685a1c287177f6f16f8e3ebc8f1ca40b630c709b
[safe/jmp/linux-2.6] / fs / ext2 / super.c
1 /*
2  *  linux/fs/ext2/super.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/inode.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  Big-endian to little-endian byte-swapping/bitmaps by
16  *        David S. Miller (davem@caip.rutgers.edu), 1995
17  */
18
19 #include <linux/module.h>
20 #include <linux/string.h>
21 #include <linux/fs.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/blkdev.h>
25 #include <linux/parser.h>
26 #include <linux/random.h>
27 #include <linux/buffer_head.h>
28 #include <linux/smp_lock.h>
29 #include <linux/vfs.h>
30 #include <linux/seq_file.h>
31 #include <linux/mount.h>
32 #include <asm/uaccess.h>
33 #include "ext2.h"
34 #include "xattr.h"
35 #include "acl.h"
36 #include "xip.h"
37
38 static void ext2_sync_super(struct super_block *sb,
39                             struct ext2_super_block *es);
40 static int ext2_remount (struct super_block * sb, int * flags, char * data);
41 static int ext2_statfs (struct dentry * dentry, struct kstatfs * buf);
42
43 void ext2_error (struct super_block * sb, const char * function,
44                  const char * fmt, ...)
45 {
46         va_list args;
47         struct ext2_sb_info *sbi = EXT2_SB(sb);
48         struct ext2_super_block *es = sbi->s_es;
49
50         if (!(sb->s_flags & MS_RDONLY)) {
51                 sbi->s_mount_state |= EXT2_ERROR_FS;
52                 es->s_state =
53                         cpu_to_le16(le16_to_cpu(es->s_state) | EXT2_ERROR_FS);
54                 ext2_sync_super(sb, es);
55         }
56
57         va_start(args, fmt);
58         printk(KERN_CRIT "EXT2-fs error (device %s): %s: ",sb->s_id, function);
59         vprintk(fmt, args);
60         printk("\n");
61         va_end(args);
62
63         if (test_opt(sb, ERRORS_PANIC))
64                 panic("EXT2-fs panic from previous error\n");
65         if (test_opt(sb, ERRORS_RO)) {
66                 printk("Remounting filesystem read-only\n");
67                 sb->s_flags |= MS_RDONLY;
68         }
69 }
70
71 void ext2_warning (struct super_block * sb, const char * function,
72                    const char * fmt, ...)
73 {
74         va_list args;
75
76         va_start(args, fmt);
77         printk(KERN_WARNING "EXT2-fs warning (device %s): %s: ",
78                sb->s_id, function);
79         vprintk(fmt, args);
80         printk("\n");
81         va_end(args);
82 }
83
84 void ext2_update_dynamic_rev(struct super_block *sb)
85 {
86         struct ext2_super_block *es = EXT2_SB(sb)->s_es;
87
88         if (le32_to_cpu(es->s_rev_level) > EXT2_GOOD_OLD_REV)
89                 return;
90
91         ext2_warning(sb, __FUNCTION__,
92                      "updating to rev %d because of new feature flag, "
93                      "running e2fsck is recommended",
94                      EXT2_DYNAMIC_REV);
95
96         es->s_first_ino = cpu_to_le32(EXT2_GOOD_OLD_FIRST_INO);
97         es->s_inode_size = cpu_to_le16(EXT2_GOOD_OLD_INODE_SIZE);
98         es->s_rev_level = cpu_to_le32(EXT2_DYNAMIC_REV);
99         /* leave es->s_feature_*compat flags alone */
100         /* es->s_uuid will be set by e2fsck if empty */
101
102         /*
103          * The rest of the superblock fields should be zero, and if not it
104          * means they are likely already in use, so leave them alone.  We
105          * can leave it up to e2fsck to clean up any inconsistencies there.
106          */
107 }
108
109 static void ext2_put_super (struct super_block * sb)
110 {
111         int db_count;
112         int i;
113         struct ext2_sb_info *sbi = EXT2_SB(sb);
114
115         ext2_xattr_put_super(sb);
116         if (!(sb->s_flags & MS_RDONLY)) {
117                 struct ext2_super_block *es = sbi->s_es;
118
119                 es->s_state = cpu_to_le16(sbi->s_mount_state);
120                 ext2_sync_super(sb, es);
121         }
122         db_count = sbi->s_gdb_count;
123         for (i = 0; i < db_count; i++)
124                 if (sbi->s_group_desc[i])
125                         brelse (sbi->s_group_desc[i]);
126         kfree(sbi->s_group_desc);
127         kfree(sbi->s_debts);
128         percpu_counter_destroy(&sbi->s_freeblocks_counter);
129         percpu_counter_destroy(&sbi->s_freeinodes_counter);
130         percpu_counter_destroy(&sbi->s_dirs_counter);
131         brelse (sbi->s_sbh);
132         sb->s_fs_info = NULL;
133         kfree(sbi);
134
135         return;
136 }
137
138 static struct kmem_cache * ext2_inode_cachep;
139
140 static struct inode *ext2_alloc_inode(struct super_block *sb)
141 {
142         struct ext2_inode_info *ei;
143         ei = (struct ext2_inode_info *)kmem_cache_alloc(ext2_inode_cachep, GFP_KERNEL);
144         if (!ei)
145                 return NULL;
146 #ifdef CONFIG_EXT2_FS_POSIX_ACL
147         ei->i_acl = EXT2_ACL_NOT_CACHED;
148         ei->i_default_acl = EXT2_ACL_NOT_CACHED;
149 #endif
150         ei->vfs_inode.i_version = 1;
151         return &ei->vfs_inode;
152 }
153
154 static void ext2_destroy_inode(struct inode *inode)
155 {
156         kmem_cache_free(ext2_inode_cachep, EXT2_I(inode));
157 }
158
159 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
160 {
161         struct ext2_inode_info *ei = (struct ext2_inode_info *) foo;
162
163         if (flags & SLAB_CTOR_CONSTRUCTOR) {
164                 rwlock_init(&ei->i_meta_lock);
165 #ifdef CONFIG_EXT2_FS_XATTR
166                 init_rwsem(&ei->xattr_sem);
167 #endif
168                 inode_init_once(&ei->vfs_inode);
169         }
170 }
171  
172 static int init_inodecache(void)
173 {
174         ext2_inode_cachep = kmem_cache_create("ext2_inode_cache",
175                                              sizeof(struct ext2_inode_info),
176                                              0, (SLAB_RECLAIM_ACCOUNT|
177                                                 SLAB_MEM_SPREAD),
178                                              init_once, NULL);
179         if (ext2_inode_cachep == NULL)
180                 return -ENOMEM;
181         return 0;
182 }
183
184 static void destroy_inodecache(void)
185 {
186         kmem_cache_destroy(ext2_inode_cachep);
187 }
188
189 static void ext2_clear_inode(struct inode *inode)
190 {
191 #ifdef CONFIG_EXT2_FS_POSIX_ACL
192         struct ext2_inode_info *ei = EXT2_I(inode);
193
194         if (ei->i_acl && ei->i_acl != EXT2_ACL_NOT_CACHED) {
195                 posix_acl_release(ei->i_acl);
196                 ei->i_acl = EXT2_ACL_NOT_CACHED;
197         }
198         if (ei->i_default_acl && ei->i_default_acl != EXT2_ACL_NOT_CACHED) {
199                 posix_acl_release(ei->i_default_acl);
200                 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
201         }
202 #endif
203 }
204
205 static int ext2_show_options(struct seq_file *seq, struct vfsmount *vfs)
206 {
207         struct ext2_sb_info *sbi = EXT2_SB(vfs->mnt_sb);
208
209         if (sbi->s_mount_opt & EXT2_MOUNT_GRPID)
210                 seq_puts(seq, ",grpid");
211
212 #if defined(CONFIG_QUOTA)
213         if (sbi->s_mount_opt & EXT2_MOUNT_USRQUOTA)
214                 seq_puts(seq, ",usrquota");
215
216         if (sbi->s_mount_opt & EXT2_MOUNT_GRPQUOTA)
217                 seq_puts(seq, ",grpquota");
218 #endif
219
220 #if defined(CONFIG_EXT2_FS_XIP)
221         if (sbi->s_mount_opt & EXT2_MOUNT_XIP)
222                 seq_puts(seq, ",xip");
223 #endif
224
225         return 0;
226 }
227
228 #ifdef CONFIG_QUOTA
229 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data, size_t len, loff_t off);
230 static ssize_t ext2_quota_write(struct super_block *sb, int type, const char *data, size_t len, loff_t off);
231 #endif
232
233 static const struct super_operations ext2_sops = {
234         .alloc_inode    = ext2_alloc_inode,
235         .destroy_inode  = ext2_destroy_inode,
236         .read_inode     = ext2_read_inode,
237         .write_inode    = ext2_write_inode,
238         .put_inode      = ext2_put_inode,
239         .delete_inode   = ext2_delete_inode,
240         .put_super      = ext2_put_super,
241         .write_super    = ext2_write_super,
242         .statfs         = ext2_statfs,
243         .remount_fs     = ext2_remount,
244         .clear_inode    = ext2_clear_inode,
245         .show_options   = ext2_show_options,
246 #ifdef CONFIG_QUOTA
247         .quota_read     = ext2_quota_read,
248         .quota_write    = ext2_quota_write,
249 #endif
250 };
251
252 static struct dentry *ext2_get_dentry(struct super_block *sb, void *vobjp)
253 {
254         __u32 *objp = vobjp;
255         unsigned long ino = objp[0];
256         __u32 generation = objp[1];
257         struct inode *inode;
258         struct dentry *result;
259
260         if (ino < EXT2_FIRST_INO(sb) && ino != EXT2_ROOT_INO)
261                 return ERR_PTR(-ESTALE);
262         if (ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
263                 return ERR_PTR(-ESTALE);
264
265         /* iget isn't really right if the inode is currently unallocated!!
266          * ext2_read_inode currently does appropriate checks, but
267          * it might be "neater" to call ext2_get_inode first and check
268          * if the inode is valid.....
269          */
270         inode = iget(sb, ino);
271         if (inode == NULL)
272                 return ERR_PTR(-ENOMEM);
273         if (is_bad_inode(inode) ||
274             (generation && inode->i_generation != generation)) {
275                 /* we didn't find the right inode.. */
276                 iput(inode);
277                 return ERR_PTR(-ESTALE);
278         }
279         /* now to find a dentry.
280          * If possible, get a well-connected one
281          */
282         result = d_alloc_anon(inode);
283         if (!result) {
284                 iput(inode);
285                 return ERR_PTR(-ENOMEM);
286         }
287         return result;
288 }
289
290 /* Yes, most of these are left as NULL!!
291  * A NULL value implies the default, which works with ext2-like file
292  * systems, but can be improved upon.
293  * Currently only get_parent is required.
294  */
295 static struct export_operations ext2_export_ops = {
296         .get_parent = ext2_get_parent,
297         .get_dentry = ext2_get_dentry,
298 };
299
300 static unsigned long get_sb_block(void **data)
301 {
302         unsigned long   sb_block;
303         char            *options = (char *) *data;
304
305         if (!options || strncmp(options, "sb=", 3) != 0)
306                 return 1;       /* Default location */
307         options += 3;
308         sb_block = simple_strtoul(options, &options, 0);
309         if (*options && *options != ',') {
310                 printk("EXT2-fs: Invalid sb specification: %s\n",
311                        (char *) *data);
312                 return 1;
313         }
314         if (*options == ',')
315                 options++;
316         *data = (void *) options;
317         return sb_block;
318 }
319
320 enum {
321         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
322         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic,
323         Opt_err_ro, Opt_nouid32, Opt_nocheck, Opt_debug,
324         Opt_oldalloc, Opt_orlov, Opt_nobh, Opt_user_xattr, Opt_nouser_xattr,
325         Opt_acl, Opt_noacl, Opt_xip, Opt_ignore, Opt_err, Opt_quota,
326         Opt_usrquota, Opt_grpquota
327 };
328
329 static match_table_t tokens = {
330         {Opt_bsd_df, "bsddf"},
331         {Opt_minix_df, "minixdf"},
332         {Opt_grpid, "grpid"},
333         {Opt_grpid, "bsdgroups"},
334         {Opt_nogrpid, "nogrpid"},
335         {Opt_nogrpid, "sysvgroups"},
336         {Opt_resgid, "resgid=%u"},
337         {Opt_resuid, "resuid=%u"},
338         {Opt_sb, "sb=%u"},
339         {Opt_err_cont, "errors=continue"},
340         {Opt_err_panic, "errors=panic"},
341         {Opt_err_ro, "errors=remount-ro"},
342         {Opt_nouid32, "nouid32"},
343         {Opt_nocheck, "check=none"},
344         {Opt_nocheck, "nocheck"},
345         {Opt_debug, "debug"},
346         {Opt_oldalloc, "oldalloc"},
347         {Opt_orlov, "orlov"},
348         {Opt_nobh, "nobh"},
349         {Opt_user_xattr, "user_xattr"},
350         {Opt_nouser_xattr, "nouser_xattr"},
351         {Opt_acl, "acl"},
352         {Opt_noacl, "noacl"},
353         {Opt_xip, "xip"},
354         {Opt_grpquota, "grpquota"},
355         {Opt_ignore, "noquota"},
356         {Opt_quota, "quota"},
357         {Opt_usrquota, "usrquota"},
358         {Opt_err, NULL}
359 };
360
361 static int parse_options (char * options,
362                           struct ext2_sb_info *sbi)
363 {
364         char * p;
365         substring_t args[MAX_OPT_ARGS];
366         int option;
367
368         if (!options)
369                 return 1;
370
371         while ((p = strsep (&options, ",")) != NULL) {
372                 int token;
373                 if (!*p)
374                         continue;
375
376                 token = match_token(p, tokens, args);
377                 switch (token) {
378                 case Opt_bsd_df:
379                         clear_opt (sbi->s_mount_opt, MINIX_DF);
380                         break;
381                 case Opt_minix_df:
382                         set_opt (sbi->s_mount_opt, MINIX_DF);
383                         break;
384                 case Opt_grpid:
385                         set_opt (sbi->s_mount_opt, GRPID);
386                         break;
387                 case Opt_nogrpid:
388                         clear_opt (sbi->s_mount_opt, GRPID);
389                         break;
390                 case Opt_resuid:
391                         if (match_int(&args[0], &option))
392                                 return 0;
393                         sbi->s_resuid = option;
394                         break;
395                 case Opt_resgid:
396                         if (match_int(&args[0], &option))
397                                 return 0;
398                         sbi->s_resgid = option;
399                         break;
400                 case Opt_sb:
401                         /* handled by get_sb_block() instead of here */
402                         /* *sb_block = match_int(&args[0]); */
403                         break;
404                 case Opt_err_panic:
405                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
406                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
407                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
408                         break;
409                 case Opt_err_ro:
410                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
411                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
412                         set_opt (sbi->s_mount_opt, ERRORS_RO);
413                         break;
414                 case Opt_err_cont:
415                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
416                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
417                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
418                         break;
419                 case Opt_nouid32:
420                         set_opt (sbi->s_mount_opt, NO_UID32);
421                         break;
422                 case Opt_nocheck:
423                         clear_opt (sbi->s_mount_opt, CHECK);
424                         break;
425                 case Opt_debug:
426                         set_opt (sbi->s_mount_opt, DEBUG);
427                         break;
428                 case Opt_oldalloc:
429                         set_opt (sbi->s_mount_opt, OLDALLOC);
430                         break;
431                 case Opt_orlov:
432                         clear_opt (sbi->s_mount_opt, OLDALLOC);
433                         break;
434                 case Opt_nobh:
435                         set_opt (sbi->s_mount_opt, NOBH);
436                         break;
437 #ifdef CONFIG_EXT2_FS_XATTR
438                 case Opt_user_xattr:
439                         set_opt (sbi->s_mount_opt, XATTR_USER);
440                         break;
441                 case Opt_nouser_xattr:
442                         clear_opt (sbi->s_mount_opt, XATTR_USER);
443                         break;
444 #else
445                 case Opt_user_xattr:
446                 case Opt_nouser_xattr:
447                         printk("EXT2 (no)user_xattr options not supported\n");
448                         break;
449 #endif
450 #ifdef CONFIG_EXT2_FS_POSIX_ACL
451                 case Opt_acl:
452                         set_opt(sbi->s_mount_opt, POSIX_ACL);
453                         break;
454                 case Opt_noacl:
455                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
456                         break;
457 #else
458                 case Opt_acl:
459                 case Opt_noacl:
460                         printk("EXT2 (no)acl options not supported\n");
461                         break;
462 #endif
463                 case Opt_xip:
464 #ifdef CONFIG_EXT2_FS_XIP
465                         set_opt (sbi->s_mount_opt, XIP);
466 #else
467                         printk("EXT2 xip option not supported\n");
468 #endif
469                         break;
470
471 #if defined(CONFIG_QUOTA)
472                 case Opt_quota:
473                 case Opt_usrquota:
474                         set_opt(sbi->s_mount_opt, USRQUOTA);
475                         break;
476
477                 case Opt_grpquota:
478                         set_opt(sbi->s_mount_opt, GRPQUOTA);
479                         break;
480 #else
481                 case Opt_quota:
482                 case Opt_usrquota:
483                 case Opt_grpquota:
484                         printk(KERN_ERR
485                                 "EXT2-fs: quota operations not supported.\n");
486
487                         break;
488 #endif
489
490                 case Opt_ignore:
491                         break;
492                 default:
493                         return 0;
494                 }
495         }
496         return 1;
497 }
498
499 static int ext2_setup_super (struct super_block * sb,
500                               struct ext2_super_block * es,
501                               int read_only)
502 {
503         int res = 0;
504         struct ext2_sb_info *sbi = EXT2_SB(sb);
505
506         if (le32_to_cpu(es->s_rev_level) > EXT2_MAX_SUPP_REV) {
507                 printk ("EXT2-fs warning: revision level too high, "
508                         "forcing read-only mode\n");
509                 res = MS_RDONLY;
510         }
511         if (read_only)
512                 return res;
513         if (!(sbi->s_mount_state & EXT2_VALID_FS))
514                 printk ("EXT2-fs warning: mounting unchecked fs, "
515                         "running e2fsck is recommended\n");
516         else if ((sbi->s_mount_state & EXT2_ERROR_FS))
517                 printk ("EXT2-fs warning: mounting fs with errors, "
518                         "running e2fsck is recommended\n");
519         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
520                  le16_to_cpu(es->s_mnt_count) >=
521                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
522                 printk ("EXT2-fs warning: maximal mount count reached, "
523                         "running e2fsck is recommended\n");
524         else if (le32_to_cpu(es->s_checkinterval) &&
525                 (le32_to_cpu(es->s_lastcheck) + le32_to_cpu(es->s_checkinterval) <= get_seconds()))
526                 printk ("EXT2-fs warning: checktime reached, "
527                         "running e2fsck is recommended\n");
528         if (!le16_to_cpu(es->s_max_mnt_count))
529                 es->s_max_mnt_count = cpu_to_le16(EXT2_DFL_MAX_MNT_COUNT);
530         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
531         ext2_write_super(sb);
532         if (test_opt (sb, DEBUG))
533                 printk ("[EXT II FS %s, %s, bs=%lu, fs=%lu, gc=%lu, "
534                         "bpg=%lu, ipg=%lu, mo=%04lx]\n",
535                         EXT2FS_VERSION, EXT2FS_DATE, sb->s_blocksize,
536                         sbi->s_frag_size,
537                         sbi->s_groups_count,
538                         EXT2_BLOCKS_PER_GROUP(sb),
539                         EXT2_INODES_PER_GROUP(sb),
540                         sbi->s_mount_opt);
541         return res;
542 }
543
544 static int ext2_check_descriptors (struct super_block * sb)
545 {
546         int i;
547         int desc_block = 0;
548         struct ext2_sb_info *sbi = EXT2_SB(sb);
549         unsigned long first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
550         unsigned long last_block;
551         struct ext2_group_desc * gdp = NULL;
552
553         ext2_debug ("Checking group descriptors");
554
555         for (i = 0; i < sbi->s_groups_count; i++)
556         {
557                 if (i == sbi->s_groups_count - 1)
558                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
559                 else
560                         last_block = first_block +
561                                 (EXT2_BLOCKS_PER_GROUP(sb) - 1);
562
563                 if ((i % EXT2_DESC_PER_BLOCK(sb)) == 0)
564                         gdp = (struct ext2_group_desc *) sbi->s_group_desc[desc_block++]->b_data;
565                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
566                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
567                 {
568                         ext2_error (sb, "ext2_check_descriptors",
569                                     "Block bitmap for group %d"
570                                     " not in group (block %lu)!",
571                                     i, (unsigned long) le32_to_cpu(gdp->bg_block_bitmap));
572                         return 0;
573                 }
574                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
575                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
576                 {
577                         ext2_error (sb, "ext2_check_descriptors",
578                                     "Inode bitmap for group %d"
579                                     " not in group (block %lu)!",
580                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_bitmap));
581                         return 0;
582                 }
583                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
584                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >
585                     last_block)
586                 {
587                         ext2_error (sb, "ext2_check_descriptors",
588                                     "Inode table for group %d"
589                                     " not in group (block %lu)!",
590                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_table));
591                         return 0;
592                 }
593                 first_block += EXT2_BLOCKS_PER_GROUP(sb);
594                 gdp++;
595         }
596         return 1;
597 }
598
599 /*
600  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
601  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
602  * We need to be 1 filesystem block less than the 2^32 sector limit.
603  */
604 static loff_t ext2_max_size(int bits)
605 {
606         loff_t res = EXT2_NDIR_BLOCKS;
607         /* This constant is calculated to be the largest file size for a
608          * dense, 4k-blocksize file such that the total number of
609          * sectors in the file, including data and all indirect blocks,
610          * does not exceed 2^32. */
611         const loff_t upper_limit = 0x1ff7fffd000LL;
612
613         res += 1LL << (bits-2);
614         res += 1LL << (2*(bits-2));
615         res += 1LL << (3*(bits-2));
616         res <<= bits;
617         if (res > upper_limit)
618                 res = upper_limit;
619         return res;
620 }
621
622 static unsigned long descriptor_loc(struct super_block *sb,
623                                     unsigned long logic_sb_block,
624                                     int nr)
625 {
626         struct ext2_sb_info *sbi = EXT2_SB(sb);
627         unsigned long bg, first_data_block, first_meta_bg;
628         int has_super = 0;
629         
630         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
631         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
632
633         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_META_BG) ||
634             nr < first_meta_bg)
635                 return (logic_sb_block + nr + 1);
636         bg = sbi->s_desc_per_block * nr;
637         if (ext2_bg_has_super(sb, bg))
638                 has_super = 1;
639         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
640 }
641
642 static int ext2_fill_super(struct super_block *sb, void *data, int silent)
643 {
644         struct buffer_head * bh;
645         struct ext2_sb_info * sbi;
646         struct ext2_super_block * es;
647         struct inode *root;
648         unsigned long block;
649         unsigned long sb_block = get_sb_block(&data);
650         unsigned long logic_sb_block;
651         unsigned long offset = 0;
652         unsigned long def_mount_opts;
653         int blocksize = BLOCK_SIZE;
654         int db_count;
655         int i, j;
656         __le32 features;
657
658         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
659         if (!sbi)
660                 return -ENOMEM;
661         sb->s_fs_info = sbi;
662
663         /*
664          * See what the current blocksize for the device is, and
665          * use that as the blocksize.  Otherwise (or if the blocksize
666          * is smaller than the default) use the default.
667          * This is important for devices that have a hardware
668          * sectorsize that is larger than the default.
669          */
670         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
671         if (!blocksize) {
672                 printk ("EXT2-fs: unable to set blocksize\n");
673                 goto failed_sbi;
674         }
675
676         /*
677          * If the superblock doesn't start on a hardware sector boundary,
678          * calculate the offset.  
679          */
680         if (blocksize != BLOCK_SIZE) {
681                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
682                 offset = (sb_block*BLOCK_SIZE) % blocksize;
683         } else {
684                 logic_sb_block = sb_block;
685         }
686
687         if (!(bh = sb_bread(sb, logic_sb_block))) {
688                 printk ("EXT2-fs: unable to read superblock\n");
689                 goto failed_sbi;
690         }
691         /*
692          * Note: s_es must be initialized as soon as possible because
693          *       some ext2 macro-instructions depend on its value
694          */
695         es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
696         sbi->s_es = es;
697         sb->s_magic = le16_to_cpu(es->s_magic);
698
699         if (sb->s_magic != EXT2_SUPER_MAGIC)
700                 goto cantfind_ext2;
701
702         /* Set defaults before we parse the mount options */
703         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
704         if (def_mount_opts & EXT2_DEFM_DEBUG)
705                 set_opt(sbi->s_mount_opt, DEBUG);
706         if (def_mount_opts & EXT2_DEFM_BSDGROUPS)
707                 set_opt(sbi->s_mount_opt, GRPID);
708         if (def_mount_opts & EXT2_DEFM_UID16)
709                 set_opt(sbi->s_mount_opt, NO_UID32);
710 #ifdef CONFIG_EXT2_FS_XATTR
711         if (def_mount_opts & EXT2_DEFM_XATTR_USER)
712                 set_opt(sbi->s_mount_opt, XATTR_USER);
713 #endif
714 #ifdef CONFIG_EXT2_FS_POSIX_ACL
715         if (def_mount_opts & EXT2_DEFM_ACL)
716                 set_opt(sbi->s_mount_opt, POSIX_ACL);
717 #endif
718         
719         if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_PANIC)
720                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
721         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_RO)
722                 set_opt(sbi->s_mount_opt, ERRORS_RO);
723         else
724                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
725
726         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
727         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
728         
729         if (!parse_options ((char *) data, sbi))
730                 goto failed_mount;
731
732         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
733                 ((EXT2_SB(sb)->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ?
734                  MS_POSIXACL : 0);
735
736         ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
737                                     EXT2_MOUNT_XIP if not */
738
739         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV &&
740             (EXT2_HAS_COMPAT_FEATURE(sb, ~0U) ||
741              EXT2_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
742              EXT2_HAS_INCOMPAT_FEATURE(sb, ~0U)))
743                 printk("EXT2-fs warning: feature flags set on rev 0 fs, "
744                        "running e2fsck is recommended\n");
745         /*
746          * Check feature flags regardless of the revision level, since we
747          * previously didn't change the revision level when setting the flags,
748          * so there is a chance incompat flags are set on a rev 0 filesystem.
749          */
750         features = EXT2_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP);
751         if (features) {
752                 printk("EXT2-fs: %s: couldn't mount because of "
753                        "unsupported optional features (%x).\n",
754                        sb->s_id, le32_to_cpu(features));
755                 goto failed_mount;
756         }
757         if (!(sb->s_flags & MS_RDONLY) &&
758             (features = EXT2_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))){
759                 printk("EXT2-fs: %s: couldn't mount RDWR because of "
760                        "unsupported optional features (%x).\n",
761                        sb->s_id, le32_to_cpu(features));
762                 goto failed_mount;
763         }
764
765         blocksize = BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
766
767         if ((ext2_use_xip(sb)) && ((blocksize != PAGE_SIZE) ||
768                                   (sb->s_blocksize != blocksize))) {
769                 if (!silent)
770                         printk("XIP: Unsupported blocksize\n");
771                 goto failed_mount;
772         }
773
774         /* If the blocksize doesn't match, re-read the thing.. */
775         if (sb->s_blocksize != blocksize) {
776                 brelse(bh);
777
778                 if (!sb_set_blocksize(sb, blocksize)) {
779                         printk(KERN_ERR "EXT2-fs: blocksize too small for device.\n");
780                         goto failed_sbi;
781                 }
782
783                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
784                 offset = (sb_block*BLOCK_SIZE) % blocksize;
785                 bh = sb_bread(sb, logic_sb_block);
786                 if(!bh) {
787                         printk("EXT2-fs: Couldn't read superblock on "
788                                "2nd try.\n");
789                         goto failed_sbi;
790                 }
791                 es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
792                 sbi->s_es = es;
793                 if (es->s_magic != cpu_to_le16(EXT2_SUPER_MAGIC)) {
794                         printk ("EXT2-fs: Magic mismatch, very weird !\n");
795                         goto failed_mount;
796                 }
797         }
798
799         sb->s_maxbytes = ext2_max_size(sb->s_blocksize_bits);
800
801         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV) {
802                 sbi->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
803                 sbi->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
804         } else {
805                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
806                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
807                 if ((sbi->s_inode_size < EXT2_GOOD_OLD_INODE_SIZE) ||
808                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
809                     (sbi->s_inode_size > blocksize)) {
810                         printk ("EXT2-fs: unsupported inode size: %d\n",
811                                 sbi->s_inode_size);
812                         goto failed_mount;
813                 }
814         }
815
816         sbi->s_frag_size = EXT2_MIN_FRAG_SIZE <<
817                                    le32_to_cpu(es->s_log_frag_size);
818         if (sbi->s_frag_size == 0)
819                 goto cantfind_ext2;
820         sbi->s_frags_per_block = sb->s_blocksize / sbi->s_frag_size;
821
822         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
823         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
824         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
825
826         if (EXT2_INODE_SIZE(sb) == 0)
827                 goto cantfind_ext2;
828         sbi->s_inodes_per_block = sb->s_blocksize / EXT2_INODE_SIZE(sb);
829         if (sbi->s_inodes_per_block == 0 || sbi->s_inodes_per_group == 0)
830                 goto cantfind_ext2;
831         sbi->s_itb_per_group = sbi->s_inodes_per_group /
832                                         sbi->s_inodes_per_block;
833         sbi->s_desc_per_block = sb->s_blocksize /
834                                         sizeof (struct ext2_group_desc);
835         sbi->s_sbh = bh;
836         sbi->s_mount_state = le16_to_cpu(es->s_state);
837         sbi->s_addr_per_block_bits =
838                 ilog2 (EXT2_ADDR_PER_BLOCK(sb));
839         sbi->s_desc_per_block_bits =
840                 ilog2 (EXT2_DESC_PER_BLOCK(sb));
841
842         if (sb->s_magic != EXT2_SUPER_MAGIC)
843                 goto cantfind_ext2;
844
845         if (sb->s_blocksize != bh->b_size) {
846                 if (!silent)
847                         printk ("VFS: Unsupported blocksize on dev "
848                                 "%s.\n", sb->s_id);
849                 goto failed_mount;
850         }
851
852         if (sb->s_blocksize != sbi->s_frag_size) {
853                 printk ("EXT2-fs: fragsize %lu != blocksize %lu (not supported yet)\n",
854                         sbi->s_frag_size, sb->s_blocksize);
855                 goto failed_mount;
856         }
857
858         if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
859                 printk ("EXT2-fs: #blocks per group too big: %lu\n",
860                         sbi->s_blocks_per_group);
861                 goto failed_mount;
862         }
863         if (sbi->s_frags_per_group > sb->s_blocksize * 8) {
864                 printk ("EXT2-fs: #fragments per group too big: %lu\n",
865                         sbi->s_frags_per_group);
866                 goto failed_mount;
867         }
868         if (sbi->s_inodes_per_group > sb->s_blocksize * 8) {
869                 printk ("EXT2-fs: #inodes per group too big: %lu\n",
870                         sbi->s_inodes_per_group);
871                 goto failed_mount;
872         }
873
874         if (EXT2_BLOCKS_PER_GROUP(sb) == 0)
875                 goto cantfind_ext2;
876         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
877                                 le32_to_cpu(es->s_first_data_block) - 1)
878                                         / EXT2_BLOCKS_PER_GROUP(sb)) + 1;
879         db_count = (sbi->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) /
880                    EXT2_DESC_PER_BLOCK(sb);
881         sbi->s_group_desc = kmalloc (db_count * sizeof (struct buffer_head *), GFP_KERNEL);
882         if (sbi->s_group_desc == NULL) {
883                 printk ("EXT2-fs: not enough memory\n");
884                 goto failed_mount;
885         }
886         bgl_lock_init(&sbi->s_blockgroup_lock);
887         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(*sbi->s_debts),
888                                GFP_KERNEL);
889         if (!sbi->s_debts) {
890                 printk ("EXT2-fs: not enough memory\n");
891                 goto failed_mount_group_desc;
892         }
893         memset(sbi->s_debts, 0, sbi->s_groups_count * sizeof(*sbi->s_debts));
894         for (i = 0; i < db_count; i++) {
895                 block = descriptor_loc(sb, logic_sb_block, i);
896                 sbi->s_group_desc[i] = sb_bread(sb, block);
897                 if (!sbi->s_group_desc[i]) {
898                         for (j = 0; j < i; j++)
899                                 brelse (sbi->s_group_desc[j]);
900                         printk ("EXT2-fs: unable to read group descriptors\n");
901                         goto failed_mount_group_desc;
902                 }
903         }
904         if (!ext2_check_descriptors (sb)) {
905                 printk ("EXT2-fs: group descriptors corrupted!\n");
906                 goto failed_mount2;
907         }
908         sbi->s_gdb_count = db_count;
909         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
910         spin_lock_init(&sbi->s_next_gen_lock);
911
912         percpu_counter_init(&sbi->s_freeblocks_counter,
913                                 ext2_count_free_blocks(sb));
914         percpu_counter_init(&sbi->s_freeinodes_counter,
915                                 ext2_count_free_inodes(sb));
916         percpu_counter_init(&sbi->s_dirs_counter,
917                                 ext2_count_dirs(sb));
918         /*
919          * set up enough so that it can read an inode
920          */
921         sb->s_op = &ext2_sops;
922         sb->s_export_op = &ext2_export_ops;
923         sb->s_xattr = ext2_xattr_handlers;
924         root = iget(sb, EXT2_ROOT_INO);
925         sb->s_root = d_alloc_root(root);
926         if (!sb->s_root) {
927                 iput(root);
928                 printk(KERN_ERR "EXT2-fs: get root inode failed\n");
929                 goto failed_mount3;
930         }
931         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
932                 dput(sb->s_root);
933                 sb->s_root = NULL;
934                 printk(KERN_ERR "EXT2-fs: corrupt root inode, run e2fsck\n");
935                 goto failed_mount3;
936         }
937         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL))
938                 ext2_warning(sb, __FUNCTION__,
939                         "mounting ext3 filesystem as ext2");
940         ext2_setup_super (sb, es, sb->s_flags & MS_RDONLY);
941         return 0;
942
943 cantfind_ext2:
944         if (!silent)
945                 printk("VFS: Can't find an ext2 filesystem on dev %s.\n",
946                        sb->s_id);
947         goto failed_mount;
948 failed_mount3:
949         percpu_counter_destroy(&sbi->s_freeblocks_counter);
950         percpu_counter_destroy(&sbi->s_freeinodes_counter);
951         percpu_counter_destroy(&sbi->s_dirs_counter);
952 failed_mount2:
953         for (i = 0; i < db_count; i++)
954                 brelse(sbi->s_group_desc[i]);
955 failed_mount_group_desc:
956         kfree(sbi->s_group_desc);
957         kfree(sbi->s_debts);
958 failed_mount:
959         brelse(bh);
960 failed_sbi:
961         sb->s_fs_info = NULL;
962         kfree(sbi);
963         return -EINVAL;
964 }
965
966 static void ext2_commit_super (struct super_block * sb,
967                                struct ext2_super_block * es)
968 {
969         es->s_wtime = cpu_to_le32(get_seconds());
970         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
971         sb->s_dirt = 0;
972 }
973
974 static void ext2_sync_super(struct super_block *sb, struct ext2_super_block *es)
975 {
976         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
977         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
978         es->s_wtime = cpu_to_le32(get_seconds());
979         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
980         sync_dirty_buffer(EXT2_SB(sb)->s_sbh);
981         sb->s_dirt = 0;
982 }
983
984 /*
985  * In the second extended file system, it is not necessary to
986  * write the super block since we use a mapping of the
987  * disk super block in a buffer.
988  *
989  * However, this function is still used to set the fs valid
990  * flags to 0.  We need to set this flag to 0 since the fs
991  * may have been checked while mounted and e2fsck may have
992  * set s_state to EXT2_VALID_FS after some corrections.
993  */
994
995 void ext2_write_super (struct super_block * sb)
996 {
997         struct ext2_super_block * es;
998         lock_kernel();
999         if (!(sb->s_flags & MS_RDONLY)) {
1000                 es = EXT2_SB(sb)->s_es;
1001
1002                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS) {
1003                         ext2_debug ("setting valid to 0\n");
1004                         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) &
1005                                                   ~EXT2_VALID_FS);
1006                         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
1007                         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
1008                         es->s_mtime = cpu_to_le32(get_seconds());
1009                         ext2_sync_super(sb, es);
1010                 } else
1011                         ext2_commit_super (sb, es);
1012         }
1013         sb->s_dirt = 0;
1014         unlock_kernel();
1015 }
1016
1017 static int ext2_remount (struct super_block * sb, int * flags, char * data)
1018 {
1019         struct ext2_sb_info * sbi = EXT2_SB(sb);
1020         struct ext2_super_block * es;
1021         unsigned long old_mount_opt = sbi->s_mount_opt;
1022         struct ext2_mount_options old_opts;
1023         unsigned long old_sb_flags;
1024         int err;
1025
1026         /* Store the old options */
1027         old_sb_flags = sb->s_flags;
1028         old_opts.s_mount_opt = sbi->s_mount_opt;
1029         old_opts.s_resuid = sbi->s_resuid;
1030         old_opts.s_resgid = sbi->s_resgid;
1031
1032         /*
1033          * Allow the "check" option to be passed as a remount option.
1034          */
1035         if (!parse_options (data, sbi)) {
1036                 err = -EINVAL;
1037                 goto restore_opts;
1038         }
1039
1040         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1041                 ((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1042
1043         es = sbi->s_es;
1044         if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
1045             (old_mount_opt & EXT2_MOUNT_XIP)) &&
1046             invalidate_inodes(sb))
1047                 ext2_warning(sb, __FUNCTION__, "busy inodes while remounting "\
1048                              "xip remain in cache (no functional problem)");
1049         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1050                 return 0;
1051         if (*flags & MS_RDONLY) {
1052                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS ||
1053                     !(sbi->s_mount_state & EXT2_VALID_FS))
1054                         return 0;
1055                 /*
1056                  * OK, we are remounting a valid rw partition rdonly, so set
1057                  * the rdonly flag and then mark the partition as valid again.
1058                  */
1059                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1060                 es->s_mtime = cpu_to_le32(get_seconds());
1061         } else {
1062                 __le32 ret = EXT2_HAS_RO_COMPAT_FEATURE(sb,
1063                                                ~EXT2_FEATURE_RO_COMPAT_SUPP);
1064                 if (ret) {
1065                         printk("EXT2-fs: %s: couldn't remount RDWR because of "
1066                                "unsupported optional features (%x).\n",
1067                                sb->s_id, le32_to_cpu(ret));
1068                         err = -EROFS;
1069                         goto restore_opts;
1070                 }
1071                 /*
1072                  * Mounting a RDONLY partition read-write, so reread and
1073                  * store the current valid flag.  (It may have been changed
1074                  * by e2fsck since we originally mounted the partition.)
1075                  */
1076                 sbi->s_mount_state = le16_to_cpu(es->s_state);
1077                 if (!ext2_setup_super (sb, es, 0))
1078                         sb->s_flags &= ~MS_RDONLY;
1079         }
1080         ext2_sync_super(sb, es);
1081         return 0;
1082 restore_opts:
1083         sbi->s_mount_opt = old_opts.s_mount_opt;
1084         sbi->s_resuid = old_opts.s_resuid;
1085         sbi->s_resgid = old_opts.s_resgid;
1086         sb->s_flags = old_sb_flags;
1087         return err;
1088 }
1089
1090 static int ext2_statfs (struct dentry * dentry, struct kstatfs * buf)
1091 {
1092         struct super_block *sb = dentry->d_sb;
1093         struct ext2_sb_info *sbi = EXT2_SB(sb);
1094         struct ext2_super_block *es = sbi->s_es;
1095         unsigned long overhead;
1096         int i;
1097         u64 fsid;
1098
1099         if (test_opt (sb, MINIX_DF))
1100                 overhead = 0;
1101         else {
1102                 /*
1103                  * Compute the overhead (FS structures)
1104                  */
1105
1106                 /*
1107                  * All of the blocks before first_data_block are
1108                  * overhead
1109                  */
1110                 overhead = le32_to_cpu(es->s_first_data_block);
1111
1112                 /*
1113                  * Add the overhead attributed to the superblock and
1114                  * block group descriptors.  If the sparse superblocks
1115                  * feature is turned on, then not all groups have this.
1116                  */
1117                 for (i = 0; i < sbi->s_groups_count; i++)
1118                         overhead += ext2_bg_has_super(sb, i) +
1119                                 ext2_bg_num_gdb(sb, i);
1120
1121                 /*
1122                  * Every block group has an inode bitmap, a block
1123                  * bitmap, and an inode table.
1124                  */
1125                 overhead += (sbi->s_groups_count *
1126                              (2 + sbi->s_itb_per_group));
1127         }
1128
1129         buf->f_type = EXT2_SUPER_MAGIC;
1130         buf->f_bsize = sb->s_blocksize;
1131         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
1132         buf->f_bfree = ext2_count_free_blocks(sb);
1133         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
1134         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
1135                 buf->f_bavail = 0;
1136         buf->f_files = le32_to_cpu(es->s_inodes_count);
1137         buf->f_ffree = ext2_count_free_inodes(sb);
1138         buf->f_namelen = EXT2_NAME_LEN;
1139         fsid = le64_to_cpup((void *)es->s_uuid) ^
1140                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
1141         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1142         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1143         return 0;
1144 }
1145
1146 static int ext2_get_sb(struct file_system_type *fs_type,
1147         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1148 {
1149         return get_sb_bdev(fs_type, flags, dev_name, data, ext2_fill_super, mnt);
1150 }
1151
1152 #ifdef CONFIG_QUOTA
1153
1154 /* Read data from quotafile - avoid pagecache and such because we cannot afford
1155  * acquiring the locks... As quota files are never truncated and quota code
1156  * itself serializes the operations (and noone else should touch the files)
1157  * we don't have to be afraid of races */
1158 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data,
1159                                size_t len, loff_t off)
1160 {
1161         struct inode *inode = sb_dqopt(sb)->files[type];
1162         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1163         int err = 0;
1164         int offset = off & (sb->s_blocksize - 1);
1165         int tocopy;
1166         size_t toread;
1167         struct buffer_head tmp_bh;
1168         struct buffer_head *bh;
1169         loff_t i_size = i_size_read(inode);
1170
1171         if (off > i_size)
1172                 return 0;
1173         if (off+len > i_size)
1174                 len = i_size-off;
1175         toread = len;
1176         while (toread > 0) {
1177                 tocopy = sb->s_blocksize - offset < toread ?
1178                                 sb->s_blocksize - offset : toread;
1179
1180                 tmp_bh.b_state = 0;
1181                 err = ext2_get_block(inode, blk, &tmp_bh, 0);
1182                 if (err)
1183                         return err;
1184                 if (!buffer_mapped(&tmp_bh))    /* A hole? */
1185                         memset(data, 0, tocopy);
1186                 else {
1187                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1188                         if (!bh)
1189                                 return -EIO;
1190                         memcpy(data, bh->b_data+offset, tocopy);
1191                         brelse(bh);
1192                 }
1193                 offset = 0;
1194                 toread -= tocopy;
1195                 data += tocopy;
1196                 blk++;
1197         }
1198         return len;
1199 }
1200
1201 /* Write to quotafile */
1202 static ssize_t ext2_quota_write(struct super_block *sb, int type,
1203                                 const char *data, size_t len, loff_t off)
1204 {
1205         struct inode *inode = sb_dqopt(sb)->files[type];
1206         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1207         int err = 0;
1208         int offset = off & (sb->s_blocksize - 1);
1209         int tocopy;
1210         size_t towrite = len;
1211         struct buffer_head tmp_bh;
1212         struct buffer_head *bh;
1213
1214         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
1215         while (towrite > 0) {
1216                 tocopy = sb->s_blocksize - offset < towrite ?
1217                                 sb->s_blocksize - offset : towrite;
1218
1219                 tmp_bh.b_state = 0;
1220                 err = ext2_get_block(inode, blk, &tmp_bh, 1);
1221                 if (err)
1222                         goto out;
1223                 if (offset || tocopy != EXT2_BLOCK_SIZE(sb))
1224                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1225                 else
1226                         bh = sb_getblk(sb, tmp_bh.b_blocknr);
1227                 if (!bh) {
1228                         err = -EIO;
1229                         goto out;
1230                 }
1231                 lock_buffer(bh);
1232                 memcpy(bh->b_data+offset, data, tocopy);
1233                 flush_dcache_page(bh->b_page);
1234                 set_buffer_uptodate(bh);
1235                 mark_buffer_dirty(bh);
1236                 unlock_buffer(bh);
1237                 brelse(bh);
1238                 offset = 0;
1239                 towrite -= tocopy;
1240                 data += tocopy;
1241                 blk++;
1242         }
1243 out:
1244         if (len == towrite)
1245                 return err;
1246         if (inode->i_size < off+len-towrite)
1247                 i_size_write(inode, off+len-towrite);
1248         inode->i_version++;
1249         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1250         mark_inode_dirty(inode);
1251         mutex_unlock(&inode->i_mutex);
1252         return len - towrite;
1253 }
1254
1255 #endif
1256
1257 static struct file_system_type ext2_fs_type = {
1258         .owner          = THIS_MODULE,
1259         .name           = "ext2",
1260         .get_sb         = ext2_get_sb,
1261         .kill_sb        = kill_block_super,
1262         .fs_flags       = FS_REQUIRES_DEV,
1263 };
1264
1265 static int __init init_ext2_fs(void)
1266 {
1267         int err = init_ext2_xattr();
1268         if (err)
1269                 return err;
1270         err = init_inodecache();
1271         if (err)
1272                 goto out1;
1273         err = register_filesystem(&ext2_fs_type);
1274         if (err)
1275                 goto out;
1276         return 0;
1277 out:
1278         destroy_inodecache();
1279 out1:
1280         exit_ext2_xattr();
1281         return err;
1282 }
1283
1284 static void __exit exit_ext2_fs(void)
1285 {
1286         unregister_filesystem(&ext2_fs_type);
1287         destroy_inodecache();
1288         exit_ext2_xattr();
1289 }
1290
1291 module_init(init_ext2_fs)
1292 module_exit(exit_ext2_fs)