Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/**
3 * inode.c - NTFS kernel inode handling.
4 *
5 * Copyright (c) 2001-2014 Anton Altaparmakov and Tuxera Inc.
6 */
7
8#include <linux/buffer_head.h>
9#include <linux/fs.h>
10#include <linux/mm.h>
11#include <linux/mount.h>
12#include <linux/mutex.h>
13#include <linux/pagemap.h>
14#include <linux/quotaops.h>
15#include <linux/slab.h>
16#include <linux/log2.h>
17
18#include "aops.h"
19#include "attrib.h"
20#include "bitmap.h"
21#include "dir.h"
22#include "debug.h"
23#include "inode.h"
24#include "lcnalloc.h"
25#include "malloc.h"
26#include "mft.h"
27#include "time.h"
28#include "ntfs.h"
29
30/**
31 * ntfs_test_inode - compare two (possibly fake) inodes for equality
32 * @vi: vfs inode which to test
33 * @data: data which is being tested with
34 *
35 * Compare the ntfs attribute embedded in the ntfs specific part of the vfs
36 * inode @vi for equality with the ntfs attribute @data.
37 *
38 * If searching for the normal file/directory inode, set @na->type to AT_UNUSED.
39 * @na->name and @na->name_len are then ignored.
40 *
41 * Return 1 if the attributes match and 0 if not.
42 *
43 * NOTE: This function runs with the inode_hash_lock spin lock held so it is not
44 * allowed to sleep.
45 */
46int ntfs_test_inode(struct inode *vi, void *data)
47{
48 ntfs_attr *na = (ntfs_attr *)data;
49 ntfs_inode *ni;
50
51 if (vi->i_ino != na->mft_no)
52 return 0;
53 ni = NTFS_I(vi);
54 /* If !NInoAttr(ni), @vi is a normal file or directory inode. */
55 if (likely(!NInoAttr(ni))) {
56 /* If not looking for a normal inode this is a mismatch. */
57 if (unlikely(na->type != AT_UNUSED))
58 return 0;
59 } else {
60 /* A fake inode describing an attribute. */
61 if (ni->type != na->type)
62 return 0;
63 if (ni->name_len != na->name_len)
64 return 0;
65 if (na->name_len && memcmp(ni->name, na->name,
66 na->name_len * sizeof(ntfschar)))
67 return 0;
68 }
69 /* Match! */
70 return 1;
71}
72
73/**
74 * ntfs_init_locked_inode - initialize an inode
75 * @vi: vfs inode to initialize
76 * @data: data which to initialize @vi to
77 *
78 * Initialize the vfs inode @vi with the values from the ntfs attribute @data in
79 * order to enable ntfs_test_inode() to do its work.
80 *
81 * If initializing the normal file/directory inode, set @na->type to AT_UNUSED.
82 * In that case, @na->name and @na->name_len should be set to NULL and 0,
83 * respectively. Although that is not strictly necessary as
84 * ntfs_read_locked_inode() will fill them in later.
85 *
86 * Return 0 on success and -errno on error.
87 *
88 * NOTE: This function runs with the inode->i_lock spin lock held so it is not
89 * allowed to sleep. (Hence the GFP_ATOMIC allocation.)
90 */
91static int ntfs_init_locked_inode(struct inode *vi, void *data)
92{
93 ntfs_attr *na = (ntfs_attr *)data;
94 ntfs_inode *ni = NTFS_I(vi);
95
96 vi->i_ino = na->mft_no;
97
98 ni->type = na->type;
99 if (na->type == AT_INDEX_ALLOCATION)
100 NInoSetMstProtected(ni);
101
102 ni->name = na->name;
103 ni->name_len = na->name_len;
104
105 /* If initializing a normal inode, we are done. */
106 if (likely(na->type == AT_UNUSED)) {
107 BUG_ON(na->name);
108 BUG_ON(na->name_len);
109 return 0;
110 }
111
112 /* It is a fake inode. */
113 NInoSetAttr(ni);
114
115 /*
116 * We have I30 global constant as an optimization as it is the name
117 * in >99.9% of named attributes! The other <0.1% incur a GFP_ATOMIC
118 * allocation but that is ok. And most attributes are unnamed anyway,
119 * thus the fraction of named attributes with name != I30 is actually
120 * absolutely tiny.
121 */
122 if (na->name_len && na->name != I30) {
123 unsigned int i;
124
125 BUG_ON(!na->name);
126 i = na->name_len * sizeof(ntfschar);
127 ni->name = kmalloc(i + sizeof(ntfschar), GFP_ATOMIC);
128 if (!ni->name)
129 return -ENOMEM;
130 memcpy(ni->name, na->name, i);
131 ni->name[na->name_len] = 0;
132 }
133 return 0;
134}
135
136static int ntfs_read_locked_inode(struct inode *vi);
137static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi);
138static int ntfs_read_locked_index_inode(struct inode *base_vi,
139 struct inode *vi);
140
141/**
142 * ntfs_iget - obtain a struct inode corresponding to a specific normal inode
143 * @sb: super block of mounted volume
144 * @mft_no: mft record number / inode number to obtain
145 *
146 * Obtain the struct inode corresponding to a specific normal inode (i.e. a
147 * file or directory).
148 *
149 * If the inode is in the cache, it is just returned with an increased
150 * reference count. Otherwise, a new struct inode is allocated and initialized,
151 * and finally ntfs_read_locked_inode() is called to read in the inode and
152 * fill in the remainder of the inode structure.
153 *
154 * Return the struct inode on success. Check the return value with IS_ERR() and
155 * if true, the function failed and the error code is obtained from PTR_ERR().
156 */
157struct inode *ntfs_iget(struct super_block *sb, unsigned long mft_no)
158{
159 struct inode *vi;
160 int err;
161 ntfs_attr na;
162
163 na.mft_no = mft_no;
164 na.type = AT_UNUSED;
165 na.name = NULL;
166 na.name_len = 0;
167
168 vi = iget5_locked(sb, mft_no, ntfs_test_inode,
169 ntfs_init_locked_inode, &na);
170 if (unlikely(!vi))
171 return ERR_PTR(-ENOMEM);
172
173 err = 0;
174
175 /* If this is a freshly allocated inode, need to read it now. */
176 if (vi->i_state & I_NEW) {
177 err = ntfs_read_locked_inode(vi);
178 unlock_new_inode(vi);
179 }
180 /*
181 * There is no point in keeping bad inodes around if the failure was
182 * due to ENOMEM. We want to be able to retry again later.
183 */
184 if (unlikely(err == -ENOMEM)) {
185 iput(vi);
186 vi = ERR_PTR(err);
187 }
188 return vi;
189}
190
191/**
192 * ntfs_attr_iget - obtain a struct inode corresponding to an attribute
193 * @base_vi: vfs base inode containing the attribute
194 * @type: attribute type
195 * @name: Unicode name of the attribute (NULL if unnamed)
196 * @name_len: length of @name in Unicode characters (0 if unnamed)
197 *
198 * Obtain the (fake) struct inode corresponding to the attribute specified by
199 * @type, @name, and @name_len, which is present in the base mft record
200 * specified by the vfs inode @base_vi.
201 *
202 * If the attribute inode is in the cache, it is just returned with an
203 * increased reference count. Otherwise, a new struct inode is allocated and
204 * initialized, and finally ntfs_read_locked_attr_inode() is called to read the
205 * attribute and fill in the inode structure.
206 *
207 * Note, for index allocation attributes, you need to use ntfs_index_iget()
208 * instead of ntfs_attr_iget() as working with indices is a lot more complex.
209 *
210 * Return the struct inode of the attribute inode on success. Check the return
211 * value with IS_ERR() and if true, the function failed and the error code is
212 * obtained from PTR_ERR().
213 */
214struct inode *ntfs_attr_iget(struct inode *base_vi, ATTR_TYPE type,
215 ntfschar *name, u32 name_len)
216{
217 struct inode *vi;
218 int err;
219 ntfs_attr na;
220
221 /* Make sure no one calls ntfs_attr_iget() for indices. */
222 BUG_ON(type == AT_INDEX_ALLOCATION);
223
224 na.mft_no = base_vi->i_ino;
225 na.type = type;
226 na.name = name;
227 na.name_len = name_len;
228
229 vi = iget5_locked(base_vi->i_sb, na.mft_no, ntfs_test_inode,
230 ntfs_init_locked_inode, &na);
231 if (unlikely(!vi))
232 return ERR_PTR(-ENOMEM);
233
234 err = 0;
235
236 /* If this is a freshly allocated inode, need to read it now. */
237 if (vi->i_state & I_NEW) {
238 err = ntfs_read_locked_attr_inode(base_vi, vi);
239 unlock_new_inode(vi);
240 }
241 /*
242 * There is no point in keeping bad attribute inodes around. This also
243 * simplifies things in that we never need to check for bad attribute
244 * inodes elsewhere.
245 */
246 if (unlikely(err)) {
247 iput(vi);
248 vi = ERR_PTR(err);
249 }
250 return vi;
251}
252
253/**
254 * ntfs_index_iget - obtain a struct inode corresponding to an index
255 * @base_vi: vfs base inode containing the index related attributes
256 * @name: Unicode name of the index
257 * @name_len: length of @name in Unicode characters
258 *
259 * Obtain the (fake) struct inode corresponding to the index specified by @name
260 * and @name_len, which is present in the base mft record specified by the vfs
261 * inode @base_vi.
262 *
263 * If the index inode is in the cache, it is just returned with an increased
264 * reference count. Otherwise, a new struct inode is allocated and
265 * initialized, and finally ntfs_read_locked_index_inode() is called to read
266 * the index related attributes and fill in the inode structure.
267 *
268 * Return the struct inode of the index inode on success. Check the return
269 * value with IS_ERR() and if true, the function failed and the error code is
270 * obtained from PTR_ERR().
271 */
272struct inode *ntfs_index_iget(struct inode *base_vi, ntfschar *name,
273 u32 name_len)
274{
275 struct inode *vi;
276 int err;
277 ntfs_attr na;
278
279 na.mft_no = base_vi->i_ino;
280 na.type = AT_INDEX_ALLOCATION;
281 na.name = name;
282 na.name_len = name_len;
283
284 vi = iget5_locked(base_vi->i_sb, na.mft_no, ntfs_test_inode,
285 ntfs_init_locked_inode, &na);
286 if (unlikely(!vi))
287 return ERR_PTR(-ENOMEM);
288
289 err = 0;
290
291 /* If this is a freshly allocated inode, need to read it now. */
292 if (vi->i_state & I_NEW) {
293 err = ntfs_read_locked_index_inode(base_vi, vi);
294 unlock_new_inode(vi);
295 }
296 /*
297 * There is no point in keeping bad index inodes around. This also
298 * simplifies things in that we never need to check for bad index
299 * inodes elsewhere.
300 */
301 if (unlikely(err)) {
302 iput(vi);
303 vi = ERR_PTR(err);
304 }
305 return vi;
306}
307
308struct inode *ntfs_alloc_big_inode(struct super_block *sb)
309{
310 ntfs_inode *ni;
311
312 ntfs_debug("Entering.");
313 ni = kmem_cache_alloc(ntfs_big_inode_cache, GFP_NOFS);
314 if (likely(ni != NULL)) {
315 ni->state = 0;
316 return VFS_I(ni);
317 }
318 ntfs_error(sb, "Allocation of NTFS big inode structure failed.");
319 return NULL;
320}
321
322void ntfs_free_big_inode(struct inode *inode)
323{
324 kmem_cache_free(ntfs_big_inode_cache, NTFS_I(inode));
325}
326
327static inline ntfs_inode *ntfs_alloc_extent_inode(void)
328{
329 ntfs_inode *ni;
330
331 ntfs_debug("Entering.");
332 ni = kmem_cache_alloc(ntfs_inode_cache, GFP_NOFS);
333 if (likely(ni != NULL)) {
334 ni->state = 0;
335 return ni;
336 }
337 ntfs_error(NULL, "Allocation of NTFS inode structure failed.");
338 return NULL;
339}
340
341static void ntfs_destroy_extent_inode(ntfs_inode *ni)
342{
343 ntfs_debug("Entering.");
344 BUG_ON(ni->page);
345 if (!atomic_dec_and_test(&ni->count))
346 BUG();
347 kmem_cache_free(ntfs_inode_cache, ni);
348}
349
350/*
351 * The attribute runlist lock has separate locking rules from the
352 * normal runlist lock, so split the two lock-classes:
353 */
354static struct lock_class_key attr_list_rl_lock_class;
355
356/**
357 * __ntfs_init_inode - initialize ntfs specific part of an inode
358 * @sb: super block of mounted volume
359 * @ni: freshly allocated ntfs inode which to initialize
360 *
361 * Initialize an ntfs inode to defaults.
362 *
363 * NOTE: ni->mft_no, ni->state, ni->type, ni->name, and ni->name_len are left
364 * untouched. Make sure to initialize them elsewhere.
365 *
366 * Return zero on success and -ENOMEM on error.
367 */
368void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
369{
370 ntfs_debug("Entering.");
371 rwlock_init(&ni->size_lock);
372 ni->initialized_size = ni->allocated_size = 0;
373 ni->seq_no = 0;
374 atomic_set(&ni->count, 1);
375 ni->vol = NTFS_SB(sb);
376 ntfs_init_runlist(&ni->runlist);
377 mutex_init(&ni->mrec_lock);
378 ni->page = NULL;
379 ni->page_ofs = 0;
380 ni->attr_list_size = 0;
381 ni->attr_list = NULL;
382 ntfs_init_runlist(&ni->attr_list_rl);
383 lockdep_set_class(&ni->attr_list_rl.lock,
384 &attr_list_rl_lock_class);
385 ni->itype.index.block_size = 0;
386 ni->itype.index.vcn_size = 0;
387 ni->itype.index.collation_rule = 0;
388 ni->itype.index.block_size_bits = 0;
389 ni->itype.index.vcn_size_bits = 0;
390 mutex_init(&ni->extent_lock);
391 ni->nr_extents = 0;
392 ni->ext.base_ntfs_ino = NULL;
393}
394
395/*
396 * Extent inodes get MFT-mapped in a nested way, while the base inode
397 * is still mapped. Teach this nesting to the lock validator by creating
398 * a separate class for nested inode's mrec_lock's:
399 */
400static struct lock_class_key extent_inode_mrec_lock_key;
401
402inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
403 unsigned long mft_no)
404{
405 ntfs_inode *ni = ntfs_alloc_extent_inode();
406
407 ntfs_debug("Entering.");
408 if (likely(ni != NULL)) {
409 __ntfs_init_inode(sb, ni);
410 lockdep_set_class(&ni->mrec_lock, &extent_inode_mrec_lock_key);
411 ni->mft_no = mft_no;
412 ni->type = AT_UNUSED;
413 ni->name = NULL;
414 ni->name_len = 0;
415 }
416 return ni;
417}
418
419/**
420 * ntfs_is_extended_system_file - check if a file is in the $Extend directory
421 * @ctx: initialized attribute search context
422 *
423 * Search all file name attributes in the inode described by the attribute
424 * search context @ctx and check if any of the names are in the $Extend system
425 * directory.
426 *
427 * Return values:
428 * 1: file is in $Extend directory
429 * 0: file is not in $Extend directory
430 * -errno: failed to determine if the file is in the $Extend directory
431 */
432static int ntfs_is_extended_system_file(ntfs_attr_search_ctx *ctx)
433{
434 int nr_links, err;
435
436 /* Restart search. */
437 ntfs_attr_reinit_search_ctx(ctx);
438
439 /* Get number of hard links. */
440 nr_links = le16_to_cpu(ctx->mrec->link_count);
441
442 /* Loop through all hard links. */
443 while (!(err = ntfs_attr_lookup(AT_FILE_NAME, NULL, 0, 0, 0, NULL, 0,
444 ctx))) {
445 FILE_NAME_ATTR *file_name_attr;
446 ATTR_RECORD *attr = ctx->attr;
447 u8 *p, *p2;
448
449 nr_links--;
450 /*
451 * Maximum sanity checking as we are called on an inode that
452 * we suspect might be corrupt.
453 */
454 p = (u8*)attr + le32_to_cpu(attr->length);
455 if (p < (u8*)ctx->mrec || (u8*)p > (u8*)ctx->mrec +
456 le32_to_cpu(ctx->mrec->bytes_in_use)) {
457err_corrupt_attr:
458 ntfs_error(ctx->ntfs_ino->vol->sb, "Corrupt file name "
459 "attribute. You should run chkdsk.");
460 return -EIO;
461 }
462 if (attr->non_resident) {
463 ntfs_error(ctx->ntfs_ino->vol->sb, "Non-resident file "
464 "name. You should run chkdsk.");
465 return -EIO;
466 }
467 if (attr->flags) {
468 ntfs_error(ctx->ntfs_ino->vol->sb, "File name with "
469 "invalid flags. You should run "
470 "chkdsk.");
471 return -EIO;
472 }
473 if (!(attr->data.resident.flags & RESIDENT_ATTR_IS_INDEXED)) {
474 ntfs_error(ctx->ntfs_ino->vol->sb, "Unindexed file "
475 "name. You should run chkdsk.");
476 return -EIO;
477 }
478 file_name_attr = (FILE_NAME_ATTR*)((u8*)attr +
479 le16_to_cpu(attr->data.resident.value_offset));
480 p2 = (u8*)attr + le32_to_cpu(attr->data.resident.value_length);
481 if (p2 < (u8*)attr || p2 > p)
482 goto err_corrupt_attr;
483 /* This attribute is ok, but is it in the $Extend directory? */
484 if (MREF_LE(file_name_attr->parent_directory) == FILE_Extend)
485 return 1; /* YES, it's an extended system file. */
486 }
487 if (unlikely(err != -ENOENT))
488 return err;
489 if (unlikely(nr_links)) {
490 ntfs_error(ctx->ntfs_ino->vol->sb, "Inode hard link count "
491 "doesn't match number of name attributes. You "
492 "should run chkdsk.");
493 return -EIO;
494 }
495 return 0; /* NO, it is not an extended system file. */
496}
497
498/**
499 * ntfs_read_locked_inode - read an inode from its device
500 * @vi: inode to read
501 *
502 * ntfs_read_locked_inode() is called from ntfs_iget() to read the inode
503 * described by @vi into memory from the device.
504 *
505 * The only fields in @vi that we need to/can look at when the function is
506 * called are i_sb, pointing to the mounted device's super block, and i_ino,
507 * the number of the inode to load.
508 *
509 * ntfs_read_locked_inode() maps, pins and locks the mft record number i_ino
510 * for reading and sets up the necessary @vi fields as well as initializing
511 * the ntfs inode.
512 *
513 * Q: What locks are held when the function is called?
514 * A: i_state has I_NEW set, hence the inode is locked, also
515 * i_count is set to 1, so it is not going to go away
516 * i_flags is set to 0 and we have no business touching it. Only an ioctl()
517 * is allowed to write to them. We should of course be honouring them but
518 * we need to do that using the IS_* macros defined in include/linux/fs.h.
519 * In any case ntfs_read_locked_inode() has nothing to do with i_flags.
520 *
521 * Return 0 on success and -errno on error. In the error case, the inode will
522 * have had make_bad_inode() executed on it.
523 */
524static int ntfs_read_locked_inode(struct inode *vi)
525{
526 ntfs_volume *vol = NTFS_SB(vi->i_sb);
527 ntfs_inode *ni;
528 struct inode *bvi;
529 MFT_RECORD *m;
530 ATTR_RECORD *a;
531 STANDARD_INFORMATION *si;
532 ntfs_attr_search_ctx *ctx;
533 int err = 0;
534
535 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
536
537 /* Setup the generic vfs inode parts now. */
538 vi->i_uid = vol->uid;
539 vi->i_gid = vol->gid;
540 vi->i_mode = 0;
541
542 /*
543 * Initialize the ntfs specific part of @vi special casing
544 * FILE_MFT which we need to do at mount time.
545 */
546 if (vi->i_ino != FILE_MFT)
547 ntfs_init_big_inode(vi);
548 ni = NTFS_I(vi);
549
550 m = map_mft_record(ni);
551 if (IS_ERR(m)) {
552 err = PTR_ERR(m);
553 goto err_out;
554 }
555 ctx = ntfs_attr_get_search_ctx(ni, m);
556 if (!ctx) {
557 err = -ENOMEM;
558 goto unm_err_out;
559 }
560
561 if (!(m->flags & MFT_RECORD_IN_USE)) {
562 ntfs_error(vi->i_sb, "Inode is not in use!");
563 goto unm_err_out;
564 }
565 if (m->base_mft_record) {
566 ntfs_error(vi->i_sb, "Inode is an extent inode!");
567 goto unm_err_out;
568 }
569
570 /* Transfer information from mft record into vfs and ntfs inodes. */
571 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
572
573 /*
574 * FIXME: Keep in mind that link_count is two for files which have both
575 * a long file name and a short file name as separate entries, so if
576 * we are hiding short file names this will be too high. Either we need
577 * to account for the short file names by subtracting them or we need
578 * to make sure we delete files even though i_nlink is not zero which
579 * might be tricky due to vfs interactions. Need to think about this
580 * some more when implementing the unlink command.
581 */
582 set_nlink(vi, le16_to_cpu(m->link_count));
583 /*
584 * FIXME: Reparse points can have the directory bit set even though
585 * they would be S_IFLNK. Need to deal with this further below when we
586 * implement reparse points / symbolic links but it will do for now.
587 * Also if not a directory, it could be something else, rather than
588 * a regular file. But again, will do for now.
589 */
590 /* Everyone gets all permissions. */
591 vi->i_mode |= S_IRWXUGO;
592 /* If read-only, no one gets write permissions. */
593 if (IS_RDONLY(vi))
594 vi->i_mode &= ~S_IWUGO;
595 if (m->flags & MFT_RECORD_IS_DIRECTORY) {
596 vi->i_mode |= S_IFDIR;
597 /*
598 * Apply the directory permissions mask set in the mount
599 * options.
600 */
601 vi->i_mode &= ~vol->dmask;
602 /* Things break without this kludge! */
603 if (vi->i_nlink > 1)
604 set_nlink(vi, 1);
605 } else {
606 vi->i_mode |= S_IFREG;
607 /* Apply the file permissions mask set in the mount options. */
608 vi->i_mode &= ~vol->fmask;
609 }
610 /*
611 * Find the standard information attribute in the mft record. At this
612 * stage we haven't setup the attribute list stuff yet, so this could
613 * in fact fail if the standard information is in an extent record, but
614 * I don't think this actually ever happens.
615 */
616 err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0, 0, 0, NULL, 0,
617 ctx);
618 if (unlikely(err)) {
619 if (err == -ENOENT) {
620 /*
621 * TODO: We should be performing a hot fix here (if the
622 * recover mount option is set) by creating a new
623 * attribute.
624 */
625 ntfs_error(vi->i_sb, "$STANDARD_INFORMATION attribute "
626 "is missing.");
627 }
628 goto unm_err_out;
629 }
630 a = ctx->attr;
631 /* Get the standard information attribute value. */
632 si = (STANDARD_INFORMATION*)((u8*)a +
633 le16_to_cpu(a->data.resident.value_offset));
634
635 /* Transfer information from the standard information into vi. */
636 /*
637 * Note: The i_?times do not quite map perfectly onto the NTFS times,
638 * but they are close enough, and in the end it doesn't really matter
639 * that much...
640 */
641 /*
642 * mtime is the last change of the data within the file. Not changed
643 * when only metadata is changed, e.g. a rename doesn't affect mtime.
644 */
645 vi->i_mtime = ntfs2utc(si->last_data_change_time);
646 /*
647 * ctime is the last change of the metadata of the file. This obviously
648 * always changes, when mtime is changed. ctime can be changed on its
649 * own, mtime is then not changed, e.g. when a file is renamed.
650 */
651 vi->i_ctime = ntfs2utc(si->last_mft_change_time);
652 /*
653 * Last access to the data within the file. Not changed during a rename
654 * for example but changed whenever the file is written to.
655 */
656 vi->i_atime = ntfs2utc(si->last_access_time);
657
658 /* Find the attribute list attribute if present. */
659 ntfs_attr_reinit_search_ctx(ctx);
660 err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
661 if (err) {
662 if (unlikely(err != -ENOENT)) {
663 ntfs_error(vi->i_sb, "Failed to lookup attribute list "
664 "attribute.");
665 goto unm_err_out;
666 }
667 } else /* if (!err) */ {
668 if (vi->i_ino == FILE_MFT)
669 goto skip_attr_list_load;
670 ntfs_debug("Attribute list found in inode 0x%lx.", vi->i_ino);
671 NInoSetAttrList(ni);
672 a = ctx->attr;
673 if (a->flags & ATTR_COMPRESSION_MASK) {
674 ntfs_error(vi->i_sb, "Attribute list attribute is "
675 "compressed.");
676 goto unm_err_out;
677 }
678 if (a->flags & ATTR_IS_ENCRYPTED ||
679 a->flags & ATTR_IS_SPARSE) {
680 if (a->non_resident) {
681 ntfs_error(vi->i_sb, "Non-resident attribute "
682 "list attribute is encrypted/"
683 "sparse.");
684 goto unm_err_out;
685 }
686 ntfs_warning(vi->i_sb, "Resident attribute list "
687 "attribute in inode 0x%lx is marked "
688 "encrypted/sparse which is not true. "
689 "However, Windows allows this and "
690 "chkdsk does not detect or correct it "
691 "so we will just ignore the invalid "
692 "flags and pretend they are not set.",
693 vi->i_ino);
694 }
695 /* Now allocate memory for the attribute list. */
696 ni->attr_list_size = (u32)ntfs_attr_size(a);
697 ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
698 if (!ni->attr_list) {
699 ntfs_error(vi->i_sb, "Not enough memory to allocate "
700 "buffer for attribute list.");
701 err = -ENOMEM;
702 goto unm_err_out;
703 }
704 if (a->non_resident) {
705 NInoSetAttrListNonResident(ni);
706 if (a->data.non_resident.lowest_vcn) {
707 ntfs_error(vi->i_sb, "Attribute list has non "
708 "zero lowest_vcn.");
709 goto unm_err_out;
710 }
711 /*
712 * Setup the runlist. No need for locking as we have
713 * exclusive access to the inode at this time.
714 */
715 ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
716 a, NULL);
717 if (IS_ERR(ni->attr_list_rl.rl)) {
718 err = PTR_ERR(ni->attr_list_rl.rl);
719 ni->attr_list_rl.rl = NULL;
720 ntfs_error(vi->i_sb, "Mapping pairs "
721 "decompression failed.");
722 goto unm_err_out;
723 }
724 /* Now load the attribute list. */
725 if ((err = load_attribute_list(vol, &ni->attr_list_rl,
726 ni->attr_list, ni->attr_list_size,
727 sle64_to_cpu(a->data.non_resident.
728 initialized_size)))) {
729 ntfs_error(vi->i_sb, "Failed to load "
730 "attribute list attribute.");
731 goto unm_err_out;
732 }
733 } else /* if (!a->non_resident) */ {
734 if ((u8*)a + le16_to_cpu(a->data.resident.value_offset)
735 + le32_to_cpu(
736 a->data.resident.value_length) >
737 (u8*)ctx->mrec + vol->mft_record_size) {
738 ntfs_error(vi->i_sb, "Corrupt attribute list "
739 "in inode.");
740 goto unm_err_out;
741 }
742 /* Now copy the attribute list. */
743 memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
744 a->data.resident.value_offset),
745 le32_to_cpu(
746 a->data.resident.value_length));
747 }
748 }
749skip_attr_list_load:
750 /*
751 * If an attribute list is present we now have the attribute list value
752 * in ntfs_ino->attr_list and it is ntfs_ino->attr_list_size bytes.
753 */
754 if (S_ISDIR(vi->i_mode)) {
755 loff_t bvi_size;
756 ntfs_inode *bni;
757 INDEX_ROOT *ir;
758 u8 *ir_end, *index_end;
759
760 /* It is a directory, find index root attribute. */
761 ntfs_attr_reinit_search_ctx(ctx);
762 err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE,
763 0, NULL, 0, ctx);
764 if (unlikely(err)) {
765 if (err == -ENOENT) {
766 // FIXME: File is corrupt! Hot-fix with empty
767 // index root attribute if recovery option is
768 // set.
769 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute "
770 "is missing.");
771 }
772 goto unm_err_out;
773 }
774 a = ctx->attr;
775 /* Set up the state. */
776 if (unlikely(a->non_resident)) {
777 ntfs_error(vol->sb, "$INDEX_ROOT attribute is not "
778 "resident.");
779 goto unm_err_out;
780 }
781 /* Ensure the attribute name is placed before the value. */
782 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
783 le16_to_cpu(a->data.resident.value_offset)))) {
784 ntfs_error(vol->sb, "$INDEX_ROOT attribute name is "
785 "placed after the attribute value.");
786 goto unm_err_out;
787 }
788 /*
789 * Compressed/encrypted index root just means that the newly
790 * created files in that directory should be created compressed/
791 * encrypted. However index root cannot be both compressed and
792 * encrypted.
793 */
794 if (a->flags & ATTR_COMPRESSION_MASK)
795 NInoSetCompressed(ni);
796 if (a->flags & ATTR_IS_ENCRYPTED) {
797 if (a->flags & ATTR_COMPRESSION_MASK) {
798 ntfs_error(vi->i_sb, "Found encrypted and "
799 "compressed attribute.");
800 goto unm_err_out;
801 }
802 NInoSetEncrypted(ni);
803 }
804 if (a->flags & ATTR_IS_SPARSE)
805 NInoSetSparse(ni);
806 ir = (INDEX_ROOT*)((u8*)a +
807 le16_to_cpu(a->data.resident.value_offset));
808 ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
809 if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
810 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
811 "corrupt.");
812 goto unm_err_out;
813 }
814 index_end = (u8*)&ir->index +
815 le32_to_cpu(ir->index.index_length);
816 if (index_end > ir_end) {
817 ntfs_error(vi->i_sb, "Directory index is corrupt.");
818 goto unm_err_out;
819 }
820 if (ir->type != AT_FILE_NAME) {
821 ntfs_error(vi->i_sb, "Indexed attribute is not "
822 "$FILE_NAME.");
823 goto unm_err_out;
824 }
825 if (ir->collation_rule != COLLATION_FILE_NAME) {
826 ntfs_error(vi->i_sb, "Index collation rule is not "
827 "COLLATION_FILE_NAME.");
828 goto unm_err_out;
829 }
830 ni->itype.index.collation_rule = ir->collation_rule;
831 ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
832 if (ni->itype.index.block_size &
833 (ni->itype.index.block_size - 1)) {
834 ntfs_error(vi->i_sb, "Index block size (%u) is not a "
835 "power of two.",
836 ni->itype.index.block_size);
837 goto unm_err_out;
838 }
839 if (ni->itype.index.block_size > PAGE_SIZE) {
840 ntfs_error(vi->i_sb, "Index block size (%u) > "
841 "PAGE_SIZE (%ld) is not "
842 "supported. Sorry.",
843 ni->itype.index.block_size,
844 PAGE_SIZE);
845 err = -EOPNOTSUPP;
846 goto unm_err_out;
847 }
848 if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
849 ntfs_error(vi->i_sb, "Index block size (%u) < "
850 "NTFS_BLOCK_SIZE (%i) is not "
851 "supported. Sorry.",
852 ni->itype.index.block_size,
853 NTFS_BLOCK_SIZE);
854 err = -EOPNOTSUPP;
855 goto unm_err_out;
856 }
857 ni->itype.index.block_size_bits =
858 ffs(ni->itype.index.block_size) - 1;
859 /* Determine the size of a vcn in the directory index. */
860 if (vol->cluster_size <= ni->itype.index.block_size) {
861 ni->itype.index.vcn_size = vol->cluster_size;
862 ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
863 } else {
864 ni->itype.index.vcn_size = vol->sector_size;
865 ni->itype.index.vcn_size_bits = vol->sector_size_bits;
866 }
867
868 /* Setup the index allocation attribute, even if not present. */
869 NInoSetMstProtected(ni);
870 ni->type = AT_INDEX_ALLOCATION;
871 ni->name = I30;
872 ni->name_len = 4;
873
874 if (!(ir->index.flags & LARGE_INDEX)) {
875 /* No index allocation. */
876 vi->i_size = ni->initialized_size =
877 ni->allocated_size = 0;
878 /* We are done with the mft record, so we release it. */
879 ntfs_attr_put_search_ctx(ctx);
880 unmap_mft_record(ni);
881 m = NULL;
882 ctx = NULL;
883 goto skip_large_dir_stuff;
884 } /* LARGE_INDEX: Index allocation present. Setup state. */
885 NInoSetIndexAllocPresent(ni);
886 /* Find index allocation attribute. */
887 ntfs_attr_reinit_search_ctx(ctx);
888 err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, I30, 4,
889 CASE_SENSITIVE, 0, NULL, 0, ctx);
890 if (unlikely(err)) {
891 if (err == -ENOENT)
892 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION "
893 "attribute is not present but "
894 "$INDEX_ROOT indicated it is.");
895 else
896 ntfs_error(vi->i_sb, "Failed to lookup "
897 "$INDEX_ALLOCATION "
898 "attribute.");
899 goto unm_err_out;
900 }
901 a = ctx->attr;
902 if (!a->non_resident) {
903 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
904 "is resident.");
905 goto unm_err_out;
906 }
907 /*
908 * Ensure the attribute name is placed before the mapping pairs
909 * array.
910 */
911 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
912 le16_to_cpu(
913 a->data.non_resident.mapping_pairs_offset)))) {
914 ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name "
915 "is placed after the mapping pairs "
916 "array.");
917 goto unm_err_out;
918 }
919 if (a->flags & ATTR_IS_ENCRYPTED) {
920 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
921 "is encrypted.");
922 goto unm_err_out;
923 }
924 if (a->flags & ATTR_IS_SPARSE) {
925 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
926 "is sparse.");
927 goto unm_err_out;
928 }
929 if (a->flags & ATTR_COMPRESSION_MASK) {
930 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
931 "is compressed.");
932 goto unm_err_out;
933 }
934 if (a->data.non_resident.lowest_vcn) {
935 ntfs_error(vi->i_sb, "First extent of "
936 "$INDEX_ALLOCATION attribute has non "
937 "zero lowest_vcn.");
938 goto unm_err_out;
939 }
940 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
941 ni->initialized_size = sle64_to_cpu(
942 a->data.non_resident.initialized_size);
943 ni->allocated_size = sle64_to_cpu(
944 a->data.non_resident.allocated_size);
945 /*
946 * We are done with the mft record, so we release it. Otherwise
947 * we would deadlock in ntfs_attr_iget().
948 */
949 ntfs_attr_put_search_ctx(ctx);
950 unmap_mft_record(ni);
951 m = NULL;
952 ctx = NULL;
953 /* Get the index bitmap attribute inode. */
954 bvi = ntfs_attr_iget(vi, AT_BITMAP, I30, 4);
955 if (IS_ERR(bvi)) {
956 ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
957 err = PTR_ERR(bvi);
958 goto unm_err_out;
959 }
960 bni = NTFS_I(bvi);
961 if (NInoCompressed(bni) || NInoEncrypted(bni) ||
962 NInoSparse(bni)) {
963 ntfs_error(vi->i_sb, "$BITMAP attribute is compressed "
964 "and/or encrypted and/or sparse.");
965 goto iput_unm_err_out;
966 }
967 /* Consistency check bitmap size vs. index allocation size. */
968 bvi_size = i_size_read(bvi);
969 if ((bvi_size << 3) < (vi->i_size >>
970 ni->itype.index.block_size_bits)) {
971 ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) "
972 "for index allocation (0x%llx).",
973 bvi_size << 3, vi->i_size);
974 goto iput_unm_err_out;
975 }
976 /* No longer need the bitmap attribute inode. */
977 iput(bvi);
978skip_large_dir_stuff:
979 /* Setup the operations for this inode. */
980 vi->i_op = &ntfs_dir_inode_ops;
981 vi->i_fop = &ntfs_dir_ops;
982 vi->i_mapping->a_ops = &ntfs_mst_aops;
983 } else {
984 /* It is a file. */
985 ntfs_attr_reinit_search_ctx(ctx);
986
987 /* Setup the data attribute, even if not present. */
988 ni->type = AT_DATA;
989 ni->name = NULL;
990 ni->name_len = 0;
991
992 /* Find first extent of the unnamed data attribute. */
993 err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, 0, NULL, 0, ctx);
994 if (unlikely(err)) {
995 vi->i_size = ni->initialized_size =
996 ni->allocated_size = 0;
997 if (err != -ENOENT) {
998 ntfs_error(vi->i_sb, "Failed to lookup $DATA "
999 "attribute.");
1000 goto unm_err_out;
1001 }
1002 /*
1003 * FILE_Secure does not have an unnamed $DATA
1004 * attribute, so we special case it here.
1005 */
1006 if (vi->i_ino == FILE_Secure)
1007 goto no_data_attr_special_case;
1008 /*
1009 * Most if not all the system files in the $Extend
1010 * system directory do not have unnamed data
1011 * attributes so we need to check if the parent
1012 * directory of the file is FILE_Extend and if it is
1013 * ignore this error. To do this we need to get the
1014 * name of this inode from the mft record as the name
1015 * contains the back reference to the parent directory.
1016 */
1017 if (ntfs_is_extended_system_file(ctx) > 0)
1018 goto no_data_attr_special_case;
1019 // FIXME: File is corrupt! Hot-fix with empty data
1020 // attribute if recovery option is set.
1021 ntfs_error(vi->i_sb, "$DATA attribute is missing.");
1022 goto unm_err_out;
1023 }
1024 a = ctx->attr;
1025 /* Setup the state. */
1026 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1027 if (a->flags & ATTR_COMPRESSION_MASK) {
1028 NInoSetCompressed(ni);
1029 if (vol->cluster_size > 4096) {
1030 ntfs_error(vi->i_sb, "Found "
1031 "compressed data but "
1032 "compression is "
1033 "disabled due to "
1034 "cluster size (%i) > "
1035 "4kiB.",
1036 vol->cluster_size);
1037 goto unm_err_out;
1038 }
1039 if ((a->flags & ATTR_COMPRESSION_MASK)
1040 != ATTR_IS_COMPRESSED) {
1041 ntfs_error(vi->i_sb, "Found unknown "
1042 "compression method "
1043 "or corrupt file.");
1044 goto unm_err_out;
1045 }
1046 }
1047 if (a->flags & ATTR_IS_SPARSE)
1048 NInoSetSparse(ni);
1049 }
1050 if (a->flags & ATTR_IS_ENCRYPTED) {
1051 if (NInoCompressed(ni)) {
1052 ntfs_error(vi->i_sb, "Found encrypted and "
1053 "compressed data.");
1054 goto unm_err_out;
1055 }
1056 NInoSetEncrypted(ni);
1057 }
1058 if (a->non_resident) {
1059 NInoSetNonResident(ni);
1060 if (NInoCompressed(ni) || NInoSparse(ni)) {
1061 if (NInoCompressed(ni) && a->data.non_resident.
1062 compression_unit != 4) {
1063 ntfs_error(vi->i_sb, "Found "
1064 "non-standard "
1065 "compression unit (%u "
1066 "instead of 4). "
1067 "Cannot handle this.",
1068 a->data.non_resident.
1069 compression_unit);
1070 err = -EOPNOTSUPP;
1071 goto unm_err_out;
1072 }
1073 if (a->data.non_resident.compression_unit) {
1074 ni->itype.compressed.block_size = 1U <<
1075 (a->data.non_resident.
1076 compression_unit +
1077 vol->cluster_size_bits);
1078 ni->itype.compressed.block_size_bits =
1079 ffs(ni->itype.
1080 compressed.
1081 block_size) - 1;
1082 ni->itype.compressed.block_clusters =
1083 1U << a->data.
1084 non_resident.
1085 compression_unit;
1086 } else {
1087 ni->itype.compressed.block_size = 0;
1088 ni->itype.compressed.block_size_bits =
1089 0;
1090 ni->itype.compressed.block_clusters =
1091 0;
1092 }
1093 ni->itype.compressed.size = sle64_to_cpu(
1094 a->data.non_resident.
1095 compressed_size);
1096 }
1097 if (a->data.non_resident.lowest_vcn) {
1098 ntfs_error(vi->i_sb, "First extent of $DATA "
1099 "attribute has non zero "
1100 "lowest_vcn.");
1101 goto unm_err_out;
1102 }
1103 vi->i_size = sle64_to_cpu(
1104 a->data.non_resident.data_size);
1105 ni->initialized_size = sle64_to_cpu(
1106 a->data.non_resident.initialized_size);
1107 ni->allocated_size = sle64_to_cpu(
1108 a->data.non_resident.allocated_size);
1109 } else { /* Resident attribute. */
1110 vi->i_size = ni->initialized_size = le32_to_cpu(
1111 a->data.resident.value_length);
1112 ni->allocated_size = le32_to_cpu(a->length) -
1113 le16_to_cpu(
1114 a->data.resident.value_offset);
1115 if (vi->i_size > ni->allocated_size) {
1116 ntfs_error(vi->i_sb, "Resident data attribute "
1117 "is corrupt (size exceeds "
1118 "allocation).");
1119 goto unm_err_out;
1120 }
1121 }
1122no_data_attr_special_case:
1123 /* We are done with the mft record, so we release it. */
1124 ntfs_attr_put_search_ctx(ctx);
1125 unmap_mft_record(ni);
1126 m = NULL;
1127 ctx = NULL;
1128 /* Setup the operations for this inode. */
1129 vi->i_op = &ntfs_file_inode_ops;
1130 vi->i_fop = &ntfs_file_ops;
1131 vi->i_mapping->a_ops = &ntfs_normal_aops;
1132 if (NInoMstProtected(ni))
1133 vi->i_mapping->a_ops = &ntfs_mst_aops;
1134 else if (NInoCompressed(ni))
1135 vi->i_mapping->a_ops = &ntfs_compressed_aops;
1136 }
1137 /*
1138 * The number of 512-byte blocks used on disk (for stat). This is in so
1139 * far inaccurate as it doesn't account for any named streams or other
1140 * special non-resident attributes, but that is how Windows works, too,
1141 * so we are at least consistent with Windows, if not entirely
1142 * consistent with the Linux Way. Doing it the Linux Way would cause a
1143 * significant slowdown as it would involve iterating over all
1144 * attributes in the mft record and adding the allocated/compressed
1145 * sizes of all non-resident attributes present to give us the Linux
1146 * correct size that should go into i_blocks (after division by 512).
1147 */
1148 if (S_ISREG(vi->i_mode) && (NInoCompressed(ni) || NInoSparse(ni)))
1149 vi->i_blocks = ni->itype.compressed.size >> 9;
1150 else
1151 vi->i_blocks = ni->allocated_size >> 9;
1152 ntfs_debug("Done.");
1153 return 0;
1154iput_unm_err_out:
1155 iput(bvi);
1156unm_err_out:
1157 if (!err)
1158 err = -EIO;
1159 if (ctx)
1160 ntfs_attr_put_search_ctx(ctx);
1161 if (m)
1162 unmap_mft_record(ni);
1163err_out:
1164 ntfs_error(vol->sb, "Failed with error code %i. Marking corrupt "
1165 "inode 0x%lx as bad. Run chkdsk.", err, vi->i_ino);
1166 make_bad_inode(vi);
1167 if (err != -EOPNOTSUPP && err != -ENOMEM)
1168 NVolSetErrors(vol);
1169 return err;
1170}
1171
1172/**
1173 * ntfs_read_locked_attr_inode - read an attribute inode from its base inode
1174 * @base_vi: base inode
1175 * @vi: attribute inode to read
1176 *
1177 * ntfs_read_locked_attr_inode() is called from ntfs_attr_iget() to read the
1178 * attribute inode described by @vi into memory from the base mft record
1179 * described by @base_ni.
1180 *
1181 * ntfs_read_locked_attr_inode() maps, pins and locks the base inode for
1182 * reading and looks up the attribute described by @vi before setting up the
1183 * necessary fields in @vi as well as initializing the ntfs inode.
1184 *
1185 * Q: What locks are held when the function is called?
1186 * A: i_state has I_NEW set, hence the inode is locked, also
1187 * i_count is set to 1, so it is not going to go away
1188 *
1189 * Return 0 on success and -errno on error. In the error case, the inode will
1190 * have had make_bad_inode() executed on it.
1191 *
1192 * Note this cannot be called for AT_INDEX_ALLOCATION.
1193 */
1194static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi)
1195{
1196 ntfs_volume *vol = NTFS_SB(vi->i_sb);
1197 ntfs_inode *ni, *base_ni;
1198 MFT_RECORD *m;
1199 ATTR_RECORD *a;
1200 ntfs_attr_search_ctx *ctx;
1201 int err = 0;
1202
1203 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1204
1205 ntfs_init_big_inode(vi);
1206
1207 ni = NTFS_I(vi);
1208 base_ni = NTFS_I(base_vi);
1209
1210 /* Just mirror the values from the base inode. */
1211 vi->i_uid = base_vi->i_uid;
1212 vi->i_gid = base_vi->i_gid;
1213 set_nlink(vi, base_vi->i_nlink);
1214 vi->i_mtime = base_vi->i_mtime;
1215 vi->i_ctime = base_vi->i_ctime;
1216 vi->i_atime = base_vi->i_atime;
1217 vi->i_generation = ni->seq_no = base_ni->seq_no;
1218
1219 /* Set inode type to zero but preserve permissions. */
1220 vi->i_mode = base_vi->i_mode & ~S_IFMT;
1221
1222 m = map_mft_record(base_ni);
1223 if (IS_ERR(m)) {
1224 err = PTR_ERR(m);
1225 goto err_out;
1226 }
1227 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1228 if (!ctx) {
1229 err = -ENOMEM;
1230 goto unm_err_out;
1231 }
1232 /* Find the attribute. */
1233 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1234 CASE_SENSITIVE, 0, NULL, 0, ctx);
1235 if (unlikely(err))
1236 goto unm_err_out;
1237 a = ctx->attr;
1238 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1239 if (a->flags & ATTR_COMPRESSION_MASK) {
1240 NInoSetCompressed(ni);
1241 if ((ni->type != AT_DATA) || (ni->type == AT_DATA &&
1242 ni->name_len)) {
1243 ntfs_error(vi->i_sb, "Found compressed "
1244 "non-data or named data "
1245 "attribute. Please report "
1246 "you saw this message to "
1247 "linux-ntfs-dev@lists."
1248 "sourceforge.net");
1249 goto unm_err_out;
1250 }
1251 if (vol->cluster_size > 4096) {
1252 ntfs_error(vi->i_sb, "Found compressed "
1253 "attribute but compression is "
1254 "disabled due to cluster size "
1255 "(%i) > 4kiB.",
1256 vol->cluster_size);
1257 goto unm_err_out;
1258 }
1259 if ((a->flags & ATTR_COMPRESSION_MASK) !=
1260 ATTR_IS_COMPRESSED) {
1261 ntfs_error(vi->i_sb, "Found unknown "
1262 "compression method.");
1263 goto unm_err_out;
1264 }
1265 }
1266 /*
1267 * The compressed/sparse flag set in an index root just means
1268 * to compress all files.
1269 */
1270 if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1271 ntfs_error(vi->i_sb, "Found mst protected attribute "
1272 "but the attribute is %s. Please "
1273 "report you saw this message to "
1274 "linux-ntfs-dev@lists.sourceforge.net",
1275 NInoCompressed(ni) ? "compressed" :
1276 "sparse");
1277 goto unm_err_out;
1278 }
1279 if (a->flags & ATTR_IS_SPARSE)
1280 NInoSetSparse(ni);
1281 }
1282 if (a->flags & ATTR_IS_ENCRYPTED) {
1283 if (NInoCompressed(ni)) {
1284 ntfs_error(vi->i_sb, "Found encrypted and compressed "
1285 "data.");
1286 goto unm_err_out;
1287 }
1288 /*
1289 * The encryption flag set in an index root just means to
1290 * encrypt all files.
1291 */
1292 if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1293 ntfs_error(vi->i_sb, "Found mst protected attribute "
1294 "but the attribute is encrypted. "
1295 "Please report you saw this message "
1296 "to linux-ntfs-dev@lists.sourceforge."
1297 "net");
1298 goto unm_err_out;
1299 }
1300 if (ni->type != AT_DATA) {
1301 ntfs_error(vi->i_sb, "Found encrypted non-data "
1302 "attribute.");
1303 goto unm_err_out;
1304 }
1305 NInoSetEncrypted(ni);
1306 }
1307 if (!a->non_resident) {
1308 /* Ensure the attribute name is placed before the value. */
1309 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1310 le16_to_cpu(a->data.resident.value_offset)))) {
1311 ntfs_error(vol->sb, "Attribute name is placed after "
1312 "the attribute value.");
1313 goto unm_err_out;
1314 }
1315 if (NInoMstProtected(ni)) {
1316 ntfs_error(vi->i_sb, "Found mst protected attribute "
1317 "but the attribute is resident. "
1318 "Please report you saw this message to "
1319 "linux-ntfs-dev@lists.sourceforge.net");
1320 goto unm_err_out;
1321 }
1322 vi->i_size = ni->initialized_size = le32_to_cpu(
1323 a->data.resident.value_length);
1324 ni->allocated_size = le32_to_cpu(a->length) -
1325 le16_to_cpu(a->data.resident.value_offset);
1326 if (vi->i_size > ni->allocated_size) {
1327 ntfs_error(vi->i_sb, "Resident attribute is corrupt "
1328 "(size exceeds allocation).");
1329 goto unm_err_out;
1330 }
1331 } else {
1332 NInoSetNonResident(ni);
1333 /*
1334 * Ensure the attribute name is placed before the mapping pairs
1335 * array.
1336 */
1337 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1338 le16_to_cpu(
1339 a->data.non_resident.mapping_pairs_offset)))) {
1340 ntfs_error(vol->sb, "Attribute name is placed after "
1341 "the mapping pairs array.");
1342 goto unm_err_out;
1343 }
1344 if (NInoCompressed(ni) || NInoSparse(ni)) {
1345 if (NInoCompressed(ni) && a->data.non_resident.
1346 compression_unit != 4) {
1347 ntfs_error(vi->i_sb, "Found non-standard "
1348 "compression unit (%u instead "
1349 "of 4). Cannot handle this.",
1350 a->data.non_resident.
1351 compression_unit);
1352 err = -EOPNOTSUPP;
1353 goto unm_err_out;
1354 }
1355 if (a->data.non_resident.compression_unit) {
1356 ni->itype.compressed.block_size = 1U <<
1357 (a->data.non_resident.
1358 compression_unit +
1359 vol->cluster_size_bits);
1360 ni->itype.compressed.block_size_bits =
1361 ffs(ni->itype.compressed.
1362 block_size) - 1;
1363 ni->itype.compressed.block_clusters = 1U <<
1364 a->data.non_resident.
1365 compression_unit;
1366 } else {
1367 ni->itype.compressed.block_size = 0;
1368 ni->itype.compressed.block_size_bits = 0;
1369 ni->itype.compressed.block_clusters = 0;
1370 }
1371 ni->itype.compressed.size = sle64_to_cpu(
1372 a->data.non_resident.compressed_size);
1373 }
1374 if (a->data.non_resident.lowest_vcn) {
1375 ntfs_error(vi->i_sb, "First extent of attribute has "
1376 "non-zero lowest_vcn.");
1377 goto unm_err_out;
1378 }
1379 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1380 ni->initialized_size = sle64_to_cpu(
1381 a->data.non_resident.initialized_size);
1382 ni->allocated_size = sle64_to_cpu(
1383 a->data.non_resident.allocated_size);
1384 }
1385 vi->i_mapping->a_ops = &ntfs_normal_aops;
1386 if (NInoMstProtected(ni))
1387 vi->i_mapping->a_ops = &ntfs_mst_aops;
1388 else if (NInoCompressed(ni))
1389 vi->i_mapping->a_ops = &ntfs_compressed_aops;
1390 if ((NInoCompressed(ni) || NInoSparse(ni)) && ni->type != AT_INDEX_ROOT)
1391 vi->i_blocks = ni->itype.compressed.size >> 9;
1392 else
1393 vi->i_blocks = ni->allocated_size >> 9;
1394 /*
1395 * Make sure the base inode does not go away and attach it to the
1396 * attribute inode.
1397 */
1398 igrab(base_vi);
1399 ni->ext.base_ntfs_ino = base_ni;
1400 ni->nr_extents = -1;
1401
1402 ntfs_attr_put_search_ctx(ctx);
1403 unmap_mft_record(base_ni);
1404
1405 ntfs_debug("Done.");
1406 return 0;
1407
1408unm_err_out:
1409 if (!err)
1410 err = -EIO;
1411 if (ctx)
1412 ntfs_attr_put_search_ctx(ctx);
1413 unmap_mft_record(base_ni);
1414err_out:
1415 ntfs_error(vol->sb, "Failed with error code %i while reading attribute "
1416 "inode (mft_no 0x%lx, type 0x%x, name_len %i). "
1417 "Marking corrupt inode and base inode 0x%lx as bad. "
1418 "Run chkdsk.", err, vi->i_ino, ni->type, ni->name_len,
1419 base_vi->i_ino);
1420 make_bad_inode(vi);
1421 if (err != -ENOMEM)
1422 NVolSetErrors(vol);
1423 return err;
1424}
1425
1426/**
1427 * ntfs_read_locked_index_inode - read an index inode from its base inode
1428 * @base_vi: base inode
1429 * @vi: index inode to read
1430 *
1431 * ntfs_read_locked_index_inode() is called from ntfs_index_iget() to read the
1432 * index inode described by @vi into memory from the base mft record described
1433 * by @base_ni.
1434 *
1435 * ntfs_read_locked_index_inode() maps, pins and locks the base inode for
1436 * reading and looks up the attributes relating to the index described by @vi
1437 * before setting up the necessary fields in @vi as well as initializing the
1438 * ntfs inode.
1439 *
1440 * Note, index inodes are essentially attribute inodes (NInoAttr() is true)
1441 * with the attribute type set to AT_INDEX_ALLOCATION. Apart from that, they
1442 * are setup like directory inodes since directories are a special case of
1443 * indices ao they need to be treated in much the same way. Most importantly,
1444 * for small indices the index allocation attribute might not actually exist.
1445 * However, the index root attribute always exists but this does not need to
1446 * have an inode associated with it and this is why we define a new inode type
1447 * index. Also, like for directories, we need to have an attribute inode for
1448 * the bitmap attribute corresponding to the index allocation attribute and we
1449 * can store this in the appropriate field of the inode, just like we do for
1450 * normal directory inodes.
1451 *
1452 * Q: What locks are held when the function is called?
1453 * A: i_state has I_NEW set, hence the inode is locked, also
1454 * i_count is set to 1, so it is not going to go away
1455 *
1456 * Return 0 on success and -errno on error. In the error case, the inode will
1457 * have had make_bad_inode() executed on it.
1458 */
1459static int ntfs_read_locked_index_inode(struct inode *base_vi, struct inode *vi)
1460{
1461 loff_t bvi_size;
1462 ntfs_volume *vol = NTFS_SB(vi->i_sb);
1463 ntfs_inode *ni, *base_ni, *bni;
1464 struct inode *bvi;
1465 MFT_RECORD *m;
1466 ATTR_RECORD *a;
1467 ntfs_attr_search_ctx *ctx;
1468 INDEX_ROOT *ir;
1469 u8 *ir_end, *index_end;
1470 int err = 0;
1471
1472 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1473 ntfs_init_big_inode(vi);
1474 ni = NTFS_I(vi);
1475 base_ni = NTFS_I(base_vi);
1476 /* Just mirror the values from the base inode. */
1477 vi->i_uid = base_vi->i_uid;
1478 vi->i_gid = base_vi->i_gid;
1479 set_nlink(vi, base_vi->i_nlink);
1480 vi->i_mtime = base_vi->i_mtime;
1481 vi->i_ctime = base_vi->i_ctime;
1482 vi->i_atime = base_vi->i_atime;
1483 vi->i_generation = ni->seq_no = base_ni->seq_no;
1484 /* Set inode type to zero but preserve permissions. */
1485 vi->i_mode = base_vi->i_mode & ~S_IFMT;
1486 /* Map the mft record for the base inode. */
1487 m = map_mft_record(base_ni);
1488 if (IS_ERR(m)) {
1489 err = PTR_ERR(m);
1490 goto err_out;
1491 }
1492 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1493 if (!ctx) {
1494 err = -ENOMEM;
1495 goto unm_err_out;
1496 }
1497 /* Find the index root attribute. */
1498 err = ntfs_attr_lookup(AT_INDEX_ROOT, ni->name, ni->name_len,
1499 CASE_SENSITIVE, 0, NULL, 0, ctx);
1500 if (unlikely(err)) {
1501 if (err == -ENOENT)
1502 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
1503 "missing.");
1504 goto unm_err_out;
1505 }
1506 a = ctx->attr;
1507 /* Set up the state. */
1508 if (unlikely(a->non_resident)) {
1509 ntfs_error(vol->sb, "$INDEX_ROOT attribute is not resident.");
1510 goto unm_err_out;
1511 }
1512 /* Ensure the attribute name is placed before the value. */
1513 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1514 le16_to_cpu(a->data.resident.value_offset)))) {
1515 ntfs_error(vol->sb, "$INDEX_ROOT attribute name is placed "
1516 "after the attribute value.");
1517 goto unm_err_out;
1518 }
1519 /*
1520 * Compressed/encrypted/sparse index root is not allowed, except for
1521 * directories of course but those are not dealt with here.
1522 */
1523 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_ENCRYPTED |
1524 ATTR_IS_SPARSE)) {
1525 ntfs_error(vi->i_sb, "Found compressed/encrypted/sparse index "
1526 "root attribute.");
1527 goto unm_err_out;
1528 }
1529 ir = (INDEX_ROOT*)((u8*)a + le16_to_cpu(a->data.resident.value_offset));
1530 ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
1531 if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
1532 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is corrupt.");
1533 goto unm_err_out;
1534 }
1535 index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
1536 if (index_end > ir_end) {
1537 ntfs_error(vi->i_sb, "Index is corrupt.");
1538 goto unm_err_out;
1539 }
1540 if (ir->type) {
1541 ntfs_error(vi->i_sb, "Index type is not 0 (type is 0x%x).",
1542 le32_to_cpu(ir->type));
1543 goto unm_err_out;
1544 }
1545 ni->itype.index.collation_rule = ir->collation_rule;
1546 ntfs_debug("Index collation rule is 0x%x.",
1547 le32_to_cpu(ir->collation_rule));
1548 ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
1549 if (!is_power_of_2(ni->itype.index.block_size)) {
1550 ntfs_error(vi->i_sb, "Index block size (%u) is not a power of "
1551 "two.", ni->itype.index.block_size);
1552 goto unm_err_out;
1553 }
1554 if (ni->itype.index.block_size > PAGE_SIZE) {
1555 ntfs_error(vi->i_sb, "Index block size (%u) > PAGE_SIZE "
1556 "(%ld) is not supported. Sorry.",
1557 ni->itype.index.block_size, PAGE_SIZE);
1558 err = -EOPNOTSUPP;
1559 goto unm_err_out;
1560 }
1561 if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
1562 ntfs_error(vi->i_sb, "Index block size (%u) < NTFS_BLOCK_SIZE "
1563 "(%i) is not supported. Sorry.",
1564 ni->itype.index.block_size, NTFS_BLOCK_SIZE);
1565 err = -EOPNOTSUPP;
1566 goto unm_err_out;
1567 }
1568 ni->itype.index.block_size_bits = ffs(ni->itype.index.block_size) - 1;
1569 /* Determine the size of a vcn in the index. */
1570 if (vol->cluster_size <= ni->itype.index.block_size) {
1571 ni->itype.index.vcn_size = vol->cluster_size;
1572 ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
1573 } else {
1574 ni->itype.index.vcn_size = vol->sector_size;
1575 ni->itype.index.vcn_size_bits = vol->sector_size_bits;
1576 }
1577 /* Check for presence of index allocation attribute. */
1578 if (!(ir->index.flags & LARGE_INDEX)) {
1579 /* No index allocation. */
1580 vi->i_size = ni->initialized_size = ni->allocated_size = 0;
1581 /* We are done with the mft record, so we release it. */
1582 ntfs_attr_put_search_ctx(ctx);
1583 unmap_mft_record(base_ni);
1584 m = NULL;
1585 ctx = NULL;
1586 goto skip_large_index_stuff;
1587 } /* LARGE_INDEX: Index allocation present. Setup state. */
1588 NInoSetIndexAllocPresent(ni);
1589 /* Find index allocation attribute. */
1590 ntfs_attr_reinit_search_ctx(ctx);
1591 err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, ni->name, ni->name_len,
1592 CASE_SENSITIVE, 0, NULL, 0, ctx);
1593 if (unlikely(err)) {
1594 if (err == -ENOENT)
1595 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1596 "not present but $INDEX_ROOT "
1597 "indicated it is.");
1598 else
1599 ntfs_error(vi->i_sb, "Failed to lookup "
1600 "$INDEX_ALLOCATION attribute.");
1601 goto unm_err_out;
1602 }
1603 a = ctx->attr;
1604 if (!a->non_resident) {
1605 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1606 "resident.");
1607 goto unm_err_out;
1608 }
1609 /*
1610 * Ensure the attribute name is placed before the mapping pairs array.
1611 */
1612 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1613 le16_to_cpu(
1614 a->data.non_resident.mapping_pairs_offset)))) {
1615 ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name is "
1616 "placed after the mapping pairs array.");
1617 goto unm_err_out;
1618 }
1619 if (a->flags & ATTR_IS_ENCRYPTED) {
1620 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1621 "encrypted.");
1622 goto unm_err_out;
1623 }
1624 if (a->flags & ATTR_IS_SPARSE) {
1625 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is sparse.");
1626 goto unm_err_out;
1627 }
1628 if (a->flags & ATTR_COMPRESSION_MASK) {
1629 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1630 "compressed.");
1631 goto unm_err_out;
1632 }
1633 if (a->data.non_resident.lowest_vcn) {
1634 ntfs_error(vi->i_sb, "First extent of $INDEX_ALLOCATION "
1635 "attribute has non zero lowest_vcn.");
1636 goto unm_err_out;
1637 }
1638 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1639 ni->initialized_size = sle64_to_cpu(
1640 a->data.non_resident.initialized_size);
1641 ni->allocated_size = sle64_to_cpu(a->data.non_resident.allocated_size);
1642 /*
1643 * We are done with the mft record, so we release it. Otherwise
1644 * we would deadlock in ntfs_attr_iget().
1645 */
1646 ntfs_attr_put_search_ctx(ctx);
1647 unmap_mft_record(base_ni);
1648 m = NULL;
1649 ctx = NULL;
1650 /* Get the index bitmap attribute inode. */
1651 bvi = ntfs_attr_iget(base_vi, AT_BITMAP, ni->name, ni->name_len);
1652 if (IS_ERR(bvi)) {
1653 ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
1654 err = PTR_ERR(bvi);
1655 goto unm_err_out;
1656 }
1657 bni = NTFS_I(bvi);
1658 if (NInoCompressed(bni) || NInoEncrypted(bni) ||
1659 NInoSparse(bni)) {
1660 ntfs_error(vi->i_sb, "$BITMAP attribute is compressed and/or "
1661 "encrypted and/or sparse.");
1662 goto iput_unm_err_out;
1663 }
1664 /* Consistency check bitmap size vs. index allocation size. */
1665 bvi_size = i_size_read(bvi);
1666 if ((bvi_size << 3) < (vi->i_size >> ni->itype.index.block_size_bits)) {
1667 ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) for "
1668 "index allocation (0x%llx).", bvi_size << 3,
1669 vi->i_size);
1670 goto iput_unm_err_out;
1671 }
1672 iput(bvi);
1673skip_large_index_stuff:
1674 /* Setup the operations for this index inode. */
1675 vi->i_mapping->a_ops = &ntfs_mst_aops;
1676 vi->i_blocks = ni->allocated_size >> 9;
1677 /*
1678 * Make sure the base inode doesn't go away and attach it to the
1679 * index inode.
1680 */
1681 igrab(base_vi);
1682 ni->ext.base_ntfs_ino = base_ni;
1683 ni->nr_extents = -1;
1684
1685 ntfs_debug("Done.");
1686 return 0;
1687iput_unm_err_out:
1688 iput(bvi);
1689unm_err_out:
1690 if (!err)
1691 err = -EIO;
1692 if (ctx)
1693 ntfs_attr_put_search_ctx(ctx);
1694 if (m)
1695 unmap_mft_record(base_ni);
1696err_out:
1697 ntfs_error(vi->i_sb, "Failed with error code %i while reading index "
1698 "inode (mft_no 0x%lx, name_len %i.", err, vi->i_ino,
1699 ni->name_len);
1700 make_bad_inode(vi);
1701 if (err != -EOPNOTSUPP && err != -ENOMEM)
1702 NVolSetErrors(vol);
1703 return err;
1704}
1705
1706/*
1707 * The MFT inode has special locking, so teach the lock validator
1708 * about this by splitting off the locking rules of the MFT from
1709 * the locking rules of other inodes. The MFT inode can never be
1710 * accessed from the VFS side (or even internally), only by the
1711 * map_mft functions.
1712 */
1713static struct lock_class_key mft_ni_runlist_lock_key, mft_ni_mrec_lock_key;
1714
1715/**
1716 * ntfs_read_inode_mount - special read_inode for mount time use only
1717 * @vi: inode to read
1718 *
1719 * Read inode FILE_MFT at mount time, only called with super_block lock
1720 * held from within the read_super() code path.
1721 *
1722 * This function exists because when it is called the page cache for $MFT/$DATA
1723 * is not initialized and hence we cannot get at the contents of mft records
1724 * by calling map_mft_record*().
1725 *
1726 * Further it needs to cope with the circular references problem, i.e. cannot
1727 * load any attributes other than $ATTRIBUTE_LIST until $DATA is loaded, because
1728 * we do not know where the other extent mft records are yet and again, because
1729 * we cannot call map_mft_record*() yet. Obviously this applies only when an
1730 * attribute list is actually present in $MFT inode.
1731 *
1732 * We solve these problems by starting with the $DATA attribute before anything
1733 * else and iterating using ntfs_attr_lookup($DATA) over all extents. As each
1734 * extent is found, we ntfs_mapping_pairs_decompress() including the implied
1735 * ntfs_runlists_merge(). Each step of the iteration necessarily provides
1736 * sufficient information for the next step to complete.
1737 *
1738 * This should work but there are two possible pit falls (see inline comments
1739 * below), but only time will tell if they are real pits or just smoke...
1740 */
1741int ntfs_read_inode_mount(struct inode *vi)
1742{
1743 VCN next_vcn, last_vcn, highest_vcn;
1744 s64 block;
1745 struct super_block *sb = vi->i_sb;
1746 ntfs_volume *vol = NTFS_SB(sb);
1747 struct buffer_head *bh;
1748 ntfs_inode *ni;
1749 MFT_RECORD *m = NULL;
1750 ATTR_RECORD *a;
1751 ntfs_attr_search_ctx *ctx;
1752 unsigned int i, nr_blocks;
1753 int err;
1754
1755 ntfs_debug("Entering.");
1756
1757 /* Initialize the ntfs specific part of @vi. */
1758 ntfs_init_big_inode(vi);
1759
1760 ni = NTFS_I(vi);
1761
1762 /* Setup the data attribute. It is special as it is mst protected. */
1763 NInoSetNonResident(ni);
1764 NInoSetMstProtected(ni);
1765 NInoSetSparseDisabled(ni);
1766 ni->type = AT_DATA;
1767 ni->name = NULL;
1768 ni->name_len = 0;
1769 /*
1770 * This sets up our little cheat allowing us to reuse the async read io
1771 * completion handler for directories.
1772 */
1773 ni->itype.index.block_size = vol->mft_record_size;
1774 ni->itype.index.block_size_bits = vol->mft_record_size_bits;
1775
1776 /* Very important! Needed to be able to call map_mft_record*(). */
1777 vol->mft_ino = vi;
1778
1779 /* Allocate enough memory to read the first mft record. */
1780 if (vol->mft_record_size > 64 * 1024) {
1781 ntfs_error(sb, "Unsupported mft record size %i (max 64kiB).",
1782 vol->mft_record_size);
1783 goto err_out;
1784 }
1785 i = vol->mft_record_size;
1786 if (i < sb->s_blocksize)
1787 i = sb->s_blocksize;
1788 m = (MFT_RECORD*)ntfs_malloc_nofs(i);
1789 if (!m) {
1790 ntfs_error(sb, "Failed to allocate buffer for $MFT record 0.");
1791 goto err_out;
1792 }
1793
1794 /* Determine the first block of the $MFT/$DATA attribute. */
1795 block = vol->mft_lcn << vol->cluster_size_bits >>
1796 sb->s_blocksize_bits;
1797 nr_blocks = vol->mft_record_size >> sb->s_blocksize_bits;
1798 if (!nr_blocks)
1799 nr_blocks = 1;
1800
1801 /* Load $MFT/$DATA's first mft record. */
1802 for (i = 0; i < nr_blocks; i++) {
1803 bh = sb_bread(sb, block++);
1804 if (!bh) {
1805 ntfs_error(sb, "Device read failed.");
1806 goto err_out;
1807 }
1808 memcpy((char*)m + (i << sb->s_blocksize_bits), bh->b_data,
1809 sb->s_blocksize);
1810 brelse(bh);
1811 }
1812
1813 /* Apply the mst fixups. */
1814 if (post_read_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size)) {
1815 /* FIXME: Try to use the $MFTMirr now. */
1816 ntfs_error(sb, "MST fixup failed. $MFT is corrupt.");
1817 goto err_out;
1818 }
1819
1820 /* Need this to sanity check attribute list references to $MFT. */
1821 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
1822
1823 /* Provides readpage() for map_mft_record(). */
1824 vi->i_mapping->a_ops = &ntfs_mst_aops;
1825
1826 ctx = ntfs_attr_get_search_ctx(ni, m);
1827 if (!ctx) {
1828 err = -ENOMEM;
1829 goto err_out;
1830 }
1831
1832 /* Find the attribute list attribute if present. */
1833 err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
1834 if (err) {
1835 if (unlikely(err != -ENOENT)) {
1836 ntfs_error(sb, "Failed to lookup attribute list "
1837 "attribute. You should run chkdsk.");
1838 goto put_err_out;
1839 }
1840 } else /* if (!err) */ {
1841 ATTR_LIST_ENTRY *al_entry, *next_al_entry;
1842 u8 *al_end;
1843 static const char *es = " Not allowed. $MFT is corrupt. "
1844 "You should run chkdsk.";
1845
1846 ntfs_debug("Attribute list attribute found in $MFT.");
1847 NInoSetAttrList(ni);
1848 a = ctx->attr;
1849 if (a->flags & ATTR_COMPRESSION_MASK) {
1850 ntfs_error(sb, "Attribute list attribute is "
1851 "compressed.%s", es);
1852 goto put_err_out;
1853 }
1854 if (a->flags & ATTR_IS_ENCRYPTED ||
1855 a->flags & ATTR_IS_SPARSE) {
1856 if (a->non_resident) {
1857 ntfs_error(sb, "Non-resident attribute list "
1858 "attribute is encrypted/"
1859 "sparse.%s", es);
1860 goto put_err_out;
1861 }
1862 ntfs_warning(sb, "Resident attribute list attribute "
1863 "in $MFT system file is marked "
1864 "encrypted/sparse which is not true. "
1865 "However, Windows allows this and "
1866 "chkdsk does not detect or correct it "
1867 "so we will just ignore the invalid "
1868 "flags and pretend they are not set.");
1869 }
1870 /* Now allocate memory for the attribute list. */
1871 ni->attr_list_size = (u32)ntfs_attr_size(a);
1872 ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
1873 if (!ni->attr_list) {
1874 ntfs_error(sb, "Not enough memory to allocate buffer "
1875 "for attribute list.");
1876 goto put_err_out;
1877 }
1878 if (a->non_resident) {
1879 NInoSetAttrListNonResident(ni);
1880 if (a->data.non_resident.lowest_vcn) {
1881 ntfs_error(sb, "Attribute list has non zero "
1882 "lowest_vcn. $MFT is corrupt. "
1883 "You should run chkdsk.");
1884 goto put_err_out;
1885 }
1886 /* Setup the runlist. */
1887 ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
1888 a, NULL);
1889 if (IS_ERR(ni->attr_list_rl.rl)) {
1890 err = PTR_ERR(ni->attr_list_rl.rl);
1891 ni->attr_list_rl.rl = NULL;
1892 ntfs_error(sb, "Mapping pairs decompression "
1893 "failed with error code %i.",
1894 -err);
1895 goto put_err_out;
1896 }
1897 /* Now load the attribute list. */
1898 if ((err = load_attribute_list(vol, &ni->attr_list_rl,
1899 ni->attr_list, ni->attr_list_size,
1900 sle64_to_cpu(a->data.
1901 non_resident.initialized_size)))) {
1902 ntfs_error(sb, "Failed to load attribute list "
1903 "attribute with error code %i.",
1904 -err);
1905 goto put_err_out;
1906 }
1907 } else /* if (!ctx.attr->non_resident) */ {
1908 if ((u8*)a + le16_to_cpu(
1909 a->data.resident.value_offset) +
1910 le32_to_cpu(
1911 a->data.resident.value_length) >
1912 (u8*)ctx->mrec + vol->mft_record_size) {
1913 ntfs_error(sb, "Corrupt attribute list "
1914 "attribute.");
1915 goto put_err_out;
1916 }
1917 /* Now copy the attribute list. */
1918 memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
1919 a->data.resident.value_offset),
1920 le32_to_cpu(
1921 a->data.resident.value_length));
1922 }
1923 /* The attribute list is now setup in memory. */
1924 /*
1925 * FIXME: I don't know if this case is actually possible.
1926 * According to logic it is not possible but I have seen too
1927 * many weird things in MS software to rely on logic... Thus we
1928 * perform a manual search and make sure the first $MFT/$DATA
1929 * extent is in the base inode. If it is not we abort with an
1930 * error and if we ever see a report of this error we will need
1931 * to do some magic in order to have the necessary mft record
1932 * loaded and in the right place in the page cache. But
1933 * hopefully logic will prevail and this never happens...
1934 */
1935 al_entry = (ATTR_LIST_ENTRY*)ni->attr_list;
1936 al_end = (u8*)al_entry + ni->attr_list_size;
1937 for (;; al_entry = next_al_entry) {
1938 /* Out of bounds check. */
1939 if ((u8*)al_entry < ni->attr_list ||
1940 (u8*)al_entry > al_end)
1941 goto em_put_err_out;
1942 /* Catch the end of the attribute list. */
1943 if ((u8*)al_entry == al_end)
1944 goto em_put_err_out;
1945 if (!al_entry->length)
1946 goto em_put_err_out;
1947 if ((u8*)al_entry + 6 > al_end || (u8*)al_entry +
1948 le16_to_cpu(al_entry->length) > al_end)
1949 goto em_put_err_out;
1950 next_al_entry = (ATTR_LIST_ENTRY*)((u8*)al_entry +
1951 le16_to_cpu(al_entry->length));
1952 if (le32_to_cpu(al_entry->type) > le32_to_cpu(AT_DATA))
1953 goto em_put_err_out;
1954 if (AT_DATA != al_entry->type)
1955 continue;
1956 /* We want an unnamed attribute. */
1957 if (al_entry->name_length)
1958 goto em_put_err_out;
1959 /* Want the first entry, i.e. lowest_vcn == 0. */
1960 if (al_entry->lowest_vcn)
1961 goto em_put_err_out;
1962 /* First entry has to be in the base mft record. */
1963 if (MREF_LE(al_entry->mft_reference) != vi->i_ino) {
1964 /* MFT references do not match, logic fails. */
1965 ntfs_error(sb, "BUG: The first $DATA extent "
1966 "of $MFT is not in the base "
1967 "mft record. Please report "
1968 "you saw this message to "
1969 "linux-ntfs-dev@lists."
1970 "sourceforge.net");
1971 goto put_err_out;
1972 } else {
1973 /* Sequence numbers must match. */
1974 if (MSEQNO_LE(al_entry->mft_reference) !=
1975 ni->seq_no)
1976 goto em_put_err_out;
1977 /* Got it. All is ok. We can stop now. */
1978 break;
1979 }
1980 }
1981 }
1982
1983 ntfs_attr_reinit_search_ctx(ctx);
1984
1985 /* Now load all attribute extents. */
1986 a = NULL;
1987 next_vcn = last_vcn = highest_vcn = 0;
1988 while (!(err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, next_vcn, NULL, 0,
1989 ctx))) {
1990 runlist_element *nrl;
1991
1992 /* Cache the current attribute. */
1993 a = ctx->attr;
1994 /* $MFT must be non-resident. */
1995 if (!a->non_resident) {
1996 ntfs_error(sb, "$MFT must be non-resident but a "
1997 "resident extent was found. $MFT is "
1998 "corrupt. Run chkdsk.");
1999 goto put_err_out;
2000 }
2001 /* $MFT must be uncompressed and unencrypted. */
2002 if (a->flags & ATTR_COMPRESSION_MASK ||
2003 a->flags & ATTR_IS_ENCRYPTED ||
2004 a->flags & ATTR_IS_SPARSE) {
2005 ntfs_error(sb, "$MFT must be uncompressed, "
2006 "non-sparse, and unencrypted but a "
2007 "compressed/sparse/encrypted extent "
2008 "was found. $MFT is corrupt. Run "
2009 "chkdsk.");
2010 goto put_err_out;
2011 }
2012 /*
2013 * Decompress the mapping pairs array of this extent and merge
2014 * the result into the existing runlist. No need for locking
2015 * as we have exclusive access to the inode at this time and we
2016 * are a mount in progress task, too.
2017 */
2018 nrl = ntfs_mapping_pairs_decompress(vol, a, ni->runlist.rl);
2019 if (IS_ERR(nrl)) {
2020 ntfs_error(sb, "ntfs_mapping_pairs_decompress() "
2021 "failed with error code %ld. $MFT is "
2022 "corrupt.", PTR_ERR(nrl));
2023 goto put_err_out;
2024 }
2025 ni->runlist.rl = nrl;
2026
2027 /* Are we in the first extent? */
2028 if (!next_vcn) {
2029 if (a->data.non_resident.lowest_vcn) {
2030 ntfs_error(sb, "First extent of $DATA "
2031 "attribute has non zero "
2032 "lowest_vcn. $MFT is corrupt. "
2033 "You should run chkdsk.");
2034 goto put_err_out;
2035 }
2036 /* Get the last vcn in the $DATA attribute. */
2037 last_vcn = sle64_to_cpu(
2038 a->data.non_resident.allocated_size)
2039 >> vol->cluster_size_bits;
2040 /* Fill in the inode size. */
2041 vi->i_size = sle64_to_cpu(
2042 a->data.non_resident.data_size);
2043 ni->initialized_size = sle64_to_cpu(
2044 a->data.non_resident.initialized_size);
2045 ni->allocated_size = sle64_to_cpu(
2046 a->data.non_resident.allocated_size);
2047 /*
2048 * Verify the number of mft records does not exceed
2049 * 2^32 - 1.
2050 */
2051 if ((vi->i_size >> vol->mft_record_size_bits) >=
2052 (1ULL << 32)) {
2053 ntfs_error(sb, "$MFT is too big! Aborting.");
2054 goto put_err_out;
2055 }
2056 /*
2057 * We have got the first extent of the runlist for
2058 * $MFT which means it is now relatively safe to call
2059 * the normal ntfs_read_inode() function.
2060 * Complete reading the inode, this will actually
2061 * re-read the mft record for $MFT, this time entering
2062 * it into the page cache with which we complete the
2063 * kick start of the volume. It should be safe to do
2064 * this now as the first extent of $MFT/$DATA is
2065 * already known and we would hope that we don't need
2066 * further extents in order to find the other
2067 * attributes belonging to $MFT. Only time will tell if
2068 * this is really the case. If not we will have to play
2069 * magic at this point, possibly duplicating a lot of
2070 * ntfs_read_inode() at this point. We will need to
2071 * ensure we do enough of its work to be able to call
2072 * ntfs_read_inode() on extents of $MFT/$DATA. But lets
2073 * hope this never happens...
2074 */
2075 ntfs_read_locked_inode(vi);
2076 if (is_bad_inode(vi)) {
2077 ntfs_error(sb, "ntfs_read_inode() of $MFT "
2078 "failed. BUG or corrupt $MFT. "
2079 "Run chkdsk and if no errors "
2080 "are found, please report you "
2081 "saw this message to "
2082 "linux-ntfs-dev@lists."
2083 "sourceforge.net");
2084 ntfs_attr_put_search_ctx(ctx);
2085 /* Revert to the safe super operations. */
2086 ntfs_free(m);
2087 return -1;
2088 }
2089 /*
2090 * Re-initialize some specifics about $MFT's inode as
2091 * ntfs_read_inode() will have set up the default ones.
2092 */
2093 /* Set uid and gid to root. */
2094 vi->i_uid = GLOBAL_ROOT_UID;
2095 vi->i_gid = GLOBAL_ROOT_GID;
2096 /* Regular file. No access for anyone. */
2097 vi->i_mode = S_IFREG;
2098 /* No VFS initiated operations allowed for $MFT. */
2099 vi->i_op = &ntfs_empty_inode_ops;
2100 vi->i_fop = &ntfs_empty_file_ops;
2101 }
2102
2103 /* Get the lowest vcn for the next extent. */
2104 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2105 next_vcn = highest_vcn + 1;
2106
2107 /* Only one extent or error, which we catch below. */
2108 if (next_vcn <= 0)
2109 break;
2110
2111 /* Avoid endless loops due to corruption. */
2112 if (next_vcn < sle64_to_cpu(
2113 a->data.non_resident.lowest_vcn)) {
2114 ntfs_error(sb, "$MFT has corrupt attribute list "
2115 "attribute. Run chkdsk.");
2116 goto put_err_out;
2117 }
2118 }
2119 if (err != -ENOENT) {
2120 ntfs_error(sb, "Failed to lookup $MFT/$DATA attribute extent. "
2121 "$MFT is corrupt. Run chkdsk.");
2122 goto put_err_out;
2123 }
2124 if (!a) {
2125 ntfs_error(sb, "$MFT/$DATA attribute not found. $MFT is "
2126 "corrupt. Run chkdsk.");
2127 goto put_err_out;
2128 }
2129 if (highest_vcn && highest_vcn != last_vcn - 1) {
2130 ntfs_error(sb, "Failed to load the complete runlist for "
2131 "$MFT/$DATA. Driver bug or corrupt $MFT. "
2132 "Run chkdsk.");
2133 ntfs_debug("highest_vcn = 0x%llx, last_vcn - 1 = 0x%llx",
2134 (unsigned long long)highest_vcn,
2135 (unsigned long long)last_vcn - 1);
2136 goto put_err_out;
2137 }
2138 ntfs_attr_put_search_ctx(ctx);
2139 ntfs_debug("Done.");
2140 ntfs_free(m);
2141
2142 /*
2143 * Split the locking rules of the MFT inode from the
2144 * locking rules of other inodes:
2145 */
2146 lockdep_set_class(&ni->runlist.lock, &mft_ni_runlist_lock_key);
2147 lockdep_set_class(&ni->mrec_lock, &mft_ni_mrec_lock_key);
2148
2149 return 0;
2150
2151em_put_err_out:
2152 ntfs_error(sb, "Couldn't find first extent of $DATA attribute in "
2153 "attribute list. $MFT is corrupt. Run chkdsk.");
2154put_err_out:
2155 ntfs_attr_put_search_ctx(ctx);
2156err_out:
2157 ntfs_error(sb, "Failed. Marking inode as bad.");
2158 make_bad_inode(vi);
2159 ntfs_free(m);
2160 return -1;
2161}
2162
2163static void __ntfs_clear_inode(ntfs_inode *ni)
2164{
2165 /* Free all alocated memory. */
2166 down_write(&ni->runlist.lock);
2167 if (ni->runlist.rl) {
2168 ntfs_free(ni->runlist.rl);
2169 ni->runlist.rl = NULL;
2170 }
2171 up_write(&ni->runlist.lock);
2172
2173 if (ni->attr_list) {
2174 ntfs_free(ni->attr_list);
2175 ni->attr_list = NULL;
2176 }
2177
2178 down_write(&ni->attr_list_rl.lock);
2179 if (ni->attr_list_rl.rl) {
2180 ntfs_free(ni->attr_list_rl.rl);
2181 ni->attr_list_rl.rl = NULL;
2182 }
2183 up_write(&ni->attr_list_rl.lock);
2184
2185 if (ni->name_len && ni->name != I30) {
2186 /* Catch bugs... */
2187 BUG_ON(!ni->name);
2188 kfree(ni->name);
2189 }
2190}
2191
2192void ntfs_clear_extent_inode(ntfs_inode *ni)
2193{
2194 ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
2195
2196 BUG_ON(NInoAttr(ni));
2197 BUG_ON(ni->nr_extents != -1);
2198
2199#ifdef NTFS_RW
2200 if (NInoDirty(ni)) {
2201 if (!is_bad_inode(VFS_I(ni->ext.base_ntfs_ino)))
2202 ntfs_error(ni->vol->sb, "Clearing dirty extent inode! "
2203 "Losing data! This is a BUG!!!");
2204 // FIXME: Do something!!!
2205 }
2206#endif /* NTFS_RW */
2207
2208 __ntfs_clear_inode(ni);
2209
2210 /* Bye, bye... */
2211 ntfs_destroy_extent_inode(ni);
2212}
2213
2214/**
2215 * ntfs_evict_big_inode - clean up the ntfs specific part of an inode
2216 * @vi: vfs inode pending annihilation
2217 *
2218 * When the VFS is going to remove an inode from memory, ntfs_clear_big_inode()
2219 * is called, which deallocates all memory belonging to the NTFS specific part
2220 * of the inode and returns.
2221 *
2222 * If the MFT record is dirty, we commit it before doing anything else.
2223 */
2224void ntfs_evict_big_inode(struct inode *vi)
2225{
2226 ntfs_inode *ni = NTFS_I(vi);
2227
2228 truncate_inode_pages_final(&vi->i_data);
2229 clear_inode(vi);
2230
2231#ifdef NTFS_RW
2232 if (NInoDirty(ni)) {
2233 bool was_bad = (is_bad_inode(vi));
2234
2235 /* Committing the inode also commits all extent inodes. */
2236 ntfs_commit_inode(vi);
2237
2238 if (!was_bad && (is_bad_inode(vi) || NInoDirty(ni))) {
2239 ntfs_error(vi->i_sb, "Failed to commit dirty inode "
2240 "0x%lx. Losing data!", vi->i_ino);
2241 // FIXME: Do something!!!
2242 }
2243 }
2244#endif /* NTFS_RW */
2245
2246 /* No need to lock at this stage as no one else has a reference. */
2247 if (ni->nr_extents > 0) {
2248 int i;
2249
2250 for (i = 0; i < ni->nr_extents; i++)
2251 ntfs_clear_extent_inode(ni->ext.extent_ntfs_inos[i]);
2252 kfree(ni->ext.extent_ntfs_inos);
2253 }
2254
2255 __ntfs_clear_inode(ni);
2256
2257 if (NInoAttr(ni)) {
2258 /* Release the base inode if we are holding it. */
2259 if (ni->nr_extents == -1) {
2260 iput(VFS_I(ni->ext.base_ntfs_ino));
2261 ni->nr_extents = 0;
2262 ni->ext.base_ntfs_ino = NULL;
2263 }
2264 }
2265 BUG_ON(ni->page);
2266 if (!atomic_dec_and_test(&ni->count))
2267 BUG();
2268 return;
2269}
2270
2271/**
2272 * ntfs_show_options - show mount options in /proc/mounts
2273 * @sf: seq_file in which to write our mount options
2274 * @root: root of the mounted tree whose mount options to display
2275 *
2276 * Called by the VFS once for each mounted ntfs volume when someone reads
2277 * /proc/mounts in order to display the NTFS specific mount options of each
2278 * mount. The mount options of fs specified by @root are written to the seq file
2279 * @sf and success is returned.
2280 */
2281int ntfs_show_options(struct seq_file *sf, struct dentry *root)
2282{
2283 ntfs_volume *vol = NTFS_SB(root->d_sb);
2284 int i;
2285
2286 seq_printf(sf, ",uid=%i", from_kuid_munged(&init_user_ns, vol->uid));
2287 seq_printf(sf, ",gid=%i", from_kgid_munged(&init_user_ns, vol->gid));
2288 if (vol->fmask == vol->dmask)
2289 seq_printf(sf, ",umask=0%o", vol->fmask);
2290 else {
2291 seq_printf(sf, ",fmask=0%o", vol->fmask);
2292 seq_printf(sf, ",dmask=0%o", vol->dmask);
2293 }
2294 seq_printf(sf, ",nls=%s", vol->nls_map->charset);
2295 if (NVolCaseSensitive(vol))
2296 seq_printf(sf, ",case_sensitive");
2297 if (NVolShowSystemFiles(vol))
2298 seq_printf(sf, ",show_sys_files");
2299 if (!NVolSparseEnabled(vol))
2300 seq_printf(sf, ",disable_sparse");
2301 for (i = 0; on_errors_arr[i].val; i++) {
2302 if (on_errors_arr[i].val & vol->on_errors)
2303 seq_printf(sf, ",errors=%s", on_errors_arr[i].str);
2304 }
2305 seq_printf(sf, ",mft_zone_multiplier=%i", vol->mft_zone_multiplier);
2306 return 0;
2307}
2308
2309#ifdef NTFS_RW
2310
2311static const char *es = " Leaving inconsistent metadata. Unmount and run "
2312 "chkdsk.";
2313
2314/**
2315 * ntfs_truncate - called when the i_size of an ntfs inode is changed
2316 * @vi: inode for which the i_size was changed
2317 *
2318 * We only support i_size changes for normal files at present, i.e. not
2319 * compressed and not encrypted. This is enforced in ntfs_setattr(), see
2320 * below.
2321 *
2322 * The kernel guarantees that @vi is a regular file (S_ISREG() is true) and
2323 * that the change is allowed.
2324 *
2325 * This implies for us that @vi is a file inode rather than a directory, index,
2326 * or attribute inode as well as that @vi is a base inode.
2327 *
2328 * Returns 0 on success or -errno on error.
2329 *
2330 * Called with ->i_mutex held.
2331 */
2332int ntfs_truncate(struct inode *vi)
2333{
2334 s64 new_size, old_size, nr_freed, new_alloc_size, old_alloc_size;
2335 VCN highest_vcn;
2336 unsigned long flags;
2337 ntfs_inode *base_ni, *ni = NTFS_I(vi);
2338 ntfs_volume *vol = ni->vol;
2339 ntfs_attr_search_ctx *ctx;
2340 MFT_RECORD *m;
2341 ATTR_RECORD *a;
2342 const char *te = " Leaving file length out of sync with i_size.";
2343 int err, mp_size, size_change, alloc_change;
2344 u32 attr_len;
2345
2346 ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2347 BUG_ON(NInoAttr(ni));
2348 BUG_ON(S_ISDIR(vi->i_mode));
2349 BUG_ON(NInoMstProtected(ni));
2350 BUG_ON(ni->nr_extents < 0);
2351retry_truncate:
2352 /*
2353 * Lock the runlist for writing and map the mft record to ensure it is
2354 * safe to mess with the attribute runlist and sizes.
2355 */
2356 down_write(&ni->runlist.lock);
2357 if (!NInoAttr(ni))
2358 base_ni = ni;
2359 else
2360 base_ni = ni->ext.base_ntfs_ino;
2361 m = map_mft_record(base_ni);
2362 if (IS_ERR(m)) {
2363 err = PTR_ERR(m);
2364 ntfs_error(vi->i_sb, "Failed to map mft record for inode 0x%lx "
2365 "(error code %d).%s", vi->i_ino, err, te);
2366 ctx = NULL;
2367 m = NULL;
2368 goto old_bad_out;
2369 }
2370 ctx = ntfs_attr_get_search_ctx(base_ni, m);
2371 if (unlikely(!ctx)) {
2372 ntfs_error(vi->i_sb, "Failed to allocate a search context for "
2373 "inode 0x%lx (not enough memory).%s",
2374 vi->i_ino, te);
2375 err = -ENOMEM;
2376 goto old_bad_out;
2377 }
2378 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
2379 CASE_SENSITIVE, 0, NULL, 0, ctx);
2380 if (unlikely(err)) {
2381 if (err == -ENOENT) {
2382 ntfs_error(vi->i_sb, "Open attribute is missing from "
2383 "mft record. Inode 0x%lx is corrupt. "
2384 "Run chkdsk.%s", vi->i_ino, te);
2385 err = -EIO;
2386 } else
2387 ntfs_error(vi->i_sb, "Failed to lookup attribute in "
2388 "inode 0x%lx (error code %d).%s",
2389 vi->i_ino, err, te);
2390 goto old_bad_out;
2391 }
2392 m = ctx->mrec;
2393 a = ctx->attr;
2394 /*
2395 * The i_size of the vfs inode is the new size for the attribute value.
2396 */
2397 new_size = i_size_read(vi);
2398 /* The current size of the attribute value is the old size. */
2399 old_size = ntfs_attr_size(a);
2400 /* Calculate the new allocated size. */
2401 if (NInoNonResident(ni))
2402 new_alloc_size = (new_size + vol->cluster_size - 1) &
2403 ~(s64)vol->cluster_size_mask;
2404 else
2405 new_alloc_size = (new_size + 7) & ~7;
2406 /* The current allocated size is the old allocated size. */
2407 read_lock_irqsave(&ni->size_lock, flags);
2408 old_alloc_size = ni->allocated_size;
2409 read_unlock_irqrestore(&ni->size_lock, flags);
2410 /*
2411 * The change in the file size. This will be 0 if no change, >0 if the
2412 * size is growing, and <0 if the size is shrinking.
2413 */
2414 size_change = -1;
2415 if (new_size - old_size >= 0) {
2416 size_change = 1;
2417 if (new_size == old_size)
2418 size_change = 0;
2419 }
2420 /* As above for the allocated size. */
2421 alloc_change = -1;
2422 if (new_alloc_size - old_alloc_size >= 0) {
2423 alloc_change = 1;
2424 if (new_alloc_size == old_alloc_size)
2425 alloc_change = 0;
2426 }
2427 /*
2428 * If neither the size nor the allocation are being changed there is
2429 * nothing to do.
2430 */
2431 if (!size_change && !alloc_change)
2432 goto unm_done;
2433 /* If the size is changing, check if new size is allowed in $AttrDef. */
2434 if (size_change) {
2435 err = ntfs_attr_size_bounds_check(vol, ni->type, new_size);
2436 if (unlikely(err)) {
2437 if (err == -ERANGE) {
2438 ntfs_error(vol->sb, "Truncate would cause the "
2439 "inode 0x%lx to %simum size "
2440 "for its attribute type "
2441 "(0x%x). Aborting truncate.",
2442 vi->i_ino,
2443 new_size > old_size ? "exceed "
2444 "the max" : "go under the min",
2445 le32_to_cpu(ni->type));
2446 err = -EFBIG;
2447 } else {
2448 ntfs_error(vol->sb, "Inode 0x%lx has unknown "
2449 "attribute type 0x%x. "
2450 "Aborting truncate.",
2451 vi->i_ino,
2452 le32_to_cpu(ni->type));
2453 err = -EIO;
2454 }
2455 /* Reset the vfs inode size to the old size. */
2456 i_size_write(vi, old_size);
2457 goto err_out;
2458 }
2459 }
2460 if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2461 ntfs_warning(vi->i_sb, "Changes in inode size are not "
2462 "supported yet for %s files, ignoring.",
2463 NInoCompressed(ni) ? "compressed" :
2464 "encrypted");
2465 err = -EOPNOTSUPP;
2466 goto bad_out;
2467 }
2468 if (a->non_resident)
2469 goto do_non_resident_truncate;
2470 BUG_ON(NInoNonResident(ni));
2471 /* Resize the attribute record to best fit the new attribute size. */
2472 if (new_size < vol->mft_record_size &&
2473 !ntfs_resident_attr_value_resize(m, a, new_size)) {
2474 /* The resize succeeded! */
2475 flush_dcache_mft_record_page(ctx->ntfs_ino);
2476 mark_mft_record_dirty(ctx->ntfs_ino);
2477 write_lock_irqsave(&ni->size_lock, flags);
2478 /* Update the sizes in the ntfs inode and all is done. */
2479 ni->allocated_size = le32_to_cpu(a->length) -
2480 le16_to_cpu(a->data.resident.value_offset);
2481 /*
2482 * Note ntfs_resident_attr_value_resize() has already done any
2483 * necessary data clearing in the attribute record. When the
2484 * file is being shrunk vmtruncate() will already have cleared
2485 * the top part of the last partial page, i.e. since this is
2486 * the resident case this is the page with index 0. However,
2487 * when the file is being expanded, the page cache page data
2488 * between the old data_size, i.e. old_size, and the new_size
2489 * has not been zeroed. Fortunately, we do not need to zero it
2490 * either since on one hand it will either already be zero due
2491 * to both readpage and writepage clearing partial page data
2492 * beyond i_size in which case there is nothing to do or in the
2493 * case of the file being mmap()ped at the same time, POSIX
2494 * specifies that the behaviour is unspecified thus we do not
2495 * have to do anything. This means that in our implementation
2496 * in the rare case that the file is mmap()ped and a write
2497 * occurred into the mmap()ped region just beyond the file size
2498 * and writepage has not yet been called to write out the page
2499 * (which would clear the area beyond the file size) and we now
2500 * extend the file size to incorporate this dirty region
2501 * outside the file size, a write of the page would result in
2502 * this data being written to disk instead of being cleared.
2503 * Given both POSIX and the Linux mmap(2) man page specify that
2504 * this corner case is undefined, we choose to leave it like
2505 * that as this is much simpler for us as we cannot lock the
2506 * relevant page now since we are holding too many ntfs locks
2507 * which would result in a lock reversal deadlock.
2508 */
2509 ni->initialized_size = new_size;
2510 write_unlock_irqrestore(&ni->size_lock, flags);
2511 goto unm_done;
2512 }
2513 /* If the above resize failed, this must be an attribute extension. */
2514 BUG_ON(size_change < 0);
2515 /*
2516 * We have to drop all the locks so we can call
2517 * ntfs_attr_make_non_resident(). This could be optimised by try-
2518 * locking the first page cache page and only if that fails dropping
2519 * the locks, locking the page, and redoing all the locking and
2520 * lookups. While this would be a huge optimisation, it is not worth
2521 * it as this is definitely a slow code path as it only ever can happen
2522 * once for any given file.
2523 */
2524 ntfs_attr_put_search_ctx(ctx);
2525 unmap_mft_record(base_ni);
2526 up_write(&ni->runlist.lock);
2527 /*
2528 * Not enough space in the mft record, try to make the attribute
2529 * non-resident and if successful restart the truncation process.
2530 */
2531 err = ntfs_attr_make_non_resident(ni, old_size);
2532 if (likely(!err))
2533 goto retry_truncate;
2534 /*
2535 * Could not make non-resident. If this is due to this not being
2536 * permitted for this attribute type or there not being enough space,
2537 * try to make other attributes non-resident. Otherwise fail.
2538 */
2539 if (unlikely(err != -EPERM && err != -ENOSPC)) {
2540 ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, attribute "
2541 "type 0x%x, because the conversion from "
2542 "resident to non-resident attribute failed "
2543 "with error code %i.", vi->i_ino,
2544 (unsigned)le32_to_cpu(ni->type), err);
2545 if (err != -ENOMEM)
2546 err = -EIO;
2547 goto conv_err_out;
2548 }
2549 /* TODO: Not implemented from here, abort. */
2550 if (err == -ENOSPC)
2551 ntfs_error(vol->sb, "Not enough space in the mft record/on "
2552 "disk for the non-resident attribute value. "
2553 "This case is not implemented yet.");
2554 else /* if (err == -EPERM) */
2555 ntfs_error(vol->sb, "This attribute type may not be "
2556 "non-resident. This case is not implemented "
2557 "yet.");
2558 err = -EOPNOTSUPP;
2559 goto conv_err_out;
2560#if 0
2561 // TODO: Attempt to make other attributes non-resident.
2562 if (!err)
2563 goto do_resident_extend;
2564 /*
2565 * Both the attribute list attribute and the standard information
2566 * attribute must remain in the base inode. Thus, if this is one of
2567 * these attributes, we have to try to move other attributes out into
2568 * extent mft records instead.
2569 */
2570 if (ni->type == AT_ATTRIBUTE_LIST ||
2571 ni->type == AT_STANDARD_INFORMATION) {
2572 // TODO: Attempt to move other attributes into extent mft
2573 // records.
2574 err = -EOPNOTSUPP;
2575 if (!err)
2576 goto do_resident_extend;
2577 goto err_out;
2578 }
2579 // TODO: Attempt to move this attribute to an extent mft record, but
2580 // only if it is not already the only attribute in an mft record in
2581 // which case there would be nothing to gain.
2582 err = -EOPNOTSUPP;
2583 if (!err)
2584 goto do_resident_extend;
2585 /* There is nothing we can do to make enough space. )-: */
2586 goto err_out;
2587#endif
2588do_non_resident_truncate:
2589 BUG_ON(!NInoNonResident(ni));
2590 if (alloc_change < 0) {
2591 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2592 if (highest_vcn > 0 &&
2593 old_alloc_size >> vol->cluster_size_bits >
2594 highest_vcn + 1) {
2595 /*
2596 * This attribute has multiple extents. Not yet
2597 * supported.
2598 */
2599 ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, "
2600 "attribute type 0x%x, because the "
2601 "attribute is highly fragmented (it "
2602 "consists of multiple extents) and "
2603 "this case is not implemented yet.",
2604 vi->i_ino,
2605 (unsigned)le32_to_cpu(ni->type));
2606 err = -EOPNOTSUPP;
2607 goto bad_out;
2608 }
2609 }
2610 /*
2611 * If the size is shrinking, need to reduce the initialized_size and
2612 * the data_size before reducing the allocation.
2613 */
2614 if (size_change < 0) {
2615 /*
2616 * Make the valid size smaller (i_size is already up-to-date).
2617 */
2618 write_lock_irqsave(&ni->size_lock, flags);
2619 if (new_size < ni->initialized_size) {
2620 ni->initialized_size = new_size;
2621 a->data.non_resident.initialized_size =
2622 cpu_to_sle64(new_size);
2623 }
2624 a->data.non_resident.data_size = cpu_to_sle64(new_size);
2625 write_unlock_irqrestore(&ni->size_lock, flags);
2626 flush_dcache_mft_record_page(ctx->ntfs_ino);
2627 mark_mft_record_dirty(ctx->ntfs_ino);
2628 /* If the allocated size is not changing, we are done. */
2629 if (!alloc_change)
2630 goto unm_done;
2631 /*
2632 * If the size is shrinking it makes no sense for the
2633 * allocation to be growing.
2634 */
2635 BUG_ON(alloc_change > 0);
2636 } else /* if (size_change >= 0) */ {
2637 /*
2638 * The file size is growing or staying the same but the
2639 * allocation can be shrinking, growing or staying the same.
2640 */
2641 if (alloc_change > 0) {
2642 /*
2643 * We need to extend the allocation and possibly update
2644 * the data size. If we are updating the data size,
2645 * since we are not touching the initialized_size we do
2646 * not need to worry about the actual data on disk.
2647 * And as far as the page cache is concerned, there
2648 * will be no pages beyond the old data size and any
2649 * partial region in the last page between the old and
2650 * new data size (or the end of the page if the new
2651 * data size is outside the page) does not need to be
2652 * modified as explained above for the resident
2653 * attribute truncate case. To do this, we simply drop
2654 * the locks we hold and leave all the work to our
2655 * friendly helper ntfs_attr_extend_allocation().
2656 */
2657 ntfs_attr_put_search_ctx(ctx);
2658 unmap_mft_record(base_ni);
2659 up_write(&ni->runlist.lock);
2660 err = ntfs_attr_extend_allocation(ni, new_size,
2661 size_change > 0 ? new_size : -1, -1);
2662 /*
2663 * ntfs_attr_extend_allocation() will have done error
2664 * output already.
2665 */
2666 goto done;
2667 }
2668 if (!alloc_change)
2669 goto alloc_done;
2670 }
2671 /* alloc_change < 0 */
2672 /* Free the clusters. */
2673 nr_freed = ntfs_cluster_free(ni, new_alloc_size >>
2674 vol->cluster_size_bits, -1, ctx);
2675 m = ctx->mrec;
2676 a = ctx->attr;
2677 if (unlikely(nr_freed < 0)) {
2678 ntfs_error(vol->sb, "Failed to release cluster(s) (error code "
2679 "%lli). Unmount and run chkdsk to recover "
2680 "the lost cluster(s).", (long long)nr_freed);
2681 NVolSetErrors(vol);
2682 nr_freed = 0;
2683 }
2684 /* Truncate the runlist. */
2685 err = ntfs_rl_truncate_nolock(vol, &ni->runlist,
2686 new_alloc_size >> vol->cluster_size_bits);
2687 /*
2688 * If the runlist truncation failed and/or the search context is no
2689 * longer valid, we cannot resize the attribute record or build the
2690 * mapping pairs array thus we mark the inode bad so that no access to
2691 * the freed clusters can happen.
2692 */
2693 if (unlikely(err || IS_ERR(m))) {
2694 ntfs_error(vol->sb, "Failed to %s (error code %li).%s",
2695 IS_ERR(m) ?
2696 "restore attribute search context" :
2697 "truncate attribute runlist",
2698 IS_ERR(m) ? PTR_ERR(m) : err, es);
2699 err = -EIO;
2700 goto bad_out;
2701 }
2702 /* Get the size for the shrunk mapping pairs array for the runlist. */
2703 mp_size = ntfs_get_size_for_mapping_pairs(vol, ni->runlist.rl, 0, -1);
2704 if (unlikely(mp_size <= 0)) {
2705 ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2706 "attribute type 0x%x, because determining the "
2707 "size for the mapping pairs failed with error "
2708 "code %i.%s", vi->i_ino,
2709 (unsigned)le32_to_cpu(ni->type), mp_size, es);
2710 err = -EIO;
2711 goto bad_out;
2712 }
2713 /*
2714 * Shrink the attribute record for the new mapping pairs array. Note,
2715 * this cannot fail since we are making the attribute smaller thus by
2716 * definition there is enough space to do so.
2717 */
2718 attr_len = le32_to_cpu(a->length);
2719 err = ntfs_attr_record_resize(m, a, mp_size +
2720 le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
2721 BUG_ON(err);
2722 /*
2723 * Generate the mapping pairs array directly into the attribute record.
2724 */
2725 err = ntfs_mapping_pairs_build(vol, (u8*)a +
2726 le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
2727 mp_size, ni->runlist.rl, 0, -1, NULL);
2728 if (unlikely(err)) {
2729 ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2730 "attribute type 0x%x, because building the "
2731 "mapping pairs failed with error code %i.%s",
2732 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
2733 err, es);
2734 err = -EIO;
2735 goto bad_out;
2736 }
2737 /* Update the allocated/compressed size as well as the highest vcn. */
2738 a->data.non_resident.highest_vcn = cpu_to_sle64((new_alloc_size >>
2739 vol->cluster_size_bits) - 1);
2740 write_lock_irqsave(&ni->size_lock, flags);
2741 ni->allocated_size = new_alloc_size;
2742 a->data.non_resident.allocated_size = cpu_to_sle64(new_alloc_size);
2743 if (NInoSparse(ni) || NInoCompressed(ni)) {
2744 if (nr_freed) {
2745 ni->itype.compressed.size -= nr_freed <<
2746 vol->cluster_size_bits;
2747 BUG_ON(ni->itype.compressed.size < 0);
2748 a->data.non_resident.compressed_size = cpu_to_sle64(
2749 ni->itype.compressed.size);
2750 vi->i_blocks = ni->itype.compressed.size >> 9;
2751 }
2752 } else
2753 vi->i_blocks = new_alloc_size >> 9;
2754 write_unlock_irqrestore(&ni->size_lock, flags);
2755 /*
2756 * We have shrunk the allocation. If this is a shrinking truncate we
2757 * have already dealt with the initialized_size and the data_size above
2758 * and we are done. If the truncate is only changing the allocation
2759 * and not the data_size, we are also done. If this is an extending
2760 * truncate, need to extend the data_size now which is ensured by the
2761 * fact that @size_change is positive.
2762 */
2763alloc_done:
2764 /*
2765 * If the size is growing, need to update it now. If it is shrinking,
2766 * we have already updated it above (before the allocation change).
2767 */
2768 if (size_change > 0)
2769 a->data.non_resident.data_size = cpu_to_sle64(new_size);
2770 /* Ensure the modified mft record is written out. */
2771 flush_dcache_mft_record_page(ctx->ntfs_ino);
2772 mark_mft_record_dirty(ctx->ntfs_ino);
2773unm_done:
2774 ntfs_attr_put_search_ctx(ctx);
2775 unmap_mft_record(base_ni);
2776 up_write(&ni->runlist.lock);
2777done:
2778 /* Update the mtime and ctime on the base inode. */
2779 /* normally ->truncate shouldn't update ctime or mtime,
2780 * but ntfs did before so it got a copy & paste version
2781 * of file_update_time. one day someone should fix this
2782 * for real.
2783 */
2784 if (!IS_NOCMTIME(VFS_I(base_ni)) && !IS_RDONLY(VFS_I(base_ni))) {
2785 struct timespec64 now = current_time(VFS_I(base_ni));
2786 int sync_it = 0;
2787
2788 if (!timespec64_equal(&VFS_I(base_ni)->i_mtime, &now) ||
2789 !timespec64_equal(&VFS_I(base_ni)->i_ctime, &now))
2790 sync_it = 1;
2791 VFS_I(base_ni)->i_mtime = now;
2792 VFS_I(base_ni)->i_ctime = now;
2793
2794 if (sync_it)
2795 mark_inode_dirty_sync(VFS_I(base_ni));
2796 }
2797
2798 if (likely(!err)) {
2799 NInoClearTruncateFailed(ni);
2800 ntfs_debug("Done.");
2801 }
2802 return err;
2803old_bad_out:
2804 old_size = -1;
2805bad_out:
2806 if (err != -ENOMEM && err != -EOPNOTSUPP)
2807 NVolSetErrors(vol);
2808 if (err != -EOPNOTSUPP)
2809 NInoSetTruncateFailed(ni);
2810 else if (old_size >= 0)
2811 i_size_write(vi, old_size);
2812err_out:
2813 if (ctx)
2814 ntfs_attr_put_search_ctx(ctx);
2815 if (m)
2816 unmap_mft_record(base_ni);
2817 up_write(&ni->runlist.lock);
2818out:
2819 ntfs_debug("Failed. Returning error code %i.", err);
2820 return err;
2821conv_err_out:
2822 if (err != -ENOMEM && err != -EOPNOTSUPP)
2823 NVolSetErrors(vol);
2824 if (err != -EOPNOTSUPP)
2825 NInoSetTruncateFailed(ni);
2826 else
2827 i_size_write(vi, old_size);
2828 goto out;
2829}
2830
2831/**
2832 * ntfs_truncate_vfs - wrapper for ntfs_truncate() that has no return value
2833 * @vi: inode for which the i_size was changed
2834 *
2835 * Wrapper for ntfs_truncate() that has no return value.
2836 *
2837 * See ntfs_truncate() description above for details.
2838 */
2839#ifdef NTFS_RW
2840void ntfs_truncate_vfs(struct inode *vi) {
2841 ntfs_truncate(vi);
2842}
2843#endif
2844
2845/**
2846 * ntfs_setattr - called from notify_change() when an attribute is being changed
2847 * @dentry: dentry whose attributes to change
2848 * @attr: structure describing the attributes and the changes
2849 *
2850 * We have to trap VFS attempts to truncate the file described by @dentry as
2851 * soon as possible, because we do not implement changes in i_size yet. So we
2852 * abort all i_size changes here.
2853 *
2854 * We also abort all changes of user, group, and mode as we do not implement
2855 * the NTFS ACLs yet.
2856 *
2857 * Called with ->i_mutex held.
2858 */
2859int ntfs_setattr(struct dentry *dentry, struct iattr *attr)
2860{
2861 struct inode *vi = d_inode(dentry);
2862 int err;
2863 unsigned int ia_valid = attr->ia_valid;
2864
2865 err = setattr_prepare(dentry, attr);
2866 if (err)
2867 goto out;
2868 /* We do not support NTFS ACLs yet. */
2869 if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
2870 ntfs_warning(vi->i_sb, "Changes in user/group/mode are not "
2871 "supported yet, ignoring.");
2872 err = -EOPNOTSUPP;
2873 goto out;
2874 }
2875 if (ia_valid & ATTR_SIZE) {
2876 if (attr->ia_size != i_size_read(vi)) {
2877 ntfs_inode *ni = NTFS_I(vi);
2878 /*
2879 * FIXME: For now we do not support resizing of
2880 * compressed or encrypted files yet.
2881 */
2882 if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2883 ntfs_warning(vi->i_sb, "Changes in inode size "
2884 "are not supported yet for "
2885 "%s files, ignoring.",
2886 NInoCompressed(ni) ?
2887 "compressed" : "encrypted");
2888 err = -EOPNOTSUPP;
2889 } else {
2890 truncate_setsize(vi, attr->ia_size);
2891 ntfs_truncate_vfs(vi);
2892 }
2893 if (err || ia_valid == ATTR_SIZE)
2894 goto out;
2895 } else {
2896 /*
2897 * We skipped the truncate but must still update
2898 * timestamps.
2899 */
2900 ia_valid |= ATTR_MTIME | ATTR_CTIME;
2901 }
2902 }
2903 if (ia_valid & ATTR_ATIME)
2904 vi->i_atime = attr->ia_atime;
2905 if (ia_valid & ATTR_MTIME)
2906 vi->i_mtime = attr->ia_mtime;
2907 if (ia_valid & ATTR_CTIME)
2908 vi->i_ctime = attr->ia_ctime;
2909 mark_inode_dirty(vi);
2910out:
2911 return err;
2912}
2913
2914/**
2915 * ntfs_write_inode - write out a dirty inode
2916 * @vi: inode to write out
2917 * @sync: if true, write out synchronously
2918 *
2919 * Write out a dirty inode to disk including any extent inodes if present.
2920 *
2921 * If @sync is true, commit the inode to disk and wait for io completion. This
2922 * is done using write_mft_record().
2923 *
2924 * If @sync is false, just schedule the write to happen but do not wait for i/o
2925 * completion. In 2.6 kernels, scheduling usually happens just by virtue of
2926 * marking the page (and in this case mft record) dirty but we do not implement
2927 * this yet as write_mft_record() largely ignores the @sync parameter and
2928 * always performs synchronous writes.
2929 *
2930 * Return 0 on success and -errno on error.
2931 */
2932int __ntfs_write_inode(struct inode *vi, int sync)
2933{
2934 sle64 nt;
2935 ntfs_inode *ni = NTFS_I(vi);
2936 ntfs_attr_search_ctx *ctx;
2937 MFT_RECORD *m;
2938 STANDARD_INFORMATION *si;
2939 int err = 0;
2940 bool modified = false;
2941
2942 ntfs_debug("Entering for %sinode 0x%lx.", NInoAttr(ni) ? "attr " : "",
2943 vi->i_ino);
2944 /*
2945 * Dirty attribute inodes are written via their real inodes so just
2946 * clean them here. Access time updates are taken care off when the
2947 * real inode is written.
2948 */
2949 if (NInoAttr(ni)) {
2950 NInoClearDirty(ni);
2951 ntfs_debug("Done.");
2952 return 0;
2953 }
2954 /* Map, pin, and lock the mft record belonging to the inode. */
2955 m = map_mft_record(ni);
2956 if (IS_ERR(m)) {
2957 err = PTR_ERR(m);
2958 goto err_out;
2959 }
2960 /* Update the access times in the standard information attribute. */
2961 ctx = ntfs_attr_get_search_ctx(ni, m);
2962 if (unlikely(!ctx)) {
2963 err = -ENOMEM;
2964 goto unm_err_out;
2965 }
2966 err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0,
2967 CASE_SENSITIVE, 0, NULL, 0, ctx);
2968 if (unlikely(err)) {
2969 ntfs_attr_put_search_ctx(ctx);
2970 goto unm_err_out;
2971 }
2972 si = (STANDARD_INFORMATION*)((u8*)ctx->attr +
2973 le16_to_cpu(ctx->attr->data.resident.value_offset));
2974 /* Update the access times if they have changed. */
2975 nt = utc2ntfs(vi->i_mtime);
2976 if (si->last_data_change_time != nt) {
2977 ntfs_debug("Updating mtime for inode 0x%lx: old = 0x%llx, "
2978 "new = 0x%llx", vi->i_ino, (long long)
2979 sle64_to_cpu(si->last_data_change_time),
2980 (long long)sle64_to_cpu(nt));
2981 si->last_data_change_time = nt;
2982 modified = true;
2983 }
2984 nt = utc2ntfs(vi->i_ctime);
2985 if (si->last_mft_change_time != nt) {
2986 ntfs_debug("Updating ctime for inode 0x%lx: old = 0x%llx, "
2987 "new = 0x%llx", vi->i_ino, (long long)
2988 sle64_to_cpu(si->last_mft_change_time),
2989 (long long)sle64_to_cpu(nt));
2990 si->last_mft_change_time = nt;
2991 modified = true;
2992 }
2993 nt = utc2ntfs(vi->i_atime);
2994 if (si->last_access_time != nt) {
2995 ntfs_debug("Updating atime for inode 0x%lx: old = 0x%llx, "
2996 "new = 0x%llx", vi->i_ino,
2997 (long long)sle64_to_cpu(si->last_access_time),
2998 (long long)sle64_to_cpu(nt));
2999 si->last_access_time = nt;
3000 modified = true;
3001 }
3002 /*
3003 * If we just modified the standard information attribute we need to
3004 * mark the mft record it is in dirty. We do this manually so that
3005 * mark_inode_dirty() is not called which would redirty the inode and
3006 * hence result in an infinite loop of trying to write the inode.
3007 * There is no need to mark the base inode nor the base mft record
3008 * dirty, since we are going to write this mft record below in any case
3009 * and the base mft record may actually not have been modified so it
3010 * might not need to be written out.
3011 * NOTE: It is not a problem when the inode for $MFT itself is being
3012 * written out as mark_ntfs_record_dirty() will only set I_DIRTY_PAGES
3013 * on the $MFT inode and hence ntfs_write_inode() will not be
3014 * re-invoked because of it which in turn is ok since the dirtied mft
3015 * record will be cleaned and written out to disk below, i.e. before
3016 * this function returns.
3017 */
3018 if (modified) {
3019 flush_dcache_mft_record_page(ctx->ntfs_ino);
3020 if (!NInoTestSetDirty(ctx->ntfs_ino))
3021 mark_ntfs_record_dirty(ctx->ntfs_ino->page,
3022 ctx->ntfs_ino->page_ofs);
3023 }
3024 ntfs_attr_put_search_ctx(ctx);
3025 /* Now the access times are updated, write the base mft record. */
3026 if (NInoDirty(ni))
3027 err = write_mft_record(ni, m, sync);
3028 /* Write all attached extent mft records. */
3029 mutex_lock(&ni->extent_lock);
3030 if (ni->nr_extents > 0) {
3031 ntfs_inode **extent_nis = ni->ext.extent_ntfs_inos;
3032 int i;
3033
3034 ntfs_debug("Writing %i extent inodes.", ni->nr_extents);
3035 for (i = 0; i < ni->nr_extents; i++) {
3036 ntfs_inode *tni = extent_nis[i];
3037
3038 if (NInoDirty(tni)) {
3039 MFT_RECORD *tm = map_mft_record(tni);
3040 int ret;
3041
3042 if (IS_ERR(tm)) {
3043 if (!err || err == -ENOMEM)
3044 err = PTR_ERR(tm);
3045 continue;
3046 }
3047 ret = write_mft_record(tni, tm, sync);
3048 unmap_mft_record(tni);
3049 if (unlikely(ret)) {
3050 if (!err || err == -ENOMEM)
3051 err = ret;
3052 }
3053 }
3054 }
3055 }
3056 mutex_unlock(&ni->extent_lock);
3057 unmap_mft_record(ni);
3058 if (unlikely(err))
3059 goto err_out;
3060 ntfs_debug("Done.");
3061 return 0;
3062unm_err_out:
3063 unmap_mft_record(ni);
3064err_out:
3065 if (err == -ENOMEM) {
3066 ntfs_warning(vi->i_sb, "Not enough memory to write inode. "
3067 "Marking the inode dirty again, so the VFS "
3068 "retries later.");
3069 mark_inode_dirty(vi);
3070 } else {
3071 ntfs_error(vi->i_sb, "Failed (error %i): Run chkdsk.", -err);
3072 NVolSetErrors(ni->vol);
3073 }
3074 return err;
3075}
3076
3077#endif /* NTFS_RW */
1/**
2 * inode.c - NTFS kernel inode handling. Part of the Linux-NTFS project.
3 *
4 * Copyright (c) 2001-2007 Anton Altaparmakov
5 *
6 * This program/include file is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as published
8 * by the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program/include file is distributed in the hope that it will be
12 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
13 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program (in the main directory of the Linux-NTFS
18 * distribution in the file COPYING); if not, write to the Free Software
19 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 */
21
22#include <linux/buffer_head.h>
23#include <linux/fs.h>
24#include <linux/mm.h>
25#include <linux/mount.h>
26#include <linux/mutex.h>
27#include <linux/pagemap.h>
28#include <linux/quotaops.h>
29#include <linux/slab.h>
30#include <linux/log2.h>
31#include <linux/aio.h>
32
33#include "aops.h"
34#include "attrib.h"
35#include "bitmap.h"
36#include "dir.h"
37#include "debug.h"
38#include "inode.h"
39#include "lcnalloc.h"
40#include "malloc.h"
41#include "mft.h"
42#include "time.h"
43#include "ntfs.h"
44
45/**
46 * ntfs_test_inode - compare two (possibly fake) inodes for equality
47 * @vi: vfs inode which to test
48 * @na: ntfs attribute which is being tested with
49 *
50 * Compare the ntfs attribute embedded in the ntfs specific part of the vfs
51 * inode @vi for equality with the ntfs attribute @na.
52 *
53 * If searching for the normal file/directory inode, set @na->type to AT_UNUSED.
54 * @na->name and @na->name_len are then ignored.
55 *
56 * Return 1 if the attributes match and 0 if not.
57 *
58 * NOTE: This function runs with the inode_hash_lock spin lock held so it is not
59 * allowed to sleep.
60 */
61int ntfs_test_inode(struct inode *vi, ntfs_attr *na)
62{
63 ntfs_inode *ni;
64
65 if (vi->i_ino != na->mft_no)
66 return 0;
67 ni = NTFS_I(vi);
68 /* If !NInoAttr(ni), @vi is a normal file or directory inode. */
69 if (likely(!NInoAttr(ni))) {
70 /* If not looking for a normal inode this is a mismatch. */
71 if (unlikely(na->type != AT_UNUSED))
72 return 0;
73 } else {
74 /* A fake inode describing an attribute. */
75 if (ni->type != na->type)
76 return 0;
77 if (ni->name_len != na->name_len)
78 return 0;
79 if (na->name_len && memcmp(ni->name, na->name,
80 na->name_len * sizeof(ntfschar)))
81 return 0;
82 }
83 /* Match! */
84 return 1;
85}
86
87/**
88 * ntfs_init_locked_inode - initialize an inode
89 * @vi: vfs inode to initialize
90 * @na: ntfs attribute which to initialize @vi to
91 *
92 * Initialize the vfs inode @vi with the values from the ntfs attribute @na in
93 * order to enable ntfs_test_inode() to do its work.
94 *
95 * If initializing the normal file/directory inode, set @na->type to AT_UNUSED.
96 * In that case, @na->name and @na->name_len should be set to NULL and 0,
97 * respectively. Although that is not strictly necessary as
98 * ntfs_read_locked_inode() will fill them in later.
99 *
100 * Return 0 on success and -errno on error.
101 *
102 * NOTE: This function runs with the inode->i_lock spin lock held so it is not
103 * allowed to sleep. (Hence the GFP_ATOMIC allocation.)
104 */
105static int ntfs_init_locked_inode(struct inode *vi, ntfs_attr *na)
106{
107 ntfs_inode *ni = NTFS_I(vi);
108
109 vi->i_ino = na->mft_no;
110
111 ni->type = na->type;
112 if (na->type == AT_INDEX_ALLOCATION)
113 NInoSetMstProtected(ni);
114
115 ni->name = na->name;
116 ni->name_len = na->name_len;
117
118 /* If initializing a normal inode, we are done. */
119 if (likely(na->type == AT_UNUSED)) {
120 BUG_ON(na->name);
121 BUG_ON(na->name_len);
122 return 0;
123 }
124
125 /* It is a fake inode. */
126 NInoSetAttr(ni);
127
128 /*
129 * We have I30 global constant as an optimization as it is the name
130 * in >99.9% of named attributes! The other <0.1% incur a GFP_ATOMIC
131 * allocation but that is ok. And most attributes are unnamed anyway,
132 * thus the fraction of named attributes with name != I30 is actually
133 * absolutely tiny.
134 */
135 if (na->name_len && na->name != I30) {
136 unsigned int i;
137
138 BUG_ON(!na->name);
139 i = na->name_len * sizeof(ntfschar);
140 ni->name = kmalloc(i + sizeof(ntfschar), GFP_ATOMIC);
141 if (!ni->name)
142 return -ENOMEM;
143 memcpy(ni->name, na->name, i);
144 ni->name[na->name_len] = 0;
145 }
146 return 0;
147}
148
149typedef int (*set_t)(struct inode *, void *);
150static int ntfs_read_locked_inode(struct inode *vi);
151static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi);
152static int ntfs_read_locked_index_inode(struct inode *base_vi,
153 struct inode *vi);
154
155/**
156 * ntfs_iget - obtain a struct inode corresponding to a specific normal inode
157 * @sb: super block of mounted volume
158 * @mft_no: mft record number / inode number to obtain
159 *
160 * Obtain the struct inode corresponding to a specific normal inode (i.e. a
161 * file or directory).
162 *
163 * If the inode is in the cache, it is just returned with an increased
164 * reference count. Otherwise, a new struct inode is allocated and initialized,
165 * and finally ntfs_read_locked_inode() is called to read in the inode and
166 * fill in the remainder of the inode structure.
167 *
168 * Return the struct inode on success. Check the return value with IS_ERR() and
169 * if true, the function failed and the error code is obtained from PTR_ERR().
170 */
171struct inode *ntfs_iget(struct super_block *sb, unsigned long mft_no)
172{
173 struct inode *vi;
174 int err;
175 ntfs_attr na;
176
177 na.mft_no = mft_no;
178 na.type = AT_UNUSED;
179 na.name = NULL;
180 na.name_len = 0;
181
182 vi = iget5_locked(sb, mft_no, (test_t)ntfs_test_inode,
183 (set_t)ntfs_init_locked_inode, &na);
184 if (unlikely(!vi))
185 return ERR_PTR(-ENOMEM);
186
187 err = 0;
188
189 /* If this is a freshly allocated inode, need to read it now. */
190 if (vi->i_state & I_NEW) {
191 err = ntfs_read_locked_inode(vi);
192 unlock_new_inode(vi);
193 }
194 /*
195 * There is no point in keeping bad inodes around if the failure was
196 * due to ENOMEM. We want to be able to retry again later.
197 */
198 if (unlikely(err == -ENOMEM)) {
199 iput(vi);
200 vi = ERR_PTR(err);
201 }
202 return vi;
203}
204
205/**
206 * ntfs_attr_iget - obtain a struct inode corresponding to an attribute
207 * @base_vi: vfs base inode containing the attribute
208 * @type: attribute type
209 * @name: Unicode name of the attribute (NULL if unnamed)
210 * @name_len: length of @name in Unicode characters (0 if unnamed)
211 *
212 * Obtain the (fake) struct inode corresponding to the attribute specified by
213 * @type, @name, and @name_len, which is present in the base mft record
214 * specified by the vfs inode @base_vi.
215 *
216 * If the attribute inode is in the cache, it is just returned with an
217 * increased reference count. Otherwise, a new struct inode is allocated and
218 * initialized, and finally ntfs_read_locked_attr_inode() is called to read the
219 * attribute and fill in the inode structure.
220 *
221 * Note, for index allocation attributes, you need to use ntfs_index_iget()
222 * instead of ntfs_attr_iget() as working with indices is a lot more complex.
223 *
224 * Return the struct inode of the attribute inode on success. Check the return
225 * value with IS_ERR() and if true, the function failed and the error code is
226 * obtained from PTR_ERR().
227 */
228struct inode *ntfs_attr_iget(struct inode *base_vi, ATTR_TYPE type,
229 ntfschar *name, u32 name_len)
230{
231 struct inode *vi;
232 int err;
233 ntfs_attr na;
234
235 /* Make sure no one calls ntfs_attr_iget() for indices. */
236 BUG_ON(type == AT_INDEX_ALLOCATION);
237
238 na.mft_no = base_vi->i_ino;
239 na.type = type;
240 na.name = name;
241 na.name_len = name_len;
242
243 vi = iget5_locked(base_vi->i_sb, na.mft_no, (test_t)ntfs_test_inode,
244 (set_t)ntfs_init_locked_inode, &na);
245 if (unlikely(!vi))
246 return ERR_PTR(-ENOMEM);
247
248 err = 0;
249
250 /* If this is a freshly allocated inode, need to read it now. */
251 if (vi->i_state & I_NEW) {
252 err = ntfs_read_locked_attr_inode(base_vi, vi);
253 unlock_new_inode(vi);
254 }
255 /*
256 * There is no point in keeping bad attribute inodes around. This also
257 * simplifies things in that we never need to check for bad attribute
258 * inodes elsewhere.
259 */
260 if (unlikely(err)) {
261 iput(vi);
262 vi = ERR_PTR(err);
263 }
264 return vi;
265}
266
267/**
268 * ntfs_index_iget - obtain a struct inode corresponding to an index
269 * @base_vi: vfs base inode containing the index related attributes
270 * @name: Unicode name of the index
271 * @name_len: length of @name in Unicode characters
272 *
273 * Obtain the (fake) struct inode corresponding to the index specified by @name
274 * and @name_len, which is present in the base mft record specified by the vfs
275 * inode @base_vi.
276 *
277 * If the index inode is in the cache, it is just returned with an increased
278 * reference count. Otherwise, a new struct inode is allocated and
279 * initialized, and finally ntfs_read_locked_index_inode() is called to read
280 * the index related attributes and fill in the inode structure.
281 *
282 * Return the struct inode of the index inode on success. Check the return
283 * value with IS_ERR() and if true, the function failed and the error code is
284 * obtained from PTR_ERR().
285 */
286struct inode *ntfs_index_iget(struct inode *base_vi, ntfschar *name,
287 u32 name_len)
288{
289 struct inode *vi;
290 int err;
291 ntfs_attr na;
292
293 na.mft_no = base_vi->i_ino;
294 na.type = AT_INDEX_ALLOCATION;
295 na.name = name;
296 na.name_len = name_len;
297
298 vi = iget5_locked(base_vi->i_sb, na.mft_no, (test_t)ntfs_test_inode,
299 (set_t)ntfs_init_locked_inode, &na);
300 if (unlikely(!vi))
301 return ERR_PTR(-ENOMEM);
302
303 err = 0;
304
305 /* If this is a freshly allocated inode, need to read it now. */
306 if (vi->i_state & I_NEW) {
307 err = ntfs_read_locked_index_inode(base_vi, vi);
308 unlock_new_inode(vi);
309 }
310 /*
311 * There is no point in keeping bad index inodes around. This also
312 * simplifies things in that we never need to check for bad index
313 * inodes elsewhere.
314 */
315 if (unlikely(err)) {
316 iput(vi);
317 vi = ERR_PTR(err);
318 }
319 return vi;
320}
321
322struct inode *ntfs_alloc_big_inode(struct super_block *sb)
323{
324 ntfs_inode *ni;
325
326 ntfs_debug("Entering.");
327 ni = kmem_cache_alloc(ntfs_big_inode_cache, GFP_NOFS);
328 if (likely(ni != NULL)) {
329 ni->state = 0;
330 return VFS_I(ni);
331 }
332 ntfs_error(sb, "Allocation of NTFS big inode structure failed.");
333 return NULL;
334}
335
336static void ntfs_i_callback(struct rcu_head *head)
337{
338 struct inode *inode = container_of(head, struct inode, i_rcu);
339 kmem_cache_free(ntfs_big_inode_cache, NTFS_I(inode));
340}
341
342void ntfs_destroy_big_inode(struct inode *inode)
343{
344 ntfs_inode *ni = NTFS_I(inode);
345
346 ntfs_debug("Entering.");
347 BUG_ON(ni->page);
348 if (!atomic_dec_and_test(&ni->count))
349 BUG();
350 call_rcu(&inode->i_rcu, ntfs_i_callback);
351}
352
353static inline ntfs_inode *ntfs_alloc_extent_inode(void)
354{
355 ntfs_inode *ni;
356
357 ntfs_debug("Entering.");
358 ni = kmem_cache_alloc(ntfs_inode_cache, GFP_NOFS);
359 if (likely(ni != NULL)) {
360 ni->state = 0;
361 return ni;
362 }
363 ntfs_error(NULL, "Allocation of NTFS inode structure failed.");
364 return NULL;
365}
366
367static void ntfs_destroy_extent_inode(ntfs_inode *ni)
368{
369 ntfs_debug("Entering.");
370 BUG_ON(ni->page);
371 if (!atomic_dec_and_test(&ni->count))
372 BUG();
373 kmem_cache_free(ntfs_inode_cache, ni);
374}
375
376/*
377 * The attribute runlist lock has separate locking rules from the
378 * normal runlist lock, so split the two lock-classes:
379 */
380static struct lock_class_key attr_list_rl_lock_class;
381
382/**
383 * __ntfs_init_inode - initialize ntfs specific part of an inode
384 * @sb: super block of mounted volume
385 * @ni: freshly allocated ntfs inode which to initialize
386 *
387 * Initialize an ntfs inode to defaults.
388 *
389 * NOTE: ni->mft_no, ni->state, ni->type, ni->name, and ni->name_len are left
390 * untouched. Make sure to initialize them elsewhere.
391 *
392 * Return zero on success and -ENOMEM on error.
393 */
394void __ntfs_init_inode(struct super_block *sb, ntfs_inode *ni)
395{
396 ntfs_debug("Entering.");
397 rwlock_init(&ni->size_lock);
398 ni->initialized_size = ni->allocated_size = 0;
399 ni->seq_no = 0;
400 atomic_set(&ni->count, 1);
401 ni->vol = NTFS_SB(sb);
402 ntfs_init_runlist(&ni->runlist);
403 mutex_init(&ni->mrec_lock);
404 ni->page = NULL;
405 ni->page_ofs = 0;
406 ni->attr_list_size = 0;
407 ni->attr_list = NULL;
408 ntfs_init_runlist(&ni->attr_list_rl);
409 lockdep_set_class(&ni->attr_list_rl.lock,
410 &attr_list_rl_lock_class);
411 ni->itype.index.block_size = 0;
412 ni->itype.index.vcn_size = 0;
413 ni->itype.index.collation_rule = 0;
414 ni->itype.index.block_size_bits = 0;
415 ni->itype.index.vcn_size_bits = 0;
416 mutex_init(&ni->extent_lock);
417 ni->nr_extents = 0;
418 ni->ext.base_ntfs_ino = NULL;
419}
420
421/*
422 * Extent inodes get MFT-mapped in a nested way, while the base inode
423 * is still mapped. Teach this nesting to the lock validator by creating
424 * a separate class for nested inode's mrec_lock's:
425 */
426static struct lock_class_key extent_inode_mrec_lock_key;
427
428inline ntfs_inode *ntfs_new_extent_inode(struct super_block *sb,
429 unsigned long mft_no)
430{
431 ntfs_inode *ni = ntfs_alloc_extent_inode();
432
433 ntfs_debug("Entering.");
434 if (likely(ni != NULL)) {
435 __ntfs_init_inode(sb, ni);
436 lockdep_set_class(&ni->mrec_lock, &extent_inode_mrec_lock_key);
437 ni->mft_no = mft_no;
438 ni->type = AT_UNUSED;
439 ni->name = NULL;
440 ni->name_len = 0;
441 }
442 return ni;
443}
444
445/**
446 * ntfs_is_extended_system_file - check if a file is in the $Extend directory
447 * @ctx: initialized attribute search context
448 *
449 * Search all file name attributes in the inode described by the attribute
450 * search context @ctx and check if any of the names are in the $Extend system
451 * directory.
452 *
453 * Return values:
454 * 1: file is in $Extend directory
455 * 0: file is not in $Extend directory
456 * -errno: failed to determine if the file is in the $Extend directory
457 */
458static int ntfs_is_extended_system_file(ntfs_attr_search_ctx *ctx)
459{
460 int nr_links, err;
461
462 /* Restart search. */
463 ntfs_attr_reinit_search_ctx(ctx);
464
465 /* Get number of hard links. */
466 nr_links = le16_to_cpu(ctx->mrec->link_count);
467
468 /* Loop through all hard links. */
469 while (!(err = ntfs_attr_lookup(AT_FILE_NAME, NULL, 0, 0, 0, NULL, 0,
470 ctx))) {
471 FILE_NAME_ATTR *file_name_attr;
472 ATTR_RECORD *attr = ctx->attr;
473 u8 *p, *p2;
474
475 nr_links--;
476 /*
477 * Maximum sanity checking as we are called on an inode that
478 * we suspect might be corrupt.
479 */
480 p = (u8*)attr + le32_to_cpu(attr->length);
481 if (p < (u8*)ctx->mrec || (u8*)p > (u8*)ctx->mrec +
482 le32_to_cpu(ctx->mrec->bytes_in_use)) {
483err_corrupt_attr:
484 ntfs_error(ctx->ntfs_ino->vol->sb, "Corrupt file name "
485 "attribute. You should run chkdsk.");
486 return -EIO;
487 }
488 if (attr->non_resident) {
489 ntfs_error(ctx->ntfs_ino->vol->sb, "Non-resident file "
490 "name. You should run chkdsk.");
491 return -EIO;
492 }
493 if (attr->flags) {
494 ntfs_error(ctx->ntfs_ino->vol->sb, "File name with "
495 "invalid flags. You should run "
496 "chkdsk.");
497 return -EIO;
498 }
499 if (!(attr->data.resident.flags & RESIDENT_ATTR_IS_INDEXED)) {
500 ntfs_error(ctx->ntfs_ino->vol->sb, "Unindexed file "
501 "name. You should run chkdsk.");
502 return -EIO;
503 }
504 file_name_attr = (FILE_NAME_ATTR*)((u8*)attr +
505 le16_to_cpu(attr->data.resident.value_offset));
506 p2 = (u8*)attr + le32_to_cpu(attr->data.resident.value_length);
507 if (p2 < (u8*)attr || p2 > p)
508 goto err_corrupt_attr;
509 /* This attribute is ok, but is it in the $Extend directory? */
510 if (MREF_LE(file_name_attr->parent_directory) == FILE_Extend)
511 return 1; /* YES, it's an extended system file. */
512 }
513 if (unlikely(err != -ENOENT))
514 return err;
515 if (unlikely(nr_links)) {
516 ntfs_error(ctx->ntfs_ino->vol->sb, "Inode hard link count "
517 "doesn't match number of name attributes. You "
518 "should run chkdsk.");
519 return -EIO;
520 }
521 return 0; /* NO, it is not an extended system file. */
522}
523
524/**
525 * ntfs_read_locked_inode - read an inode from its device
526 * @vi: inode to read
527 *
528 * ntfs_read_locked_inode() is called from ntfs_iget() to read the inode
529 * described by @vi into memory from the device.
530 *
531 * The only fields in @vi that we need to/can look at when the function is
532 * called are i_sb, pointing to the mounted device's super block, and i_ino,
533 * the number of the inode to load.
534 *
535 * ntfs_read_locked_inode() maps, pins and locks the mft record number i_ino
536 * for reading and sets up the necessary @vi fields as well as initializing
537 * the ntfs inode.
538 *
539 * Q: What locks are held when the function is called?
540 * A: i_state has I_NEW set, hence the inode is locked, also
541 * i_count is set to 1, so it is not going to go away
542 * i_flags is set to 0 and we have no business touching it. Only an ioctl()
543 * is allowed to write to them. We should of course be honouring them but
544 * we need to do that using the IS_* macros defined in include/linux/fs.h.
545 * In any case ntfs_read_locked_inode() has nothing to do with i_flags.
546 *
547 * Return 0 on success and -errno on error. In the error case, the inode will
548 * have had make_bad_inode() executed on it.
549 */
550static int ntfs_read_locked_inode(struct inode *vi)
551{
552 ntfs_volume *vol = NTFS_SB(vi->i_sb);
553 ntfs_inode *ni;
554 struct inode *bvi;
555 MFT_RECORD *m;
556 ATTR_RECORD *a;
557 STANDARD_INFORMATION *si;
558 ntfs_attr_search_ctx *ctx;
559 int err = 0;
560
561 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
562
563 /* Setup the generic vfs inode parts now. */
564
565 /*
566 * This is for checking whether an inode has changed w.r.t. a file so
567 * that the file can be updated if necessary (compare with f_version).
568 */
569 vi->i_version = 1;
570
571 vi->i_uid = vol->uid;
572 vi->i_gid = vol->gid;
573 vi->i_mode = 0;
574
575 /*
576 * Initialize the ntfs specific part of @vi special casing
577 * FILE_MFT which we need to do at mount time.
578 */
579 if (vi->i_ino != FILE_MFT)
580 ntfs_init_big_inode(vi);
581 ni = NTFS_I(vi);
582
583 m = map_mft_record(ni);
584 if (IS_ERR(m)) {
585 err = PTR_ERR(m);
586 goto err_out;
587 }
588 ctx = ntfs_attr_get_search_ctx(ni, m);
589 if (!ctx) {
590 err = -ENOMEM;
591 goto unm_err_out;
592 }
593
594 if (!(m->flags & MFT_RECORD_IN_USE)) {
595 ntfs_error(vi->i_sb, "Inode is not in use!");
596 goto unm_err_out;
597 }
598 if (m->base_mft_record) {
599 ntfs_error(vi->i_sb, "Inode is an extent inode!");
600 goto unm_err_out;
601 }
602
603 /* Transfer information from mft record into vfs and ntfs inodes. */
604 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
605
606 /*
607 * FIXME: Keep in mind that link_count is two for files which have both
608 * a long file name and a short file name as separate entries, so if
609 * we are hiding short file names this will be too high. Either we need
610 * to account for the short file names by subtracting them or we need
611 * to make sure we delete files even though i_nlink is not zero which
612 * might be tricky due to vfs interactions. Need to think about this
613 * some more when implementing the unlink command.
614 */
615 set_nlink(vi, le16_to_cpu(m->link_count));
616 /*
617 * FIXME: Reparse points can have the directory bit set even though
618 * they would be S_IFLNK. Need to deal with this further below when we
619 * implement reparse points / symbolic links but it will do for now.
620 * Also if not a directory, it could be something else, rather than
621 * a regular file. But again, will do for now.
622 */
623 /* Everyone gets all permissions. */
624 vi->i_mode |= S_IRWXUGO;
625 /* If read-only, no one gets write permissions. */
626 if (IS_RDONLY(vi))
627 vi->i_mode &= ~S_IWUGO;
628 if (m->flags & MFT_RECORD_IS_DIRECTORY) {
629 vi->i_mode |= S_IFDIR;
630 /*
631 * Apply the directory permissions mask set in the mount
632 * options.
633 */
634 vi->i_mode &= ~vol->dmask;
635 /* Things break without this kludge! */
636 if (vi->i_nlink > 1)
637 set_nlink(vi, 1);
638 } else {
639 vi->i_mode |= S_IFREG;
640 /* Apply the file permissions mask set in the mount options. */
641 vi->i_mode &= ~vol->fmask;
642 }
643 /*
644 * Find the standard information attribute in the mft record. At this
645 * stage we haven't setup the attribute list stuff yet, so this could
646 * in fact fail if the standard information is in an extent record, but
647 * I don't think this actually ever happens.
648 */
649 err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0, 0, 0, NULL, 0,
650 ctx);
651 if (unlikely(err)) {
652 if (err == -ENOENT) {
653 /*
654 * TODO: We should be performing a hot fix here (if the
655 * recover mount option is set) by creating a new
656 * attribute.
657 */
658 ntfs_error(vi->i_sb, "$STANDARD_INFORMATION attribute "
659 "is missing.");
660 }
661 goto unm_err_out;
662 }
663 a = ctx->attr;
664 /* Get the standard information attribute value. */
665 si = (STANDARD_INFORMATION*)((u8*)a +
666 le16_to_cpu(a->data.resident.value_offset));
667
668 /* Transfer information from the standard information into vi. */
669 /*
670 * Note: The i_?times do not quite map perfectly onto the NTFS times,
671 * but they are close enough, and in the end it doesn't really matter
672 * that much...
673 */
674 /*
675 * mtime is the last change of the data within the file. Not changed
676 * when only metadata is changed, e.g. a rename doesn't affect mtime.
677 */
678 vi->i_mtime = ntfs2utc(si->last_data_change_time);
679 /*
680 * ctime is the last change of the metadata of the file. This obviously
681 * always changes, when mtime is changed. ctime can be changed on its
682 * own, mtime is then not changed, e.g. when a file is renamed.
683 */
684 vi->i_ctime = ntfs2utc(si->last_mft_change_time);
685 /*
686 * Last access to the data within the file. Not changed during a rename
687 * for example but changed whenever the file is written to.
688 */
689 vi->i_atime = ntfs2utc(si->last_access_time);
690
691 /* Find the attribute list attribute if present. */
692 ntfs_attr_reinit_search_ctx(ctx);
693 err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
694 if (err) {
695 if (unlikely(err != -ENOENT)) {
696 ntfs_error(vi->i_sb, "Failed to lookup attribute list "
697 "attribute.");
698 goto unm_err_out;
699 }
700 } else /* if (!err) */ {
701 if (vi->i_ino == FILE_MFT)
702 goto skip_attr_list_load;
703 ntfs_debug("Attribute list found in inode 0x%lx.", vi->i_ino);
704 NInoSetAttrList(ni);
705 a = ctx->attr;
706 if (a->flags & ATTR_COMPRESSION_MASK) {
707 ntfs_error(vi->i_sb, "Attribute list attribute is "
708 "compressed.");
709 goto unm_err_out;
710 }
711 if (a->flags & ATTR_IS_ENCRYPTED ||
712 a->flags & ATTR_IS_SPARSE) {
713 if (a->non_resident) {
714 ntfs_error(vi->i_sb, "Non-resident attribute "
715 "list attribute is encrypted/"
716 "sparse.");
717 goto unm_err_out;
718 }
719 ntfs_warning(vi->i_sb, "Resident attribute list "
720 "attribute in inode 0x%lx is marked "
721 "encrypted/sparse which is not true. "
722 "However, Windows allows this and "
723 "chkdsk does not detect or correct it "
724 "so we will just ignore the invalid "
725 "flags and pretend they are not set.",
726 vi->i_ino);
727 }
728 /* Now allocate memory for the attribute list. */
729 ni->attr_list_size = (u32)ntfs_attr_size(a);
730 ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
731 if (!ni->attr_list) {
732 ntfs_error(vi->i_sb, "Not enough memory to allocate "
733 "buffer for attribute list.");
734 err = -ENOMEM;
735 goto unm_err_out;
736 }
737 if (a->non_resident) {
738 NInoSetAttrListNonResident(ni);
739 if (a->data.non_resident.lowest_vcn) {
740 ntfs_error(vi->i_sb, "Attribute list has non "
741 "zero lowest_vcn.");
742 goto unm_err_out;
743 }
744 /*
745 * Setup the runlist. No need for locking as we have
746 * exclusive access to the inode at this time.
747 */
748 ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
749 a, NULL);
750 if (IS_ERR(ni->attr_list_rl.rl)) {
751 err = PTR_ERR(ni->attr_list_rl.rl);
752 ni->attr_list_rl.rl = NULL;
753 ntfs_error(vi->i_sb, "Mapping pairs "
754 "decompression failed.");
755 goto unm_err_out;
756 }
757 /* Now load the attribute list. */
758 if ((err = load_attribute_list(vol, &ni->attr_list_rl,
759 ni->attr_list, ni->attr_list_size,
760 sle64_to_cpu(a->data.non_resident.
761 initialized_size)))) {
762 ntfs_error(vi->i_sb, "Failed to load "
763 "attribute list attribute.");
764 goto unm_err_out;
765 }
766 } else /* if (!a->non_resident) */ {
767 if ((u8*)a + le16_to_cpu(a->data.resident.value_offset)
768 + le32_to_cpu(
769 a->data.resident.value_length) >
770 (u8*)ctx->mrec + vol->mft_record_size) {
771 ntfs_error(vi->i_sb, "Corrupt attribute list "
772 "in inode.");
773 goto unm_err_out;
774 }
775 /* Now copy the attribute list. */
776 memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
777 a->data.resident.value_offset),
778 le32_to_cpu(
779 a->data.resident.value_length));
780 }
781 }
782skip_attr_list_load:
783 /*
784 * If an attribute list is present we now have the attribute list value
785 * in ntfs_ino->attr_list and it is ntfs_ino->attr_list_size bytes.
786 */
787 if (S_ISDIR(vi->i_mode)) {
788 loff_t bvi_size;
789 ntfs_inode *bni;
790 INDEX_ROOT *ir;
791 u8 *ir_end, *index_end;
792
793 /* It is a directory, find index root attribute. */
794 ntfs_attr_reinit_search_ctx(ctx);
795 err = ntfs_attr_lookup(AT_INDEX_ROOT, I30, 4, CASE_SENSITIVE,
796 0, NULL, 0, ctx);
797 if (unlikely(err)) {
798 if (err == -ENOENT) {
799 // FIXME: File is corrupt! Hot-fix with empty
800 // index root attribute if recovery option is
801 // set.
802 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute "
803 "is missing.");
804 }
805 goto unm_err_out;
806 }
807 a = ctx->attr;
808 /* Set up the state. */
809 if (unlikely(a->non_resident)) {
810 ntfs_error(vol->sb, "$INDEX_ROOT attribute is not "
811 "resident.");
812 goto unm_err_out;
813 }
814 /* Ensure the attribute name is placed before the value. */
815 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
816 le16_to_cpu(a->data.resident.value_offset)))) {
817 ntfs_error(vol->sb, "$INDEX_ROOT attribute name is "
818 "placed after the attribute value.");
819 goto unm_err_out;
820 }
821 /*
822 * Compressed/encrypted index root just means that the newly
823 * created files in that directory should be created compressed/
824 * encrypted. However index root cannot be both compressed and
825 * encrypted.
826 */
827 if (a->flags & ATTR_COMPRESSION_MASK)
828 NInoSetCompressed(ni);
829 if (a->flags & ATTR_IS_ENCRYPTED) {
830 if (a->flags & ATTR_COMPRESSION_MASK) {
831 ntfs_error(vi->i_sb, "Found encrypted and "
832 "compressed attribute.");
833 goto unm_err_out;
834 }
835 NInoSetEncrypted(ni);
836 }
837 if (a->flags & ATTR_IS_SPARSE)
838 NInoSetSparse(ni);
839 ir = (INDEX_ROOT*)((u8*)a +
840 le16_to_cpu(a->data.resident.value_offset));
841 ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
842 if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
843 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
844 "corrupt.");
845 goto unm_err_out;
846 }
847 index_end = (u8*)&ir->index +
848 le32_to_cpu(ir->index.index_length);
849 if (index_end > ir_end) {
850 ntfs_error(vi->i_sb, "Directory index is corrupt.");
851 goto unm_err_out;
852 }
853 if (ir->type != AT_FILE_NAME) {
854 ntfs_error(vi->i_sb, "Indexed attribute is not "
855 "$FILE_NAME.");
856 goto unm_err_out;
857 }
858 if (ir->collation_rule != COLLATION_FILE_NAME) {
859 ntfs_error(vi->i_sb, "Index collation rule is not "
860 "COLLATION_FILE_NAME.");
861 goto unm_err_out;
862 }
863 ni->itype.index.collation_rule = ir->collation_rule;
864 ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
865 if (ni->itype.index.block_size &
866 (ni->itype.index.block_size - 1)) {
867 ntfs_error(vi->i_sb, "Index block size (%u) is not a "
868 "power of two.",
869 ni->itype.index.block_size);
870 goto unm_err_out;
871 }
872 if (ni->itype.index.block_size > PAGE_CACHE_SIZE) {
873 ntfs_error(vi->i_sb, "Index block size (%u) > "
874 "PAGE_CACHE_SIZE (%ld) is not "
875 "supported. Sorry.",
876 ni->itype.index.block_size,
877 PAGE_CACHE_SIZE);
878 err = -EOPNOTSUPP;
879 goto unm_err_out;
880 }
881 if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
882 ntfs_error(vi->i_sb, "Index block size (%u) < "
883 "NTFS_BLOCK_SIZE (%i) is not "
884 "supported. Sorry.",
885 ni->itype.index.block_size,
886 NTFS_BLOCK_SIZE);
887 err = -EOPNOTSUPP;
888 goto unm_err_out;
889 }
890 ni->itype.index.block_size_bits =
891 ffs(ni->itype.index.block_size) - 1;
892 /* Determine the size of a vcn in the directory index. */
893 if (vol->cluster_size <= ni->itype.index.block_size) {
894 ni->itype.index.vcn_size = vol->cluster_size;
895 ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
896 } else {
897 ni->itype.index.vcn_size = vol->sector_size;
898 ni->itype.index.vcn_size_bits = vol->sector_size_bits;
899 }
900
901 /* Setup the index allocation attribute, even if not present. */
902 NInoSetMstProtected(ni);
903 ni->type = AT_INDEX_ALLOCATION;
904 ni->name = I30;
905 ni->name_len = 4;
906
907 if (!(ir->index.flags & LARGE_INDEX)) {
908 /* No index allocation. */
909 vi->i_size = ni->initialized_size =
910 ni->allocated_size = 0;
911 /* We are done with the mft record, so we release it. */
912 ntfs_attr_put_search_ctx(ctx);
913 unmap_mft_record(ni);
914 m = NULL;
915 ctx = NULL;
916 goto skip_large_dir_stuff;
917 } /* LARGE_INDEX: Index allocation present. Setup state. */
918 NInoSetIndexAllocPresent(ni);
919 /* Find index allocation attribute. */
920 ntfs_attr_reinit_search_ctx(ctx);
921 err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, I30, 4,
922 CASE_SENSITIVE, 0, NULL, 0, ctx);
923 if (unlikely(err)) {
924 if (err == -ENOENT)
925 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION "
926 "attribute is not present but "
927 "$INDEX_ROOT indicated it is.");
928 else
929 ntfs_error(vi->i_sb, "Failed to lookup "
930 "$INDEX_ALLOCATION "
931 "attribute.");
932 goto unm_err_out;
933 }
934 a = ctx->attr;
935 if (!a->non_resident) {
936 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
937 "is resident.");
938 goto unm_err_out;
939 }
940 /*
941 * Ensure the attribute name is placed before the mapping pairs
942 * array.
943 */
944 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
945 le16_to_cpu(
946 a->data.non_resident.mapping_pairs_offset)))) {
947 ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name "
948 "is placed after the mapping pairs "
949 "array.");
950 goto unm_err_out;
951 }
952 if (a->flags & ATTR_IS_ENCRYPTED) {
953 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
954 "is encrypted.");
955 goto unm_err_out;
956 }
957 if (a->flags & ATTR_IS_SPARSE) {
958 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
959 "is sparse.");
960 goto unm_err_out;
961 }
962 if (a->flags & ATTR_COMPRESSION_MASK) {
963 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute "
964 "is compressed.");
965 goto unm_err_out;
966 }
967 if (a->data.non_resident.lowest_vcn) {
968 ntfs_error(vi->i_sb, "First extent of "
969 "$INDEX_ALLOCATION attribute has non "
970 "zero lowest_vcn.");
971 goto unm_err_out;
972 }
973 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
974 ni->initialized_size = sle64_to_cpu(
975 a->data.non_resident.initialized_size);
976 ni->allocated_size = sle64_to_cpu(
977 a->data.non_resident.allocated_size);
978 /*
979 * We are done with the mft record, so we release it. Otherwise
980 * we would deadlock in ntfs_attr_iget().
981 */
982 ntfs_attr_put_search_ctx(ctx);
983 unmap_mft_record(ni);
984 m = NULL;
985 ctx = NULL;
986 /* Get the index bitmap attribute inode. */
987 bvi = ntfs_attr_iget(vi, AT_BITMAP, I30, 4);
988 if (IS_ERR(bvi)) {
989 ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
990 err = PTR_ERR(bvi);
991 goto unm_err_out;
992 }
993 bni = NTFS_I(bvi);
994 if (NInoCompressed(bni) || NInoEncrypted(bni) ||
995 NInoSparse(bni)) {
996 ntfs_error(vi->i_sb, "$BITMAP attribute is compressed "
997 "and/or encrypted and/or sparse.");
998 goto iput_unm_err_out;
999 }
1000 /* Consistency check bitmap size vs. index allocation size. */
1001 bvi_size = i_size_read(bvi);
1002 if ((bvi_size << 3) < (vi->i_size >>
1003 ni->itype.index.block_size_bits)) {
1004 ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) "
1005 "for index allocation (0x%llx).",
1006 bvi_size << 3, vi->i_size);
1007 goto iput_unm_err_out;
1008 }
1009 /* No longer need the bitmap attribute inode. */
1010 iput(bvi);
1011skip_large_dir_stuff:
1012 /* Setup the operations for this inode. */
1013 vi->i_op = &ntfs_dir_inode_ops;
1014 vi->i_fop = &ntfs_dir_ops;
1015 } else {
1016 /* It is a file. */
1017 ntfs_attr_reinit_search_ctx(ctx);
1018
1019 /* Setup the data attribute, even if not present. */
1020 ni->type = AT_DATA;
1021 ni->name = NULL;
1022 ni->name_len = 0;
1023
1024 /* Find first extent of the unnamed data attribute. */
1025 err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, 0, NULL, 0, ctx);
1026 if (unlikely(err)) {
1027 vi->i_size = ni->initialized_size =
1028 ni->allocated_size = 0;
1029 if (err != -ENOENT) {
1030 ntfs_error(vi->i_sb, "Failed to lookup $DATA "
1031 "attribute.");
1032 goto unm_err_out;
1033 }
1034 /*
1035 * FILE_Secure does not have an unnamed $DATA
1036 * attribute, so we special case it here.
1037 */
1038 if (vi->i_ino == FILE_Secure)
1039 goto no_data_attr_special_case;
1040 /*
1041 * Most if not all the system files in the $Extend
1042 * system directory do not have unnamed data
1043 * attributes so we need to check if the parent
1044 * directory of the file is FILE_Extend and if it is
1045 * ignore this error. To do this we need to get the
1046 * name of this inode from the mft record as the name
1047 * contains the back reference to the parent directory.
1048 */
1049 if (ntfs_is_extended_system_file(ctx) > 0)
1050 goto no_data_attr_special_case;
1051 // FIXME: File is corrupt! Hot-fix with empty data
1052 // attribute if recovery option is set.
1053 ntfs_error(vi->i_sb, "$DATA attribute is missing.");
1054 goto unm_err_out;
1055 }
1056 a = ctx->attr;
1057 /* Setup the state. */
1058 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1059 if (a->flags & ATTR_COMPRESSION_MASK) {
1060 NInoSetCompressed(ni);
1061 if (vol->cluster_size > 4096) {
1062 ntfs_error(vi->i_sb, "Found "
1063 "compressed data but "
1064 "compression is "
1065 "disabled due to "
1066 "cluster size (%i) > "
1067 "4kiB.",
1068 vol->cluster_size);
1069 goto unm_err_out;
1070 }
1071 if ((a->flags & ATTR_COMPRESSION_MASK)
1072 != ATTR_IS_COMPRESSED) {
1073 ntfs_error(vi->i_sb, "Found unknown "
1074 "compression method "
1075 "or corrupt file.");
1076 goto unm_err_out;
1077 }
1078 }
1079 if (a->flags & ATTR_IS_SPARSE)
1080 NInoSetSparse(ni);
1081 }
1082 if (a->flags & ATTR_IS_ENCRYPTED) {
1083 if (NInoCompressed(ni)) {
1084 ntfs_error(vi->i_sb, "Found encrypted and "
1085 "compressed data.");
1086 goto unm_err_out;
1087 }
1088 NInoSetEncrypted(ni);
1089 }
1090 if (a->non_resident) {
1091 NInoSetNonResident(ni);
1092 if (NInoCompressed(ni) || NInoSparse(ni)) {
1093 if (NInoCompressed(ni) && a->data.non_resident.
1094 compression_unit != 4) {
1095 ntfs_error(vi->i_sb, "Found "
1096 "non-standard "
1097 "compression unit (%u "
1098 "instead of 4). "
1099 "Cannot handle this.",
1100 a->data.non_resident.
1101 compression_unit);
1102 err = -EOPNOTSUPP;
1103 goto unm_err_out;
1104 }
1105 if (a->data.non_resident.compression_unit) {
1106 ni->itype.compressed.block_size = 1U <<
1107 (a->data.non_resident.
1108 compression_unit +
1109 vol->cluster_size_bits);
1110 ni->itype.compressed.block_size_bits =
1111 ffs(ni->itype.
1112 compressed.
1113 block_size) - 1;
1114 ni->itype.compressed.block_clusters =
1115 1U << a->data.
1116 non_resident.
1117 compression_unit;
1118 } else {
1119 ni->itype.compressed.block_size = 0;
1120 ni->itype.compressed.block_size_bits =
1121 0;
1122 ni->itype.compressed.block_clusters =
1123 0;
1124 }
1125 ni->itype.compressed.size = sle64_to_cpu(
1126 a->data.non_resident.
1127 compressed_size);
1128 }
1129 if (a->data.non_resident.lowest_vcn) {
1130 ntfs_error(vi->i_sb, "First extent of $DATA "
1131 "attribute has non zero "
1132 "lowest_vcn.");
1133 goto unm_err_out;
1134 }
1135 vi->i_size = sle64_to_cpu(
1136 a->data.non_resident.data_size);
1137 ni->initialized_size = sle64_to_cpu(
1138 a->data.non_resident.initialized_size);
1139 ni->allocated_size = sle64_to_cpu(
1140 a->data.non_resident.allocated_size);
1141 } else { /* Resident attribute. */
1142 vi->i_size = ni->initialized_size = le32_to_cpu(
1143 a->data.resident.value_length);
1144 ni->allocated_size = le32_to_cpu(a->length) -
1145 le16_to_cpu(
1146 a->data.resident.value_offset);
1147 if (vi->i_size > ni->allocated_size) {
1148 ntfs_error(vi->i_sb, "Resident data attribute "
1149 "is corrupt (size exceeds "
1150 "allocation).");
1151 goto unm_err_out;
1152 }
1153 }
1154no_data_attr_special_case:
1155 /* We are done with the mft record, so we release it. */
1156 ntfs_attr_put_search_ctx(ctx);
1157 unmap_mft_record(ni);
1158 m = NULL;
1159 ctx = NULL;
1160 /* Setup the operations for this inode. */
1161 vi->i_op = &ntfs_file_inode_ops;
1162 vi->i_fop = &ntfs_file_ops;
1163 }
1164 if (NInoMstProtected(ni))
1165 vi->i_mapping->a_ops = &ntfs_mst_aops;
1166 else
1167 vi->i_mapping->a_ops = &ntfs_aops;
1168 /*
1169 * The number of 512-byte blocks used on disk (for stat). This is in so
1170 * far inaccurate as it doesn't account for any named streams or other
1171 * special non-resident attributes, but that is how Windows works, too,
1172 * so we are at least consistent with Windows, if not entirely
1173 * consistent with the Linux Way. Doing it the Linux Way would cause a
1174 * significant slowdown as it would involve iterating over all
1175 * attributes in the mft record and adding the allocated/compressed
1176 * sizes of all non-resident attributes present to give us the Linux
1177 * correct size that should go into i_blocks (after division by 512).
1178 */
1179 if (S_ISREG(vi->i_mode) && (NInoCompressed(ni) || NInoSparse(ni)))
1180 vi->i_blocks = ni->itype.compressed.size >> 9;
1181 else
1182 vi->i_blocks = ni->allocated_size >> 9;
1183 ntfs_debug("Done.");
1184 return 0;
1185iput_unm_err_out:
1186 iput(bvi);
1187unm_err_out:
1188 if (!err)
1189 err = -EIO;
1190 if (ctx)
1191 ntfs_attr_put_search_ctx(ctx);
1192 if (m)
1193 unmap_mft_record(ni);
1194err_out:
1195 ntfs_error(vol->sb, "Failed with error code %i. Marking corrupt "
1196 "inode 0x%lx as bad. Run chkdsk.", err, vi->i_ino);
1197 make_bad_inode(vi);
1198 if (err != -EOPNOTSUPP && err != -ENOMEM)
1199 NVolSetErrors(vol);
1200 return err;
1201}
1202
1203/**
1204 * ntfs_read_locked_attr_inode - read an attribute inode from its base inode
1205 * @base_vi: base inode
1206 * @vi: attribute inode to read
1207 *
1208 * ntfs_read_locked_attr_inode() is called from ntfs_attr_iget() to read the
1209 * attribute inode described by @vi into memory from the base mft record
1210 * described by @base_ni.
1211 *
1212 * ntfs_read_locked_attr_inode() maps, pins and locks the base inode for
1213 * reading and looks up the attribute described by @vi before setting up the
1214 * necessary fields in @vi as well as initializing the ntfs inode.
1215 *
1216 * Q: What locks are held when the function is called?
1217 * A: i_state has I_NEW set, hence the inode is locked, also
1218 * i_count is set to 1, so it is not going to go away
1219 *
1220 * Return 0 on success and -errno on error. In the error case, the inode will
1221 * have had make_bad_inode() executed on it.
1222 *
1223 * Note this cannot be called for AT_INDEX_ALLOCATION.
1224 */
1225static int ntfs_read_locked_attr_inode(struct inode *base_vi, struct inode *vi)
1226{
1227 ntfs_volume *vol = NTFS_SB(vi->i_sb);
1228 ntfs_inode *ni, *base_ni;
1229 MFT_RECORD *m;
1230 ATTR_RECORD *a;
1231 ntfs_attr_search_ctx *ctx;
1232 int err = 0;
1233
1234 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1235
1236 ntfs_init_big_inode(vi);
1237
1238 ni = NTFS_I(vi);
1239 base_ni = NTFS_I(base_vi);
1240
1241 /* Just mirror the values from the base inode. */
1242 vi->i_version = base_vi->i_version;
1243 vi->i_uid = base_vi->i_uid;
1244 vi->i_gid = base_vi->i_gid;
1245 set_nlink(vi, base_vi->i_nlink);
1246 vi->i_mtime = base_vi->i_mtime;
1247 vi->i_ctime = base_vi->i_ctime;
1248 vi->i_atime = base_vi->i_atime;
1249 vi->i_generation = ni->seq_no = base_ni->seq_no;
1250
1251 /* Set inode type to zero but preserve permissions. */
1252 vi->i_mode = base_vi->i_mode & ~S_IFMT;
1253
1254 m = map_mft_record(base_ni);
1255 if (IS_ERR(m)) {
1256 err = PTR_ERR(m);
1257 goto err_out;
1258 }
1259 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1260 if (!ctx) {
1261 err = -ENOMEM;
1262 goto unm_err_out;
1263 }
1264 /* Find the attribute. */
1265 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
1266 CASE_SENSITIVE, 0, NULL, 0, ctx);
1267 if (unlikely(err))
1268 goto unm_err_out;
1269 a = ctx->attr;
1270 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_SPARSE)) {
1271 if (a->flags & ATTR_COMPRESSION_MASK) {
1272 NInoSetCompressed(ni);
1273 if ((ni->type != AT_DATA) || (ni->type == AT_DATA &&
1274 ni->name_len)) {
1275 ntfs_error(vi->i_sb, "Found compressed "
1276 "non-data or named data "
1277 "attribute. Please report "
1278 "you saw this message to "
1279 "linux-ntfs-dev@lists."
1280 "sourceforge.net");
1281 goto unm_err_out;
1282 }
1283 if (vol->cluster_size > 4096) {
1284 ntfs_error(vi->i_sb, "Found compressed "
1285 "attribute but compression is "
1286 "disabled due to cluster size "
1287 "(%i) > 4kiB.",
1288 vol->cluster_size);
1289 goto unm_err_out;
1290 }
1291 if ((a->flags & ATTR_COMPRESSION_MASK) !=
1292 ATTR_IS_COMPRESSED) {
1293 ntfs_error(vi->i_sb, "Found unknown "
1294 "compression method.");
1295 goto unm_err_out;
1296 }
1297 }
1298 /*
1299 * The compressed/sparse flag set in an index root just means
1300 * to compress all files.
1301 */
1302 if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1303 ntfs_error(vi->i_sb, "Found mst protected attribute "
1304 "but the attribute is %s. Please "
1305 "report you saw this message to "
1306 "linux-ntfs-dev@lists.sourceforge.net",
1307 NInoCompressed(ni) ? "compressed" :
1308 "sparse");
1309 goto unm_err_out;
1310 }
1311 if (a->flags & ATTR_IS_SPARSE)
1312 NInoSetSparse(ni);
1313 }
1314 if (a->flags & ATTR_IS_ENCRYPTED) {
1315 if (NInoCompressed(ni)) {
1316 ntfs_error(vi->i_sb, "Found encrypted and compressed "
1317 "data.");
1318 goto unm_err_out;
1319 }
1320 /*
1321 * The encryption flag set in an index root just means to
1322 * encrypt all files.
1323 */
1324 if (NInoMstProtected(ni) && ni->type != AT_INDEX_ROOT) {
1325 ntfs_error(vi->i_sb, "Found mst protected attribute "
1326 "but the attribute is encrypted. "
1327 "Please report you saw this message "
1328 "to linux-ntfs-dev@lists.sourceforge."
1329 "net");
1330 goto unm_err_out;
1331 }
1332 if (ni->type != AT_DATA) {
1333 ntfs_error(vi->i_sb, "Found encrypted non-data "
1334 "attribute.");
1335 goto unm_err_out;
1336 }
1337 NInoSetEncrypted(ni);
1338 }
1339 if (!a->non_resident) {
1340 /* Ensure the attribute name is placed before the value. */
1341 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1342 le16_to_cpu(a->data.resident.value_offset)))) {
1343 ntfs_error(vol->sb, "Attribute name is placed after "
1344 "the attribute value.");
1345 goto unm_err_out;
1346 }
1347 if (NInoMstProtected(ni)) {
1348 ntfs_error(vi->i_sb, "Found mst protected attribute "
1349 "but the attribute is resident. "
1350 "Please report you saw this message to "
1351 "linux-ntfs-dev@lists.sourceforge.net");
1352 goto unm_err_out;
1353 }
1354 vi->i_size = ni->initialized_size = le32_to_cpu(
1355 a->data.resident.value_length);
1356 ni->allocated_size = le32_to_cpu(a->length) -
1357 le16_to_cpu(a->data.resident.value_offset);
1358 if (vi->i_size > ni->allocated_size) {
1359 ntfs_error(vi->i_sb, "Resident attribute is corrupt "
1360 "(size exceeds allocation).");
1361 goto unm_err_out;
1362 }
1363 } else {
1364 NInoSetNonResident(ni);
1365 /*
1366 * Ensure the attribute name is placed before the mapping pairs
1367 * array.
1368 */
1369 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1370 le16_to_cpu(
1371 a->data.non_resident.mapping_pairs_offset)))) {
1372 ntfs_error(vol->sb, "Attribute name is placed after "
1373 "the mapping pairs array.");
1374 goto unm_err_out;
1375 }
1376 if (NInoCompressed(ni) || NInoSparse(ni)) {
1377 if (NInoCompressed(ni) && a->data.non_resident.
1378 compression_unit != 4) {
1379 ntfs_error(vi->i_sb, "Found non-standard "
1380 "compression unit (%u instead "
1381 "of 4). Cannot handle this.",
1382 a->data.non_resident.
1383 compression_unit);
1384 err = -EOPNOTSUPP;
1385 goto unm_err_out;
1386 }
1387 if (a->data.non_resident.compression_unit) {
1388 ni->itype.compressed.block_size = 1U <<
1389 (a->data.non_resident.
1390 compression_unit +
1391 vol->cluster_size_bits);
1392 ni->itype.compressed.block_size_bits =
1393 ffs(ni->itype.compressed.
1394 block_size) - 1;
1395 ni->itype.compressed.block_clusters = 1U <<
1396 a->data.non_resident.
1397 compression_unit;
1398 } else {
1399 ni->itype.compressed.block_size = 0;
1400 ni->itype.compressed.block_size_bits = 0;
1401 ni->itype.compressed.block_clusters = 0;
1402 }
1403 ni->itype.compressed.size = sle64_to_cpu(
1404 a->data.non_resident.compressed_size);
1405 }
1406 if (a->data.non_resident.lowest_vcn) {
1407 ntfs_error(vi->i_sb, "First extent of attribute has "
1408 "non-zero lowest_vcn.");
1409 goto unm_err_out;
1410 }
1411 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1412 ni->initialized_size = sle64_to_cpu(
1413 a->data.non_resident.initialized_size);
1414 ni->allocated_size = sle64_to_cpu(
1415 a->data.non_resident.allocated_size);
1416 }
1417 if (NInoMstProtected(ni))
1418 vi->i_mapping->a_ops = &ntfs_mst_aops;
1419 else
1420 vi->i_mapping->a_ops = &ntfs_aops;
1421 if ((NInoCompressed(ni) || NInoSparse(ni)) && ni->type != AT_INDEX_ROOT)
1422 vi->i_blocks = ni->itype.compressed.size >> 9;
1423 else
1424 vi->i_blocks = ni->allocated_size >> 9;
1425 /*
1426 * Make sure the base inode does not go away and attach it to the
1427 * attribute inode.
1428 */
1429 igrab(base_vi);
1430 ni->ext.base_ntfs_ino = base_ni;
1431 ni->nr_extents = -1;
1432
1433 ntfs_attr_put_search_ctx(ctx);
1434 unmap_mft_record(base_ni);
1435
1436 ntfs_debug("Done.");
1437 return 0;
1438
1439unm_err_out:
1440 if (!err)
1441 err = -EIO;
1442 if (ctx)
1443 ntfs_attr_put_search_ctx(ctx);
1444 unmap_mft_record(base_ni);
1445err_out:
1446 ntfs_error(vol->sb, "Failed with error code %i while reading attribute "
1447 "inode (mft_no 0x%lx, type 0x%x, name_len %i). "
1448 "Marking corrupt inode and base inode 0x%lx as bad. "
1449 "Run chkdsk.", err, vi->i_ino, ni->type, ni->name_len,
1450 base_vi->i_ino);
1451 make_bad_inode(vi);
1452 if (err != -ENOMEM)
1453 NVolSetErrors(vol);
1454 return err;
1455}
1456
1457/**
1458 * ntfs_read_locked_index_inode - read an index inode from its base inode
1459 * @base_vi: base inode
1460 * @vi: index inode to read
1461 *
1462 * ntfs_read_locked_index_inode() is called from ntfs_index_iget() to read the
1463 * index inode described by @vi into memory from the base mft record described
1464 * by @base_ni.
1465 *
1466 * ntfs_read_locked_index_inode() maps, pins and locks the base inode for
1467 * reading and looks up the attributes relating to the index described by @vi
1468 * before setting up the necessary fields in @vi as well as initializing the
1469 * ntfs inode.
1470 *
1471 * Note, index inodes are essentially attribute inodes (NInoAttr() is true)
1472 * with the attribute type set to AT_INDEX_ALLOCATION. Apart from that, they
1473 * are setup like directory inodes since directories are a special case of
1474 * indices ao they need to be treated in much the same way. Most importantly,
1475 * for small indices the index allocation attribute might not actually exist.
1476 * However, the index root attribute always exists but this does not need to
1477 * have an inode associated with it and this is why we define a new inode type
1478 * index. Also, like for directories, we need to have an attribute inode for
1479 * the bitmap attribute corresponding to the index allocation attribute and we
1480 * can store this in the appropriate field of the inode, just like we do for
1481 * normal directory inodes.
1482 *
1483 * Q: What locks are held when the function is called?
1484 * A: i_state has I_NEW set, hence the inode is locked, also
1485 * i_count is set to 1, so it is not going to go away
1486 *
1487 * Return 0 on success and -errno on error. In the error case, the inode will
1488 * have had make_bad_inode() executed on it.
1489 */
1490static int ntfs_read_locked_index_inode(struct inode *base_vi, struct inode *vi)
1491{
1492 loff_t bvi_size;
1493 ntfs_volume *vol = NTFS_SB(vi->i_sb);
1494 ntfs_inode *ni, *base_ni, *bni;
1495 struct inode *bvi;
1496 MFT_RECORD *m;
1497 ATTR_RECORD *a;
1498 ntfs_attr_search_ctx *ctx;
1499 INDEX_ROOT *ir;
1500 u8 *ir_end, *index_end;
1501 int err = 0;
1502
1503 ntfs_debug("Entering for i_ino 0x%lx.", vi->i_ino);
1504 ntfs_init_big_inode(vi);
1505 ni = NTFS_I(vi);
1506 base_ni = NTFS_I(base_vi);
1507 /* Just mirror the values from the base inode. */
1508 vi->i_version = base_vi->i_version;
1509 vi->i_uid = base_vi->i_uid;
1510 vi->i_gid = base_vi->i_gid;
1511 set_nlink(vi, base_vi->i_nlink);
1512 vi->i_mtime = base_vi->i_mtime;
1513 vi->i_ctime = base_vi->i_ctime;
1514 vi->i_atime = base_vi->i_atime;
1515 vi->i_generation = ni->seq_no = base_ni->seq_no;
1516 /* Set inode type to zero but preserve permissions. */
1517 vi->i_mode = base_vi->i_mode & ~S_IFMT;
1518 /* Map the mft record for the base inode. */
1519 m = map_mft_record(base_ni);
1520 if (IS_ERR(m)) {
1521 err = PTR_ERR(m);
1522 goto err_out;
1523 }
1524 ctx = ntfs_attr_get_search_ctx(base_ni, m);
1525 if (!ctx) {
1526 err = -ENOMEM;
1527 goto unm_err_out;
1528 }
1529 /* Find the index root attribute. */
1530 err = ntfs_attr_lookup(AT_INDEX_ROOT, ni->name, ni->name_len,
1531 CASE_SENSITIVE, 0, NULL, 0, ctx);
1532 if (unlikely(err)) {
1533 if (err == -ENOENT)
1534 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is "
1535 "missing.");
1536 goto unm_err_out;
1537 }
1538 a = ctx->attr;
1539 /* Set up the state. */
1540 if (unlikely(a->non_resident)) {
1541 ntfs_error(vol->sb, "$INDEX_ROOT attribute is not resident.");
1542 goto unm_err_out;
1543 }
1544 /* Ensure the attribute name is placed before the value. */
1545 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1546 le16_to_cpu(a->data.resident.value_offset)))) {
1547 ntfs_error(vol->sb, "$INDEX_ROOT attribute name is placed "
1548 "after the attribute value.");
1549 goto unm_err_out;
1550 }
1551 /*
1552 * Compressed/encrypted/sparse index root is not allowed, except for
1553 * directories of course but those are not dealt with here.
1554 */
1555 if (a->flags & (ATTR_COMPRESSION_MASK | ATTR_IS_ENCRYPTED |
1556 ATTR_IS_SPARSE)) {
1557 ntfs_error(vi->i_sb, "Found compressed/encrypted/sparse index "
1558 "root attribute.");
1559 goto unm_err_out;
1560 }
1561 ir = (INDEX_ROOT*)((u8*)a + le16_to_cpu(a->data.resident.value_offset));
1562 ir_end = (u8*)ir + le32_to_cpu(a->data.resident.value_length);
1563 if (ir_end > (u8*)ctx->mrec + vol->mft_record_size) {
1564 ntfs_error(vi->i_sb, "$INDEX_ROOT attribute is corrupt.");
1565 goto unm_err_out;
1566 }
1567 index_end = (u8*)&ir->index + le32_to_cpu(ir->index.index_length);
1568 if (index_end > ir_end) {
1569 ntfs_error(vi->i_sb, "Index is corrupt.");
1570 goto unm_err_out;
1571 }
1572 if (ir->type) {
1573 ntfs_error(vi->i_sb, "Index type is not 0 (type is 0x%x).",
1574 le32_to_cpu(ir->type));
1575 goto unm_err_out;
1576 }
1577 ni->itype.index.collation_rule = ir->collation_rule;
1578 ntfs_debug("Index collation rule is 0x%x.",
1579 le32_to_cpu(ir->collation_rule));
1580 ni->itype.index.block_size = le32_to_cpu(ir->index_block_size);
1581 if (!is_power_of_2(ni->itype.index.block_size)) {
1582 ntfs_error(vi->i_sb, "Index block size (%u) is not a power of "
1583 "two.", ni->itype.index.block_size);
1584 goto unm_err_out;
1585 }
1586 if (ni->itype.index.block_size > PAGE_CACHE_SIZE) {
1587 ntfs_error(vi->i_sb, "Index block size (%u) > PAGE_CACHE_SIZE "
1588 "(%ld) is not supported. Sorry.",
1589 ni->itype.index.block_size, PAGE_CACHE_SIZE);
1590 err = -EOPNOTSUPP;
1591 goto unm_err_out;
1592 }
1593 if (ni->itype.index.block_size < NTFS_BLOCK_SIZE) {
1594 ntfs_error(vi->i_sb, "Index block size (%u) < NTFS_BLOCK_SIZE "
1595 "(%i) is not supported. Sorry.",
1596 ni->itype.index.block_size, NTFS_BLOCK_SIZE);
1597 err = -EOPNOTSUPP;
1598 goto unm_err_out;
1599 }
1600 ni->itype.index.block_size_bits = ffs(ni->itype.index.block_size) - 1;
1601 /* Determine the size of a vcn in the index. */
1602 if (vol->cluster_size <= ni->itype.index.block_size) {
1603 ni->itype.index.vcn_size = vol->cluster_size;
1604 ni->itype.index.vcn_size_bits = vol->cluster_size_bits;
1605 } else {
1606 ni->itype.index.vcn_size = vol->sector_size;
1607 ni->itype.index.vcn_size_bits = vol->sector_size_bits;
1608 }
1609 /* Check for presence of index allocation attribute. */
1610 if (!(ir->index.flags & LARGE_INDEX)) {
1611 /* No index allocation. */
1612 vi->i_size = ni->initialized_size = ni->allocated_size = 0;
1613 /* We are done with the mft record, so we release it. */
1614 ntfs_attr_put_search_ctx(ctx);
1615 unmap_mft_record(base_ni);
1616 m = NULL;
1617 ctx = NULL;
1618 goto skip_large_index_stuff;
1619 } /* LARGE_INDEX: Index allocation present. Setup state. */
1620 NInoSetIndexAllocPresent(ni);
1621 /* Find index allocation attribute. */
1622 ntfs_attr_reinit_search_ctx(ctx);
1623 err = ntfs_attr_lookup(AT_INDEX_ALLOCATION, ni->name, ni->name_len,
1624 CASE_SENSITIVE, 0, NULL, 0, ctx);
1625 if (unlikely(err)) {
1626 if (err == -ENOENT)
1627 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1628 "not present but $INDEX_ROOT "
1629 "indicated it is.");
1630 else
1631 ntfs_error(vi->i_sb, "Failed to lookup "
1632 "$INDEX_ALLOCATION attribute.");
1633 goto unm_err_out;
1634 }
1635 a = ctx->attr;
1636 if (!a->non_resident) {
1637 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1638 "resident.");
1639 goto unm_err_out;
1640 }
1641 /*
1642 * Ensure the attribute name is placed before the mapping pairs array.
1643 */
1644 if (unlikely(a->name_length && (le16_to_cpu(a->name_offset) >=
1645 le16_to_cpu(
1646 a->data.non_resident.mapping_pairs_offset)))) {
1647 ntfs_error(vol->sb, "$INDEX_ALLOCATION attribute name is "
1648 "placed after the mapping pairs array.");
1649 goto unm_err_out;
1650 }
1651 if (a->flags & ATTR_IS_ENCRYPTED) {
1652 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1653 "encrypted.");
1654 goto unm_err_out;
1655 }
1656 if (a->flags & ATTR_IS_SPARSE) {
1657 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is sparse.");
1658 goto unm_err_out;
1659 }
1660 if (a->flags & ATTR_COMPRESSION_MASK) {
1661 ntfs_error(vi->i_sb, "$INDEX_ALLOCATION attribute is "
1662 "compressed.");
1663 goto unm_err_out;
1664 }
1665 if (a->data.non_resident.lowest_vcn) {
1666 ntfs_error(vi->i_sb, "First extent of $INDEX_ALLOCATION "
1667 "attribute has non zero lowest_vcn.");
1668 goto unm_err_out;
1669 }
1670 vi->i_size = sle64_to_cpu(a->data.non_resident.data_size);
1671 ni->initialized_size = sle64_to_cpu(
1672 a->data.non_resident.initialized_size);
1673 ni->allocated_size = sle64_to_cpu(a->data.non_resident.allocated_size);
1674 /*
1675 * We are done with the mft record, so we release it. Otherwise
1676 * we would deadlock in ntfs_attr_iget().
1677 */
1678 ntfs_attr_put_search_ctx(ctx);
1679 unmap_mft_record(base_ni);
1680 m = NULL;
1681 ctx = NULL;
1682 /* Get the index bitmap attribute inode. */
1683 bvi = ntfs_attr_iget(base_vi, AT_BITMAP, ni->name, ni->name_len);
1684 if (IS_ERR(bvi)) {
1685 ntfs_error(vi->i_sb, "Failed to get bitmap attribute.");
1686 err = PTR_ERR(bvi);
1687 goto unm_err_out;
1688 }
1689 bni = NTFS_I(bvi);
1690 if (NInoCompressed(bni) || NInoEncrypted(bni) ||
1691 NInoSparse(bni)) {
1692 ntfs_error(vi->i_sb, "$BITMAP attribute is compressed and/or "
1693 "encrypted and/or sparse.");
1694 goto iput_unm_err_out;
1695 }
1696 /* Consistency check bitmap size vs. index allocation size. */
1697 bvi_size = i_size_read(bvi);
1698 if ((bvi_size << 3) < (vi->i_size >> ni->itype.index.block_size_bits)) {
1699 ntfs_error(vi->i_sb, "Index bitmap too small (0x%llx) for "
1700 "index allocation (0x%llx).", bvi_size << 3,
1701 vi->i_size);
1702 goto iput_unm_err_out;
1703 }
1704 iput(bvi);
1705skip_large_index_stuff:
1706 /* Setup the operations for this index inode. */
1707 vi->i_mapping->a_ops = &ntfs_mst_aops;
1708 vi->i_blocks = ni->allocated_size >> 9;
1709 /*
1710 * Make sure the base inode doesn't go away and attach it to the
1711 * index inode.
1712 */
1713 igrab(base_vi);
1714 ni->ext.base_ntfs_ino = base_ni;
1715 ni->nr_extents = -1;
1716
1717 ntfs_debug("Done.");
1718 return 0;
1719iput_unm_err_out:
1720 iput(bvi);
1721unm_err_out:
1722 if (!err)
1723 err = -EIO;
1724 if (ctx)
1725 ntfs_attr_put_search_ctx(ctx);
1726 if (m)
1727 unmap_mft_record(base_ni);
1728err_out:
1729 ntfs_error(vi->i_sb, "Failed with error code %i while reading index "
1730 "inode (mft_no 0x%lx, name_len %i.", err, vi->i_ino,
1731 ni->name_len);
1732 make_bad_inode(vi);
1733 if (err != -EOPNOTSUPP && err != -ENOMEM)
1734 NVolSetErrors(vol);
1735 return err;
1736}
1737
1738/*
1739 * The MFT inode has special locking, so teach the lock validator
1740 * about this by splitting off the locking rules of the MFT from
1741 * the locking rules of other inodes. The MFT inode can never be
1742 * accessed from the VFS side (or even internally), only by the
1743 * map_mft functions.
1744 */
1745static struct lock_class_key mft_ni_runlist_lock_key, mft_ni_mrec_lock_key;
1746
1747/**
1748 * ntfs_read_inode_mount - special read_inode for mount time use only
1749 * @vi: inode to read
1750 *
1751 * Read inode FILE_MFT at mount time, only called with super_block lock
1752 * held from within the read_super() code path.
1753 *
1754 * This function exists because when it is called the page cache for $MFT/$DATA
1755 * is not initialized and hence we cannot get at the contents of mft records
1756 * by calling map_mft_record*().
1757 *
1758 * Further it needs to cope with the circular references problem, i.e. cannot
1759 * load any attributes other than $ATTRIBUTE_LIST until $DATA is loaded, because
1760 * we do not know where the other extent mft records are yet and again, because
1761 * we cannot call map_mft_record*() yet. Obviously this applies only when an
1762 * attribute list is actually present in $MFT inode.
1763 *
1764 * We solve these problems by starting with the $DATA attribute before anything
1765 * else and iterating using ntfs_attr_lookup($DATA) over all extents. As each
1766 * extent is found, we ntfs_mapping_pairs_decompress() including the implied
1767 * ntfs_runlists_merge(). Each step of the iteration necessarily provides
1768 * sufficient information for the next step to complete.
1769 *
1770 * This should work but there are two possible pit falls (see inline comments
1771 * below), but only time will tell if they are real pits or just smoke...
1772 */
1773int ntfs_read_inode_mount(struct inode *vi)
1774{
1775 VCN next_vcn, last_vcn, highest_vcn;
1776 s64 block;
1777 struct super_block *sb = vi->i_sb;
1778 ntfs_volume *vol = NTFS_SB(sb);
1779 struct buffer_head *bh;
1780 ntfs_inode *ni;
1781 MFT_RECORD *m = NULL;
1782 ATTR_RECORD *a;
1783 ntfs_attr_search_ctx *ctx;
1784 unsigned int i, nr_blocks;
1785 int err;
1786
1787 ntfs_debug("Entering.");
1788
1789 /* Initialize the ntfs specific part of @vi. */
1790 ntfs_init_big_inode(vi);
1791
1792 ni = NTFS_I(vi);
1793
1794 /* Setup the data attribute. It is special as it is mst protected. */
1795 NInoSetNonResident(ni);
1796 NInoSetMstProtected(ni);
1797 NInoSetSparseDisabled(ni);
1798 ni->type = AT_DATA;
1799 ni->name = NULL;
1800 ni->name_len = 0;
1801 /*
1802 * This sets up our little cheat allowing us to reuse the async read io
1803 * completion handler for directories.
1804 */
1805 ni->itype.index.block_size = vol->mft_record_size;
1806 ni->itype.index.block_size_bits = vol->mft_record_size_bits;
1807
1808 /* Very important! Needed to be able to call map_mft_record*(). */
1809 vol->mft_ino = vi;
1810
1811 /* Allocate enough memory to read the first mft record. */
1812 if (vol->mft_record_size > 64 * 1024) {
1813 ntfs_error(sb, "Unsupported mft record size %i (max 64kiB).",
1814 vol->mft_record_size);
1815 goto err_out;
1816 }
1817 i = vol->mft_record_size;
1818 if (i < sb->s_blocksize)
1819 i = sb->s_blocksize;
1820 m = (MFT_RECORD*)ntfs_malloc_nofs(i);
1821 if (!m) {
1822 ntfs_error(sb, "Failed to allocate buffer for $MFT record 0.");
1823 goto err_out;
1824 }
1825
1826 /* Determine the first block of the $MFT/$DATA attribute. */
1827 block = vol->mft_lcn << vol->cluster_size_bits >>
1828 sb->s_blocksize_bits;
1829 nr_blocks = vol->mft_record_size >> sb->s_blocksize_bits;
1830 if (!nr_blocks)
1831 nr_blocks = 1;
1832
1833 /* Load $MFT/$DATA's first mft record. */
1834 for (i = 0; i < nr_blocks; i++) {
1835 bh = sb_bread(sb, block++);
1836 if (!bh) {
1837 ntfs_error(sb, "Device read failed.");
1838 goto err_out;
1839 }
1840 memcpy((char*)m + (i << sb->s_blocksize_bits), bh->b_data,
1841 sb->s_blocksize);
1842 brelse(bh);
1843 }
1844
1845 /* Apply the mst fixups. */
1846 if (post_read_mst_fixup((NTFS_RECORD*)m, vol->mft_record_size)) {
1847 /* FIXME: Try to use the $MFTMirr now. */
1848 ntfs_error(sb, "MST fixup failed. $MFT is corrupt.");
1849 goto err_out;
1850 }
1851
1852 /* Need this to sanity check attribute list references to $MFT. */
1853 vi->i_generation = ni->seq_no = le16_to_cpu(m->sequence_number);
1854
1855 /* Provides readpage() and sync_page() for map_mft_record(). */
1856 vi->i_mapping->a_ops = &ntfs_mst_aops;
1857
1858 ctx = ntfs_attr_get_search_ctx(ni, m);
1859 if (!ctx) {
1860 err = -ENOMEM;
1861 goto err_out;
1862 }
1863
1864 /* Find the attribute list attribute if present. */
1865 err = ntfs_attr_lookup(AT_ATTRIBUTE_LIST, NULL, 0, 0, 0, NULL, 0, ctx);
1866 if (err) {
1867 if (unlikely(err != -ENOENT)) {
1868 ntfs_error(sb, "Failed to lookup attribute list "
1869 "attribute. You should run chkdsk.");
1870 goto put_err_out;
1871 }
1872 } else /* if (!err) */ {
1873 ATTR_LIST_ENTRY *al_entry, *next_al_entry;
1874 u8 *al_end;
1875 static const char *es = " Not allowed. $MFT is corrupt. "
1876 "You should run chkdsk.";
1877
1878 ntfs_debug("Attribute list attribute found in $MFT.");
1879 NInoSetAttrList(ni);
1880 a = ctx->attr;
1881 if (a->flags & ATTR_COMPRESSION_MASK) {
1882 ntfs_error(sb, "Attribute list attribute is "
1883 "compressed.%s", es);
1884 goto put_err_out;
1885 }
1886 if (a->flags & ATTR_IS_ENCRYPTED ||
1887 a->flags & ATTR_IS_SPARSE) {
1888 if (a->non_resident) {
1889 ntfs_error(sb, "Non-resident attribute list "
1890 "attribute is encrypted/"
1891 "sparse.%s", es);
1892 goto put_err_out;
1893 }
1894 ntfs_warning(sb, "Resident attribute list attribute "
1895 "in $MFT system file is marked "
1896 "encrypted/sparse which is not true. "
1897 "However, Windows allows this and "
1898 "chkdsk does not detect or correct it "
1899 "so we will just ignore the invalid "
1900 "flags and pretend they are not set.");
1901 }
1902 /* Now allocate memory for the attribute list. */
1903 ni->attr_list_size = (u32)ntfs_attr_size(a);
1904 ni->attr_list = ntfs_malloc_nofs(ni->attr_list_size);
1905 if (!ni->attr_list) {
1906 ntfs_error(sb, "Not enough memory to allocate buffer "
1907 "for attribute list.");
1908 goto put_err_out;
1909 }
1910 if (a->non_resident) {
1911 NInoSetAttrListNonResident(ni);
1912 if (a->data.non_resident.lowest_vcn) {
1913 ntfs_error(sb, "Attribute list has non zero "
1914 "lowest_vcn. $MFT is corrupt. "
1915 "You should run chkdsk.");
1916 goto put_err_out;
1917 }
1918 /* Setup the runlist. */
1919 ni->attr_list_rl.rl = ntfs_mapping_pairs_decompress(vol,
1920 a, NULL);
1921 if (IS_ERR(ni->attr_list_rl.rl)) {
1922 err = PTR_ERR(ni->attr_list_rl.rl);
1923 ni->attr_list_rl.rl = NULL;
1924 ntfs_error(sb, "Mapping pairs decompression "
1925 "failed with error code %i.",
1926 -err);
1927 goto put_err_out;
1928 }
1929 /* Now load the attribute list. */
1930 if ((err = load_attribute_list(vol, &ni->attr_list_rl,
1931 ni->attr_list, ni->attr_list_size,
1932 sle64_to_cpu(a->data.
1933 non_resident.initialized_size)))) {
1934 ntfs_error(sb, "Failed to load attribute list "
1935 "attribute with error code %i.",
1936 -err);
1937 goto put_err_out;
1938 }
1939 } else /* if (!ctx.attr->non_resident) */ {
1940 if ((u8*)a + le16_to_cpu(
1941 a->data.resident.value_offset) +
1942 le32_to_cpu(
1943 a->data.resident.value_length) >
1944 (u8*)ctx->mrec + vol->mft_record_size) {
1945 ntfs_error(sb, "Corrupt attribute list "
1946 "attribute.");
1947 goto put_err_out;
1948 }
1949 /* Now copy the attribute list. */
1950 memcpy(ni->attr_list, (u8*)a + le16_to_cpu(
1951 a->data.resident.value_offset),
1952 le32_to_cpu(
1953 a->data.resident.value_length));
1954 }
1955 /* The attribute list is now setup in memory. */
1956 /*
1957 * FIXME: I don't know if this case is actually possible.
1958 * According to logic it is not possible but I have seen too
1959 * many weird things in MS software to rely on logic... Thus we
1960 * perform a manual search and make sure the first $MFT/$DATA
1961 * extent is in the base inode. If it is not we abort with an
1962 * error and if we ever see a report of this error we will need
1963 * to do some magic in order to have the necessary mft record
1964 * loaded and in the right place in the page cache. But
1965 * hopefully logic will prevail and this never happens...
1966 */
1967 al_entry = (ATTR_LIST_ENTRY*)ni->attr_list;
1968 al_end = (u8*)al_entry + ni->attr_list_size;
1969 for (;; al_entry = next_al_entry) {
1970 /* Out of bounds check. */
1971 if ((u8*)al_entry < ni->attr_list ||
1972 (u8*)al_entry > al_end)
1973 goto em_put_err_out;
1974 /* Catch the end of the attribute list. */
1975 if ((u8*)al_entry == al_end)
1976 goto em_put_err_out;
1977 if (!al_entry->length)
1978 goto em_put_err_out;
1979 if ((u8*)al_entry + 6 > al_end || (u8*)al_entry +
1980 le16_to_cpu(al_entry->length) > al_end)
1981 goto em_put_err_out;
1982 next_al_entry = (ATTR_LIST_ENTRY*)((u8*)al_entry +
1983 le16_to_cpu(al_entry->length));
1984 if (le32_to_cpu(al_entry->type) > le32_to_cpu(AT_DATA))
1985 goto em_put_err_out;
1986 if (AT_DATA != al_entry->type)
1987 continue;
1988 /* We want an unnamed attribute. */
1989 if (al_entry->name_length)
1990 goto em_put_err_out;
1991 /* Want the first entry, i.e. lowest_vcn == 0. */
1992 if (al_entry->lowest_vcn)
1993 goto em_put_err_out;
1994 /* First entry has to be in the base mft record. */
1995 if (MREF_LE(al_entry->mft_reference) != vi->i_ino) {
1996 /* MFT references do not match, logic fails. */
1997 ntfs_error(sb, "BUG: The first $DATA extent "
1998 "of $MFT is not in the base "
1999 "mft record. Please report "
2000 "you saw this message to "
2001 "linux-ntfs-dev@lists."
2002 "sourceforge.net");
2003 goto put_err_out;
2004 } else {
2005 /* Sequence numbers must match. */
2006 if (MSEQNO_LE(al_entry->mft_reference) !=
2007 ni->seq_no)
2008 goto em_put_err_out;
2009 /* Got it. All is ok. We can stop now. */
2010 break;
2011 }
2012 }
2013 }
2014
2015 ntfs_attr_reinit_search_ctx(ctx);
2016
2017 /* Now load all attribute extents. */
2018 a = NULL;
2019 next_vcn = last_vcn = highest_vcn = 0;
2020 while (!(err = ntfs_attr_lookup(AT_DATA, NULL, 0, 0, next_vcn, NULL, 0,
2021 ctx))) {
2022 runlist_element *nrl;
2023
2024 /* Cache the current attribute. */
2025 a = ctx->attr;
2026 /* $MFT must be non-resident. */
2027 if (!a->non_resident) {
2028 ntfs_error(sb, "$MFT must be non-resident but a "
2029 "resident extent was found. $MFT is "
2030 "corrupt. Run chkdsk.");
2031 goto put_err_out;
2032 }
2033 /* $MFT must be uncompressed and unencrypted. */
2034 if (a->flags & ATTR_COMPRESSION_MASK ||
2035 a->flags & ATTR_IS_ENCRYPTED ||
2036 a->flags & ATTR_IS_SPARSE) {
2037 ntfs_error(sb, "$MFT must be uncompressed, "
2038 "non-sparse, and unencrypted but a "
2039 "compressed/sparse/encrypted extent "
2040 "was found. $MFT is corrupt. Run "
2041 "chkdsk.");
2042 goto put_err_out;
2043 }
2044 /*
2045 * Decompress the mapping pairs array of this extent and merge
2046 * the result into the existing runlist. No need for locking
2047 * as we have exclusive access to the inode at this time and we
2048 * are a mount in progress task, too.
2049 */
2050 nrl = ntfs_mapping_pairs_decompress(vol, a, ni->runlist.rl);
2051 if (IS_ERR(nrl)) {
2052 ntfs_error(sb, "ntfs_mapping_pairs_decompress() "
2053 "failed with error code %ld. $MFT is "
2054 "corrupt.", PTR_ERR(nrl));
2055 goto put_err_out;
2056 }
2057 ni->runlist.rl = nrl;
2058
2059 /* Are we in the first extent? */
2060 if (!next_vcn) {
2061 if (a->data.non_resident.lowest_vcn) {
2062 ntfs_error(sb, "First extent of $DATA "
2063 "attribute has non zero "
2064 "lowest_vcn. $MFT is corrupt. "
2065 "You should run chkdsk.");
2066 goto put_err_out;
2067 }
2068 /* Get the last vcn in the $DATA attribute. */
2069 last_vcn = sle64_to_cpu(
2070 a->data.non_resident.allocated_size)
2071 >> vol->cluster_size_bits;
2072 /* Fill in the inode size. */
2073 vi->i_size = sle64_to_cpu(
2074 a->data.non_resident.data_size);
2075 ni->initialized_size = sle64_to_cpu(
2076 a->data.non_resident.initialized_size);
2077 ni->allocated_size = sle64_to_cpu(
2078 a->data.non_resident.allocated_size);
2079 /*
2080 * Verify the number of mft records does not exceed
2081 * 2^32 - 1.
2082 */
2083 if ((vi->i_size >> vol->mft_record_size_bits) >=
2084 (1ULL << 32)) {
2085 ntfs_error(sb, "$MFT is too big! Aborting.");
2086 goto put_err_out;
2087 }
2088 /*
2089 * We have got the first extent of the runlist for
2090 * $MFT which means it is now relatively safe to call
2091 * the normal ntfs_read_inode() function.
2092 * Complete reading the inode, this will actually
2093 * re-read the mft record for $MFT, this time entering
2094 * it into the page cache with which we complete the
2095 * kick start of the volume. It should be safe to do
2096 * this now as the first extent of $MFT/$DATA is
2097 * already known and we would hope that we don't need
2098 * further extents in order to find the other
2099 * attributes belonging to $MFT. Only time will tell if
2100 * this is really the case. If not we will have to play
2101 * magic at this point, possibly duplicating a lot of
2102 * ntfs_read_inode() at this point. We will need to
2103 * ensure we do enough of its work to be able to call
2104 * ntfs_read_inode() on extents of $MFT/$DATA. But lets
2105 * hope this never happens...
2106 */
2107 ntfs_read_locked_inode(vi);
2108 if (is_bad_inode(vi)) {
2109 ntfs_error(sb, "ntfs_read_inode() of $MFT "
2110 "failed. BUG or corrupt $MFT. "
2111 "Run chkdsk and if no errors "
2112 "are found, please report you "
2113 "saw this message to "
2114 "linux-ntfs-dev@lists."
2115 "sourceforge.net");
2116 ntfs_attr_put_search_ctx(ctx);
2117 /* Revert to the safe super operations. */
2118 ntfs_free(m);
2119 return -1;
2120 }
2121 /*
2122 * Re-initialize some specifics about $MFT's inode as
2123 * ntfs_read_inode() will have set up the default ones.
2124 */
2125 /* Set uid and gid to root. */
2126 vi->i_uid = GLOBAL_ROOT_UID;
2127 vi->i_gid = GLOBAL_ROOT_GID;
2128 /* Regular file. No access for anyone. */
2129 vi->i_mode = S_IFREG;
2130 /* No VFS initiated operations allowed for $MFT. */
2131 vi->i_op = &ntfs_empty_inode_ops;
2132 vi->i_fop = &ntfs_empty_file_ops;
2133 }
2134
2135 /* Get the lowest vcn for the next extent. */
2136 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2137 next_vcn = highest_vcn + 1;
2138
2139 /* Only one extent or error, which we catch below. */
2140 if (next_vcn <= 0)
2141 break;
2142
2143 /* Avoid endless loops due to corruption. */
2144 if (next_vcn < sle64_to_cpu(
2145 a->data.non_resident.lowest_vcn)) {
2146 ntfs_error(sb, "$MFT has corrupt attribute list "
2147 "attribute. Run chkdsk.");
2148 goto put_err_out;
2149 }
2150 }
2151 if (err != -ENOENT) {
2152 ntfs_error(sb, "Failed to lookup $MFT/$DATA attribute extent. "
2153 "$MFT is corrupt. Run chkdsk.");
2154 goto put_err_out;
2155 }
2156 if (!a) {
2157 ntfs_error(sb, "$MFT/$DATA attribute not found. $MFT is "
2158 "corrupt. Run chkdsk.");
2159 goto put_err_out;
2160 }
2161 if (highest_vcn && highest_vcn != last_vcn - 1) {
2162 ntfs_error(sb, "Failed to load the complete runlist for "
2163 "$MFT/$DATA. Driver bug or corrupt $MFT. "
2164 "Run chkdsk.");
2165 ntfs_debug("highest_vcn = 0x%llx, last_vcn - 1 = 0x%llx",
2166 (unsigned long long)highest_vcn,
2167 (unsigned long long)last_vcn - 1);
2168 goto put_err_out;
2169 }
2170 ntfs_attr_put_search_ctx(ctx);
2171 ntfs_debug("Done.");
2172 ntfs_free(m);
2173
2174 /*
2175 * Split the locking rules of the MFT inode from the
2176 * locking rules of other inodes:
2177 */
2178 lockdep_set_class(&ni->runlist.lock, &mft_ni_runlist_lock_key);
2179 lockdep_set_class(&ni->mrec_lock, &mft_ni_mrec_lock_key);
2180
2181 return 0;
2182
2183em_put_err_out:
2184 ntfs_error(sb, "Couldn't find first extent of $DATA attribute in "
2185 "attribute list. $MFT is corrupt. Run chkdsk.");
2186put_err_out:
2187 ntfs_attr_put_search_ctx(ctx);
2188err_out:
2189 ntfs_error(sb, "Failed. Marking inode as bad.");
2190 make_bad_inode(vi);
2191 ntfs_free(m);
2192 return -1;
2193}
2194
2195static void __ntfs_clear_inode(ntfs_inode *ni)
2196{
2197 /* Free all alocated memory. */
2198 down_write(&ni->runlist.lock);
2199 if (ni->runlist.rl) {
2200 ntfs_free(ni->runlist.rl);
2201 ni->runlist.rl = NULL;
2202 }
2203 up_write(&ni->runlist.lock);
2204
2205 if (ni->attr_list) {
2206 ntfs_free(ni->attr_list);
2207 ni->attr_list = NULL;
2208 }
2209
2210 down_write(&ni->attr_list_rl.lock);
2211 if (ni->attr_list_rl.rl) {
2212 ntfs_free(ni->attr_list_rl.rl);
2213 ni->attr_list_rl.rl = NULL;
2214 }
2215 up_write(&ni->attr_list_rl.lock);
2216
2217 if (ni->name_len && ni->name != I30) {
2218 /* Catch bugs... */
2219 BUG_ON(!ni->name);
2220 kfree(ni->name);
2221 }
2222}
2223
2224void ntfs_clear_extent_inode(ntfs_inode *ni)
2225{
2226 ntfs_debug("Entering for inode 0x%lx.", ni->mft_no);
2227
2228 BUG_ON(NInoAttr(ni));
2229 BUG_ON(ni->nr_extents != -1);
2230
2231#ifdef NTFS_RW
2232 if (NInoDirty(ni)) {
2233 if (!is_bad_inode(VFS_I(ni->ext.base_ntfs_ino)))
2234 ntfs_error(ni->vol->sb, "Clearing dirty extent inode! "
2235 "Losing data! This is a BUG!!!");
2236 // FIXME: Do something!!!
2237 }
2238#endif /* NTFS_RW */
2239
2240 __ntfs_clear_inode(ni);
2241
2242 /* Bye, bye... */
2243 ntfs_destroy_extent_inode(ni);
2244}
2245
2246/**
2247 * ntfs_evict_big_inode - clean up the ntfs specific part of an inode
2248 * @vi: vfs inode pending annihilation
2249 *
2250 * When the VFS is going to remove an inode from memory, ntfs_clear_big_inode()
2251 * is called, which deallocates all memory belonging to the NTFS specific part
2252 * of the inode and returns.
2253 *
2254 * If the MFT record is dirty, we commit it before doing anything else.
2255 */
2256void ntfs_evict_big_inode(struct inode *vi)
2257{
2258 ntfs_inode *ni = NTFS_I(vi);
2259
2260 truncate_inode_pages_final(&vi->i_data);
2261 clear_inode(vi);
2262
2263#ifdef NTFS_RW
2264 if (NInoDirty(ni)) {
2265 bool was_bad = (is_bad_inode(vi));
2266
2267 /* Committing the inode also commits all extent inodes. */
2268 ntfs_commit_inode(vi);
2269
2270 if (!was_bad && (is_bad_inode(vi) || NInoDirty(ni))) {
2271 ntfs_error(vi->i_sb, "Failed to commit dirty inode "
2272 "0x%lx. Losing data!", vi->i_ino);
2273 // FIXME: Do something!!!
2274 }
2275 }
2276#endif /* NTFS_RW */
2277
2278 /* No need to lock at this stage as no one else has a reference. */
2279 if (ni->nr_extents > 0) {
2280 int i;
2281
2282 for (i = 0; i < ni->nr_extents; i++)
2283 ntfs_clear_extent_inode(ni->ext.extent_ntfs_inos[i]);
2284 kfree(ni->ext.extent_ntfs_inos);
2285 }
2286
2287 __ntfs_clear_inode(ni);
2288
2289 if (NInoAttr(ni)) {
2290 /* Release the base inode if we are holding it. */
2291 if (ni->nr_extents == -1) {
2292 iput(VFS_I(ni->ext.base_ntfs_ino));
2293 ni->nr_extents = 0;
2294 ni->ext.base_ntfs_ino = NULL;
2295 }
2296 }
2297 return;
2298}
2299
2300/**
2301 * ntfs_show_options - show mount options in /proc/mounts
2302 * @sf: seq_file in which to write our mount options
2303 * @root: root of the mounted tree whose mount options to display
2304 *
2305 * Called by the VFS once for each mounted ntfs volume when someone reads
2306 * /proc/mounts in order to display the NTFS specific mount options of each
2307 * mount. The mount options of fs specified by @root are written to the seq file
2308 * @sf and success is returned.
2309 */
2310int ntfs_show_options(struct seq_file *sf, struct dentry *root)
2311{
2312 ntfs_volume *vol = NTFS_SB(root->d_sb);
2313 int i;
2314
2315 seq_printf(sf, ",uid=%i", from_kuid_munged(&init_user_ns, vol->uid));
2316 seq_printf(sf, ",gid=%i", from_kgid_munged(&init_user_ns, vol->gid));
2317 if (vol->fmask == vol->dmask)
2318 seq_printf(sf, ",umask=0%o", vol->fmask);
2319 else {
2320 seq_printf(sf, ",fmask=0%o", vol->fmask);
2321 seq_printf(sf, ",dmask=0%o", vol->dmask);
2322 }
2323 seq_printf(sf, ",nls=%s", vol->nls_map->charset);
2324 if (NVolCaseSensitive(vol))
2325 seq_printf(sf, ",case_sensitive");
2326 if (NVolShowSystemFiles(vol))
2327 seq_printf(sf, ",show_sys_files");
2328 if (!NVolSparseEnabled(vol))
2329 seq_printf(sf, ",disable_sparse");
2330 for (i = 0; on_errors_arr[i].val; i++) {
2331 if (on_errors_arr[i].val & vol->on_errors)
2332 seq_printf(sf, ",errors=%s", on_errors_arr[i].str);
2333 }
2334 seq_printf(sf, ",mft_zone_multiplier=%i", vol->mft_zone_multiplier);
2335 return 0;
2336}
2337
2338#ifdef NTFS_RW
2339
2340static const char *es = " Leaving inconsistent metadata. Unmount and run "
2341 "chkdsk.";
2342
2343/**
2344 * ntfs_truncate - called when the i_size of an ntfs inode is changed
2345 * @vi: inode for which the i_size was changed
2346 *
2347 * We only support i_size changes for normal files at present, i.e. not
2348 * compressed and not encrypted. This is enforced in ntfs_setattr(), see
2349 * below.
2350 *
2351 * The kernel guarantees that @vi is a regular file (S_ISREG() is true) and
2352 * that the change is allowed.
2353 *
2354 * This implies for us that @vi is a file inode rather than a directory, index,
2355 * or attribute inode as well as that @vi is a base inode.
2356 *
2357 * Returns 0 on success or -errno on error.
2358 *
2359 * Called with ->i_mutex held.
2360 */
2361int ntfs_truncate(struct inode *vi)
2362{
2363 s64 new_size, old_size, nr_freed, new_alloc_size, old_alloc_size;
2364 VCN highest_vcn;
2365 unsigned long flags;
2366 ntfs_inode *base_ni, *ni = NTFS_I(vi);
2367 ntfs_volume *vol = ni->vol;
2368 ntfs_attr_search_ctx *ctx;
2369 MFT_RECORD *m;
2370 ATTR_RECORD *a;
2371 const char *te = " Leaving file length out of sync with i_size.";
2372 int err, mp_size, size_change, alloc_change;
2373 u32 attr_len;
2374
2375 ntfs_debug("Entering for inode 0x%lx.", vi->i_ino);
2376 BUG_ON(NInoAttr(ni));
2377 BUG_ON(S_ISDIR(vi->i_mode));
2378 BUG_ON(NInoMstProtected(ni));
2379 BUG_ON(ni->nr_extents < 0);
2380retry_truncate:
2381 /*
2382 * Lock the runlist for writing and map the mft record to ensure it is
2383 * safe to mess with the attribute runlist and sizes.
2384 */
2385 down_write(&ni->runlist.lock);
2386 if (!NInoAttr(ni))
2387 base_ni = ni;
2388 else
2389 base_ni = ni->ext.base_ntfs_ino;
2390 m = map_mft_record(base_ni);
2391 if (IS_ERR(m)) {
2392 err = PTR_ERR(m);
2393 ntfs_error(vi->i_sb, "Failed to map mft record for inode 0x%lx "
2394 "(error code %d).%s", vi->i_ino, err, te);
2395 ctx = NULL;
2396 m = NULL;
2397 goto old_bad_out;
2398 }
2399 ctx = ntfs_attr_get_search_ctx(base_ni, m);
2400 if (unlikely(!ctx)) {
2401 ntfs_error(vi->i_sb, "Failed to allocate a search context for "
2402 "inode 0x%lx (not enough memory).%s",
2403 vi->i_ino, te);
2404 err = -ENOMEM;
2405 goto old_bad_out;
2406 }
2407 err = ntfs_attr_lookup(ni->type, ni->name, ni->name_len,
2408 CASE_SENSITIVE, 0, NULL, 0, ctx);
2409 if (unlikely(err)) {
2410 if (err == -ENOENT) {
2411 ntfs_error(vi->i_sb, "Open attribute is missing from "
2412 "mft record. Inode 0x%lx is corrupt. "
2413 "Run chkdsk.%s", vi->i_ino, te);
2414 err = -EIO;
2415 } else
2416 ntfs_error(vi->i_sb, "Failed to lookup attribute in "
2417 "inode 0x%lx (error code %d).%s",
2418 vi->i_ino, err, te);
2419 goto old_bad_out;
2420 }
2421 m = ctx->mrec;
2422 a = ctx->attr;
2423 /*
2424 * The i_size of the vfs inode is the new size for the attribute value.
2425 */
2426 new_size = i_size_read(vi);
2427 /* The current size of the attribute value is the old size. */
2428 old_size = ntfs_attr_size(a);
2429 /* Calculate the new allocated size. */
2430 if (NInoNonResident(ni))
2431 new_alloc_size = (new_size + vol->cluster_size - 1) &
2432 ~(s64)vol->cluster_size_mask;
2433 else
2434 new_alloc_size = (new_size + 7) & ~7;
2435 /* The current allocated size is the old allocated size. */
2436 read_lock_irqsave(&ni->size_lock, flags);
2437 old_alloc_size = ni->allocated_size;
2438 read_unlock_irqrestore(&ni->size_lock, flags);
2439 /*
2440 * The change in the file size. This will be 0 if no change, >0 if the
2441 * size is growing, and <0 if the size is shrinking.
2442 */
2443 size_change = -1;
2444 if (new_size - old_size >= 0) {
2445 size_change = 1;
2446 if (new_size == old_size)
2447 size_change = 0;
2448 }
2449 /* As above for the allocated size. */
2450 alloc_change = -1;
2451 if (new_alloc_size - old_alloc_size >= 0) {
2452 alloc_change = 1;
2453 if (new_alloc_size == old_alloc_size)
2454 alloc_change = 0;
2455 }
2456 /*
2457 * If neither the size nor the allocation are being changed there is
2458 * nothing to do.
2459 */
2460 if (!size_change && !alloc_change)
2461 goto unm_done;
2462 /* If the size is changing, check if new size is allowed in $AttrDef. */
2463 if (size_change) {
2464 err = ntfs_attr_size_bounds_check(vol, ni->type, new_size);
2465 if (unlikely(err)) {
2466 if (err == -ERANGE) {
2467 ntfs_error(vol->sb, "Truncate would cause the "
2468 "inode 0x%lx to %simum size "
2469 "for its attribute type "
2470 "(0x%x). Aborting truncate.",
2471 vi->i_ino,
2472 new_size > old_size ? "exceed "
2473 "the max" : "go under the min",
2474 le32_to_cpu(ni->type));
2475 err = -EFBIG;
2476 } else {
2477 ntfs_error(vol->sb, "Inode 0x%lx has unknown "
2478 "attribute type 0x%x. "
2479 "Aborting truncate.",
2480 vi->i_ino,
2481 le32_to_cpu(ni->type));
2482 err = -EIO;
2483 }
2484 /* Reset the vfs inode size to the old size. */
2485 i_size_write(vi, old_size);
2486 goto err_out;
2487 }
2488 }
2489 if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2490 ntfs_warning(vi->i_sb, "Changes in inode size are not "
2491 "supported yet for %s files, ignoring.",
2492 NInoCompressed(ni) ? "compressed" :
2493 "encrypted");
2494 err = -EOPNOTSUPP;
2495 goto bad_out;
2496 }
2497 if (a->non_resident)
2498 goto do_non_resident_truncate;
2499 BUG_ON(NInoNonResident(ni));
2500 /* Resize the attribute record to best fit the new attribute size. */
2501 if (new_size < vol->mft_record_size &&
2502 !ntfs_resident_attr_value_resize(m, a, new_size)) {
2503 /* The resize succeeded! */
2504 flush_dcache_mft_record_page(ctx->ntfs_ino);
2505 mark_mft_record_dirty(ctx->ntfs_ino);
2506 write_lock_irqsave(&ni->size_lock, flags);
2507 /* Update the sizes in the ntfs inode and all is done. */
2508 ni->allocated_size = le32_to_cpu(a->length) -
2509 le16_to_cpu(a->data.resident.value_offset);
2510 /*
2511 * Note ntfs_resident_attr_value_resize() has already done any
2512 * necessary data clearing in the attribute record. When the
2513 * file is being shrunk vmtruncate() will already have cleared
2514 * the top part of the last partial page, i.e. since this is
2515 * the resident case this is the page with index 0. However,
2516 * when the file is being expanded, the page cache page data
2517 * between the old data_size, i.e. old_size, and the new_size
2518 * has not been zeroed. Fortunately, we do not need to zero it
2519 * either since on one hand it will either already be zero due
2520 * to both readpage and writepage clearing partial page data
2521 * beyond i_size in which case there is nothing to do or in the
2522 * case of the file being mmap()ped at the same time, POSIX
2523 * specifies that the behaviour is unspecified thus we do not
2524 * have to do anything. This means that in our implementation
2525 * in the rare case that the file is mmap()ped and a write
2526 * occurred into the mmap()ped region just beyond the file size
2527 * and writepage has not yet been called to write out the page
2528 * (which would clear the area beyond the file size) and we now
2529 * extend the file size to incorporate this dirty region
2530 * outside the file size, a write of the page would result in
2531 * this data being written to disk instead of being cleared.
2532 * Given both POSIX and the Linux mmap(2) man page specify that
2533 * this corner case is undefined, we choose to leave it like
2534 * that as this is much simpler for us as we cannot lock the
2535 * relevant page now since we are holding too many ntfs locks
2536 * which would result in a lock reversal deadlock.
2537 */
2538 ni->initialized_size = new_size;
2539 write_unlock_irqrestore(&ni->size_lock, flags);
2540 goto unm_done;
2541 }
2542 /* If the above resize failed, this must be an attribute extension. */
2543 BUG_ON(size_change < 0);
2544 /*
2545 * We have to drop all the locks so we can call
2546 * ntfs_attr_make_non_resident(). This could be optimised by try-
2547 * locking the first page cache page and only if that fails dropping
2548 * the locks, locking the page, and redoing all the locking and
2549 * lookups. While this would be a huge optimisation, it is not worth
2550 * it as this is definitely a slow code path as it only ever can happen
2551 * once for any given file.
2552 */
2553 ntfs_attr_put_search_ctx(ctx);
2554 unmap_mft_record(base_ni);
2555 up_write(&ni->runlist.lock);
2556 /*
2557 * Not enough space in the mft record, try to make the attribute
2558 * non-resident and if successful restart the truncation process.
2559 */
2560 err = ntfs_attr_make_non_resident(ni, old_size);
2561 if (likely(!err))
2562 goto retry_truncate;
2563 /*
2564 * Could not make non-resident. If this is due to this not being
2565 * permitted for this attribute type or there not being enough space,
2566 * try to make other attributes non-resident. Otherwise fail.
2567 */
2568 if (unlikely(err != -EPERM && err != -ENOSPC)) {
2569 ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, attribute "
2570 "type 0x%x, because the conversion from "
2571 "resident to non-resident attribute failed "
2572 "with error code %i.", vi->i_ino,
2573 (unsigned)le32_to_cpu(ni->type), err);
2574 if (err != -ENOMEM)
2575 err = -EIO;
2576 goto conv_err_out;
2577 }
2578 /* TODO: Not implemented from here, abort. */
2579 if (err == -ENOSPC)
2580 ntfs_error(vol->sb, "Not enough space in the mft record/on "
2581 "disk for the non-resident attribute value. "
2582 "This case is not implemented yet.");
2583 else /* if (err == -EPERM) */
2584 ntfs_error(vol->sb, "This attribute type may not be "
2585 "non-resident. This case is not implemented "
2586 "yet.");
2587 err = -EOPNOTSUPP;
2588 goto conv_err_out;
2589#if 0
2590 // TODO: Attempt to make other attributes non-resident.
2591 if (!err)
2592 goto do_resident_extend;
2593 /*
2594 * Both the attribute list attribute and the standard information
2595 * attribute must remain in the base inode. Thus, if this is one of
2596 * these attributes, we have to try to move other attributes out into
2597 * extent mft records instead.
2598 */
2599 if (ni->type == AT_ATTRIBUTE_LIST ||
2600 ni->type == AT_STANDARD_INFORMATION) {
2601 // TODO: Attempt to move other attributes into extent mft
2602 // records.
2603 err = -EOPNOTSUPP;
2604 if (!err)
2605 goto do_resident_extend;
2606 goto err_out;
2607 }
2608 // TODO: Attempt to move this attribute to an extent mft record, but
2609 // only if it is not already the only attribute in an mft record in
2610 // which case there would be nothing to gain.
2611 err = -EOPNOTSUPP;
2612 if (!err)
2613 goto do_resident_extend;
2614 /* There is nothing we can do to make enough space. )-: */
2615 goto err_out;
2616#endif
2617do_non_resident_truncate:
2618 BUG_ON(!NInoNonResident(ni));
2619 if (alloc_change < 0) {
2620 highest_vcn = sle64_to_cpu(a->data.non_resident.highest_vcn);
2621 if (highest_vcn > 0 &&
2622 old_alloc_size >> vol->cluster_size_bits >
2623 highest_vcn + 1) {
2624 /*
2625 * This attribute has multiple extents. Not yet
2626 * supported.
2627 */
2628 ntfs_error(vol->sb, "Cannot truncate inode 0x%lx, "
2629 "attribute type 0x%x, because the "
2630 "attribute is highly fragmented (it "
2631 "consists of multiple extents) and "
2632 "this case is not implemented yet.",
2633 vi->i_ino,
2634 (unsigned)le32_to_cpu(ni->type));
2635 err = -EOPNOTSUPP;
2636 goto bad_out;
2637 }
2638 }
2639 /*
2640 * If the size is shrinking, need to reduce the initialized_size and
2641 * the data_size before reducing the allocation.
2642 */
2643 if (size_change < 0) {
2644 /*
2645 * Make the valid size smaller (i_size is already up-to-date).
2646 */
2647 write_lock_irqsave(&ni->size_lock, flags);
2648 if (new_size < ni->initialized_size) {
2649 ni->initialized_size = new_size;
2650 a->data.non_resident.initialized_size =
2651 cpu_to_sle64(new_size);
2652 }
2653 a->data.non_resident.data_size = cpu_to_sle64(new_size);
2654 write_unlock_irqrestore(&ni->size_lock, flags);
2655 flush_dcache_mft_record_page(ctx->ntfs_ino);
2656 mark_mft_record_dirty(ctx->ntfs_ino);
2657 /* If the allocated size is not changing, we are done. */
2658 if (!alloc_change)
2659 goto unm_done;
2660 /*
2661 * If the size is shrinking it makes no sense for the
2662 * allocation to be growing.
2663 */
2664 BUG_ON(alloc_change > 0);
2665 } else /* if (size_change >= 0) */ {
2666 /*
2667 * The file size is growing or staying the same but the
2668 * allocation can be shrinking, growing or staying the same.
2669 */
2670 if (alloc_change > 0) {
2671 /*
2672 * We need to extend the allocation and possibly update
2673 * the data size. If we are updating the data size,
2674 * since we are not touching the initialized_size we do
2675 * not need to worry about the actual data on disk.
2676 * And as far as the page cache is concerned, there
2677 * will be no pages beyond the old data size and any
2678 * partial region in the last page between the old and
2679 * new data size (or the end of the page if the new
2680 * data size is outside the page) does not need to be
2681 * modified as explained above for the resident
2682 * attribute truncate case. To do this, we simply drop
2683 * the locks we hold and leave all the work to our
2684 * friendly helper ntfs_attr_extend_allocation().
2685 */
2686 ntfs_attr_put_search_ctx(ctx);
2687 unmap_mft_record(base_ni);
2688 up_write(&ni->runlist.lock);
2689 err = ntfs_attr_extend_allocation(ni, new_size,
2690 size_change > 0 ? new_size : -1, -1);
2691 /*
2692 * ntfs_attr_extend_allocation() will have done error
2693 * output already.
2694 */
2695 goto done;
2696 }
2697 if (!alloc_change)
2698 goto alloc_done;
2699 }
2700 /* alloc_change < 0 */
2701 /* Free the clusters. */
2702 nr_freed = ntfs_cluster_free(ni, new_alloc_size >>
2703 vol->cluster_size_bits, -1, ctx);
2704 m = ctx->mrec;
2705 a = ctx->attr;
2706 if (unlikely(nr_freed < 0)) {
2707 ntfs_error(vol->sb, "Failed to release cluster(s) (error code "
2708 "%lli). Unmount and run chkdsk to recover "
2709 "the lost cluster(s).", (long long)nr_freed);
2710 NVolSetErrors(vol);
2711 nr_freed = 0;
2712 }
2713 /* Truncate the runlist. */
2714 err = ntfs_rl_truncate_nolock(vol, &ni->runlist,
2715 new_alloc_size >> vol->cluster_size_bits);
2716 /*
2717 * If the runlist truncation failed and/or the search context is no
2718 * longer valid, we cannot resize the attribute record or build the
2719 * mapping pairs array thus we mark the inode bad so that no access to
2720 * the freed clusters can happen.
2721 */
2722 if (unlikely(err || IS_ERR(m))) {
2723 ntfs_error(vol->sb, "Failed to %s (error code %li).%s",
2724 IS_ERR(m) ?
2725 "restore attribute search context" :
2726 "truncate attribute runlist",
2727 IS_ERR(m) ? PTR_ERR(m) : err, es);
2728 err = -EIO;
2729 goto bad_out;
2730 }
2731 /* Get the size for the shrunk mapping pairs array for the runlist. */
2732 mp_size = ntfs_get_size_for_mapping_pairs(vol, ni->runlist.rl, 0, -1);
2733 if (unlikely(mp_size <= 0)) {
2734 ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2735 "attribute type 0x%x, because determining the "
2736 "size for the mapping pairs failed with error "
2737 "code %i.%s", vi->i_ino,
2738 (unsigned)le32_to_cpu(ni->type), mp_size, es);
2739 err = -EIO;
2740 goto bad_out;
2741 }
2742 /*
2743 * Shrink the attribute record for the new mapping pairs array. Note,
2744 * this cannot fail since we are making the attribute smaller thus by
2745 * definition there is enough space to do so.
2746 */
2747 attr_len = le32_to_cpu(a->length);
2748 err = ntfs_attr_record_resize(m, a, mp_size +
2749 le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
2750 BUG_ON(err);
2751 /*
2752 * Generate the mapping pairs array directly into the attribute record.
2753 */
2754 err = ntfs_mapping_pairs_build(vol, (u8*)a +
2755 le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
2756 mp_size, ni->runlist.rl, 0, -1, NULL);
2757 if (unlikely(err)) {
2758 ntfs_error(vol->sb, "Cannot shrink allocation of inode 0x%lx, "
2759 "attribute type 0x%x, because building the "
2760 "mapping pairs failed with error code %i.%s",
2761 vi->i_ino, (unsigned)le32_to_cpu(ni->type),
2762 err, es);
2763 err = -EIO;
2764 goto bad_out;
2765 }
2766 /* Update the allocated/compressed size as well as the highest vcn. */
2767 a->data.non_resident.highest_vcn = cpu_to_sle64((new_alloc_size >>
2768 vol->cluster_size_bits) - 1);
2769 write_lock_irqsave(&ni->size_lock, flags);
2770 ni->allocated_size = new_alloc_size;
2771 a->data.non_resident.allocated_size = cpu_to_sle64(new_alloc_size);
2772 if (NInoSparse(ni) || NInoCompressed(ni)) {
2773 if (nr_freed) {
2774 ni->itype.compressed.size -= nr_freed <<
2775 vol->cluster_size_bits;
2776 BUG_ON(ni->itype.compressed.size < 0);
2777 a->data.non_resident.compressed_size = cpu_to_sle64(
2778 ni->itype.compressed.size);
2779 vi->i_blocks = ni->itype.compressed.size >> 9;
2780 }
2781 } else
2782 vi->i_blocks = new_alloc_size >> 9;
2783 write_unlock_irqrestore(&ni->size_lock, flags);
2784 /*
2785 * We have shrunk the allocation. If this is a shrinking truncate we
2786 * have already dealt with the initialized_size and the data_size above
2787 * and we are done. If the truncate is only changing the allocation
2788 * and not the data_size, we are also done. If this is an extending
2789 * truncate, need to extend the data_size now which is ensured by the
2790 * fact that @size_change is positive.
2791 */
2792alloc_done:
2793 /*
2794 * If the size is growing, need to update it now. If it is shrinking,
2795 * we have already updated it above (before the allocation change).
2796 */
2797 if (size_change > 0)
2798 a->data.non_resident.data_size = cpu_to_sle64(new_size);
2799 /* Ensure the modified mft record is written out. */
2800 flush_dcache_mft_record_page(ctx->ntfs_ino);
2801 mark_mft_record_dirty(ctx->ntfs_ino);
2802unm_done:
2803 ntfs_attr_put_search_ctx(ctx);
2804 unmap_mft_record(base_ni);
2805 up_write(&ni->runlist.lock);
2806done:
2807 /* Update the mtime and ctime on the base inode. */
2808 /* normally ->truncate shouldn't update ctime or mtime,
2809 * but ntfs did before so it got a copy & paste version
2810 * of file_update_time. one day someone should fix this
2811 * for real.
2812 */
2813 if (!IS_NOCMTIME(VFS_I(base_ni)) && !IS_RDONLY(VFS_I(base_ni))) {
2814 struct timespec now = current_fs_time(VFS_I(base_ni)->i_sb);
2815 int sync_it = 0;
2816
2817 if (!timespec_equal(&VFS_I(base_ni)->i_mtime, &now) ||
2818 !timespec_equal(&VFS_I(base_ni)->i_ctime, &now))
2819 sync_it = 1;
2820 VFS_I(base_ni)->i_mtime = now;
2821 VFS_I(base_ni)->i_ctime = now;
2822
2823 if (sync_it)
2824 mark_inode_dirty_sync(VFS_I(base_ni));
2825 }
2826
2827 if (likely(!err)) {
2828 NInoClearTruncateFailed(ni);
2829 ntfs_debug("Done.");
2830 }
2831 return err;
2832old_bad_out:
2833 old_size = -1;
2834bad_out:
2835 if (err != -ENOMEM && err != -EOPNOTSUPP)
2836 NVolSetErrors(vol);
2837 if (err != -EOPNOTSUPP)
2838 NInoSetTruncateFailed(ni);
2839 else if (old_size >= 0)
2840 i_size_write(vi, old_size);
2841err_out:
2842 if (ctx)
2843 ntfs_attr_put_search_ctx(ctx);
2844 if (m)
2845 unmap_mft_record(base_ni);
2846 up_write(&ni->runlist.lock);
2847out:
2848 ntfs_debug("Failed. Returning error code %i.", err);
2849 return err;
2850conv_err_out:
2851 if (err != -ENOMEM && err != -EOPNOTSUPP)
2852 NVolSetErrors(vol);
2853 if (err != -EOPNOTSUPP)
2854 NInoSetTruncateFailed(ni);
2855 else
2856 i_size_write(vi, old_size);
2857 goto out;
2858}
2859
2860/**
2861 * ntfs_truncate_vfs - wrapper for ntfs_truncate() that has no return value
2862 * @vi: inode for which the i_size was changed
2863 *
2864 * Wrapper for ntfs_truncate() that has no return value.
2865 *
2866 * See ntfs_truncate() description above for details.
2867 */
2868#ifdef NTFS_RW
2869void ntfs_truncate_vfs(struct inode *vi) {
2870 ntfs_truncate(vi);
2871}
2872#endif
2873
2874/**
2875 * ntfs_setattr - called from notify_change() when an attribute is being changed
2876 * @dentry: dentry whose attributes to change
2877 * @attr: structure describing the attributes and the changes
2878 *
2879 * We have to trap VFS attempts to truncate the file described by @dentry as
2880 * soon as possible, because we do not implement changes in i_size yet. So we
2881 * abort all i_size changes here.
2882 *
2883 * We also abort all changes of user, group, and mode as we do not implement
2884 * the NTFS ACLs yet.
2885 *
2886 * Called with ->i_mutex held.
2887 */
2888int ntfs_setattr(struct dentry *dentry, struct iattr *attr)
2889{
2890 struct inode *vi = dentry->d_inode;
2891 int err;
2892 unsigned int ia_valid = attr->ia_valid;
2893
2894 err = inode_change_ok(vi, attr);
2895 if (err)
2896 goto out;
2897 /* We do not support NTFS ACLs yet. */
2898 if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
2899 ntfs_warning(vi->i_sb, "Changes in user/group/mode are not "
2900 "supported yet, ignoring.");
2901 err = -EOPNOTSUPP;
2902 goto out;
2903 }
2904 if (ia_valid & ATTR_SIZE) {
2905 if (attr->ia_size != i_size_read(vi)) {
2906 ntfs_inode *ni = NTFS_I(vi);
2907 /*
2908 * FIXME: For now we do not support resizing of
2909 * compressed or encrypted files yet.
2910 */
2911 if (NInoCompressed(ni) || NInoEncrypted(ni)) {
2912 ntfs_warning(vi->i_sb, "Changes in inode size "
2913 "are not supported yet for "
2914 "%s files, ignoring.",
2915 NInoCompressed(ni) ?
2916 "compressed" : "encrypted");
2917 err = -EOPNOTSUPP;
2918 } else {
2919 truncate_setsize(vi, attr->ia_size);
2920 ntfs_truncate_vfs(vi);
2921 }
2922 if (err || ia_valid == ATTR_SIZE)
2923 goto out;
2924 } else {
2925 /*
2926 * We skipped the truncate but must still update
2927 * timestamps.
2928 */
2929 ia_valid |= ATTR_MTIME | ATTR_CTIME;
2930 }
2931 }
2932 if (ia_valid & ATTR_ATIME)
2933 vi->i_atime = timespec_trunc(attr->ia_atime,
2934 vi->i_sb->s_time_gran);
2935 if (ia_valid & ATTR_MTIME)
2936 vi->i_mtime = timespec_trunc(attr->ia_mtime,
2937 vi->i_sb->s_time_gran);
2938 if (ia_valid & ATTR_CTIME)
2939 vi->i_ctime = timespec_trunc(attr->ia_ctime,
2940 vi->i_sb->s_time_gran);
2941 mark_inode_dirty(vi);
2942out:
2943 return err;
2944}
2945
2946/**
2947 * ntfs_write_inode - write out a dirty inode
2948 * @vi: inode to write out
2949 * @sync: if true, write out synchronously
2950 *
2951 * Write out a dirty inode to disk including any extent inodes if present.
2952 *
2953 * If @sync is true, commit the inode to disk and wait for io completion. This
2954 * is done using write_mft_record().
2955 *
2956 * If @sync is false, just schedule the write to happen but do not wait for i/o
2957 * completion. In 2.6 kernels, scheduling usually happens just by virtue of
2958 * marking the page (and in this case mft record) dirty but we do not implement
2959 * this yet as write_mft_record() largely ignores the @sync parameter and
2960 * always performs synchronous writes.
2961 *
2962 * Return 0 on success and -errno on error.
2963 */
2964int __ntfs_write_inode(struct inode *vi, int sync)
2965{
2966 sle64 nt;
2967 ntfs_inode *ni = NTFS_I(vi);
2968 ntfs_attr_search_ctx *ctx;
2969 MFT_RECORD *m;
2970 STANDARD_INFORMATION *si;
2971 int err = 0;
2972 bool modified = false;
2973
2974 ntfs_debug("Entering for %sinode 0x%lx.", NInoAttr(ni) ? "attr " : "",
2975 vi->i_ino);
2976 /*
2977 * Dirty attribute inodes are written via their real inodes so just
2978 * clean them here. Access time updates are taken care off when the
2979 * real inode is written.
2980 */
2981 if (NInoAttr(ni)) {
2982 NInoClearDirty(ni);
2983 ntfs_debug("Done.");
2984 return 0;
2985 }
2986 /* Map, pin, and lock the mft record belonging to the inode. */
2987 m = map_mft_record(ni);
2988 if (IS_ERR(m)) {
2989 err = PTR_ERR(m);
2990 goto err_out;
2991 }
2992 /* Update the access times in the standard information attribute. */
2993 ctx = ntfs_attr_get_search_ctx(ni, m);
2994 if (unlikely(!ctx)) {
2995 err = -ENOMEM;
2996 goto unm_err_out;
2997 }
2998 err = ntfs_attr_lookup(AT_STANDARD_INFORMATION, NULL, 0,
2999 CASE_SENSITIVE, 0, NULL, 0, ctx);
3000 if (unlikely(err)) {
3001 ntfs_attr_put_search_ctx(ctx);
3002 goto unm_err_out;
3003 }
3004 si = (STANDARD_INFORMATION*)((u8*)ctx->attr +
3005 le16_to_cpu(ctx->attr->data.resident.value_offset));
3006 /* Update the access times if they have changed. */
3007 nt = utc2ntfs(vi->i_mtime);
3008 if (si->last_data_change_time != nt) {
3009 ntfs_debug("Updating mtime for inode 0x%lx: old = 0x%llx, "
3010 "new = 0x%llx", vi->i_ino, (long long)
3011 sle64_to_cpu(si->last_data_change_time),
3012 (long long)sle64_to_cpu(nt));
3013 si->last_data_change_time = nt;
3014 modified = true;
3015 }
3016 nt = utc2ntfs(vi->i_ctime);
3017 if (si->last_mft_change_time != nt) {
3018 ntfs_debug("Updating ctime for inode 0x%lx: old = 0x%llx, "
3019 "new = 0x%llx", vi->i_ino, (long long)
3020 sle64_to_cpu(si->last_mft_change_time),
3021 (long long)sle64_to_cpu(nt));
3022 si->last_mft_change_time = nt;
3023 modified = true;
3024 }
3025 nt = utc2ntfs(vi->i_atime);
3026 if (si->last_access_time != nt) {
3027 ntfs_debug("Updating atime for inode 0x%lx: old = 0x%llx, "
3028 "new = 0x%llx", vi->i_ino,
3029 (long long)sle64_to_cpu(si->last_access_time),
3030 (long long)sle64_to_cpu(nt));
3031 si->last_access_time = nt;
3032 modified = true;
3033 }
3034 /*
3035 * If we just modified the standard information attribute we need to
3036 * mark the mft record it is in dirty. We do this manually so that
3037 * mark_inode_dirty() is not called which would redirty the inode and
3038 * hence result in an infinite loop of trying to write the inode.
3039 * There is no need to mark the base inode nor the base mft record
3040 * dirty, since we are going to write this mft record below in any case
3041 * and the base mft record may actually not have been modified so it
3042 * might not need to be written out.
3043 * NOTE: It is not a problem when the inode for $MFT itself is being
3044 * written out as mark_ntfs_record_dirty() will only set I_DIRTY_PAGES
3045 * on the $MFT inode and hence ntfs_write_inode() will not be
3046 * re-invoked because of it which in turn is ok since the dirtied mft
3047 * record will be cleaned and written out to disk below, i.e. before
3048 * this function returns.
3049 */
3050 if (modified) {
3051 flush_dcache_mft_record_page(ctx->ntfs_ino);
3052 if (!NInoTestSetDirty(ctx->ntfs_ino))
3053 mark_ntfs_record_dirty(ctx->ntfs_ino->page,
3054 ctx->ntfs_ino->page_ofs);
3055 }
3056 ntfs_attr_put_search_ctx(ctx);
3057 /* Now the access times are updated, write the base mft record. */
3058 if (NInoDirty(ni))
3059 err = write_mft_record(ni, m, sync);
3060 /* Write all attached extent mft records. */
3061 mutex_lock(&ni->extent_lock);
3062 if (ni->nr_extents > 0) {
3063 ntfs_inode **extent_nis = ni->ext.extent_ntfs_inos;
3064 int i;
3065
3066 ntfs_debug("Writing %i extent inodes.", ni->nr_extents);
3067 for (i = 0; i < ni->nr_extents; i++) {
3068 ntfs_inode *tni = extent_nis[i];
3069
3070 if (NInoDirty(tni)) {
3071 MFT_RECORD *tm = map_mft_record(tni);
3072 int ret;
3073
3074 if (IS_ERR(tm)) {
3075 if (!err || err == -ENOMEM)
3076 err = PTR_ERR(tm);
3077 continue;
3078 }
3079 ret = write_mft_record(tni, tm, sync);
3080 unmap_mft_record(tni);
3081 if (unlikely(ret)) {
3082 if (!err || err == -ENOMEM)
3083 err = ret;
3084 }
3085 }
3086 }
3087 }
3088 mutex_unlock(&ni->extent_lock);
3089 unmap_mft_record(ni);
3090 if (unlikely(err))
3091 goto err_out;
3092 ntfs_debug("Done.");
3093 return 0;
3094unm_err_out:
3095 unmap_mft_record(ni);
3096err_out:
3097 if (err == -ENOMEM) {
3098 ntfs_warning(vi->i_sb, "Not enough memory to write inode. "
3099 "Marking the inode dirty again, so the VFS "
3100 "retries later.");
3101 mark_inode_dirty(vi);
3102 } else {
3103 ntfs_error(vi->i_sb, "Failed (error %i): Run chkdsk.", -err);
3104 NVolSetErrors(ni->vol);
3105 }
3106 return err;
3107}
3108
3109#endif /* NTFS_RW */