_chain);
320         obd_disconnect(sbi->ll_dt_exp, 0);
321
322         lprocfs_unregister_mountpoint(sbi);
323         if (sbi->ll_proc_root) {
324                 lprocfs_remove(sbi->ll_proc_root);
325                 sbi->ll_proc_root = NULL;
326         }
327
328         obd_disconnect(sbi->ll_md_exp, 0);
329
330         // We do this to get rid of orphaned dentries. That is not really trw.
331         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
332                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
333                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
334                       dentry->d_name.len, dentry->d_name.name, dentry, next);
335                 shrink_dcache_parent(dentry);
336         }
337         EXIT;
338 }
339
340 char *ll_read_opt(const char *opt, char *data)
341 {
342         char *value;
343         char *retval;
344         ENTRY;
345
346         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
347         if (strncmp(opt, data, strlen(opt)))
348                 RETURN(NULL);
349         if ((value = strchr(data, '=')) == NULL)
350                 RETURN(NULL);
351
352         value++;
353         OBD_ALLOC(retval, strlen(value) + 1);
354         if (!retval) {
355                 CERROR("out of memory!\n");
356                 RETURN(NULL);
357         }
358
359         memcpy(retval, value, strlen(value)+1);
360         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
361         RETURN(retval);
362 }
363
364 int ll_set_opt(const char *opt, char *data, int fl)
365 {
366         ENTRY;
367
368         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
369         if (strncmp(opt, data, strlen(opt)))
370                 RETURN(0);
371         else
372                 RETURN(fl);
373 }
374
375 void ll_options(char *options, char **lov, char **lmv, char **sec, 
376                 int *async, int *flags)
377 {
378         char *this_char;
379 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
380         char *opt_ptr = options;
381 #endif
382         ENTRY;
383
384         if (!options) {
385                 EXIT;
386                 return;
387         }
388
389         *async = 0;
390 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
391         for (this_char = strtok (options, ",");
392              this_char != NULL;
393              this_char = strtok (NULL, ",")) {
394 #else
395         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
396 #endif
397                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
398                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
399                         continue;
400                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
401                         continue;
402                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
403                         *async = 1;
404                         continue;
405                 }
406                 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
407                         continue;
408                 if (!(*flags & LL_SBI_NOLCK) &&
409                     ((*flags) = (*flags) |
410                                 ll_set_opt("nolock", this_char,
411                                            LL_SBI_NOLCK)))
412                         continue;
413         }
414         
415         EXIT;
416 }
417
418 void ll_lli_init(struct ll_inode_info *lli)
419 {
420         sema_init(&lli->lli_open_sem, 1);
421         sema_init(&lli->lli_size_sem, 1);
422         lli->lli_flags = 0;
423         lli->lli_size_pid = 0;
424         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
425         spin_lock_init(&lli->lli_lock);
426         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
427         lli->lli_inode_magic = LLI_INODE_MAGIC;
428         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
429         sema_init(&lli->lli_och_sem, 1);
430         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
431         lli->lli_mds_exec_och = NULL;
432         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
433         lli->lli_open_fd_exec_count = 0;
434 }
435
436 int ll_fill_super(struct super_block *sb, void *data, int silent)
437 {
438         struct ll_sb_info *sbi;
439         char *lov = NULL;
440         char *lmv = NULL;
441         int async, err;
442         char *sec = NULL;
443         __u32 nllu[2] = { 99, 99 };
444         ENTRY;
445
446         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
447
448         sbi = lustre_init_sbi(sb);
449         if (!sbi)
450                 RETURN(-ENOMEM);
451
452         sbi->ll_flags |= LL_SBI_READAHEAD;
453         ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
454
455         if (!lov) {
456                 CERROR("no osc\n");
457                 GOTO(out, err = -EINVAL);
458         }
459
460         if (!lmv) {
461                 CERROR("no mdc\n");
462                 GOTO(out, err = -EINVAL);
463         }
464
465         err = lustre_common_fill_super(sb, lmv, lov, sec, nllu, async);
466         EXIT;
467 out:
468         if (err)
469                 lustre_free_sbi(sb);
470
471         if (sec)
472                 OBD_FREE(sec, strlen(sec) + 1);
473         if (lmv)
474                 OBD_FREE(lmv, strlen(lmv) + 1);
475         if (lov)
476                 OBD_FREE(lov, strlen(lov) + 1);
477         return err;
478 } /* ll_read_super */
479
480 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
481                               struct config_llog_instance *cfg, int allow_recov)
482 {
483         struct lustre_cfg lcfg;
484         struct portals_cfg pcfg;
485         char *peer = "MDS_PEER_UUID";
486         struct obd_device *obd;
487         struct lustre_handle md_conn = {0, };
488         struct obd_export *exp;
489         char *name = "mdc_dev";
490         class_uuid_t uuid;
491         struct obd_uuid lmv_uuid;
492         struct llog_ctxt *ctxt;
493         int rc = 0, err = 0;
494         ENTRY;
495
496         if (lmd_bad_magic(lmd))
497                 RETURN(-EINVAL);
498
499         generate_random_uuid(uuid);
500         class_uuid_unparse(uuid, &lmv_uuid);
501
502         if (lmd->lmd_local_nid) {
503                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
504                 pcfg.pcfg_nal = lmd->lmd_nal;
505                 pcfg.pcfg_nid = lmd->lmd_local_nid;
506                 rc = libcfs_nal_cmd(&pcfg);
507                 if (rc < 0)
508                         GOTO(out, rc);
509         }
510
511         if (lmd->lmd_nal == SOCKNAL ||
512             lmd->lmd_nal == OPENIBNAL ||
513             lmd->lmd_nal == IIBNAL ||
514             lmd->lmd_nal == VIBNAL ||
515             lmd->lmd_nal == RANAL) {
516                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
517                 pcfg.pcfg_nal     = lmd->lmd_nal;
518                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
519                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
520                 pcfg.pcfg_misc    = lmd->lmd_port;
521                 rc = libcfs_nal_cmd(&pcfg);
522                 if (rc < 0)
523                         GOTO(out, rc);
524         }
525
526         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
527         lcfg.lcfg_nid = lmd->lmd_server_nid;
528         lcfg.lcfg_inllen1 = strlen(peer) + 1;
529         lcfg.lcfg_inlbuf1 = peer;
530         lcfg.lcfg_nal = lmd->lmd_nal;
531         rc = class_process_config(&lcfg);
532         if (rc < 0)
533                 GOTO(out_del_conn, rc);
534
535         LCFG_INIT(lcfg, LCFG_ATTACH, name);
536         lcfg.lcfg_inlbuf1 = "mdc";
537         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
538         lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
539         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
540         err = class_process_config(&lcfg);
541         if (rc < 0)
542                 GOTO(out_del_uuid, rc);
543
544         LCFG_INIT(lcfg, LCFG_SETUP, name);
545         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
546         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
547         lcfg.lcfg_inlbuf2 = peer;
548         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
549         rc = class_process_config(&lcfg);
550         if (rc < 0)
551                 GOTO(out_detach, rc);
552
553         obd = class_name2obd(name);
554         if (obd == NULL)
555                 GOTO(out_cleanup, rc = -EINVAL);
556
557         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
558                           strlen(lmd->lmd_security), lmd->lmd_security);
559         if (rc)
560                 GOTO(out_cleanup, rc);
561
562         /* Disable initial recovery on this import */
563         rc = obd_set_info(obd->obd_self_export,
564                           strlen("initial_recov"), "initial_recov",
565                           sizeof(allow_recov), &allow_recov);
566         if (rc)
567                 GOTO(out_cleanup, rc);
568
569         rc = obd_connect(&md_conn, obd, &lmv_uuid, 0);
570         if (rc) {
571                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
572                 GOTO(out_cleanup, rc);
573         }
574
575         exp = class_conn2export(&md_conn);
576
577         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
578         rc = class_config_process_llog(ctxt, profile, cfg);
579         if (rc)
580                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
581
582         err = obd_disconnect(exp, 0);
583         
584         EXIT;
585 out_cleanup:
586         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
587         err = class_process_config(&lcfg);
588         if (err < 0)
589                 GOTO(out, err);
590
591 out_detach:
592         LCFG_INIT(lcfg, LCFG_DETACH, name);
593         err = class_process_config(&lcfg);
594         if (err < 0)
595                 GOTO(out, err);
596
597 out_del_uuid:
598         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
599         lcfg.lcfg_inllen1 = strlen(peer) + 1;
600         lcfg.lcfg_inlbuf1 = peer;
601         err = class_process_config(&lcfg);
602
603 out_del_conn:
604         if (lmd->lmd_nal == SOCKNAL ||
605             lmd->lmd_nal == OPENIBNAL ||
606             lmd->lmd_nal == IIBNAL ||
607             lmd->lmd_nal == VIBNAL ||
608             lmd->lmd_nal == RANAL) {
609                 int err2;
610
611                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
612                 pcfg.pcfg_nal     = lmd->lmd_nal;
613                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
614                 pcfg.pcfg_flags   = 1;          /* single_share */
615                 err2 = libcfs_nal_cmd(&pcfg);
616                 if (err2 && !err)
617                         err = err2;
618                 if (err < 0)
619                         GOTO(out, err);
620         }
621 out:
622         if (rc == 0)
623                 rc = err;
624
625         return rc;
626 }
627
628 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
629 {
630         struct lustre_cfg lcfg;
631         struct obd_device *obd;
632         int next = 0;
633
634         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
635         {
636                 int err;
637
638                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
639                 err = class_process_config(&lcfg);
640                 if (err) {
641                         CERROR("cleanup failed: %s\n", obd->obd_name);
642                         //continue;
643                 }
644
645                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
646                 err = class_process_config(&lcfg);
647                 if (err) {
648                         CERROR("detach failed: %s\n", obd->obd_name);
649                         //continue;
650                 }
651         }
652
653         if (sbi->ll_lmd != NULL)
654                 class_del_profile(sbi->ll_lmd->lmd_profile);
655 }
656
657 int lustre_fill_super(struct super_block *sb, void *data, int silent)
658 {
659         struct lustre_mount_data * lmd = data;
660         char *lov = NULL, *lmv = NULL;
661         struct ll_sb_info *sbi;
662         int err;
663         ENTRY;
664
665         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
666         if (lmd_bad_magic(lmd))
667                 RETURN(-EINVAL);
668
669         sbi = lustre_init_sbi(sb);
670         if (!sbi)
671                 RETURN(-ENOMEM);
672
673         sbi->ll_flags |= LL_SBI_READAHEAD;
674
675         if (lmd->lmd_profile) {
676                 struct lustre_profile *lprof;
677                 struct config_llog_instance cfg;
678                 int len;
679
680                 if (lmd->lmd_mds[0] == '\0') {
681                         CERROR("no mds name\n");
682                         GOTO(out_free, err = -EINVAL);
683                 }
684                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
685
686                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
687                 if (sbi->ll_lmd == NULL)
688                         GOTO(out_free, err = -ENOMEM);
689                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
690
691                 /* generate a string unique to this super, let's try
692                  the address of the super itself.*/
693                 len = (sizeof(sb) * 2) + 1;
694                 OBD_ALLOC(sbi->ll_instance, len);
695                 if (sbi->ll_instance == NULL)
696                         GOTO(out_free, err = -ENOMEM);
697                 sprintf(sbi->ll_instance, "%p", sb);
698
699                 cfg.cfg_instance = sbi->ll_instance;
700                 cfg.cfg_uuid = sbi->ll_sb_uuid;
701                 cfg.cfg_local_nid = lmd->lmd_local_nid;
702                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
703                 if (err < 0) {
704                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
705                         GOTO(out_free, err);
706                 }
707
708                 lprof = class_get_profile(lmd->lmd_profile);
709                 if (lprof == NULL) {
710                         CERROR("No profile found: %s\n", lmd->lmd_profile);
711                         GOTO(out_free, err = -EINVAL);
712                 }
713                 if (lov)
714                         OBD_FREE(lov, strlen(lov) + 1);
715                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
716                           strlen(sbi->ll_instance) + 2);
717                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
718
719                 if (lmv)
720                         OBD_FREE(lmv, strlen(lmv) + 1);
721                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
722                           strlen(sbi->ll_instance) + 2);
723                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
724         }
725
726         if (!lov) {
727                 CERROR("no osc\n");
728                 GOTO(out_free, err = -EINVAL);
729         }
730
731         if (!lmv) {
732                 CERROR("no mdc\n");
733                 GOTO(out_free, err = -EINVAL);
734         }
735
736         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_security,
737                                        &lmd->lmd_nllu, lmd->lmd_async);
738
739         if (err)
740                 GOTO(out_free, err);
741         
742 out_dev:
743         if (lmv)
744                 OBD_FREE(lmv, strlen(lmv) + 1);
745         if (lov)
746                 OBD_FREE(lov, strlen(lov) + 1);
747
748         RETURN(err);
749
750 out_free:
751         if (sbi->ll_lmd) {
752                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
753                 int err;
754
755                 if (sbi->ll_instance != NULL) {
756                         struct lustre_mount_data *lmd = sbi->ll_lmd;
757                         struct config_llog_instance cfg;
758                         char *cl_prof;
759
760                         cfg.cfg_instance = sbi->ll_instance;
761                         cfg.cfg_uuid = sbi->ll_sb_uuid;
762
763                         OBD_ALLOC(cl_prof, len);
764                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
765                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
766                         if (err < 0) {
767                                 CERROR("Unable to process log: %s\n", cl_prof);
768                                 lustre_manual_cleanup(sbi);
769                         }
770                         OBD_FREE(cl_prof, len);
771                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
772                 }
773                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
774         }
775         lustre_free_sbi(sb);
776         goto out_dev;
777 } /* lustre_fill_super */
778
779 void lustre_put_super(struct super_block *sb)
780 {
781         struct obd_device *obd;
782         struct ll_sb_info *sbi = ll_s2sbi(sb);
783         int force_umount = 0;
784         ENTRY;
785
786         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
787         obd = class_exp2obd(sbi->ll_md_exp);
788         if (obd)
789                 force_umount = obd->obd_no_recov;
790         obd = NULL;
791
792         lustre_common_put_super(sb);
793         if (sbi->ll_lmd != NULL) {
794                 char *cl_prof;
795                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
796                 int err;
797                 struct config_llog_instance cfg;
798
799                 if (force_umount) {
800                         CERROR("force umount, doing manual cleanup\n");
801                         lustre_manual_cleanup(sbi);
802                         GOTO(free_lmd, 0);
803                 }
804
805                 cfg.cfg_instance = sbi->ll_instance;
806                 cfg.cfg_uuid = sbi->ll_sb_uuid;
807
808                 OBD_ALLOC(cl_prof, len);
809                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
810                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
811                 if (err < 0) {
812                         CERROR("Unable to process log: %s, doing manual cleanup"
813                                "\n", cl_prof);
814                         lustre_manual_cleanup(sbi);
815                 }
816
817                 OBD_FREE(cl_prof, len);
818         free_lmd:
819                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
820                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
821         }
822
823         lustre_free_sbi(sb);
824
825         EXIT;
826 } /* lustre_put_super */
827
828 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
829 {
830         struct obd_export *md_exp = sbi->ll_md_exp;
831         struct lustre_mount_data *lmd = sbi->ll_lmd;
832         struct llog_ctxt *ctxt;
833         struct config_llog_instance cfg;
834         char *profile = lmd->lmd_profile, *name = NULL;
835         int rc, namelen =  0, version;
836         ENTRY;
837
838         if (profile == NULL)
839                 RETURN(0);
840         if (lmd == NULL) {
841                 CERROR("Client not mounted with zero-conf; cannot process "
842                        "update log.\n");
843                 RETURN(0);
844         }
845
846         rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
847                                     LDLM_FL_CONFIG_CHANGE, NULL);
848         if (rc != 0)
849                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
850
851         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
852                                NULL);
853         if (rc != 0)
854                 CWARN("obd_cancel_unused(lov): %d\n", rc);
855
856         cfg.cfg_instance = sbi->ll_instance;
857         cfg.cfg_uuid = sbi->ll_sb_uuid;
858         cfg.cfg_local_nid = lmd->lmd_local_nid;
859
860         namelen = strlen(profile) + 20; /* -clean-######### */
861         OBD_ALLOC(name, namelen);
862         if (name == NULL)
863                 RETURN(-ENOMEM);
864
865         if (clean) {
866                 version = sbi->ll_config_version - 1;
867                 sprintf(name, "%s-clean-%d", profile, version);
868         } else {
869                 version = sbi->ll_config_version + 1;
870                 sprintf(name, "%s-%d", profile, version);
871         }
872
873         CWARN("Applying configuration log %s\n", name);
874
875         ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
876                                 LLOG_CONFIG_REPL_CTXT);
877         rc = class_config_process_llog(ctxt, name, &cfg);
878         if (rc == 0)
879                 sbi->ll_config_version = version;
880         CWARN("Finished applying configuration log %s: %d\n", name, rc);
881
882         if (rc == 0 && clean == 0) {
883                 struct lov_desc desc;
884                 int rc, valsize;
885                 valsize = sizeof(desc);
886                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
887                                   "lovdesc", &valsize, &desc);
888
889                 rc = obd_init_ea_size(md_exp,
890                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
891                                       (desc.ld_tgt_count *
892                                        sizeof(struct llog_cookie)));
893         }
894         OBD_FREE(name, namelen);
895         RETURN(rc);
896 }
897
898 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
899 {
900         struct inode *inode = NULL;
901         l_lock(&lock->l_resource->lr_namespace->ns_lock);
902         if (lock->l_ast_data) {
903                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
904                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
905                         inode = igrab(lock->l_ast_data);
906                 } else {
907                         inode = lock->l_ast_data;
908                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
909                                "l_ast_data %p is bogus: magic %0x8\n",
910                                lock->l_ast_data, lli->lli_inode_magic);
911                         inode = NULL;
912                 }
913         }
914         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
915         return inode;
916 }
917
918 int null_if_equal(struct ldlm_lock *lock, void *data)
919 {
920         if (data == lock->l_ast_data) {
921                 lock->l_ast_data = NULL;
922
923                 if (lock->l_req_mode != lock->l_granted_mode)
924                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
925         }
926
927         return LDLM_ITER_CONTINUE;
928 }
929
930 void ll_clear_inode(struct inode *inode)
931 {
932         struct lustre_id id;
933         struct ll_inode_info *lli = ll_i2info(inode);
934         struct ll_sb_info *sbi = ll_i2sbi(inode);
935         ENTRY;
936
937         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
938                inode->i_generation, inode);
939
940         lli->lli_inode_magic = LLI_INODE_DEAD;
941         ll_inode2id(&id, inode);
942         
943         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
944         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
945
946         LASSERT(!lli->lli_open_fd_write_count);
947         LASSERT(!lli->lli_open_fd_read_count);
948         LASSERT(!lli->lli_open_fd_exec_count);
949
950         if (lli->lli_mds_write_och)
951                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
952         if (lli->lli_mds_exec_och)
953                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
954         if (lli->lli_mds_read_och)
955                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
956
957         if (lli->lli_smd)
958                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
959                                   null_if_equal, inode);
960
961         if (lli->lli_smd) {
962                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
963                 lli->lli_smd = NULL;
964         }
965
966         if (lli->lli_mea) {
967                 obd_free_memmd(sbi->ll_md_exp,
968                                (struct lov_stripe_md **) &lli->lli_mea);
969                 lli->lli_mea = NULL;
970         }
971
972         if (lli->lli_symlink_name) {
973                 OBD_FREE(lli->lli_symlink_name,
974                          strlen(lli->lli_symlink_name) + 1);
975                 lli->lli_symlink_name = NULL;
976         }
977         lli->lli_inode_magic = LLI_INODE_DEAD;
978
979         EXIT;
980 }
981
982 /* If this inode has objects allocated to it (lsm != NULL), then the OST
983  * object(s) determine the file size and mtime.  Otherwise, the MDS will
984  * keep these values until such a time that objects are allocated for it.
985  * We do the MDS operations first, as it is checking permissions for us.
986  * We don't to the MDS RPC if there is nothing that we want to store there,
987  * otherwise there is no harm in updating mtime/atime on the MDS if we are
988  * going to do an RPC anyways.
989  *
990  * If we are doing a truncate, we will send the mtime and ctime updates
991  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
992  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
993  * at the same time.
994  */
995 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
996 {
997         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
998         struct ll_sb_info *sbi = ll_i2sbi(inode);
999         struct ptlrpc_request *request = NULL;
1000         struct mdc_op_data *op_data;
1001         int ia_valid = attr->ia_valid;
1002         int rc = 0;
1003         ENTRY;
1004
1005         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1006         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1007
1008         if (ia_valid & ATTR_SIZE) {
1009                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1010                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1011                                attr->ia_size, ll_file_maxbytes(inode));
1012                         RETURN(-EFBIG);
1013                 }
1014
1015                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1016         }
1017
1018         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1019         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1020                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1021                         RETURN(-EPERM);
1022         }
1023
1024         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1025         if (attr->ia_valid & ATTR_CTIME) {
1026                 attr->ia_ctime = CURRENT_TIME;
1027                 attr->ia_valid |= ATTR_CTIME_SET;
1028         }
1029         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1030                 attr->ia_atime = CURRENT_TIME;
1031                 attr->ia_valid |= ATTR_ATIME_SET;
1032         }
1033         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1034                 attr->ia_mtime = CURRENT_TIME;
1035                 attr->ia_valid |= ATTR_MTIME_SET;
1036         }
1037
1038         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1039                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1040                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1041                        LTIME_S(CURRENT_TIME));
1042         if (lsm)
1043                 attr->ia_valid &= ~ATTR_SIZE;
1044
1045         /* If only OST attributes being set on objects, don't do MDS RPC.
1046          * In that case, we need to check permissions and update the local
1047          * inode ourselves so we can call obdo_from_inode() always. */
1048         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1049                 struct lustre_md md;
1050
1051                 OBD_ALLOC(op_data, sizeof(*op_data));
1052                 if (op_data == NULL)
1053                         RETURN(-ENOMEM);
1054                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1055
1056                 rc = md_setattr(sbi->ll_md_exp, op_data,
1057                                 attr, NULL, 0, NULL, 0, &request);
1058                 OBD_FREE(op_data, sizeof(*op_data));
1059                 if (rc) {
1060                         ptlrpc_req_finished(request);
1061                         if (rc != -EPERM && rc != -EACCES)
1062                                 CERROR("md_setattr fails: rc = %d\n", rc);
1063                         RETURN(rc);
1064                 }
1065
1066                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0, 
1067                                        sbi->ll_dt_exp, &md);
1068                 if (rc) {
1069                         ptlrpc_req_finished(request);
1070                         RETURN(rc);
1071                 }
1072
1073                 /* We call inode_setattr to adjust timestamps, but we first
1074                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1075                  *
1076                  * NB: ATTR_SIZE will only be set at this point if the size
1077                  * resides on the MDS, ie, this file has no objects. */
1078                 attr->ia_valid &= ~ATTR_SIZE;
1079                 inode_setattr(inode, attr);
1080                  
1081                 ll_update_inode(inode, &md);
1082                 ptlrpc_req_finished(request);
1083
1084                 if (!lsm || !S_ISREG(inode->i_mode)) {
1085                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1086                         RETURN(0);
1087                 }
1088         } else {
1089                 /* The OST doesn't check permissions, but the alternative is
1090                  * a gratuitous RPC to the MDS.  We already rely on the client
1091                  * to do read/write/truncate permission checks, so is mtime OK?
1092                  */
1093                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1094                         /* from sys_utime() */
1095                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1096                                 if (current->fsuid != inode->i_uid &&
1097                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1098                                         RETURN(rc);
1099                         } else {
1100                                 /* from inode_change_ok() */
1101                                 if (current->fsuid != inode->i_uid &&
1102                                     !capable(CAP_FOWNER))
1103                                         RETURN(-EPERM);
1104                         }
1105                 }
1106
1107                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1108                 inode_setattr(inode, attr);
1109         }
1110
1111         /* We really need to get our PW lock before we change inode->i_size.
1112          * If we don't we can race with other i_size updaters on our node, like
1113          * ll_file_read.  We can also race with i_size propogation to other
1114          * nodes through dirtying and writeback of final cached pages.  This
1115          * last one is especially bad for racing o_append users on other
1116          * nodes. */
1117         if (ia_valid & ATTR_SIZE) {
1118                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1119                                                            OBD_OBJECT_EOF } };
1120                 struct lustre_handle lockh = { 0 };
1121                 struct ll_inode_info *lli = ll_i2info(inode);
1122                 int err, ast_flags = 0;
1123                 /* XXX when we fix the AST intents to pass the discard-range
1124                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1125                  * XXX here. */
1126                 if (attr->ia_size == 0)
1127                         ast_flags = LDLM_AST_DISCARD_DATA;
1128
1129                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1130                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1131
1132                 if (rc != 0)
1133                         RETURN(rc);
1134
1135                 down(&lli->lli_size_sem);
1136                 lli->lli_size_pid = current->pid;
1137                 rc = vmtruncate(inode, attr->ia_size);
1138                 if (rc != 0) {
1139                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1140                         lli->lli_size_pid = 0;
1141                         up(&lli->lli_size_sem);
1142                 }
1143
1144                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1145                 if (err) {
1146                         CERROR("ll_extent_unlock failed: %d\n", err);
1147                         if (!rc)
1148                                 rc = err;
1149                 }
1150         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1151                 struct obdo *oa = NULL;
1152
1153                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1154                        inode->i_ino, LTIME_S(attr->ia_mtime));
1155
1156                 oa = obdo_alloc();
1157                 if (oa == NULL)
1158                         RETURN(-ENOMEM);
1159
1160                 oa->o_id = lsm->lsm_object_id;
1161                 oa->o_gr = lsm->lsm_object_gr;
1162                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1163                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1164                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1165                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1166                 obdo_free(oa);
1167                 if (rc)
1168                         CERROR("obd_setattr fails: rc = %d\n", rc);
1169         }
1170         RETURN(rc);
1171 }
1172
1173 int ll_setattr(struct dentry *de, struct iattr *attr)
1174 {
1175         LASSERT(de->d_inode);
1176         return ll_setattr_raw(de->d_inode, attr);
1177 }
1178
1179 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1180                        unsigned long max_age)
1181 {
1182         struct ll_sb_info *sbi = ll_s2sbi(sb);
1183         struct obd_statfs obd_osfs;
1184         int rc;
1185         ENTRY;
1186
1187         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1188         if (rc) {
1189                 CERROR("obd_statfs fails: rc = %d\n", rc);
1190                 RETURN(rc);
1191         }
1192
1193         osfs->os_type = sb->s_magic;
1194
1195         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1196                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1197
1198         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1199         if (rc) {
1200                 CERROR("obd_statfs fails: rc = %d\n", rc);
1201                 RETURN(rc);
1202         }
1203
1204         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1205                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1206                obd_osfs.os_files);
1207
1208         osfs->os_blocks = obd_osfs.os_blocks;
1209         osfs->os_bfree = obd_osfs.os_bfree;
1210         osfs->os_bavail = obd_osfs.os_bavail;
1211
1212         /* If we don't have as many objects free on the OST as inodes
1213          * on the MDS, we reduce the total number of inodes to
1214          * compensate, so that the "inodes in use" number is correct.
1215          */
1216         if (obd_osfs.os_ffree < osfs->os_ffree) {
1217                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1218                         obd_osfs.os_ffree;
1219                 osfs->os_ffree = obd_osfs.os_ffree;
1220         }
1221
1222         RETURN(rc);
1223 }
1224
1225 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1226 {
1227         struct obd_statfs osfs;
1228         int rc;
1229
1230         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1231         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1232
1233         /* For now we will always get up-to-date statfs values, but in the
1234          * future we may allow some amount of caching on the client (e.g.
1235          * from QOS or lprocfs updates). */
1236         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1237         if (rc)
1238                 return rc;
1239
1240         statfs_unpack(sfs, &osfs);
1241
1242         if (sizeof(sfs->f_blocks) == 4) {
1243                 while (osfs.os_blocks > ~0UL) {
1244                         sfs->f_bsize <<= 1;
1245
1246                         osfs.os_blocks >>= 1;
1247                         osfs.os_bfree >>= 1;
1248                         osfs.os_bavail >>= 1;
1249                 }
1250         }
1251
1252         sfs->f_blocks = osfs.os_blocks;
1253         sfs->f_bfree = osfs.os_bfree;
1254         sfs->f_bavail = osfs.os_bavail;
1255
1256         return 0;
1257 }
1258
1259 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1260 {
1261         struct ll_inode_info *lli = ll_i2info(inode);
1262         struct lov_stripe_md *lsm = md->lsm;
1263         struct mds_body *body = md->body;
1264         struct mea *mea = md->mea;
1265         struct posix_acl *ll_acl_access = md->acl_access;
1266         ENTRY;
1267
1268         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1269
1270         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1271                 /* check for default striping info for dir. */
1272                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1273         }
1274         
1275         if (lsm != NULL) {
1276                 LASSERT(lsm->lsm_object_gr > 0);
1277                 if (lli->lli_smd == NULL) {
1278                         lli->lli_smd = lsm;
1279                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1280                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1281                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1282                 } else {
1283                         int i;
1284                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1285                                 CERROR("lsm mismatch for inode %ld\n",
1286                                        inode->i_ino);
1287                                 CERROR("lli_smd:\n");
1288                                 dump_lsm(D_ERROR, lli->lli_smd);
1289                                 CERROR("lsm:\n");
1290                                 dump_lsm(D_ERROR, lsm);
1291                                 LBUG();
1292                         }
1293                         /* XXX FIXME -- We should decide on a safer (atomic) and
1294                          * more elegant way to update the lsm */
1295                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1296                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1297                                         lsm->lsm_oinfo[i].loi_id;
1298                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1299                                         lsm->lsm_oinfo[i].loi_gr;
1300                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1301                                         lsm->lsm_oinfo[i].loi_ost_idx;
1302                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1303                                         lsm->lsm_oinfo[i].loi_ost_gen;
1304                         }
1305                 }
1306                 /* bug 2844 - limit i_blksize for broken user-space apps */
1307                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1308                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1309                 if (lli->lli_smd != lsm)
1310                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1311         }
1312
1313         if (mea != NULL) {
1314                 if (lli->lli_mea == NULL) {
1315                         lli->lli_mea = mea;
1316                 } else {
1317                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1318                                 CERROR("mea mismatch for inode %lu\n",
1319                                         inode->i_ino);
1320                                 LBUG();
1321                         }
1322                 }
1323                 if (lli->lli_mea != mea)
1324                         obd_free_memmd(ll_i2mdexp(inode),
1325                                        (struct lov_stripe_md **) &mea);
1326         }
1327
1328         if (body->valid & OBD_MD_FID)
1329                 id_assign_fid(&lli->lli_id, &body->id1);
1330         
1331         if (body->valid & OBD_MD_FLID)
1332                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1333
1334         if (body->valid & OBD_MD_FLGENER)
1335                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1336
1337         spin_lock(&lli->lli_lock);
1338         if (ll_acl_access != NULL) {
1339                 if (lli->lli_acl_access != NULL)
1340                         posix_acl_release(lli->lli_acl_access);
1341                 lli->lli_acl_access = ll_acl_access;
1342         }
1343         spin_unlock(&lli->lli_lock);
1344  
1345         if (body->valid & OBD_MD_FLID)
1346                 inode->i_ino = id_ino(&body->id1);
1347         if (body->valid & OBD_MD_FLGENER)
1348                 inode->i_generation = id_gen(&body->id1);
1349         if (body->valid & OBD_MD_FLATIME)
1350                 LTIME_S(inode->i_atime) = body->atime;
1351         if (body->valid & OBD_MD_FLMTIME &&
1352             body->mtime > LTIME_S(inode->i_mtime)) {
1353                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1354                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1355                 LTIME_S(inode->i_mtime) = body->mtime;
1356         }
1357         if (body->valid & OBD_MD_FLCTIME &&
1358             body->ctime > LTIME_S(inode->i_ctime))
1359                 LTIME_S(inode->i_ctime) = body->ctime;
1360         if (body->valid & OBD_MD_FLMODE) {
1361                 inode->i_mode = (inode->i_mode & S_IFMT) |
1362                         (body->mode & ~S_IFMT);
1363         }
1364         if (body->valid & OBD_MD_FLTYPE) {
1365                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1366                         (body->mode & S_IFMT);
1367         }
1368         if (body->valid & OBD_MD_FLUID)
1369                 inode->i_uid = body->uid;
1370         if (body->valid & OBD_MD_FLGID)
1371                 inode->i_gid = body->gid;
1372         if (body->valid & OBD_MD_FLFLAGS)
1373                 inode->i_flags = body->flags;
1374         if (body->valid & OBD_MD_FLNLINK)
1375                 inode->i_nlink = body->nlink;
1376         if (body->valid & OBD_MD_FLRDEV)
1377 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1378                 inode->i_rdev = body->rdev;
1379 #else
1380                 inode->i_rdev = old_decode_dev(body->rdev);
1381 #endif
1382         if (body->valid & OBD_MD_FLSIZE)
1383                 inode->i_size = body->size;
1384         if (body->valid & OBD_MD_FLBLOCKS)
1385                 inode->i_blocks = body->blocks;
1386
1387         if (body->valid & OBD_MD_FLSIZE)
1388                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1389
1390 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1391         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1392 #endif
1393         LASSERT(id_fid(&lli->lli_id) != 0);
1394 }
1395
1396 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1397 static struct backing_dev_info ll_backing_dev_info = {
1398         .ra_pages       = 0,    /* No readahead */
1399         .memory_backed  = 0,    /* Does contribute to dirty memory */
1400 };
1401 #endif
1402
1403 void ll_read_inode2(struct inode *inode, void *opaque)
1404 {
1405         struct lustre_md *md = opaque;
1406         struct ll_inode_info *lli = ll_i2info(inode);
1407         ENTRY;
1408
1409         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1410                inode->i_generation, inode);
1411
1412         ll_lli_init(lli);
1413
1414         LASSERT(!lli->lli_smd);
1415
1416         /* Core attributes from the MDS first.  This is a new inode, and
1417          * the VFS doesn't zero times in the core inode so we have to do
1418          * it ourselves.  They will be overwritten by either MDS or OST
1419          * attributes - we just need to make sure they aren't newer. */
1420         LTIME_S(inode->i_mtime) = 0;
1421         LTIME_S(inode->i_atime) = 0;
1422         LTIME_S(inode->i_ctime) = 0;
1423
1424         inode->i_rdev = 0;
1425         ll_update_inode(inode, md);
1426
1427         /* OIDEBUG(inode); */
1428
1429         if (S_ISREG(inode->i_mode)) {
1430                 inode->i_op = &ll_file_inode_operations;
1431                 inode->i_fop = &ll_file_operations;
1432                 inode->i_mapping->a_ops = &ll_aops;
1433                 EXIT;
1434         } else if (S_ISDIR(inode->i_mode)) {
1435                 inode->i_op = &ll_dir_inode_operations;
1436                 inode->i_fop = &ll_dir_operations;
1437                 inode->i_mapping->a_ops = &ll_dir_aops;
1438                 EXIT;
1439         } else if (S_ISLNK(inode->i_mode)) {
1440                 inode->i_op = &ll_fast_symlink_inode_operations;
1441                 EXIT;
1442         } else {
1443                 inode->i_op = &ll_special_inode_operations;
1444
1445 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1446                 init_special_inode(inode, inode->i_mode,
1447                                    kdev_t_to_nr(inode->i_rdev));
1448
1449                 /* initializing backing dev info. */
1450                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1451 #else
1452                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1453 #endif
1454                 lli->ll_save_ifop = inode->i_fop;
1455
1456                 if (S_ISCHR(inode->i_mode))
1457                         inode->i_fop = &ll_special_chr_inode_fops;
1458                 else if (S_ISBLK(inode->i_mode))
1459                         inode->i_fop = &ll_special_blk_inode_fops;
1460                 else if (S_ISFIFO(inode->i_mode))
1461                         inode->i_fop = &ll_special_fifo_inode_fops;
1462                 else if (S_ISSOCK(inode->i_mode))
1463                         inode->i_fop = &ll_special_sock_inode_fops;
1464
1465                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1466                       inode->i_fop);
1467
1468                 if (lli->ll_save_ifop->owner) {
1469                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1470                               lli->ll_save_ifop->owner);
1471                 }
1472                 EXIT;
1473         }
1474 }
1475
1476 void ll_delete_inode(struct inode *inode)
1477 {
1478         struct ll_sb_info *sbi = ll_i2sbi(inode);
1479         struct lustre_id id;
1480         int rc;
1481         ENTRY;
1482
1483         ll_inode2id(&id, inode);
1484
1485         rc = md_delete_inode(sbi->ll_md_exp, &id);
1486         if (rc) {
1487                 CERROR("md_delete_inode() failed, error %d\n", 
1488                        rc);
1489         }
1490
1491         clear_inode(inode);
1492         EXIT;
1493 }
1494
1495 int ll_iocontrol(struct inode *inode, struct file *file,
1496                  unsigned int cmd, unsigned long arg)
1497 {
1498         struct ll_sb_info *sbi = ll_i2sbi(inode);
1499         struct ptlrpc_request *req = NULL;
1500         int rc, flags = 0;
1501         ENTRY;
1502
1503         switch(cmd) {
1504         case EXT3_IOC_GETFLAGS: {
1505                 struct lustre_id id;
1506                 __u64 valid = OBD_MD_FLFLAGS;
1507                 struct mds_body *body;
1508
1509                 ll_inode2id(&id, inode);
1510                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1511                 if (rc) {
1512                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1513                         RETURN(-abs(rc));
1514                 }
1515
1516                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1517
1518                 if (body->flags & S_APPEND)
1519                         flags |= EXT3_APPEND_FL;
1520                 if (body->flags & S_IMMUTABLE)
1521                         flags |= EXT3_IMMUTABLE_FL;
1522                 if (body->flags & S_NOATIME)
1523                         flags |= EXT3_NOATIME_FL;
1524
1525                 ptlrpc_req_finished (req);
1526
1527                 RETURN(put_user(flags, (int *)arg));
1528         }
1529         case EXT3_IOC_SETFLAGS: {
1530                 struct mdc_op_data *op_data;
1531                 struct iattr attr;
1532                 struct obdo *oa;
1533                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1534
1535                 if (get_user(flags, (int *)arg))
1536                         RETURN(-EFAULT);
1537
1538                 oa = obdo_alloc();
1539                 if (!oa)
1540                         RETURN(-ENOMEM);
1541
1542                 OBD_ALLOC(op_data, sizeof(*op_data));
1543                 if (op_data == NULL) {
1544                         obdo_free(oa);
1545                         RETURN(-ENOMEM);
1546                 }
1547                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1548
1549                 memset(&attr, 0x0, sizeof(attr));
1550                 attr.ia_attr_flags = flags;
1551                 attr.ia_valid |= ATTR_ATTR_FLAG;
1552
1553                 rc = md_setattr(sbi->ll_md_exp, op_data,
1554                                 &attr, NULL, 0, NULL, 0, &req);
1555                 OBD_FREE(op_data, sizeof(*op_data));
1556                 if (rc) {
1557                         ptlrpc_req_finished(req);
1558                         if (rc != -EPERM && rc != -EACCES)
1559                                 CERROR("md_setattr fails: rc = %d\n", rc);
1560                         obdo_free(oa);
1561                         RETURN(rc);
1562                 }
1563                 ptlrpc_req_finished(req);
1564
1565                 oa->o_id = lsm->lsm_object_id;
1566                 oa->o_gr = lsm->lsm_object_gr;
1567                 oa->o_flags = flags;
1568                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1569
1570                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1571                 obdo_free(oa);
1572                 if (rc) {
1573                         if (rc != -EPERM && rc != -EACCES)
1574                                 CERROR("md_setattr fails: rc = %d\n", rc);
1575                         RETURN(rc);
1576                 }
1577
1578                 if (flags & EXT3_APPEND_FL)
1579                         inode->i_flags |= S_APPEND;
1580                 else
1581                         inode->i_flags &= ~S_APPEND;
1582                 if (flags & EXT3_IMMUTABLE_FL)
1583                         inode->i_flags |= S_IMMUTABLE;
1584                 else
1585                         inode->i_flags &= ~S_IMMUTABLE;
1586                 if (flags & EXT3_NOATIME_FL)
1587                         inode->i_flags |= S_NOATIME;
1588                 else
1589                         inode->i_flags &= ~S_NOATIME;
1590
1591                 RETURN(0);
1592         }
1593         default:
1594                 RETURN(-ENOSYS);
1595         }
1596
1597         RETURN(0);
1598 }
1599
1600 /* this is only called in the case of forced umount. */
1601 void ll_umount_begin(struct super_block *sb)
1602 {
1603         struct ll_sb_info *sbi = ll_s2sbi(sb);
1604         struct obd_ioctl_data ioc_data = { 0 };
1605         struct obd_device *obd;
1606         ENTRY;
1607      
1608         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1609                sb->s_count, atomic_read(&sb->s_active));
1610         
1611         obd = class_exp2obd(sbi->ll_md_exp);
1612         if (obd == NULL) {
1613                 CERROR("Invalid MDC connection handle "LPX64"\n",
1614                        sbi->ll_md_exp->exp_handle.h_cookie);
1615                 EXIT;
1616                 return;
1617         }
1618         obd->obd_no_recov = 1;
1619         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1620                       sizeof(ioc_data), &ioc_data, NULL);
1621
1622         obd = class_exp2obd(sbi->ll_dt_exp);
1623         if (obd == NULL) {
1624                 CERROR("Invalid LOV connection handle "LPX64"\n",
1625                        sbi->ll_dt_exp->exp_handle.h_cookie);
1626                 EXIT;
1627                 return;
1628         }
1629
1630         obd->obd_no_recov = 1;
1631         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1632                       sizeof(ioc_data), &ioc_data, NULL);
1633
1634         /*
1635          * really, we'd like to wait until there are no requests outstanding,
1636          * and then continue.  For now, we just invalidate the requests,
1637          * schedule, and hope.
1638          */
1639         schedule();
1640
1641         EXIT;
1642 }
1643
1644 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1645                   struct inode **inode, struct ptlrpc_request *req,
1646                   int offset, struct super_block *sb)
1647 {
1648         struct lustre_md md;
1649         int rc = 0;
1650
1651         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1652         if (rc)
1653                 RETURN(rc);
1654
1655         if (*inode) {
1656                 ll_update_inode(*inode, &md);
1657         } else {
1658                 LASSERT(sb);
1659                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1660                 if (*inode == NULL || is_bad_inode(*inode)) {
1661                         /* free the lsm if we allocated one above */
1662                         if (md.lsm != NULL)
1663                                 obd_free_memmd(dt_exp, &md.lsm);
1664                         if (md.mea != NULL)
1665                                 obd_free_memmd(md_exp,
1666                                                (struct lov_stripe_md**)&md.mea);
1667                         rc = -ENOMEM;
1668                         CERROR("new_inode -fatal: rc %d\n", rc);
1669                 }
1670         }
1671
1672         RETURN(rc);
1673 }
1674
1675 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1676                char *filename, struct lustre_id *ret)
1677 {
1678         struct ptlrpc_request *request = NULL;
1679         struct mds_body *body;
1680         int rc;
1681
1682         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1683                              OBD_MD_FID, 0, &request);
1684         if (rc < 0) {
1685                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1686                        filename, rc);
1687                 return rc;
1688         }
1689
1690         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1691         LASSERT(body != NULL);
1692         LASSERT_REPSWABBED(request, 0);
1693
1694         *ret = body->id1;
1695         ptlrpc_req_finished(request);
1696
1697         return rc;
1698 }