[PATCH] EXT2: Remove superblock lock contention in ext2_statfs
[linux-2.6.git] / 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 kmem_cache_t * 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, SLAB_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, kmem_cache_t * 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         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
214 #if defined(CONFIG_QUOTA)
215         if (sbi->s_mount_opt & EXT2_MOUNT_USRQUOTA)
216                 seq_puts(seq, ",usrquota");
217
218         if (sbi->s_mount_opt & EXT2_MOUNT_GRPQUOTA)
219                 seq_puts(seq, ",grpquota");
220 #endif
221
222 #if defined(CONFIG_EXT2_FS_XIP)
223         if (sbi->s_mount_opt & EXT2_MOUNT_XIP)
224                 seq_puts(seq, ",xip");
225 #endif
226
227         return 0;
228 }
229
230 #ifdef CONFIG_QUOTA
231 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data, size_t len, loff_t off);
232 static ssize_t ext2_quota_write(struct super_block *sb, int type, const char *data, size_t len, loff_t off);
233 #endif
234
235 static struct super_operations ext2_sops = {
236         .alloc_inode    = ext2_alloc_inode,
237         .destroy_inode  = ext2_destroy_inode,
238         .read_inode     = ext2_read_inode,
239         .write_inode    = ext2_write_inode,
240         .put_inode      = ext2_put_inode,
241         .delete_inode   = ext2_delete_inode,
242         .put_super      = ext2_put_super,
243         .write_super    = ext2_write_super,
244         .statfs         = ext2_statfs,
245         .remount_fs     = ext2_remount,
246         .clear_inode    = ext2_clear_inode,
247         .show_options   = ext2_show_options,
248 #ifdef CONFIG_QUOTA
249         .quota_read     = ext2_quota_read,
250         .quota_write    = ext2_quota_write,
251 #endif
252 };
253
254 static struct dentry *ext2_get_dentry(struct super_block *sb, void *vobjp)
255 {
256         __u32 *objp = vobjp;
257         unsigned long ino = objp[0];
258         __u32 generation = objp[1];
259         struct inode *inode;
260         struct dentry *result;
261
262         if (ino < EXT2_FIRST_INO(sb) && ino != EXT2_ROOT_INO)
263                 return ERR_PTR(-ESTALE);
264         if (ino > le32_to_cpu(EXT2_SB(sb)->s_es->s_inodes_count))
265                 return ERR_PTR(-ESTALE);
266
267         /* iget isn't really right if the inode is currently unallocated!!
268          * ext2_read_inode currently does appropriate checks, but
269          * it might be "neater" to call ext2_get_inode first and check
270          * if the inode is valid.....
271          */
272         inode = iget(sb, ino);
273         if (inode == NULL)
274                 return ERR_PTR(-ENOMEM);
275         if (is_bad_inode(inode) ||
276             (generation && inode->i_generation != generation)) {
277                 /* we didn't find the right inode.. */
278                 iput(inode);
279                 return ERR_PTR(-ESTALE);
280         }
281         /* now to find a dentry.
282          * If possible, get a well-connected one
283          */
284         result = d_alloc_anon(inode);
285         if (!result) {
286                 iput(inode);
287                 return ERR_PTR(-ENOMEM);
288         }
289         return result;
290 }
291
292 /* Yes, most of these are left as NULL!!
293  * A NULL value implies the default, which works with ext2-like file
294  * systems, but can be improved upon.
295  * Currently only get_parent is required.
296  */
297 static struct export_operations ext2_export_ops = {
298         .get_parent = ext2_get_parent,
299         .get_dentry = ext2_get_dentry,
300 };
301
302 static unsigned long get_sb_block(void **data)
303 {
304         unsigned long   sb_block;
305         char            *options = (char *) *data;
306
307         if (!options || strncmp(options, "sb=", 3) != 0)
308                 return 1;       /* Default location */
309         options += 3;
310         sb_block = simple_strtoul(options, &options, 0);
311         if (*options && *options != ',') {
312                 printk("EXT2-fs: Invalid sb specification: %s\n",
313                        (char *) *data);
314                 return 1;
315         }
316         if (*options == ',')
317                 options++;
318         *data = (void *) options;
319         return sb_block;
320 }
321
322 enum {
323         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
324         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic,
325         Opt_err_ro, Opt_nouid32, Opt_nocheck, Opt_debug,
326         Opt_oldalloc, Opt_orlov, Opt_nobh, Opt_user_xattr, Opt_nouser_xattr,
327         Opt_acl, Opt_noacl, Opt_xip, Opt_ignore, Opt_err, Opt_quota,
328         Opt_usrquota, Opt_grpquota
329 };
330
331 static match_table_t tokens = {
332         {Opt_bsd_df, "bsddf"},
333         {Opt_minix_df, "minixdf"},
334         {Opt_grpid, "grpid"},
335         {Opt_grpid, "bsdgroups"},
336         {Opt_nogrpid, "nogrpid"},
337         {Opt_nogrpid, "sysvgroups"},
338         {Opt_resgid, "resgid=%u"},
339         {Opt_resuid, "resuid=%u"},
340         {Opt_sb, "sb=%u"},
341         {Opt_err_cont, "errors=continue"},
342         {Opt_err_panic, "errors=panic"},
343         {Opt_err_ro, "errors=remount-ro"},
344         {Opt_nouid32, "nouid32"},
345         {Opt_nocheck, "check=none"},
346         {Opt_nocheck, "nocheck"},
347         {Opt_debug, "debug"},
348         {Opt_oldalloc, "oldalloc"},
349         {Opt_orlov, "orlov"},
350         {Opt_nobh, "nobh"},
351         {Opt_user_xattr, "user_xattr"},
352         {Opt_nouser_xattr, "nouser_xattr"},
353         {Opt_acl, "acl"},
354         {Opt_noacl, "noacl"},
355         {Opt_xip, "xip"},
356         {Opt_grpquota, "grpquota"},
357         {Opt_ignore, "noquota"},
358         {Opt_quota, "quota"},
359         {Opt_usrquota, "usrquota"},
360         {Opt_err, NULL}
361 };
362
363 static int parse_options (char * options,
364                           struct ext2_sb_info *sbi)
365 {
366         char * p;
367         substring_t args[MAX_OPT_ARGS];
368         unsigned long kind = EXT2_MOUNT_ERRORS_CONT;
369         int option;
370
371         if (!options)
372                 return 1;
373
374         while ((p = strsep (&options, ",")) != NULL) {
375                 int token;
376                 if (!*p)
377                         continue;
378
379                 token = match_token(p, tokens, args);
380                 switch (token) {
381                 case Opt_bsd_df:
382                         clear_opt (sbi->s_mount_opt, MINIX_DF);
383                         break;
384                 case Opt_minix_df:
385                         set_opt (sbi->s_mount_opt, MINIX_DF);
386                         break;
387                 case Opt_grpid:
388                         set_opt (sbi->s_mount_opt, GRPID);
389                         break;
390                 case Opt_nogrpid:
391                         clear_opt (sbi->s_mount_opt, GRPID);
392                         break;
393                 case Opt_resuid:
394                         if (match_int(&args[0], &option))
395                                 return 0;
396                         sbi->s_resuid = option;
397                         break;
398                 case Opt_resgid:
399                         if (match_int(&args[0], &option))
400                                 return 0;
401                         sbi->s_resgid = option;
402                         break;
403                 case Opt_sb:
404                         /* handled by get_sb_block() instead of here */
405                         /* *sb_block = match_int(&args[0]); */
406                         break;
407                 case Opt_err_panic:
408                         kind = EXT2_MOUNT_ERRORS_PANIC;
409                         break;
410                 case Opt_err_ro:
411                         kind = EXT2_MOUNT_ERRORS_RO;
412                         break;
413                 case Opt_err_cont:
414                         kind = EXT2_MOUNT_ERRORS_CONT;
415                         break;
416                 case Opt_nouid32:
417                         set_opt (sbi->s_mount_opt, NO_UID32);
418                         break;
419                 case Opt_nocheck:
420                         clear_opt (sbi->s_mount_opt, CHECK);
421                         break;
422                 case Opt_debug:
423                         set_opt (sbi->s_mount_opt, DEBUG);
424                         break;
425                 case Opt_oldalloc:
426                         set_opt (sbi->s_mount_opt, OLDALLOC);
427                         break;
428                 case Opt_orlov:
429                         clear_opt (sbi->s_mount_opt, OLDALLOC);
430                         break;
431                 case Opt_nobh:
432                         set_opt (sbi->s_mount_opt, NOBH);
433                         break;
434 #ifdef CONFIG_EXT2_FS_XATTR
435                 case Opt_user_xattr:
436                         set_opt (sbi->s_mount_opt, XATTR_USER);
437                         break;
438                 case Opt_nouser_xattr:
439                         clear_opt (sbi->s_mount_opt, XATTR_USER);
440                         break;
441 #else
442                 case Opt_user_xattr:
443                 case Opt_nouser_xattr:
444                         printk("EXT2 (no)user_xattr options not supported\n");
445                         break;
446 #endif
447 #ifdef CONFIG_EXT2_FS_POSIX_ACL
448                 case Opt_acl:
449                         set_opt(sbi->s_mount_opt, POSIX_ACL);
450                         break;
451                 case Opt_noacl:
452                         clear_opt(sbi->s_mount_opt, POSIX_ACL);
453                         break;
454 #else
455                 case Opt_acl:
456                 case Opt_noacl:
457                         printk("EXT2 (no)acl options not supported\n");
458                         break;
459 #endif
460                 case Opt_xip:
461 #ifdef CONFIG_EXT2_FS_XIP
462                         set_opt (sbi->s_mount_opt, XIP);
463 #else
464                         printk("EXT2 xip option not supported\n");
465 #endif
466                         break;
467
468 #if defined(CONFIG_QUOTA)
469                 case Opt_quota:
470                 case Opt_usrquota:
471                         set_opt(sbi->s_mount_opt, USRQUOTA);
472                         break;
473
474                 case Opt_grpquota:
475                         set_opt(sbi->s_mount_opt, GRPQUOTA);
476                         break;
477 #else
478                 case Opt_quota:
479                 case Opt_usrquota:
480                 case Opt_grpquota:
481                         printk(KERN_ERR
482                                 "EXT2-fs: quota operations not supported.\n");
483
484                         break;
485 #endif
486
487                 case Opt_ignore:
488                         break;
489                 default:
490                         return 0;
491                 }
492         }
493         sbi->s_mount_opt |= kind;
494         return 1;
495 }
496
497 static int ext2_setup_super (struct super_block * sb,
498                               struct ext2_super_block * es,
499                               int read_only)
500 {
501         int res = 0;
502         struct ext2_sb_info *sbi = EXT2_SB(sb);
503
504         if (le32_to_cpu(es->s_rev_level) > EXT2_MAX_SUPP_REV) {
505                 printk ("EXT2-fs warning: revision level too high, "
506                         "forcing read-only mode\n");
507                 res = MS_RDONLY;
508         }
509         if (read_only)
510                 return res;
511         if (!(sbi->s_mount_state & EXT2_VALID_FS))
512                 printk ("EXT2-fs warning: mounting unchecked fs, "
513                         "running e2fsck is recommended\n");
514         else if ((sbi->s_mount_state & EXT2_ERROR_FS))
515                 printk ("EXT2-fs warning: mounting fs with errors, "
516                         "running e2fsck is recommended\n");
517         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) >= 0 &&
518                  le16_to_cpu(es->s_mnt_count) >=
519                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
520                 printk ("EXT2-fs warning: maximal mount count reached, "
521                         "running e2fsck is recommended\n");
522         else if (le32_to_cpu(es->s_checkinterval) &&
523                 (le32_to_cpu(es->s_lastcheck) + le32_to_cpu(es->s_checkinterval) <= get_seconds()))
524                 printk ("EXT2-fs warning: checktime reached, "
525                         "running e2fsck is recommended\n");
526         if (!le16_to_cpu(es->s_max_mnt_count))
527                 es->s_max_mnt_count = cpu_to_le16(EXT2_DFL_MAX_MNT_COUNT);
528         es->s_mnt_count=cpu_to_le16(le16_to_cpu(es->s_mnt_count) + 1);
529         ext2_write_super(sb);
530         if (test_opt (sb, DEBUG))
531                 printk ("[EXT II FS %s, %s, bs=%lu, fs=%lu, gc=%lu, "
532                         "bpg=%lu, ipg=%lu, mo=%04lx]\n",
533                         EXT2FS_VERSION, EXT2FS_DATE, sb->s_blocksize,
534                         sbi->s_frag_size,
535                         sbi->s_groups_count,
536                         EXT2_BLOCKS_PER_GROUP(sb),
537                         EXT2_INODES_PER_GROUP(sb),
538                         sbi->s_mount_opt);
539         return res;
540 }
541
542 static int ext2_check_descriptors (struct super_block * sb)
543 {
544         int i;
545         int desc_block = 0;
546         struct ext2_sb_info *sbi = EXT2_SB(sb);
547         unsigned long block = le32_to_cpu(sbi->s_es->s_first_data_block);
548         struct ext2_group_desc * gdp = NULL;
549
550         ext2_debug ("Checking group descriptors");
551
552         for (i = 0; i < sbi->s_groups_count; i++)
553         {
554                 if ((i % EXT2_DESC_PER_BLOCK(sb)) == 0)
555                         gdp = (struct ext2_group_desc *) sbi->s_group_desc[desc_block++]->b_data;
556                 if (le32_to_cpu(gdp->bg_block_bitmap) < block ||
557                     le32_to_cpu(gdp->bg_block_bitmap) >= block + EXT2_BLOCKS_PER_GROUP(sb))
558                 {
559                         ext2_error (sb, "ext2_check_descriptors",
560                                     "Block bitmap for group %d"
561                                     " not in group (block %lu)!",
562                                     i, (unsigned long) le32_to_cpu(gdp->bg_block_bitmap));
563                         return 0;
564                 }
565                 if (le32_to_cpu(gdp->bg_inode_bitmap) < block ||
566                     le32_to_cpu(gdp->bg_inode_bitmap) >= block + EXT2_BLOCKS_PER_GROUP(sb))
567                 {
568                         ext2_error (sb, "ext2_check_descriptors",
569                                     "Inode bitmap for group %d"
570                                     " not in group (block %lu)!",
571                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_bitmap));
572                         return 0;
573                 }
574                 if (le32_to_cpu(gdp->bg_inode_table) < block ||
575                     le32_to_cpu(gdp->bg_inode_table) + sbi->s_itb_per_group >=
576                     block + EXT2_BLOCKS_PER_GROUP(sb))
577                 {
578                         ext2_error (sb, "ext2_check_descriptors",
579                                     "Inode table for group %d"
580                                     " not in group (block %lu)!",
581                                     i, (unsigned long) le32_to_cpu(gdp->bg_inode_table));
582                         return 0;
583                 }
584                 block += EXT2_BLOCKS_PER_GROUP(sb);
585                 gdp++;
586         }
587         return 1;
588 }
589
590 #define log2(n) ffz(~(n))
591  
592 /*
593  * Maximal file size.  There is a direct, and {,double-,triple-}indirect
594  * block limit, and also a limit of (2^32 - 1) 512-byte sectors in i_blocks.
595  * We need to be 1 filesystem block less than the 2^32 sector limit.
596  */
597 static loff_t ext2_max_size(int bits)
598 {
599         loff_t res = EXT2_NDIR_BLOCKS;
600         /* This constant is calculated to be the largest file size for a
601          * dense, 4k-blocksize file such that the total number of
602          * sectors in the file, including data and all indirect blocks,
603          * does not exceed 2^32. */
604         const loff_t upper_limit = 0x1ff7fffd000LL;
605
606         res += 1LL << (bits-2);
607         res += 1LL << (2*(bits-2));
608         res += 1LL << (3*(bits-2));
609         res <<= bits;
610         if (res > upper_limit)
611                 res = upper_limit;
612         return res;
613 }
614
615 static unsigned long descriptor_loc(struct super_block *sb,
616                                     unsigned long logic_sb_block,
617                                     int nr)
618 {
619         struct ext2_sb_info *sbi = EXT2_SB(sb);
620         unsigned long bg, first_data_block, first_meta_bg;
621         int has_super = 0;
622         
623         first_data_block = le32_to_cpu(sbi->s_es->s_first_data_block);
624         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
625
626         if (!EXT2_HAS_INCOMPAT_FEATURE(sb, EXT2_FEATURE_INCOMPAT_META_BG) ||
627             nr < first_meta_bg)
628                 return (logic_sb_block + nr + 1);
629         bg = sbi->s_desc_per_block * nr;
630         if (ext2_bg_has_super(sb, bg))
631                 has_super = 1;
632         return (first_data_block + has_super + (bg * sbi->s_blocks_per_group));
633 }
634
635 static int ext2_fill_super(struct super_block *sb, void *data, int silent)
636 {
637         struct buffer_head * bh;
638         struct ext2_sb_info * sbi;
639         struct ext2_super_block * es;
640         struct inode *root;
641         unsigned long block;
642         unsigned long sb_block = get_sb_block(&data);
643         unsigned long logic_sb_block;
644         unsigned long offset = 0;
645         unsigned long def_mount_opts;
646         int blocksize = BLOCK_SIZE;
647         int db_count;
648         int i, j;
649         __le32 features;
650
651         sbi = kmalloc(sizeof(*sbi), GFP_KERNEL);
652         if (!sbi)
653                 return -ENOMEM;
654         sb->s_fs_info = sbi;
655         memset(sbi, 0, sizeof(*sbi));
656
657         /*
658          * See what the current blocksize for the device is, and
659          * use that as the blocksize.  Otherwise (or if the blocksize
660          * is smaller than the default) use the default.
661          * This is important for devices that have a hardware
662          * sectorsize that is larger than the default.
663          */
664         blocksize = sb_min_blocksize(sb, BLOCK_SIZE);
665         if (!blocksize) {
666                 printk ("EXT2-fs: unable to set blocksize\n");
667                 goto failed_sbi;
668         }
669
670         /*
671          * If the superblock doesn't start on a hardware sector boundary,
672          * calculate the offset.  
673          */
674         if (blocksize != BLOCK_SIZE) {
675                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
676                 offset = (sb_block*BLOCK_SIZE) % blocksize;
677         } else {
678                 logic_sb_block = sb_block;
679         }
680
681         if (!(bh = sb_bread(sb, logic_sb_block))) {
682                 printk ("EXT2-fs: unable to read superblock\n");
683                 goto failed_sbi;
684         }
685         /*
686          * Note: s_es must be initialized as soon as possible because
687          *       some ext2 macro-instructions depend on its value
688          */
689         es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
690         sbi->s_es = es;
691         sb->s_magic = le16_to_cpu(es->s_magic);
692
693         if (sb->s_magic != EXT2_SUPER_MAGIC)
694                 goto cantfind_ext2;
695
696         /* Set defaults before we parse the mount options */
697         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
698         if (def_mount_opts & EXT2_DEFM_DEBUG)
699                 set_opt(sbi->s_mount_opt, DEBUG);
700         if (def_mount_opts & EXT2_DEFM_BSDGROUPS)
701                 set_opt(sbi->s_mount_opt, GRPID);
702         if (def_mount_opts & EXT2_DEFM_UID16)
703                 set_opt(sbi->s_mount_opt, NO_UID32);
704         if (def_mount_opts & EXT2_DEFM_XATTR_USER)
705                 set_opt(sbi->s_mount_opt, XATTR_USER);
706         if (def_mount_opts & EXT2_DEFM_ACL)
707                 set_opt(sbi->s_mount_opt, POSIX_ACL);
708         
709         if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_PANIC)
710                 set_opt(sbi->s_mount_opt, ERRORS_PANIC);
711         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT2_ERRORS_RO)
712                 set_opt(sbi->s_mount_opt, ERRORS_RO);
713
714         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
715         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
716         
717         if (!parse_options ((char *) data, sbi))
718                 goto failed_mount;
719
720         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
721                 ((EXT2_SB(sb)->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ?
722                  MS_POSIXACL : 0);
723
724         ext2_xip_verify_sb(sb); /* see if bdev supports xip, unset
725                                     EXT2_MOUNT_XIP if not */
726
727         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV &&
728             (EXT2_HAS_COMPAT_FEATURE(sb, ~0U) ||
729              EXT2_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
730              EXT2_HAS_INCOMPAT_FEATURE(sb, ~0U)))
731                 printk("EXT2-fs warning: feature flags set on rev 0 fs, "
732                        "running e2fsck is recommended\n");
733         /*
734          * Check feature flags regardless of the revision level, since we
735          * previously didn't change the revision level when setting the flags,
736          * so there is a chance incompat flags are set on a rev 0 filesystem.
737          */
738         features = EXT2_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP);
739         if (features) {
740                 printk("EXT2-fs: %s: couldn't mount because of "
741                        "unsupported optional features (%x).\n",
742                        sb->s_id, le32_to_cpu(features));
743                 goto failed_mount;
744         }
745         if (!(sb->s_flags & MS_RDONLY) &&
746             (features = EXT2_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))){
747                 printk("EXT2-fs: %s: couldn't mount RDWR because of "
748                        "unsupported optional features (%x).\n",
749                        sb->s_id, le32_to_cpu(features));
750                 goto failed_mount;
751         }
752
753         blocksize = BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
754
755         if ((ext2_use_xip(sb)) && ((blocksize != PAGE_SIZE) ||
756                                   (sb->s_blocksize != blocksize))) {
757                 if (!silent)
758                         printk("XIP: Unsupported blocksize\n");
759                 goto failed_mount;
760         }
761
762         /* If the blocksize doesn't match, re-read the thing.. */
763         if (sb->s_blocksize != blocksize) {
764                 brelse(bh);
765
766                 if (!sb_set_blocksize(sb, blocksize)) {
767                         printk(KERN_ERR "EXT2-fs: blocksize too small for device.\n");
768                         goto failed_sbi;
769                 }
770
771                 logic_sb_block = (sb_block*BLOCK_SIZE) / blocksize;
772                 offset = (sb_block*BLOCK_SIZE) % blocksize;
773                 bh = sb_bread(sb, logic_sb_block);
774                 if(!bh) {
775                         printk("EXT2-fs: Couldn't read superblock on "
776                                "2nd try.\n");
777                         goto failed_sbi;
778                 }
779                 es = (struct ext2_super_block *) (((char *)bh->b_data) + offset);
780                 sbi->s_es = es;
781                 if (es->s_magic != cpu_to_le16(EXT2_SUPER_MAGIC)) {
782                         printk ("EXT2-fs: Magic mismatch, very weird !\n");
783                         goto failed_mount;
784                 }
785         }
786
787         sb->s_maxbytes = ext2_max_size(sb->s_blocksize_bits);
788
789         if (le32_to_cpu(es->s_rev_level) == EXT2_GOOD_OLD_REV) {
790                 sbi->s_inode_size = EXT2_GOOD_OLD_INODE_SIZE;
791                 sbi->s_first_ino = EXT2_GOOD_OLD_FIRST_INO;
792         } else {
793                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
794                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
795                 if ((sbi->s_inode_size < EXT2_GOOD_OLD_INODE_SIZE) ||
796                     (sbi->s_inode_size & (sbi->s_inode_size - 1)) ||
797                     (sbi->s_inode_size > blocksize)) {
798                         printk ("EXT2-fs: unsupported inode size: %d\n",
799                                 sbi->s_inode_size);
800                         goto failed_mount;
801                 }
802         }
803
804         sbi->s_frag_size = EXT2_MIN_FRAG_SIZE <<
805                                    le32_to_cpu(es->s_log_frag_size);
806         if (sbi->s_frag_size == 0)
807                 goto cantfind_ext2;
808         sbi->s_frags_per_block = sb->s_blocksize / sbi->s_frag_size;
809
810         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
811         sbi->s_frags_per_group = le32_to_cpu(es->s_frags_per_group);
812         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
813
814         if (EXT2_INODE_SIZE(sb) == 0)
815                 goto cantfind_ext2;
816         sbi->s_inodes_per_block = sb->s_blocksize / EXT2_INODE_SIZE(sb);
817         if (sbi->s_inodes_per_block == 0 || sbi->s_inodes_per_group == 0)
818                 goto cantfind_ext2;
819         sbi->s_itb_per_group = sbi->s_inodes_per_group /
820                                         sbi->s_inodes_per_block;
821         sbi->s_desc_per_block = sb->s_blocksize /
822                                         sizeof (struct ext2_group_desc);
823         sbi->s_sbh = bh;
824         sbi->s_mount_state = le16_to_cpu(es->s_state);
825         sbi->s_addr_per_block_bits =
826                 log2 (EXT2_ADDR_PER_BLOCK(sb));
827         sbi->s_desc_per_block_bits =
828                 log2 (EXT2_DESC_PER_BLOCK(sb));
829
830         if (sb->s_magic != EXT2_SUPER_MAGIC)
831                 goto cantfind_ext2;
832
833         if (sb->s_blocksize != bh->b_size) {
834                 if (!silent)
835                         printk ("VFS: Unsupported blocksize on dev "
836                                 "%s.\n", sb->s_id);
837                 goto failed_mount;
838         }
839
840         if (sb->s_blocksize != sbi->s_frag_size) {
841                 printk ("EXT2-fs: fragsize %lu != blocksize %lu (not supported yet)\n",
842                         sbi->s_frag_size, sb->s_blocksize);
843                 goto failed_mount;
844         }
845
846         if (sbi->s_blocks_per_group > sb->s_blocksize * 8) {
847                 printk ("EXT2-fs: #blocks per group too big: %lu\n",
848                         sbi->s_blocks_per_group);
849                 goto failed_mount;
850         }
851         if (sbi->s_frags_per_group > sb->s_blocksize * 8) {
852                 printk ("EXT2-fs: #fragments per group too big: %lu\n",
853                         sbi->s_frags_per_group);
854                 goto failed_mount;
855         }
856         if (sbi->s_inodes_per_group > sb->s_blocksize * 8) {
857                 printk ("EXT2-fs: #inodes per group too big: %lu\n",
858                         sbi->s_inodes_per_group);
859                 goto failed_mount;
860         }
861
862         if (EXT2_BLOCKS_PER_GROUP(sb) == 0)
863                 goto cantfind_ext2;
864         sbi->s_groups_count = (le32_to_cpu(es->s_blocks_count) -
865                                         le32_to_cpu(es->s_first_data_block) +
866                                        EXT2_BLOCKS_PER_GROUP(sb) - 1) /
867                                        EXT2_BLOCKS_PER_GROUP(sb);
868         db_count = (sbi->s_groups_count + EXT2_DESC_PER_BLOCK(sb) - 1) /
869                    EXT2_DESC_PER_BLOCK(sb);
870         sbi->s_group_desc = kmalloc (db_count * sizeof (struct buffer_head *), GFP_KERNEL);
871         if (sbi->s_group_desc == NULL) {
872                 printk ("EXT2-fs: not enough memory\n");
873                 goto failed_mount;
874         }
875         bgl_lock_init(&sbi->s_blockgroup_lock);
876         sbi->s_debts = kmalloc(sbi->s_groups_count * sizeof(*sbi->s_debts),
877                                GFP_KERNEL);
878         if (!sbi->s_debts) {
879                 printk ("EXT2-fs: not enough memory\n");
880                 goto failed_mount_group_desc;
881         }
882         memset(sbi->s_debts, 0, sbi->s_groups_count * sizeof(*sbi->s_debts));
883         for (i = 0; i < db_count; i++) {
884                 block = descriptor_loc(sb, logic_sb_block, i);
885                 sbi->s_group_desc[i] = sb_bread(sb, block);
886                 if (!sbi->s_group_desc[i]) {
887                         for (j = 0; j < i; j++)
888                                 brelse (sbi->s_group_desc[j]);
889                         printk ("EXT2-fs: unable to read group descriptors\n");
890                         goto failed_mount_group_desc;
891                 }
892         }
893         if (!ext2_check_descriptors (sb)) {
894                 printk ("EXT2-fs: group descriptors corrupted!\n");
895                 goto failed_mount2;
896         }
897         sbi->s_gdb_count = db_count;
898         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
899         spin_lock_init(&sbi->s_next_gen_lock);
900
901         percpu_counter_init(&sbi->s_freeblocks_counter,
902                                 ext2_count_free_blocks(sb));
903         percpu_counter_init(&sbi->s_freeinodes_counter,
904                                 ext2_count_free_inodes(sb));
905         percpu_counter_init(&sbi->s_dirs_counter,
906                                 ext2_count_dirs(sb));
907         /*
908          * set up enough so that it can read an inode
909          */
910         sb->s_op = &ext2_sops;
911         sb->s_export_op = &ext2_export_ops;
912         sb->s_xattr = ext2_xattr_handlers;
913         root = iget(sb, EXT2_ROOT_INO);
914         sb->s_root = d_alloc_root(root);
915         if (!sb->s_root) {
916                 iput(root);
917                 printk(KERN_ERR "EXT2-fs: get root inode failed\n");
918                 goto failed_mount3;
919         }
920         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
921                 dput(sb->s_root);
922                 sb->s_root = NULL;
923                 printk(KERN_ERR "EXT2-fs: corrupt root inode, run e2fsck\n");
924                 goto failed_mount3;
925         }
926         if (EXT2_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_HAS_JOURNAL))
927                 ext2_warning(sb, __FUNCTION__,
928                         "mounting ext3 filesystem as ext2");
929         ext2_setup_super (sb, es, sb->s_flags & MS_RDONLY);
930         return 0;
931
932 cantfind_ext2:
933         if (!silent)
934                 printk("VFS: Can't find an ext2 filesystem on dev %s.\n",
935                        sb->s_id);
936         goto failed_mount;
937 failed_mount3:
938         percpu_counter_destroy(&sbi->s_freeblocks_counter);
939         percpu_counter_destroy(&sbi->s_freeinodes_counter);
940         percpu_counter_destroy(&sbi->s_dirs_counter);
941 failed_mount2:
942         for (i = 0; i < db_count; i++)
943                 brelse(sbi->s_group_desc[i]);
944 failed_mount_group_desc:
945         kfree(sbi->s_group_desc);
946         kfree(sbi->s_debts);
947 failed_mount:
948         brelse(bh);
949 failed_sbi:
950         sb->s_fs_info = NULL;
951         kfree(sbi);
952         return -EINVAL;
953 }
954
955 static void ext2_commit_super (struct super_block * sb,
956                                struct ext2_super_block * es)
957 {
958         es->s_wtime = cpu_to_le32(get_seconds());
959         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
960         sb->s_dirt = 0;
961 }
962
963 static void ext2_sync_super(struct super_block *sb, struct ext2_super_block *es)
964 {
965         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
966         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
967         es->s_wtime = cpu_to_le32(get_seconds());
968         mark_buffer_dirty(EXT2_SB(sb)->s_sbh);
969         sync_dirty_buffer(EXT2_SB(sb)->s_sbh);
970         sb->s_dirt = 0;
971 }
972
973 /*
974  * In the second extended file system, it is not necessary to
975  * write the super block since we use a mapping of the
976  * disk super block in a buffer.
977  *
978  * However, this function is still used to set the fs valid
979  * flags to 0.  We need to set this flag to 0 since the fs
980  * may have been checked while mounted and e2fsck may have
981  * set s_state to EXT2_VALID_FS after some corrections.
982  */
983
984 void ext2_write_super (struct super_block * sb)
985 {
986         struct ext2_super_block * es;
987         lock_kernel();
988         if (!(sb->s_flags & MS_RDONLY)) {
989                 es = EXT2_SB(sb)->s_es;
990
991                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS) {
992                         ext2_debug ("setting valid to 0\n");
993                         es->s_state = cpu_to_le16(le16_to_cpu(es->s_state) &
994                                                   ~EXT2_VALID_FS);
995                         es->s_free_blocks_count = cpu_to_le32(ext2_count_free_blocks(sb));
996                         es->s_free_inodes_count = cpu_to_le32(ext2_count_free_inodes(sb));
997                         es->s_mtime = cpu_to_le32(get_seconds());
998                         ext2_sync_super(sb, es);
999                 } else
1000                         ext2_commit_super (sb, es);
1001         }
1002         sb->s_dirt = 0;
1003         unlock_kernel();
1004 }
1005
1006 static int ext2_remount (struct super_block * sb, int * flags, char * data)
1007 {
1008         struct ext2_sb_info * sbi = EXT2_SB(sb);
1009         struct ext2_super_block * es;
1010         unsigned long old_mount_opt = sbi->s_mount_opt;
1011         struct ext2_mount_options old_opts;
1012         unsigned long old_sb_flags;
1013         int err;
1014
1015         /* Store the old options */
1016         old_sb_flags = sb->s_flags;
1017         old_opts.s_mount_opt = sbi->s_mount_opt;
1018         old_opts.s_resuid = sbi->s_resuid;
1019         old_opts.s_resgid = sbi->s_resgid;
1020
1021         /*
1022          * Allow the "check" option to be passed as a remount option.
1023          */
1024         if (!parse_options (data, sbi)) {
1025                 err = -EINVAL;
1026                 goto restore_opts;
1027         }
1028
1029         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
1030                 ((sbi->s_mount_opt & EXT2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
1031
1032         es = sbi->s_es;
1033         if (((sbi->s_mount_opt & EXT2_MOUNT_XIP) !=
1034             (old_mount_opt & EXT2_MOUNT_XIP)) &&
1035             invalidate_inodes(sb))
1036                 ext2_warning(sb, __FUNCTION__, "busy inodes while remounting "\
1037                              "xip remain in cache (no functional problem)");
1038         if ((*flags & MS_RDONLY) == (sb->s_flags & MS_RDONLY))
1039                 return 0;
1040         if (*flags & MS_RDONLY) {
1041                 if (le16_to_cpu(es->s_state) & EXT2_VALID_FS ||
1042                     !(sbi->s_mount_state & EXT2_VALID_FS))
1043                         return 0;
1044                 /*
1045                  * OK, we are remounting a valid rw partition rdonly, so set
1046                  * the rdonly flag and then mark the partition as valid again.
1047                  */
1048                 es->s_state = cpu_to_le16(sbi->s_mount_state);
1049                 es->s_mtime = cpu_to_le32(get_seconds());
1050         } else {
1051                 __le32 ret = EXT2_HAS_RO_COMPAT_FEATURE(sb,
1052                                                ~EXT2_FEATURE_RO_COMPAT_SUPP);
1053                 if (ret) {
1054                         printk("EXT2-fs: %s: couldn't remount RDWR because of "
1055                                "unsupported optional features (%x).\n",
1056                                sb->s_id, le32_to_cpu(ret));
1057                         err = -EROFS;
1058                         goto restore_opts;
1059                 }
1060                 /*
1061                  * Mounting a RDONLY partition read-write, so reread and
1062                  * store the current valid flag.  (It may have been changed
1063                  * by e2fsck since we originally mounted the partition.)
1064                  */
1065                 sbi->s_mount_state = le16_to_cpu(es->s_state);
1066                 if (!ext2_setup_super (sb, es, 0))
1067                         sb->s_flags &= ~MS_RDONLY;
1068         }
1069         ext2_sync_super(sb, es);
1070         return 0;
1071 restore_opts:
1072         sbi->s_mount_opt = old_opts.s_mount_opt;
1073         sbi->s_resuid = old_opts.s_resuid;
1074         sbi->s_resgid = old_opts.s_resgid;
1075         sb->s_flags = old_sb_flags;
1076         return err;
1077 }
1078
1079 static int ext2_statfs (struct dentry * dentry, struct kstatfs * buf)
1080 {
1081         struct super_block *sb = dentry->d_sb;
1082         struct ext2_sb_info *sbi = EXT2_SB(sb);
1083         unsigned long overhead;
1084         int i;
1085
1086         if (test_opt (sb, MINIX_DF))
1087                 overhead = 0;
1088         else {
1089                 /*
1090                  * Compute the overhead (FS structures)
1091                  */
1092
1093                 /*
1094                  * All of the blocks before first_data_block are
1095                  * overhead
1096                  */
1097                 overhead = le32_to_cpu(sbi->s_es->s_first_data_block);
1098
1099                 /*
1100                  * Add the overhead attributed to the superblock and
1101                  * block group descriptors.  If the sparse superblocks
1102                  * feature is turned on, then not all groups have this.
1103                  */
1104                 for (i = 0; i < sbi->s_groups_count; i++)
1105                         overhead += ext2_bg_has_super(sb, i) +
1106                                 ext2_bg_num_gdb(sb, i);
1107
1108                 /*
1109                  * Every block group has an inode bitmap, a block
1110                  * bitmap, and an inode table.
1111                  */
1112                 overhead += (sbi->s_groups_count *
1113                              (2 + sbi->s_itb_per_group));
1114         }
1115
1116         buf->f_type = EXT2_SUPER_MAGIC;
1117         buf->f_bsize = sb->s_blocksize;
1118         buf->f_blocks = le32_to_cpu(sbi->s_es->s_blocks_count) - overhead;
1119         buf->f_bfree = ext2_count_free_blocks(sb);
1120         buf->f_bavail = buf->f_bfree - le32_to_cpu(sbi->s_es->s_r_blocks_count);
1121         if (buf->f_bfree < le32_to_cpu(sbi->s_es->s_r_blocks_count))
1122                 buf->f_bavail = 0;
1123         buf->f_files = le32_to_cpu(sbi->s_es->s_inodes_count);
1124         buf->f_ffree = ext2_count_free_inodes (sb);
1125         buf->f_namelen = EXT2_NAME_LEN;
1126         return 0;
1127 }
1128
1129 static int ext2_get_sb(struct file_system_type *fs_type,
1130         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1131 {
1132         return get_sb_bdev(fs_type, flags, dev_name, data, ext2_fill_super, mnt);
1133 }
1134
1135 #ifdef CONFIG_QUOTA
1136
1137 /* Read data from quotafile - avoid pagecache and such because we cannot afford
1138  * acquiring the locks... As quota files are never truncated and quota code
1139  * itself serializes the operations (and noone else should touch the files)
1140  * we don't have to be afraid of races */
1141 static ssize_t ext2_quota_read(struct super_block *sb, int type, char *data,
1142                                size_t len, loff_t off)
1143 {
1144         struct inode *inode = sb_dqopt(sb)->files[type];
1145         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1146         int err = 0;
1147         int offset = off & (sb->s_blocksize - 1);
1148         int tocopy;
1149         size_t toread;
1150         struct buffer_head tmp_bh;
1151         struct buffer_head *bh;
1152         loff_t i_size = i_size_read(inode);
1153
1154         if (off > i_size)
1155                 return 0;
1156         if (off+len > i_size)
1157                 len = i_size-off;
1158         toread = len;
1159         while (toread > 0) {
1160                 tocopy = sb->s_blocksize - offset < toread ?
1161                                 sb->s_blocksize - offset : toread;
1162
1163                 tmp_bh.b_state = 0;
1164                 err = ext2_get_block(inode, blk, &tmp_bh, 0);
1165                 if (err)
1166                         return err;
1167                 if (!buffer_mapped(&tmp_bh))    /* A hole? */
1168                         memset(data, 0, tocopy);
1169                 else {
1170                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1171                         if (!bh)
1172                                 return -EIO;
1173                         memcpy(data, bh->b_data+offset, tocopy);
1174                         brelse(bh);
1175                 }
1176                 offset = 0;
1177                 toread -= tocopy;
1178                 data += tocopy;
1179                 blk++;
1180         }
1181         return len;
1182 }
1183
1184 /* Write to quotafile */
1185 static ssize_t ext2_quota_write(struct super_block *sb, int type,
1186                                 const char *data, size_t len, loff_t off)
1187 {
1188         struct inode *inode = sb_dqopt(sb)->files[type];
1189         sector_t blk = off >> EXT2_BLOCK_SIZE_BITS(sb);
1190         int err = 0;
1191         int offset = off & (sb->s_blocksize - 1);
1192         int tocopy;
1193         size_t towrite = len;
1194         struct buffer_head tmp_bh;
1195         struct buffer_head *bh;
1196
1197         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
1198         while (towrite > 0) {
1199                 tocopy = sb->s_blocksize - offset < towrite ?
1200                                 sb->s_blocksize - offset : towrite;
1201
1202                 tmp_bh.b_state = 0;
1203                 err = ext2_get_block(inode, blk, &tmp_bh, 1);
1204                 if (err)
1205                         goto out;
1206                 if (offset || tocopy != EXT2_BLOCK_SIZE(sb))
1207                         bh = sb_bread(sb, tmp_bh.b_blocknr);
1208                 else
1209                         bh = sb_getblk(sb, tmp_bh.b_blocknr);
1210                 if (!bh) {
1211                         err = -EIO;
1212                         goto out;
1213                 }
1214                 lock_buffer(bh);
1215                 memcpy(bh->b_data+offset, data, tocopy);
1216                 flush_dcache_page(bh->b_page);
1217                 set_buffer_uptodate(bh);
1218                 mark_buffer_dirty(bh);
1219                 unlock_buffer(bh);
1220                 brelse(bh);
1221                 offset = 0;
1222                 towrite -= tocopy;
1223                 data += tocopy;
1224                 blk++;
1225         }
1226 out:
1227         if (len == towrite)
1228                 return err;
1229         if (inode->i_size < off+len-towrite)
1230                 i_size_write(inode, off+len-towrite);
1231         inode->i_version++;
1232         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1233         mark_inode_dirty(inode);
1234         mutex_unlock(&inode->i_mutex);
1235         return len - towrite;
1236 }
1237
1238 #endif
1239
1240 static struct file_system_type ext2_fs_type = {
1241         .owner          = THIS_MODULE,
1242         .name           = "ext2",
1243         .get_sb         = ext2_get_sb,
1244         .kill_sb        = kill_block_super,
1245         .fs_flags       = FS_REQUIRES_DEV,
1246 };
1247
1248 static int __init init_ext2_fs(void)
1249 {
1250         int err = init_ext2_xattr();
1251         if (err)
1252                 return err;
1253         err = init_inodecache();
1254         if (err)
1255                 goto out1;
1256         err = register_filesystem(&ext2_fs_type);
1257         if (err)
1258                 goto out;
1259         return 0;
1260 out:
1261         destroy_inodecache();
1262 out1:
1263         exit_ext2_xattr();
1264         return err;
1265 }
1266
1267 static void __exit exit_ext2_fs(void)
1268 {
1269         unregister_filesystem(&ext2_fs_type);
1270         destroy_inodecache();
1271         exit_ext2_xattr();
1272 }
1273
1274 module_init(init_ext2_fs)
1275 module_exit(exit_ext2_fs)