[PATCH] ufs2: tindirect truncate fix
[linux-2.6.git] / fs / ufs / truncate.c
1 /*
2  *  linux/fs/ufs/truncate.c
3  *
4  * Copyright (C) 1998
5  * Daniel Pirkl <daniel.pirkl@email.cz>
6  * Charles University, Faculty of Mathematics and Physics
7  *
8  *  from
9  *
10  *  linux/fs/ext2/truncate.c
11  *
12  * Copyright (C) 1992, 1993, 1994, 1995
13  * Remy Card (card@masi.ibp.fr)
14  * Laboratoire MASI - Institut Blaise Pascal
15  * Universite Pierre et Marie Curie (Paris VI)
16  *
17  *  from
18  *
19  *  linux/fs/minix/truncate.c
20  *
21  *  Copyright (C) 1991, 1992  Linus Torvalds
22  *
23  *  Big-endian to little-endian byte-swapping/bitmaps by
24  *        David S. Miller (davem@caip.rutgers.edu), 1995
25  */
26
27 /*
28  * Real random numbers for secure rm added 94/02/18
29  * Idea from Pierre del Perugia <delperug@gla.ecoledoc.ibp.fr>
30  */
31
32 /*
33  * Adoptation to use page cache and UFS2 write support by
34  * Evgeniy Dushistov <dushistov@mail.ru>, 2006-2007
35  */
36
37 #include <linux/errno.h>
38 #include <linux/fs.h>
39 #include <linux/ufs_fs.h>
40 #include <linux/fcntl.h>
41 #include <linux/time.h>
42 #include <linux/stat.h>
43 #include <linux/string.h>
44 #include <linux/smp_lock.h>
45 #include <linux/buffer_head.h>
46 #include <linux/blkdev.h>
47 #include <linux/sched.h>
48
49 #include "swab.h"
50 #include "util.h"
51
52 /*
53  * Secure deletion currently doesn't work. It interacts very badly
54  * with buffers shared with memory mappings, and for that reason
55  * can't be done in the truncate() routines. It should instead be
56  * done separately in "release()" before calling the truncate routines
57  * that will release the actual file blocks.
58  *
59  *              Linus
60  */
61
62 #define DIRECT_BLOCK ((inode->i_size + uspi->s_bsize - 1) >> uspi->s_bshift)
63 #define DIRECT_FRAGMENT ((inode->i_size + uspi->s_fsize - 1) >> uspi->s_fshift)
64
65
66 static int ufs_trunc_direct(struct inode *inode)
67 {
68         struct ufs_inode_info *ufsi = UFS_I(inode);
69         struct super_block * sb;
70         struct ufs_sb_private_info * uspi;
71         void *p;
72         u64 frag1, frag2, frag3, frag4, block1, block2;
73         unsigned frag_to_free, free_count;
74         unsigned i, tmp;
75         int retry;
76         
77         UFSD("ENTER: ino %lu\n", inode->i_ino);
78
79         sb = inode->i_sb;
80         uspi = UFS_SB(sb)->s_uspi;
81         
82         frag_to_free = 0;
83         free_count = 0;
84         retry = 0;
85         
86         frag1 = DIRECT_FRAGMENT;
87         frag4 = min_t(u32, UFS_NDIR_FRAGMENT, ufsi->i_lastfrag);
88         frag2 = ((frag1 & uspi->s_fpbmask) ? ((frag1 | uspi->s_fpbmask) + 1) : frag1);
89         frag3 = frag4 & ~uspi->s_fpbmask;
90         block1 = block2 = 0;
91         if (frag2 > frag3) {
92                 frag2 = frag4;
93                 frag3 = frag4 = 0;
94         } else if (frag2 < frag3) {
95                 block1 = ufs_fragstoblks (frag2);
96                 block2 = ufs_fragstoblks (frag3);
97         }
98
99         UFSD("ino %lu, frag1 %llu, frag2 %llu, block1 %llu, block2 %llu,"
100              " frag3 %llu, frag4 %llu\n", inode->i_ino,
101              (unsigned long long)frag1, (unsigned long long)frag2,
102              (unsigned long long)block1, (unsigned long long)block2,
103              (unsigned long long)frag3, (unsigned long long)frag4);
104
105         if (frag1 >= frag2)
106                 goto next1;             
107
108         /*
109          * Free first free fragments
110          */
111         p = ufs_get_direct_data_ptr(uspi, ufsi, ufs_fragstoblks(frag1));
112         tmp = ufs_data_ptr_to_cpu(sb, p);
113         if (!tmp )
114                 ufs_panic (sb, "ufs_trunc_direct", "internal error");
115         frag2 -= frag1;
116         frag1 = ufs_fragnum (frag1);
117
118         ufs_free_fragments(inode, tmp + frag1, frag2);
119         mark_inode_dirty(inode);
120         frag_to_free = tmp + frag1;
121
122 next1:
123         /*
124          * Free whole blocks
125          */
126         for (i = block1 ; i < block2; i++) {
127                 p = ufs_get_direct_data_ptr(uspi, ufsi, i);
128                 tmp = ufs_data_ptr_to_cpu(sb, p);
129                 if (!tmp)
130                         continue;
131                 ufs_data_ptr_clear(uspi, p);
132
133                 if (free_count == 0) {
134                         frag_to_free = tmp;
135                         free_count = uspi->s_fpb;
136                 } else if (free_count > 0 && frag_to_free == tmp - free_count)
137                         free_count += uspi->s_fpb;
138                 else {
139                         ufs_free_blocks (inode, frag_to_free, free_count);
140                         frag_to_free = tmp;
141                         free_count = uspi->s_fpb;
142                 }
143                 mark_inode_dirty(inode);
144         }
145         
146         if (free_count > 0)
147                 ufs_free_blocks (inode, frag_to_free, free_count);
148
149         if (frag3 >= frag4)
150                 goto next3;
151
152         /*
153          * Free last free fragments
154          */
155         p = ufs_get_direct_data_ptr(uspi, ufsi, ufs_fragstoblks(frag3));
156         tmp = ufs_data_ptr_to_cpu(sb, p);
157         if (!tmp )
158                 ufs_panic(sb, "ufs_truncate_direct", "internal error");
159         frag4 = ufs_fragnum (frag4);
160         ufs_data_ptr_clear(uspi, p);
161
162         ufs_free_fragments (inode, tmp, frag4);
163         mark_inode_dirty(inode);
164  next3:
165
166         UFSD("EXIT: ino %lu\n", inode->i_ino);
167         return retry;
168 }
169
170
171 static int ufs_trunc_indirect(struct inode *inode, u64 offset, void *p)
172 {
173         struct super_block * sb;
174         struct ufs_sb_private_info * uspi;
175         struct ufs_buffer_head * ind_ubh;
176         void *ind;
177         u64 tmp, indirect_block, i, frag_to_free;
178         unsigned free_count;
179         int retry;
180
181         UFSD("ENTER: ino %lu, offset %llu, p: %p\n",
182              inode->i_ino, (unsigned long long)offset, p);
183
184         BUG_ON(!p);
185                 
186         sb = inode->i_sb;
187         uspi = UFS_SB(sb)->s_uspi;
188
189         frag_to_free = 0;
190         free_count = 0;
191         retry = 0;
192         
193         tmp = ufs_data_ptr_to_cpu(sb, p);
194         if (!tmp)
195                 return 0;
196         ind_ubh = ubh_bread(sb, tmp, uspi->s_bsize);
197         if (tmp != ufs_data_ptr_to_cpu(sb, p)) {
198                 ubh_brelse (ind_ubh);
199                 return 1;
200         }
201         if (!ind_ubh) {
202                 ufs_data_ptr_clear(uspi, p);
203                 return 0;
204         }
205
206         indirect_block = (DIRECT_BLOCK > offset) ? (DIRECT_BLOCK - offset) : 0;
207         for (i = indirect_block; i < uspi->s_apb; i++) {
208                 ind = ubh_get_data_ptr(uspi, ind_ubh, i);
209                 tmp = ufs_data_ptr_to_cpu(sb, ind);
210                 if (!tmp)
211                         continue;
212
213                 ufs_data_ptr_clear(uspi, ind);
214                 ubh_mark_buffer_dirty(ind_ubh);
215                 if (free_count == 0) {
216                         frag_to_free = tmp;
217                         free_count = uspi->s_fpb;
218                 } else if (free_count > 0 && frag_to_free == tmp - free_count)
219                         free_count += uspi->s_fpb;
220                 else {
221                         ufs_free_blocks (inode, frag_to_free, free_count);
222                         frag_to_free = tmp;
223                         free_count = uspi->s_fpb;
224                 }
225
226                 mark_inode_dirty(inode);
227         }
228
229         if (free_count > 0) {
230                 ufs_free_blocks (inode, frag_to_free, free_count);
231         }
232         for (i = 0; i < uspi->s_apb; i++)
233                 if (!ufs_is_data_ptr_zero(uspi,
234                                           ubh_get_data_ptr(uspi, ind_ubh, i)))
235                         break;
236         if (i >= uspi->s_apb) {
237                 tmp = ufs_data_ptr_to_cpu(sb, p);
238                 ufs_data_ptr_clear(uspi, p);
239
240                 ufs_free_blocks (inode, tmp, uspi->s_fpb);
241                 mark_inode_dirty(inode);
242                 ubh_bforget(ind_ubh);
243                 ind_ubh = NULL;
244         }
245         if (IS_SYNC(inode) && ind_ubh && ubh_buffer_dirty(ind_ubh)) {
246                 ubh_ll_rw_block(SWRITE, ind_ubh);
247                 ubh_wait_on_buffer (ind_ubh);
248         }
249         ubh_brelse (ind_ubh);
250         
251         UFSD("EXIT: ino %lu\n", inode->i_ino);
252         
253         return retry;
254 }
255
256 static int ufs_trunc_dindirect(struct inode *inode, u64 offset, void *p)
257 {
258         struct super_block * sb;
259         struct ufs_sb_private_info * uspi;
260         struct ufs_buffer_head *dind_bh;
261         u64 i, tmp, dindirect_block;
262         void *dind;
263         int retry = 0;
264         
265         UFSD("ENTER: ino %lu\n", inode->i_ino);
266         
267         sb = inode->i_sb;
268         uspi = UFS_SB(sb)->s_uspi;
269
270         dindirect_block = (DIRECT_BLOCK > offset) 
271                 ? ((DIRECT_BLOCK - offset) >> uspi->s_apbshift) : 0;
272         retry = 0;
273         
274         tmp = ufs_data_ptr_to_cpu(sb, p);
275         if (!tmp)
276                 return 0;
277         dind_bh = ubh_bread(sb, tmp, uspi->s_bsize);
278         if (tmp != ufs_data_ptr_to_cpu(sb, p)) {
279                 ubh_brelse (dind_bh);
280                 return 1;
281         }
282         if (!dind_bh) {
283                 ufs_data_ptr_clear(uspi, p);
284                 return 0;
285         }
286
287         for (i = dindirect_block ; i < uspi->s_apb ; i++) {
288                 dind = ubh_get_data_ptr(uspi, dind_bh, i);
289                 tmp = ufs_data_ptr_to_cpu(sb, dind);
290                 if (!tmp)
291                         continue;
292                 retry |= ufs_trunc_indirect (inode, offset + (i << uspi->s_apbshift), dind);
293                 ubh_mark_buffer_dirty(dind_bh);
294         }
295
296         for (i = 0; i < uspi->s_apb; i++)
297                 if (!ufs_is_data_ptr_zero(uspi,
298                                           ubh_get_data_ptr(uspi, dind_bh, i)))
299                         break;
300         if (i >= uspi->s_apb) {
301                 tmp = ufs_data_ptr_to_cpu(sb, p);
302                 ufs_data_ptr_clear(uspi, p);
303
304                 ufs_free_blocks(inode, tmp, uspi->s_fpb);
305                 mark_inode_dirty(inode);
306                 ubh_bforget(dind_bh);
307                 dind_bh = NULL;
308         }
309         if (IS_SYNC(inode) && dind_bh && ubh_buffer_dirty(dind_bh)) {
310                 ubh_ll_rw_block(SWRITE, dind_bh);
311                 ubh_wait_on_buffer (dind_bh);
312         }
313         ubh_brelse (dind_bh);
314         
315         UFSD("EXIT: ino %lu\n", inode->i_ino);
316         
317         return retry;
318 }
319
320 static int ufs_trunc_tindirect(struct inode *inode)
321 {
322         struct super_block *sb = inode->i_sb;
323         struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
324         struct ufs_inode_info *ufsi = UFS_I(inode);
325         struct ufs_buffer_head * tind_bh;
326         u64 tindirect_block, tmp, i;
327         void *tind, *p;
328         int retry;
329         
330         UFSD("ENTER: ino %lu\n", inode->i_ino);
331
332         retry = 0;
333         
334         tindirect_block = (DIRECT_BLOCK > (UFS_NDADDR + uspi->s_apb + uspi->s_2apb))
335                 ? ((DIRECT_BLOCK - UFS_NDADDR - uspi->s_apb - uspi->s_2apb) >> uspi->s_2apbshift) : 0;
336
337         p = ufs_get_direct_data_ptr(uspi, ufsi, UFS_TIND_BLOCK);
338         if (!(tmp = ufs_data_ptr_to_cpu(sb, p)))
339                 return 0;
340         tind_bh = ubh_bread (sb, tmp, uspi->s_bsize);
341         if (tmp != ufs_data_ptr_to_cpu(sb, p)) {
342                 ubh_brelse (tind_bh);
343                 return 1;
344         }
345         if (!tind_bh) {
346                 ufs_data_ptr_clear(uspi, p);
347                 return 0;
348         }
349
350         for (i = tindirect_block ; i < uspi->s_apb ; i++) {
351                 tind = ubh_get_data_ptr(uspi, tind_bh, i);
352                 retry |= ufs_trunc_dindirect(inode, UFS_NDADDR + 
353                         uspi->s_apb + ((i + 1) << uspi->s_2apbshift), tind);
354                 ubh_mark_buffer_dirty(tind_bh);
355         }
356         for (i = 0; i < uspi->s_apb; i++)
357                 if (!ufs_is_data_ptr_zero(uspi,
358                                           ubh_get_data_ptr(uspi, tind_bh, i)))
359                         break;
360         if (i >= uspi->s_apb) {
361                 tmp = ufs_data_ptr_to_cpu(sb, p);
362                 ufs_data_ptr_clear(uspi, p);
363
364                 ufs_free_blocks(inode, tmp, uspi->s_fpb);
365                 mark_inode_dirty(inode);
366                 ubh_bforget(tind_bh);
367                 tind_bh = NULL;
368         }
369         if (IS_SYNC(inode) && tind_bh && ubh_buffer_dirty(tind_bh)) {
370                 ubh_ll_rw_block(SWRITE, tind_bh);
371                 ubh_wait_on_buffer (tind_bh);
372         }
373         ubh_brelse (tind_bh);
374         
375         UFSD("EXIT: ino %lu\n", inode->i_ino);
376         return retry;
377 }
378
379 static int ufs_alloc_lastblock(struct inode *inode)
380 {
381         int err = 0;
382         struct super_block *sb = inode->i_sb;
383         struct address_space *mapping = inode->i_mapping;
384         struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
385         unsigned i, end;
386         sector_t lastfrag;
387         struct page *lastpage;
388         struct buffer_head *bh;
389         u64 phys64;
390
391         lastfrag = (i_size_read(inode) + uspi->s_fsize - 1) >> uspi->s_fshift;
392
393         if (!lastfrag)
394                 goto out;
395
396         lastfrag--;
397
398         lastpage = ufs_get_locked_page(mapping, lastfrag >>
399                                        (PAGE_CACHE_SHIFT - inode->i_blkbits));
400        if (IS_ERR(lastpage)) {
401                err = -EIO;
402                goto out;
403        }
404
405        end = lastfrag & ((1 << (PAGE_CACHE_SHIFT - inode->i_blkbits)) - 1);
406        bh = page_buffers(lastpage);
407        for (i = 0; i < end; ++i)
408                bh = bh->b_this_page;
409
410
411        err = ufs_getfrag_block(inode, lastfrag, bh, 1);
412
413        if (unlikely(err))
414                goto out_unlock;
415
416        if (buffer_new(bh)) {
417                clear_buffer_new(bh);
418                unmap_underlying_metadata(bh->b_bdev,
419                                          bh->b_blocknr);
420                /*
421                 * we do not zeroize fragment, because of
422                 * if it maped to hole, it already contains zeroes
423                 */
424                set_buffer_uptodate(bh);
425                mark_buffer_dirty(bh);
426                set_page_dirty(lastpage);
427        }
428
429        if (lastfrag >= UFS_IND_FRAGMENT) {
430                end = uspi->s_fpb - ufs_fragnum(lastfrag) - 1;
431                phys64 = bh->b_blocknr + 1;
432                for (i = 0; i < end; ++i) {
433                        bh = sb_getblk(sb, i + phys64);
434                        lock_buffer(bh);
435                        memset(bh->b_data, 0, sb->s_blocksize);
436                        set_buffer_uptodate(bh);
437                        mark_buffer_dirty(bh);
438                        unlock_buffer(bh);
439                        sync_dirty_buffer(bh);
440                        brelse(bh);
441                }
442        }
443 out_unlock:
444        ufs_put_locked_page(lastpage);
445 out:
446        return err;
447 }
448
449 int ufs_truncate(struct inode *inode, loff_t old_i_size)
450 {
451         struct ufs_inode_info *ufsi = UFS_I(inode);
452         struct super_block *sb = inode->i_sb;
453         struct ufs_sb_private_info *uspi = UFS_SB(sb)->s_uspi;
454         int retry, err = 0;
455         
456         UFSD("ENTER: ino %lu, i_size: %llu, old_i_size: %llu\n",
457              inode->i_ino, (unsigned long long)i_size_read(inode),
458              (unsigned long long)old_i_size);
459
460         if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
461               S_ISLNK(inode->i_mode)))
462                 return -EINVAL;
463         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
464                 return -EPERM;
465
466         err = ufs_alloc_lastblock(inode);
467
468         if (err) {
469                 i_size_write(inode, old_i_size);
470                 goto out;
471         }
472
473         block_truncate_page(inode->i_mapping, inode->i_size, ufs_getfrag_block);
474
475         lock_kernel();
476         while (1) {
477                 retry = ufs_trunc_direct(inode);
478                 retry |= ufs_trunc_indirect(inode, UFS_IND_BLOCK,
479                                             ufs_get_direct_data_ptr(uspi, ufsi,
480                                                                     UFS_IND_BLOCK));
481                 retry |= ufs_trunc_dindirect(inode, UFS_IND_BLOCK + uspi->s_apb,
482                                              ufs_get_direct_data_ptr(uspi, ufsi,
483                                                                      UFS_DIND_BLOCK));
484                 retry |= ufs_trunc_tindirect (inode);
485                 if (!retry)
486                         break;
487                 if (IS_SYNC(inode) && (inode->i_state & I_DIRTY))
488                         ufs_sync_inode (inode);
489                 blk_run_address_space(inode->i_mapping);
490                 yield();
491         }
492
493         inode->i_mtime = inode->i_ctime = CURRENT_TIME_SEC;
494         ufsi->i_lastfrag = DIRECT_FRAGMENT;
495         unlock_kernel();
496         mark_inode_dirty(inode);
497 out:
498         UFSD("EXIT: err %d\n", err);
499         return err;
500 }
501
502
503 /*
504  * We don't define our `inode->i_op->truncate', and call it here,
505  * because of:
506  * - there is no way to know old size
507  * - there is no way inform user about error, if it happens in `truncate'
508  */
509 static int ufs_setattr(struct dentry *dentry, struct iattr *attr)
510 {
511         struct inode *inode = dentry->d_inode;
512         unsigned int ia_valid = attr->ia_valid;
513         int error;
514
515         error = inode_change_ok(inode, attr);
516         if (error)
517                 return error;
518
519         if (ia_valid & ATTR_SIZE &&
520             attr->ia_size != i_size_read(inode)) {
521                 loff_t old_i_size = inode->i_size;
522                 error = vmtruncate(inode, attr->ia_size);
523                 if (error)
524                         return error;
525                 error = ufs_truncate(inode, old_i_size);
526                 if (error)
527                         return error;
528         }
529         return inode_setattr(inode, attr);
530 }
531
532 const struct inode_operations ufs_file_inode_operations = {
533         .setattr = ufs_setattr,
534 };