Merge master.kernel.org:/home/rmk/linux-2.6-serial
[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/config.h>
20 #include <linux/module.h>
21 #include <linux/string.h>
22 #include <linux/fs.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/blkdev.h>
26 #include <linux/parser.h>
27 #include <linux/random.h>
28 #include <linux/buffer_head.h>
29 #include <linux/smp_lock.h>
30 #include <linux/vfs.h>
31 #include <linux/seq_file.h>
32 #include <linux/mount.h>
33 #include <asm/uaccess.h>
34 #include "ext2.h"
35 #include "xattr.h"
36 #include "acl.h"
37 #include "xip.h"
38
39 static void ext2_sync_super(struct super_block *sb,
40                             struct ext2_super_block *es);
41 static int ext2_remount (struct super_block * sb, int * flags, char * data);
42 static int ext2_statfs (struct super_block * sb, struct kstatfs * buf);
43
44 void ext2_error (struct super_block * sb, const char * function,
45                  const char * fmt, ...)
46 {
47         va_list args;
48         struct ext2_sb_info *sbi = EXT2_SB(sb);
49         struct ext2_super_block *es = sbi->s_es;
50
51         if (!(sb->s_flags & MS_RDONLY)) {
52                 sbi->s_mount_state |= EXT2_ERROR_FS;
53                 es->s_state =
54                         cpu_to_le16(le16_to_cpu(es->s_state) | EXT2_ERROR_FS);
55                 ext2_sync_super(sb, es);
56         }
57
58         va_start(args, fmt);
59         printk(KERN_CRIT "EXT2-fs error (device %s): %s: ",sb->s_id, function);
60         vprintk(fmt, args);
61         printk("\n");
62         va_end(args);
63
64         if (test_opt(sb, ERRORS_PANIC))
65                 panic("EXT2-fs panic from previous error\n");
66         if (test_opt(sb, ERRORS_RO)) {
67                 printk("Remounting filesystem read-only\n");
68                 sb->s_flags |= MS_RDONLY;
69         }
70 }
71
72 void ext2_warning (struct super_block * sb, const char * function,
73                    const char * fmt, ...)
74 {
75         va_list args;
76
77         va_start(args, fmt);
78         printk(KERN_WARNING "EXT2-fs warning (device %s): %s: ",
79                sb->s_id, function);
80         vprintk(fmt, args);
81         printk("\n");
82         va_end(args);
83 }
84
85 void ext2_update_dynamic_rev(struct super_block *sb)
86 {
87         struct ext2_super_block *es = EXT2_SB(sb)->s_es;
88
89         if (le32_to_cpu(es->s_rev_level) > EXT2_GOOD_OLD_REV)
90                 return;
91
92         ext2_warning(sb, __FUNCTION__,
93                      "updating to rev %d because of new feature flag, "
94                      "running e2fsck is recommended",
95                      EXT2_DYNAMIC_REV);
96
97         es->s_first_ino = cpu_to_le32(EXT2_GOOD_OLD_FIRST_INO);
98         es->s_inode_size = cpu_to_le16(EXT2_GOOD_OLD_INODE_SIZE);
99         es->s_rev_level = cpu_to_le32(EXT2_DYNAMIC_REV);
100         /* leave es->s_feature_*compat flags alone */
101         /* es->s_uuid will be set by e2fsck if empty */
102
103         /*
104          * The rest of the superblock fields should be zero, and if not it
105          * means they are likely already in use, so leave them alone.  We
106          * can leave it up to e2fsck to clean up any inconsistencies there.
107          */
108 }
109
110 static void ext2_put_super (struct super_block * sb)
111 {
112         int db_count;
113         int i;
114         struct ext2_sb_info *sbi = EXT2_SB(sb);
115
116         ext2_xattr_put_super(sb);
117         if (!(sb->s_flags & MS_RDONLY)) {
118                 struct ext2_super_block *es = sbi->s_es;
119
120                 es->s_state = cpu_to_le16(sbi->s_mount_state);
121                 ext2_sync_super(sb, es);
122         }
123         db_count = sbi->s_gdb_count;
124         for (i = 0; i < db_count; i++)
125                 if (sbi->s_group_desc[i])
126                         brelse (sbi->s_group_desc[i]);
127         kfree(sbi->s_group_desc);
128         kfree(sbi->s_debts);
129         percpu_counter_destroy(&sbi->s_freeblocks_counter);
130         percpu_counter_destroy(&sbi->s_freeinodes_counter);
131         percpu_counter_destroy(&sbi->s_dirs_counter);
132         brelse (sbi->s_sbh);
133         sb->s_fs_info = NULL;
134         kfree(sbi);
135
136         return;
137 }
138
139 static kmem_cache_t * ext2_inode_cachep;
140
141 static struct inode *ext2_alloc_inode(struct super_block *sb)
142 {
143         struct ext2_inode_info *ei;
144         ei = (struct ext2_inode_info *)kmem_cache_alloc(ext2_inode_cachep, SLAB_KERNEL);
145         if (!ei)
146                 return NULL;
147 #ifdef CONFIG_EXT2_FS_POSIX_ACL
148         ei->i_acl = EXT2_ACL_NOT_CACHED;
149         ei->i_default_acl = EXT2_ACL_NOT_CACHED;
150 #endif
151         ei->vfs_inode.i_version = 1;
152         return &ei->vfs_inode;
153 }
154
155 static void ext2_destroy_inode(struct inode *inode)
156 {
157         kmem_cache_free(ext2_inode_cachep, EXT2_I(inode));
158 }
159
160 static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
161 {
162         struct ext2_inode_info *ei = (struct ext2_inode_info *) foo;
163
164         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
165             SLAB_CTOR_CONSTRUCTOR) {
166                 rwlock_init(&ei->i_meta_lock);
167 #ifdef CONFIG_EXT2_FS_XATTR
168                 init_rwsem(&ei->xattr_sem);
169 #endif
170                 inode_init_once(&ei->vfs_inode);
171         }
172 }
173  
174 static int init_inodecache(void)
175 {
176         ext2_inode_cachep = kmem_cache_create("ext2_inode_cache",
177                                              sizeof(struct ext2_inode_info),
178                                              0, SLAB_RECLAIM_ACCOUNT,
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         if (kmem_cache_destroy(ext2_inode_cachep))
188                 printk(KERN_INFO "ext2_inode_cache: not all structures were freed\n");
189 }
190
191 static void ext2_clear_inode(struct inode *inode)
192 {
193 #ifdef CONFIG_EXT2_FS_POSIX_ACL
194         struct ext2_inode_info *ei = EXT2_I(inode);
195
196         if (ei->i_acl && ei->i_acl != EXT2_ACL_NOT_CACHED) {
197                 posix_acl_release(ei->i_acl);
198                 ei->i_acl = EXT2_ACL_NOT_CACHED;
199         }
200         if (ei->i_default_acl && ei->i_default_acl != EXT2_ACL_NOT_CACHED) {
201                 posix_acl_release(ei->i_default_acl);
202                 ei->i_default_acl = EXT2_ACL_NOT_CACHED;
203         }
204 #endif
205 }
206
207 static int ext2_show_options(struct seq_file *seq, struct vfsmount *vfs)
208 {
209         struct ext2_sb_info *sbi = EXT2_SB(vfs->mnt_sb);
210
211         if (sbi->s_mount_opt & EXT2_MOUNT_GRPID)
212                 seq_puts(seq, ",grpid");
213         else
214                 seq_puts(seq, ",nogrpid");
215
216 #if defined(CONFIG_QUOTA)
217         if (sbi->s_mount_opt & EXT2_MOUNT_USRQUOTA)
218                 seq_puts(seq, ",usrquota");
219
220         if (sbi->s_mount_opt & EXT2_MOUNT_GRPQUOTA)
221                 seq_puts(seq, ",grpquota");
222 #endif
223
224         return 0;
225 }
226
227 #ifdef CONFIG_QUOTA
228 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data, size_t len, loff_t off);
229 static ssize_t ext2_quota_write(struct super_block *sb, int type, const char *data, size_t len, loff_t off);
230 #endif
231
232 static struct super_operations ext2_sops = {
233         .alloc_inode    = ext2_alloc_inode,
234         .destroy_inode  = ext2_destroy_inode,
235         .read_inode     = ext2_read_inode,
236         .write_inode    = ext2_write_inode,
237         .put_inode      = ext2_put_inode,
238         .delete_inode   = ext2_delete_inode,
239         .put_super      = ext2_put_super,
240         .write_super    = ext2_write_super,
241         .statfs         = ext2_statfs,
242         .remount_fs     = ext2_remount,
243         .clear_inode    = ext2_clear_inode,
244         .show_options   = ext2_show_options,
245 #ifdef CONFIG_QUOTA
246         .quota_read     = ext2_quota_read,
247         .quota_write    = ext2_quota_write,
248 #endif
249 };
250
251 /* Yes, most of these are left as NULL!!
252  * A NULL value implies the default, which works with ext2-like file
253  * systems, but can be improved upon.
254  * Currently only get_parent is required.
255  */
256 struct dentry *ext2_get_parent(struct dentry *child);
257 static struct export_operations ext2_export_ops = {
258         .get_parent = ext2_get_parent,
259 };
260
261 static unsigned long get_sb_block(void **data)
262 {
263         unsigned long   sb_block;
264         char            *options = (char *) *data;
265
266         if (!options || strncmp(options, "sb=", 3) != 0)
267                 return 1;       /* Default location */
268         options += 3;
269         sb_block = simple_strtoul(options, &options, 0);
270         if (*options && *options != ',') {
271                 printk("EXT2-fs: Invalid sb specification: %s\n",
272                        (char *) *data);
273                 return 1;
274         }
275         if (*options == ',')
276                 options++;
277         *data = (void *) options;
278         return sb_block;
279 }
280
281 enum {
282         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
283         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic,
284         Opt_err_ro, Opt_nouid32, Opt_check, Opt_nocheck, Opt_debug,
285         Opt_oldalloc, Opt_orlov, Opt_nobh, Opt_user_xattr, Opt_nouser_xattr,
286         Opt_acl, Opt_noacl, Opt_xip, Opt_ignore, Opt_err, Opt_quota,
287         Opt_usrquota, Opt_grpquota
288 };
289
290 static match_table_t tokens = {
291         {Opt_bsd_df, "bsddf"},
292         {Opt_minix_df, "minixdf"},
293         {Opt_grpid, "grpid"},
294         {Opt_grpid, "bsdgroups"},
295         {Opt_nogrpid, "nogrpid"},
296         {Opt_nogrpid, "sysvgroups"},
297         {Opt_resgid, "resgid=%u"},
298         {Opt_resuid, "resuid=%u"},
299         {Opt_sb, "sb=%u"},
300         {Opt_err_cont, "errors=continue"},
301         {Opt_err_panic, "errors=panic"},
302         {Opt_err_ro, "errors=remount-ro"},
303         {Opt_nouid32, "nouid32"},
304         {Opt_nocheck, "check=none"},
305         {Opt_nocheck, "nocheck"},
306         {Opt_check, "check"},
307         {Opt_debug, "debug"},
308         {Opt_oldalloc, "oldalloc"},
309         {Opt_orlov, "orlov"},
310         {Opt_nobh, "nobh"},
311         {Opt_user_xattr, "user_xattr"},
312         {Opt_nouser_xattr, "nouser_xattr"},
313         {Opt_acl, "acl"},
314         {Opt_noacl, "noacl"},
315         {Opt_xip, "xip"},
316         {Opt_grpquota, "grpquota"},
317         {Opt_ignore, "noquota"},
318         {Opt_quota, "quota"},
319         {Opt_usrquota, "usrquota"},
320         {Opt_err, NULL}
321 };
322
323 static int parse_options (char * options,
324                           struct ext2_sb_info *sbi)
325 {
326         char * p;
327         substring_t args[MAX_OPT_ARGS];
328         unsigned long kind = EXT2_MOUNT_ERRORS_CONT;
329         int option;
330
331         if (!options)
332                 return 1;
333
334         while ((p = strsep (&options, ",")) != NULL) {
335                 int token;
336                 if (!*p)
337                         continue;
338
339                 token = match_token(p, tokens, args);
340                 switch (token) {
341                 case Opt_bsd_df:
342                         clear_opt (sbi->s_mount_opt, MINIX_DF);
343                         break;
344                 case Opt_minix_df:
345                         set_opt (sbi->s_mount_opt, MINIX_DF);
346                         break;
347                 case Opt_grpid:
348                         set_opt (sbi->s_mount_opt, GRPID);
349                         break;
350                 case Opt_nogrpid:
351                         clear_opt (sbi->s_mount_opt, GRPID);
352                         break;
353                 case Opt_resuid:
354                         if (match_int(&args[0], &option))
355                                 return 0;
356                         sbi->s_resuid = option;
357                         break;
358                 case Opt_resgid:
359                         if (match_int(&args[0], &option))
360                                 return 0;
361                         sbi->s_resgid = option;
362                         break;
363                 case Opt_sb:
364                         /* handled by get_sb_block() instead of here */
365                         /* *sb_block = match_int(&args[0]); */
366                         break;
367                 case Opt_err_panic:
368                         kind = EXT2_MOUNT_ERRORS_PANIC;
369                         break;
370                 case Opt_err_ro:
371                         kind = EXT2_MOUNT_ERRORS_RO;
372                         break;
373                 case Opt_err_cont:
374                         kind = EXT2_MOUNT_ERRORS_CONT;
375                         break;
376                 case Opt_nouid32:
377                         set_opt (sbi->s_mount_opt, NO_UID32);
378                         break;
379                 case Opt_check:
380 #ifdef CONFIG_EXT2_CHECK
381                         set_opt (sbi->s_mount_opt, CHECK);
382 #else
383                         printk("EXT2 Check option not supported\n");
384 #endif
385                         break;
386                 case Opt_nocheck:
387                         clear_opt (sbi->s_mount_opt, CHECK);
388                         break;
389                 case Opt_debug:
390                         set_opt (sbi->s_mount_opt, DEBUG);
391                         break;
392                 case Opt_oldalloc:
393                         set_opt (sbi->s_mount_opt, OLDALLOC);
394                         break;
395                 case Opt_orlov:
396                         clear_opt (sbi->s_mount_opt, OLDALLOC);
397                         break;
398                 case Opt_nobh:
399                         set_opt (sbi->s_mount_opt, NOBH);
400                         break;
401 #ifdef CONFIG_EXT2_FS_XATTR
402                 case Opt_user_xattr:
403                         set_opt (sbi->s_mount_opt, XATTR_USER);
404                         break;
405                 case Opt_nouser_xattr:
406                         clear_opt (sbi->s_mount_opt, XATTR_USER);
407                         break;
408 #else
409                 case Opt_user_xattr:
410                 case Opt_nouser_xattr:
411                         printk("EXT2 (no)user_xattr options not supported\n");
412                         break;
413 #endif
414 #ifdef CONFIG_EXT2_FS_POSIX_ACL
415                 case Opt_acl:
416                         set_opt(sbi->s_mount_opt, POSIX_ACL);
417                         break;
418                 case Opt_noacl:
419                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
420                         break;
421 #else
422                 case Opt_acl:
423                 case Opt_noacl:
424                         printk("EXT2 (no)acl options not supported\n");
425                         break;
426 #endif
427                 case Opt_xip:
428 #ifdef CONFIG_EXT2_FS_XIP
429                         set_opt (sbi->s_mount_opt, XIP);
430 #else
431                         printk("EXT2 xip option not supported\n");
432 #endif
433                         break;
434
435 #if defined(CONFIG_QUOTA)
436                 case Opt_quota:
437                 case Opt_usrquota:
438                         set_opt(sbi->s_mount_opt, USRQUOTA);
439                         break;
440
441                 case Opt_grpquota:
442                         set_opt(sbi->s_mount_opt, GRPQUOTA);
443                         break;
444 #else
445                 case Opt_quota:
446                 case Opt_usrquota:
447                 case Opt_grpquota:
448                         printk(KERN_ERR
449                                 "EXT2-fs: quota operations not supported.\n");
450
451                         break;
452 #endif
453
454                 case Opt_ignore:
455                         break;
456                 default:
457                         return 0;
458                 }
459         }
460         sbi->s_mount_opt |= kind;
461         return 1;
462 }
463
464 static int ext2_setup_super (struct super_block * sb,
465                               struct ext2_super_block * es,
466                               int read_only)
467 {
468         int res = 0;
469         struct ext2_sb_info *sbi = EXT2_SB(sb);
470
471         if (le32_to_cpu(es->s_rev_level) > EXT2_MAX_SUPP_REV) {
472                 printk ("EXT2-fs warning: revision level too high, "
473                         "forcing read-only mode\n");
474                 res = MS_RDONLY;
475         }
476         if (read_only)
477                 return res;
478         if (!(sbi->s_mount_state & EXT2_VALID_FS))
479                 printk ("EXT2-fs warning: mounting unchecked fs, "
480                         "running e2fsck is recommended\n");
481         else if ((sbi->s_mount_state & EXT2_ERROR_FS))
482                 printk ("EXT2-fs warning: mounting fs with errors, "
483                         "running e2fsck is recommended\n");
484         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
485                  le16_to_cpu(es->s_mnt_count) >=
486                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
487                 printk ("EXT2-fs warning: maximal mount count reached, "
488                         "running e2fsck is recommended\n");
489         else if (le32_to_cpu(es->s_checkinterval) &&
490                 (le32_to_cpu(es->s_lastcheck) + le32_to_cpu(es->s_checkinterval) <= get_seconds()))
491                 printk ("EXT2-fs warning: checktime reached, "
492                         "running e2fsck is recommended\n");
493         if (!le16_to_cpu(es->s_max_mnt_count))
494                 es->s_max_mnt_count = cpu_to_le16(EXT2_DFL_MAX_MNT_COUNT);
495         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
496         ext2_write_super(sb);
497         if (test_opt (sb, DEBUG))
498                 printk ("[EXT II FS %s, %s, bs=%lu, fs=%lu, gc=%lu, "
499                         "bpg=%lu, ipg=%lu, mo=%04lx]\n",
500                         EXT2FS_VERSION, EXT2FS_DATE, sb->s_blocksize,
501                         sbi->s_frag_size,
502                         sbi->s_groups_count,
503                         EXT2_BLOCKS_PER_GROUP(sb),
504                         EXT2_INODES_PER_GROUP(sb),
505                         sbi->s_mount_opt);
506 #ifdef CONFIG_EXT2_CHECK
507         if (test_opt (sb, CHECK)) {
508                 ext2_check_blocks_bitmap (sb);
509                 ext2_check_inodes_bitmap (sb);
510         }
511 #endif
512         return res;
513 }
514
515 static int ext2_check_descriptors (struct super_block * sb)
516 {
517         int i;
518         int desc_block = 0;
519         struct ext2_sb_info *sbi = EXT2_SB(sb);
520         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
521         struct ext2_group_desc * gdp = NULL;
522
523         ext2_debug ("Checking group descriptors");
524
525         for (i = 0; i < sbi->s_groups_count; i++)
526         {
527                 if ((i % EXT2_DESC_PER_BLOCK(sb)) == 0)
528                         gdp = (struct ext2_group_desc *) sbi->s_group_desc[desc_block++]->b_data;
529                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
530                     le32_to_cpu(gdp->bg_block_bitmap) >= block + EXT2_BLOCKS_PER_GROUP(sb))
531                 {
532                         ext2_error (sb, "ext2_check_descriptors",
533                                     "Block bitmap for group %d"
534                                     " not in group (block %lu)!",
535                                     i, (unsigned long) le32_to_cpu(gdp->bg_block_bitmap));
536                         return 0;
537                 }
538                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
539                     le32_to_cpu(gdp->bg_inode_bitmap) >= block + EXT2_BLOCKS_PER_GROUP(sb))
540                 {
541                         ext2_error (sb, "ext2_check_descriptors",
542                                     "Inode bitmap for group %d"
543                                     " not in group (block %lu)!",
544                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_bitmap));
545                         return 0;
546                 }
547                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
548                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
549                     block + EXT2_BLOCKS_PER_GROUP(sb))
550                 {
551                         ext2_error (sb, "ext2_check_descriptors",
552                                     "Inode table for group %d"
553                                     " not in group (block %lu)!",
554                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_table));
555                         return 0;
556                 }
557                 block += EXT2_BLOCKS_PER_GROUP(sb);
558                 gdp++;
559         }
560         return 1;
561 }
562
563 #define log2(n) ffz(~(n))
564  
565 /*
566  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
567  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
568  * We need to be 1 filesystem block less than the 2^32 sector limit.
569  */
570 static loff_t ext2_max_size(int bits)
571 {
572         loff_t res = EXT2_NDIR_BLOCKS;
573         /* This constant is calculated to be the largest file size for a
574          * dense, 4k-blocksize file such that the total number of
575          * sectors in the file, including data and all indirect blocks,
576          * does not exceed 2^32. */
577         const loff_t upper_limit = 0x1ff7fffd000LL;
578
579         res += 1LL << (bits-2);
580         res += 1LL << (2*(bits-2));
581         res += 1LL << (3*(bits-2));
582         res <<= bits;
583         if (res > upper_limit)
584                 res = upper_limit;
585         return res;
586 }
587
588 static unsigned long descriptor_loc(struct super_block *sb,
589                                     unsigned long logic_sb_block,
590                                     int nr)
591 {
592         struct ext2_sb_info *sbi = EXT2_SB(sb);
593         unsigned long bg, first_data_block, first_meta_bg;
594         int has_super = 0;
595         
596         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
597         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
598
599         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_META_BG) ||
600             nr < first_meta_bg)
601                 return (logic_sb_block + nr + 1);
602         bg = sbi->s_desc_per_block * nr;
603         if (ext2_bg_has_super(sb, bg))
604                 has_super = 1;
605         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
606 }
607
608 static int ext2_fill_super(struct super_block *sb, void *data, int silent)
609 {
610         struct buffer_head * bh;
611         struct ext2_sb_info * sbi;
612         struct ext2_super_block * es;
613         struct inode *root;
614         unsigned long block;
615         unsigned long sb_block = get_sb_block(&data);
616         unsigned long logic_sb_block;
617         unsigned long offset = 0;
618         unsigned long def_mount_opts;
619         int blocksize = BLOCK_SIZE;
620         int db_count;
621         int i, j;
622         __le32 features;
623
624         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
625         if (!sbi)
626                 return -ENOMEM;
627         sb->s_fs_info = sbi;
628         memset(sbi, 0, sizeof(*sbi));
629
630         /*
631          * See what the current blocksize for the device is, and
632          * use that as the blocksize.  Otherwise (or if the blocksize
633          * is smaller than the default) use the default.
634          * This is important for devices that have a hardware
635          * sectorsize that is larger than the default.
636          */
637         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
638         if (!blocksize) {
639                 printk ("EXT2-fs: unable to set blocksize\n");
640                 goto failed_sbi;
641         }
642
643         /*
644          * If the superblock doesn't start on a hardware sector boundary,
645          * calculate the offset.  
646          */
647         if (blocksize != BLOCK_SIZE) {
648                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
649                 offset = (sb_block*BLOCK_SIZE) % blocksize;
650         } else {
651                 logic_sb_block = sb_block;
652         }
653
654         if (!(bh = sb_bread(sb, logic_sb_block))) {
655                 printk ("EXT2-fs: unable to read superblock\n");
656                 goto failed_sbi;
657         }
658         /*
659          * Note: s_es must be initialized as soon as possible because
660          *       some ext2 macro-instructions depend on its value
661          */
662         es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
663         sbi->s_es = es;
664         sb->s_magic = le16_to_cpu(es->s_magic);
665
666         if (sb->s_magic != EXT2_SUPER_MAGIC)
667                 goto cantfind_ext2;
668
669         /* Set defaults before we parse the mount options */
670         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
671         if (def_mount_opts & EXT2_DEFM_DEBUG)
672                 set_opt(sbi->s_mount_opt, DEBUG);
673         if (def_mount_opts & EXT2_DEFM_BSDGROUPS)
674                 set_opt(sbi->s_mount_opt, GRPID);
675         if (def_mount_opts & EXT2_DEFM_UID16)
676                 set_opt(sbi->s_mount_opt, NO_UID32);
677         if (def_mount_opts & EXT2_DEFM_XATTR_USER)
678                 set_opt(sbi->s_mount_opt, XATTR_USER);
679         if (def_mount_opts & EXT2_DEFM_ACL)
680                 set_opt(sbi->s_mount_opt, POSIX_ACL);
681         
682         if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_PANIC)
683                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
684         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_RO)
685                 set_opt(sbi->s_mount_opt, ERRORS_RO);
686
687         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
688         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
689         
690         if (!parse_options ((char *) data, sbi))
691                 goto failed_mount;
692
693         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
694                 ((EXT2_SB(sb)->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ?
695                  MS_POSIXACL : 0);
696
697         ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
698                                     EXT2_MOUNT_XIP if not */
699
700         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV &&
701             (EXT2_HAS_COMPAT_FEATURE(sb, ~0U) ||
702              EXT2_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
703              EXT2_HAS_INCOMPAT_FEATURE(sb, ~0U)))
704                 printk("EXT2-fs warning: feature flags set on rev 0 fs, "
705                        "running e2fsck is recommended\n");
706         /*
707          * Check feature flags regardless of the revision level, since we
708          * previously didn't change the revision level when setting the flags,
709          * so there is a chance incompat flags are set on a rev 0 filesystem.
710          */
711         features = EXT2_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP);
712         if (features) {
713                 printk("EXT2-fs: %s: couldn't mount because of "
714                        "unsupported optional features (%x).\n",
715                        sb->s_id, le32_to_cpu(features));
716                 goto failed_mount;
717         }
718         if (!(sb->s_flags & MS_RDONLY) &&
719             (features = EXT2_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))){
720                 printk("EXT2-fs: %s: couldn't mount RDWR because of "
721                        "unsupported optional features (%x).\n",
722                        sb->s_id, le32_to_cpu(features));
723                 goto failed_mount;
724         }
725
726         blocksize = BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
727
728         if ((ext2_use_xip(sb)) && ((blocksize != PAGE_SIZE) ||
729                                   (sb->s_blocksize != blocksize))) {
730                 if (!silent)
731                         printk("XIP: Unsupported blocksize\n");
732                 goto failed_mount;
733         }
734
735         /* If the blocksize doesn't match, re-read the thing.. */
736         if (sb->s_blocksize != blocksize) {
737                 brelse(bh);
738
739                 if (!sb_set_blocksize(sb, blocksize)) {
740                         printk(KERN_ERR "EXT2-fs: blocksize too small for device.\n");
741                         goto failed_sbi;
742                 }
743
744                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
745                 offset = (sb_block*BLOCK_SIZE) % blocksize;
746                 bh = sb_bread(sb, logic_sb_block);
747                 if(!bh) {
748                         printk("EXT2-fs: Couldn't read superblock on "
749                                "2nd try.\n");
750                         goto failed_sbi;
751                 }
752                 es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
753                 sbi->s_es = es;
754                 if (es->s_magic != cpu_to_le16(EXT2_SUPER_MAGIC)) {
755                         printk ("EXT2-fs: Magic mismatch, very weird !\n");
756                         goto failed_mount;
757                 }
758         }
759
760         sb->s_maxbytes = ext2_max_size(sb->s_blocksize_bits);
761
762         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV) {
763                 sbi->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
764                 sbi->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
765         } else {
766                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
767                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
768                 if ((sbi->s_inode_size < EXT2_GOOD_OLD_INODE_SIZE) ||
769                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
770                     (sbi->s_inode_size > blocksize)) {
771                         printk ("EXT2-fs: unsupported inode size: %d\n",
772                                 sbi->s_inode_size);
773                         goto failed_mount;
774                 }
775         }
776
777         sbi->s_frag_size = EXT2_MIN_FRAG_SIZE <<
778                                    le32_to_cpu(es->s_log_frag_size);
779         if (sbi->s_frag_size == 0)
780                 goto cantfind_ext2;
781         sbi->s_frags_per_block = sb->s_blocksize / sbi->s_frag_size;
782
783         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
784         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
785         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
786
787         if (EXT2_INODE_SIZE(sb) == 0)
788                 goto cantfind_ext2;
789         sbi->s_inodes_per_block = sb->s_blocksize / EXT2_INODE_SIZE(sb);
790         if (sbi->s_inodes_per_block == 0)
791                 goto cantfind_ext2;
792         sbi->s_itb_per_group = sbi->s_inodes_per_group /
793                                         sbi->s_inodes_per_block;
794         sbi->s_desc_per_block = sb->s_blocksize /
795                                         sizeof (struct ext2_group_desc);
796         sbi->s_sbh = bh;
797         sbi->s_mount_state = le16_to_cpu(es->s_state);
798         sbi->s_addr_per_block_bits =
799                 log2 (EXT2_ADDR_PER_BLOCK(sb));
800         sbi->s_desc_per_block_bits =
801                 log2 (EXT2_DESC_PER_BLOCK(sb));
802
803         if (sb->s_magic != EXT2_SUPER_MAGIC)
804                 goto cantfind_ext2;
805
806         if (sb->s_blocksize != bh->b_size) {
807                 if (!silent)
808                         printk ("VFS: Unsupported blocksize on dev "
809                                 "%s.\n", sb->s_id);
810                 goto failed_mount;
811         }
812
813         if (sb->s_blocksize != sbi->s_frag_size) {
814                 printk ("EXT2-fs: fragsize %lu != blocksize %lu (not supported yet)\n",
815                         sbi->s_frag_size, sb->s_blocksize);
816                 goto failed_mount;
817         }
818
819         if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
820                 printk ("EXT2-fs: #blocks per group too big: %lu\n",
821                         sbi->s_blocks_per_group);
822                 goto failed_mount;
823         }
824         if (sbi->s_frags_per_group > sb->s_blocksize * 8) {
825                 printk ("EXT2-fs: #fragments per group too big: %lu\n",
826                         sbi->s_frags_per_group);
827                 goto failed_mount;
828         }
829         if (sbi->s_inodes_per_group > sb->s_blocksize * 8) {
830                 printk ("EXT2-fs: #inodes per group too big: %lu\n",
831                         sbi->s_inodes_per_group);
832                 goto failed_mount;
833         }
834
835         if (EXT2_BLOCKS_PER_GROUP(sb) == 0)
836                 goto cantfind_ext2;
837         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
838                                         le32_to_cpu(es->s_first_data_block) +
839                                        EXT2_BLOCKS_PER_GROUP(sb) - 1) /
840                                        EXT2_BLOCKS_PER_GROUP(sb);
841         db_count = (sbi->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) /
842                    EXT2_DESC_PER_BLOCK(sb);
843         sbi->s_group_desc = kmalloc (db_count * sizeof (struct buffer_head *), GFP_KERNEL);
844         if (sbi->s_group_desc == NULL) {
845                 printk ("EXT2-fs: not enough memory\n");
846                 goto failed_mount;
847         }
848         percpu_counter_init(&sbi->s_freeblocks_counter);
849         percpu_counter_init(&sbi->s_freeinodes_counter);
850         percpu_counter_init(&sbi->s_dirs_counter);
851         bgl_lock_init(&sbi->s_blockgroup_lock);
852         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(*sbi->s_debts),
853                                GFP_KERNEL);
854         if (!sbi->s_debts) {
855                 printk ("EXT2-fs: not enough memory\n");
856                 goto failed_mount_group_desc;
857         }
858         memset(sbi->s_debts, 0, sbi->s_groups_count * sizeof(*sbi->s_debts));
859         for (i = 0; i < db_count; i++) {
860                 block = descriptor_loc(sb, logic_sb_block, i);
861                 sbi->s_group_desc[i] = sb_bread(sb, block);
862                 if (!sbi->s_group_desc[i]) {
863                         for (j = 0; j < i; j++)
864                                 brelse (sbi->s_group_desc[j]);
865                         printk ("EXT2-fs: unable to read group descriptors\n");
866                         goto failed_mount_group_desc;
867                 }
868         }
869         if (!ext2_check_descriptors (sb)) {
870                 printk ("EXT2-fs: group descriptors corrupted!\n");
871                 db_count = i;
872                 goto failed_mount2;
873         }
874         sbi->s_gdb_count = db_count;
875         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
876         spin_lock_init(&sbi->s_next_gen_lock);
877         /*
878          * set up enough so that it can read an inode
879          */
880         sb->s_op = &ext2_sops;
881         sb->s_export_op = &ext2_export_ops;
882         sb->s_xattr = ext2_xattr_handlers;
883         root = iget(sb, EXT2_ROOT_INO);
884         sb->s_root = d_alloc_root(root);
885         if (!sb->s_root) {
886                 iput(root);
887                 printk(KERN_ERR "EXT2-fs: get root inode failed\n");
888                 goto failed_mount2;
889         }
890         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
891                 dput(sb->s_root);
892                 sb->s_root = NULL;
893                 printk(KERN_ERR "EXT2-fs: corrupt root inode, run e2fsck\n");
894                 goto failed_mount2;
895         }
896         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL))
897                 ext2_warning(sb, __FUNCTION__,
898                         "mounting ext3 filesystem as ext2\n");
899         ext2_setup_super (sb, es, sb->s_flags & MS_RDONLY);
900         percpu_counter_mod(&sbi->s_freeblocks_counter,
901                                 ext2_count_free_blocks(sb));
902         percpu_counter_mod(&sbi->s_freeinodes_counter,
903                                 ext2_count_free_inodes(sb));
904         percpu_counter_mod(&sbi->s_dirs_counter,
905                                 ext2_count_dirs(sb));
906         return 0;
907
908 cantfind_ext2:
909         if (!silent)
910                 printk("VFS: Can't find an ext2 filesystem on dev %s.\n",
911                        sb->s_id);
912         goto failed_mount;
913
914 failed_mount2:
915         for (i = 0; i < db_count; i++)
916                 brelse(sbi->s_group_desc[i]);
917 failed_mount_group_desc:
918         kfree(sbi->s_group_desc);
919         kfree(sbi->s_debts);
920 failed_mount:
921         brelse(bh);
922 failed_sbi:
923         sb->s_fs_info = NULL;
924         kfree(sbi);
925         return -EINVAL;
926 }
927
928 static void ext2_commit_super (struct super_block * sb,
929                                struct ext2_super_block * es)
930 {
931         es->s_wtime = cpu_to_le32(get_seconds());
932         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
933         sb->s_dirt = 0;
934 }
935
936 static void ext2_sync_super(struct super_block *sb, struct ext2_super_block *es)
937 {
938         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
939         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
940         es->s_wtime = cpu_to_le32(get_seconds());
941         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
942         sync_dirty_buffer(EXT2_SB(sb)->s_sbh);
943         sb->s_dirt = 0;
944 }
945
946 /*
947  * In the second extended file system, it is not necessary to
948  * write the super block since we use a mapping of the
949  * disk super block in a buffer.
950  *
951  * However, this function is still used to set the fs valid
952  * flags to 0.  We need to set this flag to 0 since the fs
953  * may have been checked while mounted and e2fsck may have
954  * set s_state to EXT2_VALID_FS after some corrections.
955  */
956
957 void ext2_write_super (struct super_block * sb)
958 {
959         struct ext2_super_block * es;
960         lock_kernel();
961         if (!(sb->s_flags & MS_RDONLY)) {
962                 es = EXT2_SB(sb)->s_es;
963
964                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS) {
965                         ext2_debug ("setting valid to 0\n");
966                         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) &
967                                                   ~EXT2_VALID_FS);
968                         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
969                         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
970                         es->s_mtime = cpu_to_le32(get_seconds());
971                         ext2_sync_super(sb, es);
972                 } else
973                         ext2_commit_super (sb, es);
974         }
975         sb->s_dirt = 0;
976         unlock_kernel();
977 }
978
979 static int ext2_remount (struct super_block * sb, int * flags, char * data)
980 {
981         struct ext2_sb_info * sbi = EXT2_SB(sb);
982         struct ext2_super_block * es;
983         unsigned long old_mount_opt = sbi->s_mount_opt;
984         struct ext2_mount_options old_opts;
985         unsigned long old_sb_flags;
986         int err;
987
988         /* Store the old options */
989         old_sb_flags = sb->s_flags;
990         old_opts.s_mount_opt = sbi->s_mount_opt;
991         old_opts.s_resuid = sbi->s_resuid;
992         old_opts.s_resgid = sbi->s_resgid;
993
994         /*
995          * Allow the "check" option to be passed as a remount option.
996          */
997         if (!parse_options (data, sbi)) {
998                 err = -EINVAL;
999                 goto restore_opts;
1000         }
1001
1002         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1003                 ((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1004
1005         es = sbi->s_es;
1006         if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
1007             (old_mount_opt & EXT2_MOUNT_XIP)) &&
1008             invalidate_inodes(sb))
1009                 ext2_warning(sb, __FUNCTION__, "busy inodes while remounting "\
1010                              "xip remain in cache (no functional problem)");
1011         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1012                 return 0;
1013         if (*flags & MS_RDONLY) {
1014                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS ||
1015                     !(sbi->s_mount_state & EXT2_VALID_FS))
1016                         return 0;
1017                 /*
1018                  * OK, we are remounting a valid rw partition rdonly, so set
1019                  * the rdonly flag and then mark the partition as valid again.
1020                  */
1021                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1022                 es->s_mtime = cpu_to_le32(get_seconds());
1023         } else {
1024                 __le32 ret = EXT2_HAS_RO_COMPAT_FEATURE(sb,
1025                                                ~EXT2_FEATURE_RO_COMPAT_SUPP);
1026                 if (ret) {
1027                         printk("EXT2-fs: %s: couldn't remount RDWR because of "
1028                                "unsupported optional features (%x).\n",
1029                                sb->s_id, le32_to_cpu(ret));
1030                         err = -EROFS;
1031                         goto restore_opts;
1032                 }
1033                 /*
1034                  * Mounting a RDONLY partition read-write, so reread and
1035                  * store the current valid flag.  (It may have been changed
1036                  * by e2fsck since we originally mounted the partition.)
1037                  */
1038                 sbi->s_mount_state = le16_to_cpu(es->s_state);
1039                 if (!ext2_setup_super (sb, es, 0))
1040                         sb->s_flags &= ~MS_RDONLY;
1041         }
1042         ext2_sync_super(sb, es);
1043         return 0;
1044 restore_opts:
1045         sbi->s_mount_opt = old_opts.s_mount_opt;
1046         sbi->s_resuid = old_opts.s_resuid;
1047         sbi->s_resgid = old_opts.s_resgid;
1048         sb->s_flags = old_sb_flags;
1049         return err;
1050 }
1051
1052 static int ext2_statfs (struct super_block * sb, struct kstatfs * buf)
1053 {
1054         struct ext2_sb_info *sbi = EXT2_SB(sb);
1055         unsigned long overhead;
1056         int i;
1057
1058         if (test_opt (sb, MINIX_DF))
1059                 overhead = 0;
1060         else {
1061                 /*
1062                  * Compute the overhead (FS structures)
1063                  */
1064
1065                 /*
1066                  * All of the blocks before first_data_block are
1067                  * overhead
1068                  */
1069                 overhead = le32_to_cpu(sbi->s_es->s_first_data_block);
1070
1071                 /*
1072                  * Add the overhead attributed to the superblock and
1073                  * block group descriptors.  If the sparse superblocks
1074                  * feature is turned on, then not all groups have this.
1075                  */
1076                 for (i = 0; i < sbi->s_groups_count; i++)
1077                         overhead += ext2_bg_has_super(sb, i) +
1078                                 ext2_bg_num_gdb(sb, i);
1079
1080                 /*
1081                  * Every block group has an inode bitmap, a block
1082                  * bitmap, and an inode table.
1083                  */
1084                 overhead += (sbi->s_groups_count *
1085                              (2 + sbi->s_itb_per_group));
1086         }
1087
1088         buf->f_type = EXT2_SUPER_MAGIC;
1089         buf->f_bsize = sb->s_blocksize;
1090         buf->f_blocks = le32_to_cpu(sbi->s_es->s_blocks_count) - overhead;
1091         buf->f_bfree = ext2_count_free_blocks(sb);
1092         buf->f_bavail = buf->f_bfree - le32_to_cpu(sbi->s_es->s_r_blocks_count);
1093         if (buf->f_bfree < le32_to_cpu(sbi->s_es->s_r_blocks_count))
1094                 buf->f_bavail = 0;
1095         buf->f_files = le32_to_cpu(sbi->s_es->s_inodes_count);
1096         buf->f_ffree = ext2_count_free_inodes (sb);
1097         buf->f_namelen = EXT2_NAME_LEN;
1098         return 0;
1099 }
1100
1101 static struct super_block *ext2_get_sb(struct file_system_type *fs_type,
1102         int flags, const char *dev_name, void *data)
1103 {
1104         return get_sb_bdev(fs_type, flags, dev_name, data, ext2_fill_super);
1105 }
1106
1107 #ifdef CONFIG_QUOTA
1108
1109 /* Read data from quotafile - avoid pagecache and such because we cannot afford
1110  * acquiring the locks... As quota files are never truncated and quota code
1111  * itself serializes the operations (and noone else should touch the files)
1112  * we don't have to be afraid of races */
1113 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data,
1114                                size_t len, loff_t off)
1115 {
1116         struct inode *inode = sb_dqopt(sb)->files[type];
1117         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1118         int err = 0;
1119         int offset = off & (sb->s_blocksize - 1);
1120         int tocopy;
1121         size_t toread;
1122         struct buffer_head tmp_bh;
1123         struct buffer_head *bh;
1124         loff_t i_size = i_size_read(inode);
1125
1126         if (off > i_size)
1127                 return 0;
1128         if (off+len > i_size)
1129                 len = i_size-off;
1130         toread = len;
1131         while (toread > 0) {
1132                 tocopy = sb->s_blocksize - offset < toread ?
1133                                 sb->s_blocksize - offset : toread;
1134
1135                 tmp_bh.b_state = 0;
1136                 err = ext2_get_block(inode, blk, &tmp_bh, 0);
1137                 if (err)
1138                         return err;
1139                 if (!buffer_mapped(&tmp_bh))    /* A hole? */
1140                         memset(data, 0, tocopy);
1141                 else {
1142                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1143                         if (!bh)
1144                                 return -EIO;
1145                         memcpy(data, bh->b_data+offset, tocopy);
1146                         brelse(bh);
1147                 }
1148                 offset = 0;
1149                 toread -= tocopy;
1150                 data += tocopy;
1151                 blk++;
1152         }
1153         return len;
1154 }
1155
1156 /* Write to quotafile */
1157 static ssize_t ext2_quota_write(struct super_block *sb, int type,
1158                                 const char *data, 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 towrite = len;
1166         struct buffer_head tmp_bh;
1167         struct buffer_head *bh;
1168
1169         down(&inode->i_sem);
1170         while (towrite > 0) {
1171                 tocopy = sb->s_blocksize - offset < towrite ?
1172                                 sb->s_blocksize - offset : towrite;
1173
1174                 tmp_bh.b_state = 0;
1175                 err = ext2_get_block(inode, blk, &tmp_bh, 1);
1176                 if (err)
1177                         goto out;
1178                 if (offset || tocopy != EXT2_BLOCK_SIZE(sb))
1179                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1180                 else
1181                         bh = sb_getblk(sb, tmp_bh.b_blocknr);
1182                 if (!bh) {
1183                         err = -EIO;
1184                         goto out;
1185                 }
1186                 lock_buffer(bh);
1187                 memcpy(bh->b_data+offset, data, tocopy);
1188                 flush_dcache_page(bh->b_page);
1189                 set_buffer_uptodate(bh);
1190                 mark_buffer_dirty(bh);
1191                 unlock_buffer(bh);
1192                 brelse(bh);
1193                 offset = 0;
1194                 towrite -= tocopy;
1195                 data += tocopy;
1196                 blk++;
1197         }
1198 out:
1199         if (len == towrite)
1200                 return err;
1201         if (inode->i_size < off+len-towrite)
1202                 i_size_write(inode, off+len-towrite);
1203         inode->i_version++;
1204         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1205         mark_inode_dirty(inode);
1206         up(&inode->i_sem);
1207         return len - towrite;
1208 }
1209
1210 #endif
1211
1212 static struct file_system_type ext2_fs_type = {
1213         .owner          = THIS_MODULE,
1214         .name           = "ext2",
1215         .get_sb         = ext2_get_sb,
1216         .kill_sb        = kill_block_super,
1217         .fs_flags       = FS_REQUIRES_DEV,
1218 };
1219
1220 static int __init init_ext2_fs(void)
1221 {
1222         int err = init_ext2_xattr();
1223         if (err)
1224                 return err;
1225         err = init_inodecache();
1226         if (err)
1227                 goto out1;
1228         err = register_filesystem(&ext2_fs_type);
1229         if (err)
1230                 goto out;
1231         return 0;
1232 out:
1233         destroy_inodecache();
1234 out1:
1235         exit_ext2_xattr();
1236         return err;
1237 }
1238
1239 static void __exit exit_ext2_fs(void)
1240 {
1241         unregister_filesystem(&ext2_fs_type);
1242         destroy_inodecache();
1243         exit_ext2_xattr();
1244 }
1245
1246 module_init(init_ext2_fs)
1247 module_exit(exit_ext2_fs)