Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[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_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
164             SLAB_CTOR_CONSTRUCTOR) {
165                 rwlock_init(&ei->i_meta_lock);
166 #ifdef CONFIG_EXT2_FS_XATTR
167                 init_rwsem(&ei->xattr_sem);
168 #endif
169                 inode_init_once(&ei->vfs_inode);
170         }
171 }
172  
173 static int init_inodecache(void)
174 {
175         ext2_inode_cachep = kmem_cache_create("ext2_inode_cache",
176                                              sizeof(struct ext2_inode_info),
177                                              0, (SLAB_RECLAIM_ACCOUNT|
178                                                 SLAB_MEM_SPREAD),
179                                              init_once, NULL);
180         if (ext2_inode_cachep == NULL)
181                 return -ENOMEM;
182         return 0;
183 }
184
185 static void destroy_inodecache(void)
186 {
187         kmem_cache_destroy(ext2_inode_cachep);
188 }
189
190 static void ext2_clear_inode(struct inode *inode)
191 {
192 #ifdef CONFIG_EXT2_FS_POSIX_ACL
193         struct ext2_inode_info *ei = EXT2_I(inode);
194
195         if (ei->i_acl && ei->i_acl != EXT2_ACL_NOT_CACHED) {
196                 posix_acl_release(ei->i_acl);
197                 ei->i_acl = EXT2_ACL_NOT_CACHED;
198         }
199         if (ei->i_default_acl && ei->i_default_acl != EXT2_ACL_NOT_CACHED) {
200                 posix_acl_release(ei->i_default_acl);
201                 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
202         }
203 #endif
204 }
205
206 static int ext2_show_options(struct seq_file *seq, struct vfsmount *vfs)
207 {
208         struct ext2_sb_info *sbi = EXT2_SB(vfs->mnt_sb);
209
210         if (sbi->s_mount_opt & EXT2_MOUNT_GRPID)
211                 seq_puts(seq, ",grpid");
212
213 #if defined(CONFIG_QUOTA)
214         if (sbi->s_mount_opt & EXT2_MOUNT_USRQUOTA)
215                 seq_puts(seq, ",usrquota");
216
217         if (sbi->s_mount_opt & EXT2_MOUNT_GRPQUOTA)
218                 seq_puts(seq, ",grpquota");
219 #endif
220
221 #if defined(CONFIG_EXT2_FS_XIP)
222         if (sbi->s_mount_opt & EXT2_MOUNT_XIP)
223                 seq_puts(seq, ",xip");
224 #endif
225
226         return 0;
227 }
228
229 #ifdef CONFIG_QUOTA
230 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data, size_t len, loff_t off);
231 static ssize_t ext2_quota_write(struct super_block *sb, int type, const char *data, size_t len, loff_t off);
232 #endif
233
234 static struct super_operations ext2_sops = {
235         .alloc_inode    = ext2_alloc_inode,
236         .destroy_inode  = ext2_destroy_inode,
237         .read_inode     = ext2_read_inode,
238         .write_inode    = ext2_write_inode,
239         .put_inode      = ext2_put_inode,
240         .delete_inode   = ext2_delete_inode,
241         .put_super      = ext2_put_super,
242         .write_super    = ext2_write_super,
243         .statfs         = ext2_statfs,
244         .remount_fs     = ext2_remount,
245         .clear_inode    = ext2_clear_inode,
246         .show_options   = ext2_show_options,
247 #ifdef CONFIG_QUOTA
248         .quota_read     = ext2_quota_read,
249         .quota_write    = ext2_quota_write,
250 #endif
251 };
252
253 static struct dentry *ext2_get_dentry(struct super_block *sb, void *vobjp)
254 {
255         __u32 *objp = vobjp;
256         unsigned long ino = objp[0];
257         __u32 generation = objp[1];
258         struct inode *inode;
259         struct dentry *result;
260
261         if (ino < EXT2_FIRST_INO(sb) && ino != EXT2_ROOT_INO)
262                 return ERR_PTR(-ESTALE);
263         if (ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
264                 return ERR_PTR(-ESTALE);
265
266         /* iget isn't really right if the inode is currently unallocated!!
267          * ext2_read_inode currently does appropriate checks, but
268          * it might be "neater" to call ext2_get_inode first and check
269          * if the inode is valid.....
270          */
271         inode = iget(sb, ino);
272         if (inode == NULL)
273                 return ERR_PTR(-ENOMEM);
274         if (is_bad_inode(inode) ||
275             (generation && inode->i_generation != generation)) {
276                 /* we didn't find the right inode.. */
277                 iput(inode);
278                 return ERR_PTR(-ESTALE);
279         }
280         /* now to find a dentry.
281          * If possible, get a well-connected one
282          */
283         result = d_alloc_anon(inode);
284         if (!result) {
285                 iput(inode);
286                 return ERR_PTR(-ENOMEM);
287         }
288         return result;
289 }
290
291 /* Yes, most of these are left as NULL!!
292  * A NULL value implies the default, which works with ext2-like file
293  * systems, but can be improved upon.
294  * Currently only get_parent is required.
295  */
296 static struct export_operations ext2_export_ops = {
297         .get_parent = ext2_get_parent,
298         .get_dentry = ext2_get_dentry,
299 };
300
301 static unsigned long get_sb_block(void **data)
302 {
303         unsigned long   sb_block;
304         char            *options = (char *) *data;
305
306         if (!options || strncmp(options, "sb=", 3) != 0)
307                 return 1;       /* Default location */
308         options += 3;
309         sb_block = simple_strtoul(options, &options, 0);
310         if (*options && *options != ',') {
311                 printk("EXT2-fs: Invalid sb specification: %s\n",
312                        (char *) *data);
313                 return 1;
314         }
315         if (*options == ',')
316                 options++;
317         *data = (void *) options;
318         return sb_block;
319 }
320
321 enum {
322         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
323         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic,
324         Opt_err_ro, Opt_nouid32, Opt_nocheck, Opt_debug,
325         Opt_oldalloc, Opt_orlov, Opt_nobh, Opt_user_xattr, Opt_nouser_xattr,
326         Opt_acl, Opt_noacl, Opt_xip, Opt_ignore, Opt_err, Opt_quota,
327         Opt_usrquota, Opt_grpquota
328 };
329
330 static match_table_t tokens = {
331         {Opt_bsd_df, "bsddf"},
332         {Opt_minix_df, "minixdf"},
333         {Opt_grpid, "grpid"},
334         {Opt_grpid, "bsdgroups"},
335         {Opt_nogrpid, "nogrpid"},
336         {Opt_nogrpid, "sysvgroups"},
337         {Opt_resgid, "resgid=%u"},
338         {Opt_resuid, "resuid=%u"},
339         {Opt_sb, "sb=%u"},
340         {Opt_err_cont, "errors=continue"},
341         {Opt_err_panic, "errors=panic"},
342         {Opt_err_ro, "errors=remount-ro"},
343         {Opt_nouid32, "nouid32"},
344         {Opt_nocheck, "check=none"},
345         {Opt_nocheck, "nocheck"},
346         {Opt_debug, "debug"},
347         {Opt_oldalloc, "oldalloc"},
348         {Opt_orlov, "orlov"},
349         {Opt_nobh, "nobh"},
350         {Opt_user_xattr, "user_xattr"},
351         {Opt_nouser_xattr, "nouser_xattr"},
352         {Opt_acl, "acl"},
353         {Opt_noacl, "noacl"},
354         {Opt_xip, "xip"},
355         {Opt_grpquota, "grpquota"},
356         {Opt_ignore, "noquota"},
357         {Opt_quota, "quota"},
358         {Opt_usrquota, "usrquota"},
359         {Opt_err, NULL}
360 };
361
362 static int parse_options (char * options,
363                           struct ext2_sb_info *sbi)
364 {
365         char * p;
366         substring_t args[MAX_OPT_ARGS];
367         int option;
368
369         if (!options)
370                 return 1;
371
372         while ((p = strsep (&options, ",")) != NULL) {
373                 int token;
374                 if (!*p)
375                         continue;
376
377                 token = match_token(p, tokens, args);
378                 switch (token) {
379                 case Opt_bsd_df:
380                         clear_opt (sbi->s_mount_opt, MINIX_DF);
381                         break;
382                 case Opt_minix_df:
383                         set_opt (sbi->s_mount_opt, MINIX_DF);
384                         break;
385                 case Opt_grpid:
386                         set_opt (sbi->s_mount_opt, GRPID);
387                         break;
388                 case Opt_nogrpid:
389                         clear_opt (sbi->s_mount_opt, GRPID);
390                         break;
391                 case Opt_resuid:
392                         if (match_int(&args[0], &option))
393                                 return 0;
394                         sbi->s_resuid = option;
395                         break;
396                 case Opt_resgid:
397                         if (match_int(&args[0], &option))
398                                 return 0;
399                         sbi->s_resgid = option;
400                         break;
401                 case Opt_sb:
402                         /* handled by get_sb_block() instead of here */
403                         /* *sb_block = match_int(&args[0]); */
404                         break;
405                 case Opt_err_panic:
406                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
407                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
408                         set_opt (sbi->s_mount_opt, ERRORS_PANIC);
409                         break;
410                 case Opt_err_ro:
411                         clear_opt (sbi->s_mount_opt, ERRORS_CONT);
412                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
413                         set_opt (sbi->s_mount_opt, ERRORS_RO);
414                         break;
415                 case Opt_err_cont:
416                         clear_opt (sbi->s_mount_opt, ERRORS_RO);
417                         clear_opt (sbi->s_mount_opt, ERRORS_PANIC);
418                         set_opt (sbi->s_mount_opt, ERRORS_CONT);
419                         break;
420                 case Opt_nouid32:
421                         set_opt (sbi->s_mount_opt, NO_UID32);
422                         break;
423                 case Opt_nocheck:
424                         clear_opt (sbi->s_mount_opt, CHECK);
425                         break;
426                 case Opt_debug:
427                         set_opt (sbi->s_mount_opt, DEBUG);
428                         break;
429                 case Opt_oldalloc:
430                         set_opt (sbi->s_mount_opt, OLDALLOC);
431                         break;
432                 case Opt_orlov:
433                         clear_opt (sbi->s_mount_opt, OLDALLOC);
434                         break;
435                 case Opt_nobh:
436                         set_opt (sbi->s_mount_opt, NOBH);
437                         break;
438 #ifdef CONFIG_EXT2_FS_XATTR
439                 case Opt_user_xattr:
440                         set_opt (sbi->s_mount_opt, XATTR_USER);
441                         break;
442                 case Opt_nouser_xattr:
443                         clear_opt (sbi->s_mount_opt, XATTR_USER);
444                         break;
445 #else
446                 case Opt_user_xattr:
447                 case Opt_nouser_xattr:
448                         printk("EXT2 (no)user_xattr options not supported\n");
449                         break;
450 #endif
451 #ifdef CONFIG_EXT2_FS_POSIX_ACL
452                 case Opt_acl:
453                         set_opt(sbi->s_mount_opt, POSIX_ACL);
454                         break;
455                 case Opt_noacl:
456                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
457                         break;
458 #else
459                 case Opt_acl:
460                 case Opt_noacl:
461                         printk("EXT2 (no)acl options not supported\n");
462                         break;
463 #endif
464                 case Opt_xip:
465 #ifdef CONFIG_EXT2_FS_XIP
466                         set_opt (sbi->s_mount_opt, XIP);
467 #else
468                         printk("EXT2 xip option not supported\n");
469 #endif
470                         break;
471
472 #if defined(CONFIG_QUOTA)
473                 case Opt_quota:
474                 case Opt_usrquota:
475                         set_opt(sbi->s_mount_opt, USRQUOTA);
476                         break;
477
478                 case Opt_grpquota:
479                         set_opt(sbi->s_mount_opt, GRPQUOTA);
480                         break;
481 #else
482                 case Opt_quota:
483                 case Opt_usrquota:
484                 case Opt_grpquota:
485                         printk(KERN_ERR
486                                 "EXT2-fs: quota operations not supported.\n");
487
488                         break;
489 #endif
490
491                 case Opt_ignore:
492                         break;
493                 default:
494                         return 0;
495                 }
496         }
497         return 1;
498 }
499
500 static int ext2_setup_super (struct super_block * sb,
501                               struct ext2_super_block * es,
502                               int read_only)
503 {
504         int res = 0;
505         struct ext2_sb_info *sbi = EXT2_SB(sb);
506
507         if (le32_to_cpu(es->s_rev_level) > EXT2_MAX_SUPP_REV) {
508                 printk ("EXT2-fs warning: revision level too high, "
509                         "forcing read-only mode\n");
510                 res = MS_RDONLY;
511         }
512         if (read_only)
513                 return res;
514         if (!(sbi->s_mount_state & EXT2_VALID_FS))
515                 printk ("EXT2-fs warning: mounting unchecked fs, "
516                         "running e2fsck is recommended\n");
517         else if ((sbi->s_mount_state & EXT2_ERROR_FS))
518                 printk ("EXT2-fs warning: mounting fs with errors, "
519                         "running e2fsck is recommended\n");
520         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
521                  le16_to_cpu(es->s_mnt_count) >=
522                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
523                 printk ("EXT2-fs warning: maximal mount count reached, "
524                         "running e2fsck is recommended\n");
525         else if (le32_to_cpu(es->s_checkinterval) &&
526                 (le32_to_cpu(es->s_lastcheck) + le32_to_cpu(es->s_checkinterval) <= get_seconds()))
527                 printk ("EXT2-fs warning: checktime reached, "
528                         "running e2fsck is recommended\n");
529         if (!le16_to_cpu(es->s_max_mnt_count))
530                 es->s_max_mnt_count = cpu_to_le16(EXT2_DFL_MAX_MNT_COUNT);
531         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
532         ext2_write_super(sb);
533         if (test_opt (sb, DEBUG))
534                 printk ("[EXT II FS %s, %s, bs=%lu, fs=%lu, gc=%lu, "
535                         "bpg=%lu, ipg=%lu, mo=%04lx]\n",
536                         EXT2FS_VERSION, EXT2FS_DATE, sb->s_blocksize,
537                         sbi->s_frag_size,
538                         sbi->s_groups_count,
539                         EXT2_BLOCKS_PER_GROUP(sb),
540                         EXT2_INODES_PER_GROUP(sb),
541                         sbi->s_mount_opt);
542         return res;
543 }
544
545 static int ext2_check_descriptors (struct super_block * sb)
546 {
547         int i;
548         int desc_block = 0;
549         struct ext2_sb_info *sbi = EXT2_SB(sb);
550         unsigned long first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
551         unsigned long last_block;
552         struct ext2_group_desc * gdp = NULL;
553
554         ext2_debug ("Checking group descriptors");
555
556         for (i = 0; i < sbi->s_groups_count; i++)
557         {
558                 if (i == sbi->s_groups_count - 1)
559                         last_block = le32_to_cpu(sbi->s_es->s_blocks_count) - 1;
560                 else
561                         last_block = first_block +
562                                 (EXT2_BLOCKS_PER_GROUP(sb) - 1);
563
564                 if ((i % EXT2_DESC_PER_BLOCK(sb)) == 0)
565                         gdp = (struct ext2_group_desc *) sbi->s_group_desc[desc_block++]->b_data;
566                 if (le32_to_cpu(gdp->bg_block_bitmap) < first_block ||
567                     le32_to_cpu(gdp->bg_block_bitmap) > last_block)
568                 {
569                         ext2_error (sb, "ext2_check_descriptors",
570                                     "Block bitmap for group %d"
571                                     " not in group (block %lu)!",
572                                     i, (unsigned long) le32_to_cpu(gdp->bg_block_bitmap));
573                         return 0;
574                 }
575                 if (le32_to_cpu(gdp->bg_inode_bitmap) < first_block ||
576                     le32_to_cpu(gdp->bg_inode_bitmap) > last_block)
577                 {
578                         ext2_error (sb, "ext2_check_descriptors",
579                                     "Inode bitmap for group %d"
580                                     " not in group (block %lu)!",
581                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_bitmap));
582                         return 0;
583                 }
584                 if (le32_to_cpu(gdp->bg_inode_table) < first_block ||
585                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >
586                     last_block)
587                 {
588                         ext2_error (sb, "ext2_check_descriptors",
589                                     "Inode table for group %d"
590                                     " not in group (block %lu)!",
591                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_table));
592                         return 0;
593                 }
594                 first_block += EXT2_BLOCKS_PER_GROUP(sb);
595                 gdp++;
596         }
597         return 1;
598 }
599
600 #define log2(n) ffz(~(n))
601  
602 /*
603  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
604  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
605  * We need to be 1 filesystem block less than the 2^32 sector limit.
606  */
607 static loff_t ext2_max_size(int bits)
608 {
609         loff_t res = EXT2_NDIR_BLOCKS;
610         /* This constant is calculated to be the largest file size for a
611          * dense, 4k-blocksize file such that the total number of
612          * sectors in the file, including data and all indirect blocks,
613          * does not exceed 2^32. */
614         const loff_t upper_limit = 0x1ff7fffd000LL;
615
616         res += 1LL << (bits-2);
617         res += 1LL << (2*(bits-2));
618         res += 1LL << (3*(bits-2));
619         res <<= bits;
620         if (res > upper_limit)
621                 res = upper_limit;
622         return res;
623 }
624
625 static unsigned long descriptor_loc(struct super_block *sb,
626                                     unsigned long logic_sb_block,
627                                     int nr)
628 {
629         struct ext2_sb_info *sbi = EXT2_SB(sb);
630         unsigned long bg, first_data_block, first_meta_bg;
631         int has_super = 0;
632         
633         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
634         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
635
636         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_META_BG) ||
637             nr < first_meta_bg)
638                 return (logic_sb_block + nr + 1);
639         bg = sbi->s_desc_per_block * nr;
640         if (ext2_bg_has_super(sb, bg))
641                 has_super = 1;
642         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
643 }
644
645 static int ext2_fill_super(struct super_block *sb, void *data, int silent)
646 {
647         struct buffer_head * bh;
648         struct ext2_sb_info * sbi;
649         struct ext2_super_block * es;
650         struct inode *root;
651         unsigned long block;
652         unsigned long sb_block = get_sb_block(&data);
653         unsigned long logic_sb_block;
654         unsigned long offset = 0;
655         unsigned long def_mount_opts;
656         int blocksize = BLOCK_SIZE;
657         int db_count;
658         int i, j;
659         __le32 features;
660
661         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
662         if (!sbi)
663                 return -ENOMEM;
664         sb->s_fs_info = sbi;
665
666         /*
667          * See what the current blocksize for the device is, and
668          * use that as the blocksize.  Otherwise (or if the blocksize
669          * is smaller than the default) use the default.
670          * This is important for devices that have a hardware
671          * sectorsize that is larger than the default.
672          */
673         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
674         if (!blocksize) {
675                 printk ("EXT2-fs: unable to set blocksize\n");
676                 goto failed_sbi;
677         }
678
679         /*
680          * If the superblock doesn't start on a hardware sector boundary,
681          * calculate the offset.  
682          */
683         if (blocksize != BLOCK_SIZE) {
684                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
685                 offset = (sb_block*BLOCK_SIZE) % blocksize;
686         } else {
687                 logic_sb_block = sb_block;
688         }
689
690         if (!(bh = sb_bread(sb, logic_sb_block))) {
691                 printk ("EXT2-fs: unable to read superblock\n");
692                 goto failed_sbi;
693         }
694         /*
695          * Note: s_es must be initialized as soon as possible because
696          *       some ext2 macro-instructions depend on its value
697          */
698         es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
699         sbi->s_es = es;
700         sb->s_magic = le16_to_cpu(es->s_magic);
701
702         if (sb->s_magic != EXT2_SUPER_MAGIC)
703                 goto cantfind_ext2;
704
705         /* Set defaults before we parse the mount options */
706         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
707         if (def_mount_opts & EXT2_DEFM_DEBUG)
708                 set_opt(sbi->s_mount_opt, DEBUG);
709         if (def_mount_opts & EXT2_DEFM_BSDGROUPS)
710                 set_opt(sbi->s_mount_opt, GRPID);
711         if (def_mount_opts & EXT2_DEFM_UID16)
712                 set_opt(sbi->s_mount_opt, NO_UID32);
713         if (def_mount_opts & EXT2_DEFM_XATTR_USER)
714                 set_opt(sbi->s_mount_opt, XATTR_USER);
715         if (def_mount_opts & EXT2_DEFM_ACL)
716                 set_opt(sbi->s_mount_opt, POSIX_ACL);
717         
718         if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_PANIC)
719                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
720         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_RO)
721                 set_opt(sbi->s_mount_opt, ERRORS_RO);
722         else
723                 set_opt(sbi->s_mount_opt, ERRORS_CONT);
724
725         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
726         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
727         
728         if (!parse_options ((char *) data, sbi))
729                 goto failed_mount;
730
731         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
732                 ((EXT2_SB(sb)->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ?
733                  MS_POSIXACL : 0);
734
735         ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
736                                     EXT2_MOUNT_XIP if not */
737
738         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV &&
739             (EXT2_HAS_COMPAT_FEATURE(sb, ~0U) ||
740              EXT2_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
741              EXT2_HAS_INCOMPAT_FEATURE(sb, ~0U)))
742                 printk("EXT2-fs warning: feature flags set on rev 0 fs, "
743                        "running e2fsck is recommended\n");
744         /*
745          * Check feature flags regardless of the revision level, since we
746          * previously didn't change the revision level when setting the flags,
747          * so there is a chance incompat flags are set on a rev 0 filesystem.
748          */
749         features = EXT2_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP);
750         if (features) {
751                 printk("EXT2-fs: %s: couldn't mount because of "
752                        "unsupported optional features (%x).\n",
753                        sb->s_id, le32_to_cpu(features));
754                 goto failed_mount;
755         }
756         if (!(sb->s_flags & MS_RDONLY) &&
757             (features = EXT2_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))){
758                 printk("EXT2-fs: %s: couldn't mount RDWR because of "
759                        "unsupported optional features (%x).\n",
760                        sb->s_id, le32_to_cpu(features));
761                 goto failed_mount;
762         }
763
764         blocksize = BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
765
766         if ((ext2_use_xip(sb)) && ((blocksize != PAGE_SIZE) ||
767                                   (sb->s_blocksize != blocksize))) {
768                 if (!silent)
769                         printk("XIP: Unsupported blocksize\n");
770                 goto failed_mount;
771         }
772
773         /* If the blocksize doesn't match, re-read the thing.. */
774         if (sb->s_blocksize != blocksize) {
775                 brelse(bh);
776
777                 if (!sb_set_blocksize(sb, blocksize)) {
778                         printk(KERN_ERR "EXT2-fs: blocksize too small for device.\n");
779                         goto failed_sbi;
780                 }
781
782                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
783                 offset = (sb_block*BLOCK_SIZE) % blocksize;
784                 bh = sb_bread(sb, logic_sb_block);
785                 if(!bh) {
786                         printk("EXT2-fs: Couldn't read superblock on "
787                                "2nd try.\n");
788                         goto failed_sbi;
789                 }
790                 es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
791                 sbi->s_es = es;
792                 if (es->s_magic != cpu_to_le16(EXT2_SUPER_MAGIC)) {
793                         printk ("EXT2-fs: Magic mismatch, very weird !\n");
794                         goto failed_mount;
795                 }
796         }
797
798         sb->s_maxbytes = ext2_max_size(sb->s_blocksize_bits);
799
800         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV) {
801                 sbi->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
802                 sbi->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
803         } else {
804                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
805                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
806                 if ((sbi->s_inode_size < EXT2_GOOD_OLD_INODE_SIZE) ||
807                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
808                     (sbi->s_inode_size > blocksize)) {
809                         printk ("EXT2-fs: unsupported inode size: %d\n",
810                                 sbi->s_inode_size);
811                         goto failed_mount;
812                 }
813         }
814
815         sbi->s_frag_size = EXT2_MIN_FRAG_SIZE <<
816                                    le32_to_cpu(es->s_log_frag_size);
817         if (sbi->s_frag_size == 0)
818                 goto cantfind_ext2;
819         sbi->s_frags_per_block = sb->s_blocksize / sbi->s_frag_size;
820
821         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
822         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
823         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
824
825         if (EXT2_INODE_SIZE(sb) == 0)
826                 goto cantfind_ext2;
827         sbi->s_inodes_per_block = sb->s_blocksize / EXT2_INODE_SIZE(sb);
828         if (sbi->s_inodes_per_block == 0 || sbi->s_inodes_per_group == 0)
829                 goto cantfind_ext2;
830         sbi->s_itb_per_group = sbi->s_inodes_per_group /
831                                         sbi->s_inodes_per_block;
832         sbi->s_desc_per_block = sb->s_blocksize /
833                                         sizeof (struct ext2_group_desc);
834         sbi->s_sbh = bh;
835         sbi->s_mount_state = le16_to_cpu(es->s_state);
836         sbi->s_addr_per_block_bits =
837                 log2 (EXT2_ADDR_PER_BLOCK(sb));
838         sbi->s_desc_per_block_bits =
839                 log2 (EXT2_DESC_PER_BLOCK(sb));
840
841         if (sb->s_magic != EXT2_SUPER_MAGIC)
842                 goto cantfind_ext2;
843
844         if (sb->s_blocksize != bh->b_size) {
845                 if (!silent)
846                         printk ("VFS: Unsupported blocksize on dev "
847                                 "%s.\n", sb->s_id);
848                 goto failed_mount;
849         }
850
851         if (sb->s_blocksize != sbi->s_frag_size) {
852                 printk ("EXT2-fs: fragsize %lu != blocksize %lu (not supported yet)\n",
853                         sbi->s_frag_size, sb->s_blocksize);
854                 goto failed_mount;
855         }
856
857         if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
858                 printk ("EXT2-fs: #blocks per group too big: %lu\n",
859                         sbi->s_blocks_per_group);
860                 goto failed_mount;
861         }
862         if (sbi->s_frags_per_group > sb->s_blocksize * 8) {
863                 printk ("EXT2-fs: #fragments per group too big: %lu\n",
864                         sbi->s_frags_per_group);
865                 goto failed_mount;
866         }
867         if (sbi->s_inodes_per_group > sb->s_blocksize * 8) {
868                 printk ("EXT2-fs: #inodes per group too big: %lu\n",
869                         sbi->s_inodes_per_group);
870                 goto failed_mount;
871         }
872
873         if (EXT2_BLOCKS_PER_GROUP(sb) == 0)
874                 goto cantfind_ext2;
875         sbi->s_groups_count = ((le32_to_cpu(es->s_blocks_count) -
876                                 le32_to_cpu(es->s_first_data_block) - 1)
877                                         / EXT2_BLOCKS_PER_GROUP(sb)) + 1;
878         db_count = (sbi->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) /
879                    EXT2_DESC_PER_BLOCK(sb);
880         sbi->s_group_desc = kmalloc (db_count * sizeof (struct buffer_head *), GFP_KERNEL);
881         if (sbi->s_group_desc == NULL) {
882                 printk ("EXT2-fs: not enough memory\n");
883                 goto failed_mount;
884         }
885         bgl_lock_init(&sbi->s_blockgroup_lock);
886         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(*sbi->s_debts),
887                                GFP_KERNEL);
888         if (!sbi->s_debts) {
889                 printk ("EXT2-fs: not enough memory\n");
890                 goto failed_mount_group_desc;
891         }
892         memset(sbi->s_debts, 0, sbi->s_groups_count * sizeof(*sbi->s_debts));
893         for (i = 0; i < db_count; i++) {
894                 block = descriptor_loc(sb, logic_sb_block, i);
895                 sbi->s_group_desc[i] = sb_bread(sb, block);
896                 if (!sbi->s_group_desc[i]) {
897                         for (j = 0; j < i; j++)
898                                 brelse (sbi->s_group_desc[j]);
899                         printk ("EXT2-fs: unable to read group descriptors\n");
900                         goto failed_mount_group_desc;
901                 }
902         }
903         if (!ext2_check_descriptors (sb)) {
904                 printk ("EXT2-fs: group descriptors corrupted!\n");
905                 goto failed_mount2;
906         }
907         sbi->s_gdb_count = db_count;
908         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
909         spin_lock_init(&sbi->s_next_gen_lock);
910
911         percpu_counter_init(&sbi->s_freeblocks_counter,
912                                 ext2_count_free_blocks(sb));
913         percpu_counter_init(&sbi->s_freeinodes_counter,
914                                 ext2_count_free_inodes(sb));
915         percpu_counter_init(&sbi->s_dirs_counter,
916                                 ext2_count_dirs(sb));
917         /*
918          * set up enough so that it can read an inode
919          */
920         sb->s_op = &ext2_sops;
921         sb->s_export_op = &ext2_export_ops;
922         sb->s_xattr = ext2_xattr_handlers;
923         root = iget(sb, EXT2_ROOT_INO);
924         sb->s_root = d_alloc_root(root);
925         if (!sb->s_root) {
926                 iput(root);
927                 printk(KERN_ERR "EXT2-fs: get root inode failed\n");
928                 goto failed_mount3;
929         }
930         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
931                 dput(sb->s_root);
932                 sb->s_root = NULL;
933                 printk(KERN_ERR "EXT2-fs: corrupt root inode, run e2fsck\n");
934                 goto failed_mount3;
935         }
936         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL))
937                 ext2_warning(sb, __FUNCTION__,
938                         "mounting ext3 filesystem as ext2");
939         ext2_setup_super (sb, es, sb->s_flags & MS_RDONLY);
940         return 0;
941
942 cantfind_ext2:
943         if (!silent)
944                 printk("VFS: Can't find an ext2 filesystem on dev %s.\n",
945                        sb->s_id);
946         goto failed_mount;
947 failed_mount3:
948         percpu_counter_destroy(&sbi->s_freeblocks_counter);
949         percpu_counter_destroy(&sbi->s_freeinodes_counter);
950         percpu_counter_destroy(&sbi->s_dirs_counter);
951 failed_mount2:
952         for (i = 0; i < db_count; i++)
953                 brelse(sbi->s_group_desc[i]);
954 failed_mount_group_desc:
955         kfree(sbi->s_group_desc);
956         kfree(sbi->s_debts);
957 failed_mount:
958         brelse(bh);
959 failed_sbi:
960         sb->s_fs_info = NULL;
961         kfree(sbi);
962         return -EINVAL;
963 }
964
965 static void ext2_commit_super (struct super_block * sb,
966                                struct ext2_super_block * es)
967 {
968         es->s_wtime = cpu_to_le32(get_seconds());
969         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
970         sb->s_dirt = 0;
971 }
972
973 static void ext2_sync_super(struct super_block *sb, struct ext2_super_block *es)
974 {
975         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
976         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
977         es->s_wtime = cpu_to_le32(get_seconds());
978         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
979         sync_dirty_buffer(EXT2_SB(sb)->s_sbh);
980         sb->s_dirt = 0;
981 }
982
983 /*
984  * In the second extended file system, it is not necessary to
985  * write the super block since we use a mapping of the
986  * disk super block in a buffer.
987  *
988  * However, this function is still used to set the fs valid
989  * flags to 0.  We need to set this flag to 0 since the fs
990  * may have been checked while mounted and e2fsck may have
991  * set s_state to EXT2_VALID_FS after some corrections.
992  */
993
994 void ext2_write_super (struct super_block * sb)
995 {
996         struct ext2_super_block * es;
997         lock_kernel();
998         if (!(sb->s_flags & MS_RDONLY)) {
999                 es = EXT2_SB(sb)->s_es;
1000
1001                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS) {
1002                         ext2_debug ("setting valid to 0\n");
1003                         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) &
1004                                                   ~EXT2_VALID_FS);
1005                         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
1006                         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
1007                         es->s_mtime = cpu_to_le32(get_seconds());
1008                         ext2_sync_super(sb, es);
1009                 } else
1010                         ext2_commit_super (sb, es);
1011         }
1012         sb->s_dirt = 0;
1013         unlock_kernel();
1014 }
1015
1016 static int ext2_remount (struct super_block * sb, int * flags, char * data)
1017 {
1018         struct ext2_sb_info * sbi = EXT2_SB(sb);
1019         struct ext2_super_block * es;
1020         unsigned long old_mount_opt = sbi->s_mount_opt;
1021         struct ext2_mount_options old_opts;
1022         unsigned long old_sb_flags;
1023         int err;
1024
1025         /* Store the old options */
1026         old_sb_flags = sb->s_flags;
1027         old_opts.s_mount_opt = sbi->s_mount_opt;
1028         old_opts.s_resuid = sbi->s_resuid;
1029         old_opts.s_resgid = sbi->s_resgid;
1030
1031         /*
1032          * Allow the "check" option to be passed as a remount option.
1033          */
1034         if (!parse_options (data, sbi)) {
1035                 err = -EINVAL;
1036                 goto restore_opts;
1037         }
1038
1039         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1040                 ((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1041
1042         es = sbi->s_es;
1043         if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
1044             (old_mount_opt & EXT2_MOUNT_XIP)) &&
1045             invalidate_inodes(sb))
1046                 ext2_warning(sb, __FUNCTION__, "busy inodes while remounting "\
1047                              "xip remain in cache (no functional problem)");
1048         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1049                 return 0;
1050         if (*flags & MS_RDONLY) {
1051                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS ||
1052                     !(sbi->s_mount_state & EXT2_VALID_FS))
1053                         return 0;
1054                 /*
1055                  * OK, we are remounting a valid rw partition rdonly, so set
1056                  * the rdonly flag and then mark the partition as valid again.
1057                  */
1058                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1059                 es->s_mtime = cpu_to_le32(get_seconds());
1060         } else {
1061                 __le32 ret = EXT2_HAS_RO_COMPAT_FEATURE(sb,
1062                                                ~EXT2_FEATURE_RO_COMPAT_SUPP);
1063                 if (ret) {
1064                         printk("EXT2-fs: %s: couldn't remount RDWR because of "
1065                                "unsupported optional features (%x).\n",
1066                                sb->s_id, le32_to_cpu(ret));
1067                         err = -EROFS;
1068                         goto restore_opts;
1069                 }
1070                 /*
1071                  * Mounting a RDONLY partition read-write, so reread and
1072                  * store the current valid flag.  (It may have been changed
1073                  * by e2fsck since we originally mounted the partition.)
1074                  */
1075                 sbi->s_mount_state = le16_to_cpu(es->s_state);
1076                 if (!ext2_setup_super (sb, es, 0))
1077                         sb->s_flags &= ~MS_RDONLY;
1078         }
1079         ext2_sync_super(sb, es);
1080         return 0;
1081 restore_opts:
1082         sbi->s_mount_opt = old_opts.s_mount_opt;
1083         sbi->s_resuid = old_opts.s_resuid;
1084         sbi->s_resgid = old_opts.s_resgid;
1085         sb->s_flags = old_sb_flags;
1086         return err;
1087 }
1088
1089 static int ext2_statfs (struct dentry * dentry, struct kstatfs * buf)
1090 {
1091         struct super_block *sb = dentry->d_sb;
1092         struct ext2_sb_info *sbi = EXT2_SB(sb);
1093         struct ext2_super_block *es = sbi->s_es;
1094         unsigned long overhead;
1095         int i;
1096         u64 fsid;
1097
1098         if (test_opt (sb, MINIX_DF))
1099                 overhead = 0;
1100         else {
1101                 /*
1102                  * Compute the overhead (FS structures)
1103                  */
1104
1105                 /*
1106                  * All of the blocks before first_data_block are
1107                  * overhead
1108                  */
1109                 overhead = le32_to_cpu(es->s_first_data_block);
1110
1111                 /*
1112                  * Add the overhead attributed to the superblock and
1113                  * block group descriptors.  If the sparse superblocks
1114                  * feature is turned on, then not all groups have this.
1115                  */
1116                 for (i = 0; i < sbi->s_groups_count; i++)
1117                         overhead += ext2_bg_has_super(sb, i) +
1118                                 ext2_bg_num_gdb(sb, i);
1119
1120                 /*
1121                  * Every block group has an inode bitmap, a block
1122                  * bitmap, and an inode table.
1123                  */
1124                 overhead += (sbi->s_groups_count *
1125                              (2 + sbi->s_itb_per_group));
1126         }
1127
1128         buf->f_type = EXT2_SUPER_MAGIC;
1129         buf->f_bsize = sb->s_blocksize;
1130         buf->f_blocks = le32_to_cpu(es->s_blocks_count) - overhead;
1131         buf->f_bfree = ext2_count_free_blocks(sb);
1132         buf->f_bavail = buf->f_bfree - le32_to_cpu(es->s_r_blocks_count);
1133         if (buf->f_bfree < le32_to_cpu(es->s_r_blocks_count))
1134                 buf->f_bavail = 0;
1135         buf->f_files = le32_to_cpu(es->s_inodes_count);
1136         buf->f_ffree = ext2_count_free_inodes(sb);
1137         buf->f_namelen = EXT2_NAME_LEN;
1138         fsid = le64_to_cpup((void *)es->s_uuid) ^
1139                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
1140         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1141         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1142         return 0;
1143 }
1144
1145 static int ext2_get_sb(struct file_system_type *fs_type,
1146         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1147 {
1148         return get_sb_bdev(fs_type, flags, dev_name, data, ext2_fill_super, mnt);
1149 }
1150
1151 #ifdef CONFIG_QUOTA
1152
1153 /* Read data from quotafile - avoid pagecache and such because we cannot afford
1154  * acquiring the locks... As quota files are never truncated and quota code
1155  * itself serializes the operations (and noone else should touch the files)
1156  * we don't have to be afraid of races */
1157 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data,
1158                                size_t len, loff_t off)
1159 {
1160         struct inode *inode = sb_dqopt(sb)->files[type];
1161         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1162         int err = 0;
1163         int offset = off & (sb->s_blocksize - 1);
1164         int tocopy;
1165         size_t toread;
1166         struct buffer_head tmp_bh;
1167         struct buffer_head *bh;
1168         loff_t i_size = i_size_read(inode);
1169
1170         if (off > i_size)
1171                 return 0;
1172         if (off+len > i_size)
1173                 len = i_size-off;
1174         toread = len;
1175         while (toread > 0) {
1176                 tocopy = sb->s_blocksize - offset < toread ?
1177                                 sb->s_blocksize - offset : toread;
1178
1179                 tmp_bh.b_state = 0;
1180                 err = ext2_get_block(inode, blk, &tmp_bh, 0);
1181                 if (err)
1182                         return err;
1183                 if (!buffer_mapped(&tmp_bh))    /* A hole? */
1184                         memset(data, 0, tocopy);
1185                 else {
1186                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1187                         if (!bh)
1188                                 return -EIO;
1189                         memcpy(data, bh->b_data+offset, tocopy);
1190                         brelse(bh);
1191                 }
1192                 offset = 0;
1193                 toread -= tocopy;
1194                 data += tocopy;
1195                 blk++;
1196         }
1197         return len;
1198 }
1199
1200 /* Write to quotafile */
1201 static ssize_t ext2_quota_write(struct super_block *sb, int type,
1202                                 const char *data, size_t len, loff_t off)
1203 {
1204         struct inode *inode = sb_dqopt(sb)->files[type];
1205         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1206         int err = 0;
1207         int offset = off & (sb->s_blocksize - 1);
1208         int tocopy;
1209         size_t towrite = len;
1210         struct buffer_head tmp_bh;
1211         struct buffer_head *bh;
1212
1213         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
1214         while (towrite > 0) {
1215                 tocopy = sb->s_blocksize - offset < towrite ?
1216                                 sb->s_blocksize - offset : towrite;
1217
1218                 tmp_bh.b_state = 0;
1219                 err = ext2_get_block(inode, blk, &tmp_bh, 1);
1220                 if (err)
1221                         goto out;
1222                 if (offset || tocopy != EXT2_BLOCK_SIZE(sb))
1223                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1224                 else
1225                         bh = sb_getblk(sb, tmp_bh.b_blocknr);
1226                 if (!bh) {
1227                         err = -EIO;
1228                         goto out;
1229                 }
1230                 lock_buffer(bh);
1231                 memcpy(bh->b_data+offset, data, tocopy);
1232                 flush_dcache_page(bh->b_page);
1233                 set_buffer_uptodate(bh);
1234                 mark_buffer_dirty(bh);
1235                 unlock_buffer(bh);
1236                 brelse(bh);
1237                 offset = 0;
1238                 towrite -= tocopy;
1239                 data += tocopy;
1240                 blk++;
1241         }
1242 out:
1243         if (len == towrite)
1244                 return err;
1245         if (inode->i_size < off+len-towrite)
1246                 i_size_write(inode, off+len-towrite);
1247         inode->i_version++;
1248         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1249         mark_inode_dirty(inode);
1250         mutex_unlock(&inode->i_mutex);
1251         return len - towrite;
1252 }
1253
1254 #endif
1255
1256 static struct file_system_type ext2_fs_type = {
1257         .owner          = THIS_MODULE,
1258         .name           = "ext2",
1259         .get_sb         = ext2_get_sb,
1260         .kill_sb        = kill_block_super,
1261         .fs_flags       = FS_REQUIRES_DEV,
1262 };
1263
1264 static int __init init_ext2_fs(void)
1265 {
1266         int err = init_ext2_xattr();
1267         if (err)
1268                 return err;
1269         err = init_inodecache();
1270         if (err)
1271                 goto out1;
1272         err = register_filesystem(&ext2_fs_type);
1273         if (err)
1274                 goto out;
1275         return 0;
1276 out:
1277         destroy_inodecache();
1278 out1:
1279         exit_ext2_xattr();
1280         return err;
1281 }
1282
1283 static void __exit exit_ext2_fs(void)
1284 {
1285         unregister_filesystem(&ext2_fs_type);
1286         destroy_inodecache();
1287         exit_ext2_xattr();
1288 }
1289
1290 module_init(init_ext2_fs)
1291 module_exit(exit_ext2_fs)