1	// SPDX-License-Identifier: GPL-2.0-or-later
2	/*
3	* attrib.c - NTFS attribute operations. Part of the Linux-NTFS project.
4	*
5	* Copyright (c) 2001-2012 Anton Altaparmakov and Tuxera Inc.
6	* Copyright (c) 2002 Richard Russon
7	*/
8
9	#include <linux/buffer_head.h>
10	#include <linux/sched.h>
11	#include <linux/slab.h>
12	#include <linux/swap.h>
13	#include <linux/writeback.h>
14
15	#include "attrib.h"
16	#include "debug.h"
17	#include "layout.h"
18	#include "lcnalloc.h"
19	#include "malloc.h"
20	#include "mft.h"
21	#include "ntfs.h"
22	#include "types.h"
23
24	/**
25	* ntfs_map_runlist_nolock - map (a part of) a runlist of an ntfs inode
26	* @ni: ntfs inode for which to map (part of) a runlist
27	* @vcn: map runlist part containing this vcn
28	* @ctx: active attribute search context if present or NULL if not
29	*
30	* Map the part of a runlist containing the @vcn of the ntfs inode @ni.
31	*
32	* If @ctx is specified, it is an active search context of @ni and its base mft
33	* record. This is needed when ntfs_map_runlist_nolock() encounters unmapped
34	* runlist fragments and allows their mapping. If you do not have the mft
35	* record mapped, you can specify @ctx as NULL and ntfs_map_runlist_nolock()
36	* will perform the necessary mapping and unmapping.
37	*
38	* Note, ntfs_map_runlist_nolock() saves the state of @ctx on entry and
39	* restores it before returning. Thus, @ctx will be left pointing to the same
40	* attribute on return as on entry. However, the actual pointers in @ctx may
41	* point to different memory locations on return, so you must remember to reset
42	* any cached pointers from the @ctx, i.e. after the call to
43	* ntfs_map_runlist_nolock(), you will probably want to do:
44	* m = ctx->mrec;
45	* a = ctx->attr;
46	* Assuming you cache ctx->attr in a variable @a of type ATTR_RECORD * and that
47	* you cache ctx->mrec in a variable @m of type MFT_RECORD *.
48	*
49	* Return 0 on success and -errno on error. There is one special error code
50	* which is not an error as such. This is -ENOENT. It means that @vcn is out
51	* of bounds of the runlist.
52	*
53	* Note the runlist can be NULL after this function returns if @vcn is zero and
54	* the attribute has zero allocated size, i.e. there simply is no runlist.
55	*
56	* WARNING: If @ctx is supplied, regardless of whether success or failure is
57	* returned, you need to check IS_ERR(@ctx->mrec) and if 'true' the @ctx
58	* is no longer valid, i.e. you need to either call
59	* ntfs_attr_reinit_search_ctx() or ntfs_attr_put_search_ctx() on it.
60	* In that case PTR_ERR(@ctx->mrec) will give you the error code for
61	* why the mapping of the old inode failed.
62	*
63	* Locking: - The runlist described by @ni must be locked for writing on entry
64	* and is locked on return. Note the runlist will be modified.
65	* - If @ctx is NULL, the base mft record of @ni must not be mapped on
66	* entry and it will be left unmapped on return.
67	* - If @ctx is not NULL, the base mft record must be mapped on entry
68	* and it will be left mapped on return.
69	*/
70	int ntfs_map_runlist_nolock(ntfs_inode *ni, VCN vcn, ntfs_attr_search_ctx *ctx)
71	{
72	VCN end_vcn;
73	unsigned long flags;
74	ntfs_inode *base_ni;
75	MFT_RECORD *m;
76	ATTR_RECORD *a;
77	runlist_element *rl;
78	struct page *put_this_page = NULL;
79	int err = 0;
80	bool ctx_is_temporary, ctx_needs_reset;
81	ntfs_attr_search_ctx old_ctx = { NULL, };
82
83	ntfs_debug("Mapping runlist part containing vcn 0x%llx.",
84	(unsigned long long)vcn);
85	if (!NInoAttr(ni))
86	base_ni = ni;
87	else
88	base_ni = ni->ext.base_ntfs_ino;
89	if (!ctx) {
90	ctx_is_temporary = ctx_needs_reset = true;
91	m = map_mft_record(ni: base_ni);
92	if (IS_ERR(ptr: m))
93	return PTR_ERR(ptr: m);
94	ctx = ntfs_attr_get_search_ctx(ni: base_ni, mrec: m);
95	if (unlikely(!ctx)) {
96	err = -ENOMEM;
97	goto err_out;
98	}
99	} else {
100	VCN allocated_size_vcn;
101
102	BUG_ON(IS_ERR(ctx->mrec));
103	a = ctx->attr;
104	BUG_ON(!a->non_resident);
105	ctx_is_temporary = false;
106	end_vcn = sle64_to_cpu(x: a->data.non_resident.highest_vcn);
107	read_lock_irqsave(&ni->size_lock, flags);
108	allocated_size_vcn = ni->allocated_size >>
109	ni->vol->cluster_size_bits;
110	read_unlock_irqrestore(&ni->size_lock, flags);
111	if (!a->data.non_resident.lowest_vcn && end_vcn <= 0)
112	end_vcn = allocated_size_vcn - 1;
113	/*
114	* If we already have the attribute extent containing @vcn in
115	* @ctx, no need to look it up again. We slightly cheat in
116	* that if vcn exceeds the allocated size, we will refuse to
117	* map the runlist below, so there is definitely no need to get
118	* the right attribute extent.
119	*/
120	if (vcn >= allocated_size_vcn || (a->type == ni->type &&
121	a->name_length == ni->name_len &&
122	!memcmp(p: (u8*)a + le16_to_cpu(a->name_offset),
123	q: ni->name, size: ni->name_len) &&
124	sle64_to_cpu(x: a->data.non_resident.lowest_vcn)
125	<= vcn && end_vcn >= vcn))
126	ctx_needs_reset = false;
127	else {
128	/* Save the old search context. */
129	old_ctx = *ctx;
130	/*
131	* If the currently mapped (extent) inode is not the
132	* base inode we will unmap it when we reinitialize the
133	* search context which means we need to get a
134	* reference to the page containing the mapped mft
135	* record so we do not accidentally drop changes to the
136	* mft record when it has not been marked dirty yet.
137	*/
138	if (old_ctx.base_ntfs_ino && old_ctx.ntfs_ino !=
139	old_ctx.base_ntfs_ino) {
140	put_this_page = old_ctx.ntfs_ino->page;
141	get_page(page: put_this_page);
142	}
143	/*
144	* Reinitialize the search context so we can lookup the
145	* needed attribute extent.
146	*/
147	ntfs_attr_reinit_search_ctx(ctx);
148	ctx_needs_reset = true;
149	}
150	}
151	if (ctx_needs_reset) {
152	err = ntfs_attr_lookup(type: ni->type, name: ni->name, name_len: ni->name_len,
153	ic: CASE_SENSITIVE, lowest_vcn: vcn, NULL, val_len: 0, ctx);
154	if (unlikely(err)) {
155	if (err == -ENOENT)
156	err = -EIO;
157	goto err_out;
158	}
159	BUG_ON(!ctx->attr->non_resident);
160	}
161	a = ctx->attr;
162	/*
163	* Only decompress the mapping pairs if @vcn is inside it. Otherwise
164	* we get into problems when we try to map an out of bounds vcn because
165	* we then try to map the already mapped runlist fragment and
166	* ntfs_mapping_pairs_decompress() fails.
167	*/
168	end_vcn = sle64_to_cpu(x: a->data.non_resident.highest_vcn) + 1;
169	if (unlikely(vcn && vcn >= end_vcn)) {
170	err = -ENOENT;
171	goto err_out;
172	}
173	rl = ntfs_mapping_pairs_decompress(vol: ni->vol, attr: a, old_rl: ni->runlist.rl);
174	if (IS_ERR(ptr: rl))
175	err = PTR_ERR(ptr: rl);
176	else
177	ni->runlist.rl = rl;
178	err_out:
179	if (ctx_is_temporary) {
180	if (likely(ctx))
181	ntfs_attr_put_search_ctx(ctx);
182	unmap_mft_record(ni: base_ni);
183	} else if (ctx_needs_reset) {
184	/*
185	* If there is no attribute list, restoring the search context
186	* is accomplished simply by copying the saved context back over
187	* the caller supplied context. If there is an attribute list,
188	* things are more complicated as we need to deal with mapping
189	* of mft records and resulting potential changes in pointers.
190	*/
191	if (NInoAttrList(ni: base_ni)) {
192	/*
193	* If the currently mapped (extent) inode is not the
194	* one we had before, we need to unmap it and map the
195	* old one.
196	*/
197	if (ctx->ntfs_ino != old_ctx.ntfs_ino) {
198	/*
199	* If the currently mapped inode is not the
200	* base inode, unmap it.
201	*/
202	if (ctx->base_ntfs_ino && ctx->ntfs_ino !=
203	ctx->base_ntfs_ino) {
204	unmap_extent_mft_record(ni: ctx->ntfs_ino);
205	ctx->mrec = ctx->base_mrec;
206	BUG_ON(!ctx->mrec);
207	}
208	/*
209	* If the old mapped inode is not the base
210	* inode, map it.
211	*/
212	if (old_ctx.base_ntfs_ino &&
213	old_ctx.ntfs_ino !=
214	old_ctx.base_ntfs_ino) {
215	retry_map:
216	ctx->mrec = map_mft_record(
217	ni: old_ctx.ntfs_ino);
218	/*
219	* Something bad has happened. If out
220	* of memory retry till it succeeds.
221	* Any other errors are fatal and we
222	* return the error code in ctx->mrec.
223	* Let the caller deal with it... We
224	* just need to fudge things so the
225	* caller can reinit and/or put the
226	* search context safely.
227	*/
228	if (IS_ERR(ptr: ctx->mrec)) {
229	if (PTR_ERR(ptr: ctx->mrec) ==
230	-ENOMEM) {
231	schedule();
232	goto retry_map;
233	} else
234	old_ctx.ntfs_ino =
235	old_ctx.
236	base_ntfs_ino;
237	}
238	}
239	}
240	/* Update the changed pointers in the saved context. */
241	if (ctx->mrec != old_ctx.mrec) {
242	if (!IS_ERR(ptr: ctx->mrec))
243	old_ctx.attr = (ATTR_RECORD*)(
244	(u8*)ctx->mrec +
245	((u8*)old_ctx.attr -
246	(u8*)old_ctx.mrec));
247	old_ctx.mrec = ctx->mrec;
248	}
249	}
250	/* Restore the search context to the saved one. */
251	*ctx = old_ctx;
252	/*
253	* We drop the reference on the page we took earlier. In the
254	* case that IS_ERR(ctx->mrec) is true this means we might lose
255	* some changes to the mft record that had been made between
256	* the last time it was marked dirty/written out and now. This
257	* at this stage is not a problem as the mapping error is fatal
258	* enough that the mft record cannot be written out anyway and
259	* the caller is very likely to shutdown the whole inode
260	* immediately and mark the volume dirty for chkdsk to pick up
261	* the pieces anyway.
262	*/
263	if (put_this_page)
264	put_page(page: put_this_page);
265	}
266	return err;
267	}
268
269	/**
270	* ntfs_map_runlist - map (a part of) a runlist of an ntfs inode
271	* @ni: ntfs inode for which to map (part of) a runlist
272	* @vcn: map runlist part containing this vcn
273	*
274	* Map the part of a runlist containing the @vcn of the ntfs inode @ni.
275	*
276	* Return 0 on success and -errno on error. There is one special error code
277	* which is not an error as such. This is -ENOENT. It means that @vcn is out
278	* of bounds of the runlist.
279	*
280	* Locking: - The runlist must be unlocked on entry and is unlocked on return.
281	* - This function takes the runlist lock for writing and may modify
282	* the runlist.
283	*/
284	int ntfs_map_runlist(ntfs_inode *ni, VCN vcn)
285	{
286	int err = 0;
287
288	down_write(sem: &ni->runlist.lock);
289	/* Make sure someone else didn't do the work while we were sleeping. */
290	if (likely(ntfs_rl_vcn_to_lcn(ni->runlist.rl, vcn) <=
291	LCN_RL_NOT_MAPPED))
292	err = ntfs_map_runlist_nolock(ni, vcn, NULL);
293	up_write(sem: &ni->runlist.lock);
294	return err;
295	}
296
297	/**
298	* ntfs_attr_vcn_to_lcn_nolock - convert a vcn into a lcn given an ntfs inode
299	* @ni: ntfs inode of the attribute whose runlist to search
300	* @vcn: vcn to convert
301	* @write_locked: true if the runlist is locked for writing
302	*
303	* Find the virtual cluster number @vcn in the runlist of the ntfs attribute
304	* described by the ntfs inode @ni and return the corresponding logical cluster
305	* number (lcn).
306	*
307	* If the @vcn is not mapped yet, the attempt is made to map the attribute
308	* extent containing the @vcn and the vcn to lcn conversion is retried.
309	*
310	* If @write_locked is true the caller has locked the runlist for writing and
311	* if false for reading.
312	*
313	* Since lcns must be >= 0, we use negative return codes with special meaning:
314	*
315	* Return code Meaning / Description
316	* ==========================================
317	* LCN_HOLE Hole / not allocated on disk.
318	* LCN_ENOENT There is no such vcn in the runlist, i.e. @vcn is out of bounds.
319	* LCN_ENOMEM Not enough memory to map runlist.
320	* LCN_EIO Critical error (runlist/file is corrupt, i/o error, etc).
321	*
322	* Locking: - The runlist must be locked on entry and is left locked on return.
323	* - If @write_locked is 'false', i.e. the runlist is locked for reading,
324	* the lock may be dropped inside the function so you cannot rely on
325	* the runlist still being the same when this function returns.
326	*/
327	LCN ntfs_attr_vcn_to_lcn_nolock(ntfs_inode *ni, const VCN vcn,
328	const bool write_locked)
329	{
330	LCN lcn;
331	unsigned long flags;
332	bool is_retry = false;
333
334	BUG_ON(!ni);
335	ntfs_debug("Entering for i_ino 0x%lx, vcn 0x%llx, %s_locked.",
336	ni->mft_no, (unsigned long long)vcn,
337	write_locked ? "write" : "read");
338	BUG_ON(!NInoNonResident(ni));
339	BUG_ON(vcn < 0);
340	if (!ni->runlist.rl) {
341	read_lock_irqsave(&ni->size_lock, flags);
342	if (!ni->allocated_size) {
343	read_unlock_irqrestore(&ni->size_lock, flags);
344	return LCN_ENOENT;
345	}
346	read_unlock_irqrestore(&ni->size_lock, flags);
347	}
348	retry_remap:
349	/* Convert vcn to lcn. If that fails map the runlist and retry once. */
350	lcn = ntfs_rl_vcn_to_lcn(rl: ni->runlist.rl, vcn);
351	if (likely(lcn >= LCN_HOLE)) {
352	ntfs_debug("Done, lcn 0x%llx.", (long long)lcn);
353	return lcn;
354	}
355	if (lcn != LCN_RL_NOT_MAPPED) {
356	if (lcn != LCN_ENOENT)
357	lcn = LCN_EIO;
358	} else if (!is_retry) {
359	int err;
360
361	if (!write_locked) {
362	up_read(sem: &ni->runlist.lock);
363	down_write(sem: &ni->runlist.lock);
364	if (unlikely(ntfs_rl_vcn_to_lcn(ni->runlist.rl, vcn) !=
365	LCN_RL_NOT_MAPPED)) {
366	up_write(sem: &ni->runlist.lock);
367	down_read(sem: &ni->runlist.lock);
368	goto retry_remap;
369	}
370	}
371	err = ntfs_map_runlist_nolock(ni, vcn, NULL);
372	if (!write_locked) {
373	up_write(sem: &ni->runlist.lock);
374	down_read(sem: &ni->runlist.lock);
375	}
376	if (likely(!err)) {
377	is_retry = true;
378	goto retry_remap;
379	}
380	if (err == -ENOENT)
381	lcn = LCN_ENOENT;
382	else if (err == -ENOMEM)
383	lcn = LCN_ENOMEM;
384	else
385	lcn = LCN_EIO;
386	}
387	if (lcn != LCN_ENOENT)
388	ntfs_error(ni->vol->sb, "Failed with error code %lli.",
389	(long long)lcn);
390	return lcn;
391	}
392
393	/**
394	* ntfs_attr_find_vcn_nolock - find a vcn in the runlist of an ntfs inode
395	* @ni: ntfs inode describing the runlist to search
396	* @vcn: vcn to find
397	* @ctx: active attribute search context if present or NULL if not
398	*
399	* Find the virtual cluster number @vcn in the runlist described by the ntfs
400	* inode @ni and return the address of the runlist element containing the @vcn.
401	*
402	* If the @vcn is not mapped yet, the attempt is made to map the attribute
403	* extent containing the @vcn and the vcn to lcn conversion is retried.
404	*
405	* If @ctx is specified, it is an active search context of @ni and its base mft
406	* record. This is needed when ntfs_attr_find_vcn_nolock() encounters unmapped
407	* runlist fragments and allows their mapping. If you do not have the mft
408	* record mapped, you can specify @ctx as NULL and ntfs_attr_find_vcn_nolock()
409	* will perform the necessary mapping and unmapping.
410	*
411	* Note, ntfs_attr_find_vcn_nolock() saves the state of @ctx on entry and
412	* restores it before returning. Thus, @ctx will be left pointing to the same
413	* attribute on return as on entry. However, the actual pointers in @ctx may
414	* point to different memory locations on return, so you must remember to reset
415	* any cached pointers from the @ctx, i.e. after the call to
416	* ntfs_attr_find_vcn_nolock(), you will probably want to do:
417	* m = ctx->mrec;
418	* a = ctx->attr;
419	* Assuming you cache ctx->attr in a variable @a of type ATTR_RECORD * and that
420	* you cache ctx->mrec in a variable @m of type MFT_RECORD *.
421	* Note you need to distinguish between the lcn of the returned runlist element
422	* being >= 0 and LCN_HOLE. In the later case you have to return zeroes on
423	* read and allocate clusters on write.
424	*
425	* Return the runlist element containing the @vcn on success and
426	* ERR_PTR(-errno) on error. You need to test the return value with IS_ERR()
427	* to decide if the return is success or failure and PTR_ERR() to get to the
428	* error code if IS_ERR() is true.
429	*
430	* The possible error return codes are:
431	* -ENOENT - No such vcn in the runlist, i.e. @vcn is out of bounds.
432	* -ENOMEM - Not enough memory to map runlist.
433	* -EIO - Critical error (runlist/file is corrupt, i/o error, etc).
434	*
435	* WARNING: If @ctx is supplied, regardless of whether success or failure is
436	* returned, you need to check IS_ERR(@ctx->mrec) and if 'true' the @ctx
437	* is no longer valid, i.e. you need to either call
438	* ntfs_attr_reinit_search_ctx() or ntfs_attr_put_search_ctx() on it.
439	* In that case PTR_ERR(@ctx->mrec) will give you the error code for
440	* why the mapping of the old inode failed.
441	*
442	* Locking: - The runlist described by @ni must be locked for writing on entry
443	* and is locked on return. Note the runlist may be modified when
444	* needed runlist fragments need to be mapped.
445	* - If @ctx is NULL, the base mft record of @ni must not be mapped on
446	* entry and it will be left unmapped on return.
447	* - If @ctx is not NULL, the base mft record must be mapped on entry
448	* and it will be left mapped on return.
449	*/
450	runlist_element *ntfs_attr_find_vcn_nolock(ntfs_inode *ni, const VCN vcn,
451	ntfs_attr_search_ctx *ctx)
452	{
453	unsigned long flags;
454	runlist_element *rl;
455	int err = 0;
456	bool is_retry = false;
457
458	BUG_ON(!ni);
459	ntfs_debug("Entering for i_ino 0x%lx, vcn 0x%llx, with%s ctx.",
460	ni->mft_no, (unsigned long long)vcn, ctx ? "" : "out");
461	BUG_ON(!NInoNonResident(ni));
462	BUG_ON(vcn < 0);
463	if (!ni->runlist.rl) {
464	read_lock_irqsave(&ni->size_lock, flags);
465	if (!ni->allocated_size) {
466	read_unlock_irqrestore(&ni->size_lock, flags);
467	return ERR_PTR(error: -ENOENT);
468	}
469	read_unlock_irqrestore(&ni->size_lock, flags);
470	}
471	retry_remap:
472	rl = ni->runlist.rl;
473	if (likely(rl && vcn >= rl[0].vcn)) {
474	while (likely(rl->length)) {
475	if (unlikely(vcn < rl[1].vcn)) {
476	if (likely(rl->lcn >= LCN_HOLE)) {
477	ntfs_debug("Done.");
478	return rl;
479	}
480	break;
481	}
482	rl++;
483	}
484	if (likely(rl->lcn != LCN_RL_NOT_MAPPED)) {
485	if (likely(rl->lcn == LCN_ENOENT))
486	err = -ENOENT;
487	else
488	err = -EIO;
489	}
490	}
491	if (!err && !is_retry) {
492	/*
493	* If the search context is invalid we cannot map the unmapped
494	* region.
495	*/
496	if (IS_ERR(ptr: ctx->mrec))
497	err = PTR_ERR(ptr: ctx->mrec);
498	else {
499	/*
500	* The @vcn is in an unmapped region, map the runlist
501	* and retry.
502	*/
503	err = ntfs_map_runlist_nolock(ni, vcn, ctx);
504	if (likely(!err)) {
505	is_retry = true;
506	goto retry_remap;
507	}
508	}
509	if (err == -EINVAL)
510	err = -EIO;
511	} else if (!err)
512	err = -EIO;
513	if (err != -ENOENT)
514	ntfs_error(ni->vol->sb, "Failed with error code %i.", err);
515	return ERR_PTR(error: err);
516	}
517
518	/**
519	* ntfs_attr_find - find (next) attribute in mft record
520	* @type: attribute type to find
521	* @name: attribute name to find (optional, i.e. NULL means don't care)
522	* @name_len: attribute name length (only needed if @name present)
523	* @ic: IGNORE_CASE or CASE_SENSITIVE (ignored if @name not present)
524	* @val: attribute value to find (optional, resident attributes only)
525	* @val_len: attribute value length
526	* @ctx: search context with mft record and attribute to search from
527	*
528	* You should not need to call this function directly. Use ntfs_attr_lookup()
529	* instead.
530	*
531	* ntfs_attr_find() takes a search context @ctx as parameter and searches the
532	* mft record specified by @ctx->mrec, beginning at @ctx->attr, for an
533	* attribute of @type, optionally @name and @val.
534	*
535	* If the attribute is found, ntfs_attr_find() returns 0 and @ctx->attr will
536	* point to the found attribute.
537	*
538	* If the attribute is not found, ntfs_attr_find() returns -ENOENT and
539	* @ctx->attr will point to the attribute before which the attribute being
540	* searched for would need to be inserted if such an action were to be desired.
541	*
542	* On actual error, ntfs_attr_find() returns -EIO. In this case @ctx->attr is
543	* undefined and in particular do not rely on it not changing.
544	*
545	* If @ctx->is_first is 'true', the search begins with @ctx->attr itself. If it
546	* is 'false', the search begins after @ctx->attr.
547	*
548	* If @ic is IGNORE_CASE, the @name comparisson is not case sensitive and
549	* @ctx->ntfs_ino must be set to the ntfs inode to which the mft record
550	* @ctx->mrec belongs. This is so we can get at the ntfs volume and hence at
551	* the upcase table. If @ic is CASE_SENSITIVE, the comparison is case
552	* sensitive. When @name is present, @name_len is the @name length in Unicode
553	* characters.
554	*
555	* If @name is not present (NULL), we assume that the unnamed attribute is
556	* being searched for.
557	*
558	* Finally, the resident attribute value @val is looked for, if present. If
559	* @val is not present (NULL), @val_len is ignored.
560	*
561	* ntfs_attr_find() only searches the specified mft record and it ignores the
562	* presence of an attribute list attribute (unless it is the one being searched
563	* for, obviously). If you need to take attribute lists into consideration,
564	* use ntfs_attr_lookup() instead (see below). This also means that you cannot
565	* use ntfs_attr_find() to search for extent records of non-resident
566	* attributes, as extents with lowest_vcn != 0 are usually described by the
567	* attribute list attribute only. - Note that it is possible that the first
568	* extent is only in the attribute list while the last extent is in the base
569	* mft record, so do not rely on being able to find the first extent in the
570	* base mft record.
571	*
572	* Warning: Never use @val when looking for attribute types which can be
573	* non-resident as this most likely will result in a crash!
574	*/
575	static int ntfs_attr_find(const ATTR_TYPE type, const ntfschar *name,
576	const u32 name_len, const IGNORE_CASE_BOOL ic,
577	const u8 *val, const u32 val_len, ntfs_attr_search_ctx *ctx)
578	{
579	ATTR_RECORD *a;
580	ntfs_volume *vol = ctx->ntfs_ino->vol;
581	ntfschar *upcase = vol->upcase;
582	u32 upcase_len = vol->upcase_len;
583
584	/*
585	* Iterate over attributes in mft record starting at @ctx->attr, or the
586	* attribute following that, if @ctx->is_first is 'true'.
587	*/
588	if (ctx->is_first) {
589	a = ctx->attr;
590	ctx->is_first = false;
591	} else
592	a = (ATTR_RECORD*)((u8*)ctx->attr +
593	le32_to_cpu(ctx->attr->length));
594	for (;; a = (ATTR_RECORD*)((u8*)a + le32_to_cpu(a->length))) {
595	u8 *mrec_end = (u8 *)ctx->mrec +
596	le32_to_cpu(ctx->mrec->bytes_allocated);
597	u8 *name_end;
598
599	/* check whether ATTR_RECORD wrap */
600	if ((u8 *)a < (u8 *)ctx->mrec)
601	break;
602
603	/* check whether Attribute Record Header is within bounds */
604	if ((u8 *)a > mrec_end ||
605	(u8 *)a + sizeof(ATTR_RECORD) > mrec_end)
606	break;
607
608	/* check whether ATTR_RECORD's name is within bounds */
609	name_end = (u8 *)a + le16_to_cpu(a->name_offset) +
610	a->name_length * sizeof(ntfschar);
611	if (name_end > mrec_end)
612	break;
613
614	ctx->attr = a;
615	if (unlikely(le32_to_cpu(a->type) > le32_to_cpu(type) ||
616	a->type == AT_END))
617	return -ENOENT;
618	if (unlikely(!a->length))
619	break;
620
621	/* check whether ATTR_RECORD's length wrap */
622	if ((u8 *)a + le32_to_cpu(a->length) < (u8 *)a)
623	break;
624	/* check whether ATTR_RECORD's length is within bounds */
625	if ((u8 *)a + le32_to_cpu(a->length) > mrec_end)
626	break;
627
628	if (a->type != type)
629	continue;
630	/*
631	* If @name is present, compare the two names. If @name is
632	* missing, assume we want an unnamed attribute.
633	*/
634	if (!name) {
635	/* The search failed if the found attribute is named. */
636	if (a->name_length)
637	return -ENOENT;
638	} else if (!ntfs_are_names_equal(s1: name, s1_len: name_len,
639	s2: (ntfschar*)((u8*)a + le16_to_cpu(a->name_offset)),
640	s2_len: a->name_length, ic, upcase, upcase_size: upcase_len)) {
641	register int rc;
642
643	rc = ntfs_collate_names(name1: name, name1_len: name_len,
644	name2: (ntfschar*)((u8*)a +
645	le16_to_cpu(a->name_offset)),
646	name2_len: a->name_length, err_val: 1, ic: IGNORE_CASE,
647	upcase, upcase_len);
648	/*
649	* If @name collates before a->name, there is no
650	* matching attribute.
651	*/
652	if (rc == -1)
653	return -ENOENT;
654	/* If the strings are not equal, continue search. */
655	if (rc)
656	continue;
657	rc = ntfs_collate_names(name1: name, name1_len: name_len,
658	name2: (ntfschar*)((u8*)a +
659	le16_to_cpu(a->name_offset)),
660	name2_len: a->name_length, err_val: 1, ic: CASE_SENSITIVE,
661	upcase, upcase_len);
662	if (rc == -1)
663	return -ENOENT;
664	if (rc)
665	continue;
666	}
667	/*
668	* The names match or @name not present and attribute is
669	* unnamed. If no @val specified, we have found the attribute
670	* and are done.
671	*/
672	if (!val)
673	return 0;
674	/* @val is present; compare values. */
675	else {
676	register int rc;
677
678	rc = memcmp(p: val, q: (u8*)a + le16_to_cpu(
679	a->data.resident.value_offset),
680	min_t(u32, val_len, le32_to_cpu(
681	a->data.resident.value_length)));
682	/*
683	* If @val collates before the current attribute's
684	* value, there is no matching attribute.
685	*/
686	if (!rc) {
687	register u32 avl;
688
689	avl = le32_to_cpu(
690	a->data.resident.value_length);
691	if (val_len == avl)
692	return 0;
693	if (val_len < avl)
694	return -ENOENT;
695	} else if (rc < 0)
696	return -ENOENT;
697	}
698	}
699	ntfs_error(vol->sb, "Inode is corrupt. Run chkdsk.");
700	NVolSetErrors(vol);
701	return -EIO;
702	}
703
704	/**
705	* load_attribute_list - load an attribute list into memory
706	* @vol: ntfs volume from which to read
707	* @runlist: runlist of the attribute list
708	* @al_start: destination buffer
709	* @size: size of the destination buffer in bytes
710	* @initialized_size: initialized size of the attribute list
711	*
712	* Walk the runlist @runlist and load all clusters from it copying them into
713	* the linear buffer @al. The maximum number of bytes copied to @al is @size
714	* bytes. Note, @size does not need to be a multiple of the cluster size. If
715	* @initialized_size is less than @size, the region in @al between
716	* @initialized_size and @size will be zeroed and not read from disk.
717	*
718	* Return 0 on success or -errno on error.
719	*/
720	int load_attribute_list(ntfs_volume *vol, runlist *runlist, u8 *al_start,
721	const s64 size, const s64 initialized_size)
722	{
723	LCN lcn;
724	u8 *al = al_start;
725	u8 *al_end = al + initialized_size;
726	runlist_element *rl;
727	struct buffer_head *bh;
728	struct super_block *sb;
729	unsigned long block_size;
730	unsigned long block, max_block;
731	int err = 0;
732	unsigned char block_size_bits;
733
734	ntfs_debug("Entering.");
735	if (!vol || !runlist || !al || size <= 0 || initialized_size < 0 ||
736	initialized_size > size)
737	return -EINVAL;
738	if (!initialized_size) {
739	memset(al, 0, size);
740	return 0;
741	}
742	sb = vol->sb;
743	block_size = sb->s_blocksize;
744	block_size_bits = sb->s_blocksize_bits;
745	down_read(sem: &runlist->lock);
746	rl = runlist->rl;
747	if (!rl) {
748	ntfs_error(sb, "Cannot read attribute list since runlist is "
749	"missing.");
750	goto err_out;
751	}
752	/* Read all clusters specified by the runlist one run at a time. */
753	while (rl->length) {
754	lcn = ntfs_rl_vcn_to_lcn(rl, vcn: rl->vcn);
755	ntfs_debug("Reading vcn = 0x%llx, lcn = 0x%llx.",
756	(unsigned long long)rl->vcn,
757	(unsigned long long)lcn);
758	/* The attribute list cannot be sparse. */
759	if (lcn < 0) {
760	ntfs_error(sb, "ntfs_rl_vcn_to_lcn() failed. Cannot "
761	"read attribute list.");
762	goto err_out;
763	}
764	block = lcn << vol->cluster_size_bits >> block_size_bits;
765	/* Read the run from device in chunks of block_size bytes. */
766	max_block = block + (rl->length << vol->cluster_size_bits >>
767	block_size_bits);
768	ntfs_debug("max_block = 0x%lx.", max_block);
769	do {
770	ntfs_debug("Reading block = 0x%lx.", block);
771	bh = sb_bread(sb, block);
772	if (!bh) {
773	ntfs_error(sb, "sb_bread() failed. Cannot "
774	"read attribute list.");
775	goto err_out;
776	}
777	if (al + block_size >= al_end)
778	goto do_final;
779	memcpy(al, bh->b_data, block_size);
780	brelse(bh);
781	al += block_size;
782	} while (++block < max_block);
783	rl++;
784	}
785	if (initialized_size < size) {
786	initialize:
787	memset(al_start + initialized_size, 0, size - initialized_size);
788	}
789	done:
790	up_read(sem: &runlist->lock);
791	return err;
792	do_final:
793	if (al < al_end) {
794	/*
795	* Partial block.
796	*
797	* Note: The attribute list can be smaller than its allocation
798	* by multiple clusters. This has been encountered by at least
799	* two people running Windows XP, thus we cannot do any
800	* truncation sanity checking here. (AIA)
801	*/
802	memcpy(al, bh->b_data, al_end - al);
803	brelse(bh);
804	if (initialized_size < size)
805	goto initialize;
806	goto done;
807	}
808	brelse(bh);
809	/* Real overflow! */
810	ntfs_error(sb, "Attribute list buffer overflow. Read attribute list "
811	"is truncated.");
812	err_out:
813	err = -EIO;
814	goto done;
815	}
816
817	/**
818	* ntfs_external_attr_find - find an attribute in the attribute list of an inode
819	* @type: attribute type to find
820	* @name: attribute name to find (optional, i.e. NULL means don't care)
821	* @name_len: attribute name length (only needed if @name present)
822	* @ic: IGNORE_CASE or CASE_SENSITIVE (ignored if @name not present)
823	* @lowest_vcn: lowest vcn to find (optional, non-resident attributes only)
824	* @val: attribute value to find (optional, resident attributes only)
825	* @val_len: attribute value length
826	* @ctx: search context with mft record and attribute to search from
827	*
828	* You should not need to call this function directly. Use ntfs_attr_lookup()
829	* instead.
830	*
831	* Find an attribute by searching the attribute list for the corresponding
832	* attribute list entry. Having found the entry, map the mft record if the
833	* attribute is in a different mft record/inode, ntfs_attr_find() the attribute
834	* in there and return it.
835	*
836	* On first search @ctx->ntfs_ino must be the base mft record and @ctx must
837	* have been obtained from a call to ntfs_attr_get_search_ctx(). On subsequent
838	* calls @ctx->ntfs_ino can be any extent inode, too (@ctx->base_ntfs_ino is
839	* then the base inode).
840	*
841	* After finishing with the attribute/mft record you need to call
842	* ntfs_attr_put_search_ctx() to cleanup the search context (unmapping any
843	* mapped inodes, etc).
844	*
845	* If the attribute is found, ntfs_external_attr_find() returns 0 and
846	* @ctx->attr will point to the found attribute. @ctx->mrec will point to the
847	* mft record in which @ctx->attr is located and @ctx->al_entry will point to
848	* the attribute list entry for the attribute.
849	*
850	* If the attribute is not found, ntfs_external_attr_find() returns -ENOENT and
851	* @ctx->attr will point to the attribute in the base mft record before which
852	* the attribute being searched for would need to be inserted if such an action
853	* were to be desired. @ctx->mrec will point to the mft record in which
854	* @ctx->attr is located and @ctx->al_entry will point to the attribute list
855	* entry of the attribute before which the attribute being searched for would
856	* need to be inserted if such an action were to be desired.
857	*
858	* Thus to insert the not found attribute, one wants to add the attribute to
859	* @ctx->mrec (the base mft record) and if there is not enough space, the
860	* attribute should be placed in a newly allocated extent mft record. The
861	* attribute list entry for the inserted attribute should be inserted in the
862	* attribute list attribute at @ctx->al_entry.
863	*
864	* On actual error, ntfs_external_attr_find() returns -EIO. In this case
865	* @ctx->attr is undefined and in particular do not rely on it not changing.
866	*/
867	static int ntfs_external_attr_find(const ATTR_TYPE type,
868	const ntfschar *name, const u32 name_len,
869	const IGNORE_CASE_BOOL ic, const VCN lowest_vcn,
870	const u8 *val, const u32 val_len, ntfs_attr_search_ctx *ctx)
871	{
872	ntfs_inode *base_ni, *ni;
873	ntfs_volume *vol;
874	ATTR_LIST_ENTRY *al_entry, *next_al_entry;
875	u8 *al_start, *al_end;
876	ATTR_RECORD *a;
877	ntfschar *al_name;
878	u32 al_name_len;
879	int err = 0;
880	static const char *es = " Unmount and run chkdsk.";
881
882	ni = ctx->ntfs_ino;
883	base_ni = ctx->base_ntfs_ino;
884	ntfs_debug("Entering for inode 0x%lx, type 0x%x.", ni->mft_no, type);
885	if (!base_ni) {
886	/* First call happens with the base mft record. */
887	base_ni = ctx->base_ntfs_ino = ctx->ntfs_ino;
888	ctx->base_mrec = ctx->mrec;
889	}
890	if (ni == base_ni)
891	ctx->base_attr = ctx->attr;
892	if (type == AT_END)
893	goto not_found;
894	vol = base_ni->vol;
895	al_start = base_ni->attr_list;
896	al_end = al_start + base_ni->attr_list_size;
897	if (!ctx->al_entry)
898	ctx->al_entry = (ATTR_LIST_ENTRY*)al_start;
899	/*
900	* Iterate over entries in attribute list starting at @ctx->al_entry,
901	* or the entry following that, if @ctx->is_first is 'true'.
902	*/
903	if (ctx->is_first) {
904	al_entry = ctx->al_entry;
905	ctx->is_first = false;
906	} else
907	al_entry = (ATTR_LIST_ENTRY*)((u8*)ctx->al_entry +
908	le16_to_cpu(ctx->al_entry->length));
909	for (;; al_entry = next_al_entry) {
910	/* Out of bounds check. */
911	if ((u8*)al_entry < base_ni->attr_list ||
912	(u8*)al_entry > al_end)
913	break; /* Inode is corrupt. */
914	ctx->al_entry = al_entry;
915	/* Catch the end of the attribute list. */
916	if ((u8*)al_entry == al_end)
917	goto not_found;
918	if (!al_entry->length)
919	break;
920	if ((u8*)al_entry + 6 > al_end || (u8*)al_entry +
921	le16_to_cpu(al_entry->length) > al_end)
922	break;
923	next_al_entry = (ATTR_LIST_ENTRY*)((u8*)al_entry +
924	le16_to_cpu(al_entry->length));
925	if (le32_to_cpu(al_entry->type) > le32_to_cpu(type))
926	goto not_found;
927	if (type != al_entry->type)
928	continue;
929	/*
930	* If @name is present, compare the two names. If @name is
931	* missing, assume we want an unnamed attribute.
932	*/
933	al_name_len = al_entry->name_length;
934	al_name = (ntfschar*)((u8*)al_entry + al_entry->name_offset);
935	if (!name) {
936	if (al_name_len)
937	goto not_found;
938	} else if (!ntfs_are_names_equal(s1: al_name, s1_len: al_name_len, s2: name,
939	s2_len: name_len, ic, upcase: vol->upcase, upcase_size: vol->upcase_len)) {
940	register int rc;
941
942	rc = ntfs_collate_names(name1: name, name1_len: name_len, name2: al_name,
943	name2_len: al_name_len, err_val: 1, ic: IGNORE_CASE,
944	upcase: vol->upcase, upcase_len: vol->upcase_len);
945	/*
946	* If @name collates before al_name, there is no
947	* matching attribute.
948	*/
949	if (rc == -1)
950	goto not_found;
951	/* If the strings are not equal, continue search. */
952	if (rc)
953	continue;
954	/*
955	* FIXME: Reverse engineering showed 0, IGNORE_CASE but
956	* that is inconsistent with ntfs_attr_find(). The
957	* subsequent rc checks were also different. Perhaps I
958	* made a mistake in one of the two. Need to recheck
959	* which is correct or at least see what is going on...
960	* (AIA)
961	*/
962	rc = ntfs_collate_names(name1: name, name1_len: name_len, name2: al_name,
963	name2_len: al_name_len, err_val: 1, ic: CASE_SENSITIVE,
964	upcase: vol->upcase, upcase_len: vol->upcase_len);
965	if (rc == -1)
966	goto not_found;
967	if (rc)
968	continue;
969	}
970	/*
971	* The names match or @name not present and attribute is
972	* unnamed. Now check @lowest_vcn. Continue search if the
973	* next attribute list entry still fits @lowest_vcn. Otherwise
974	* we have reached the right one or the search has failed.
975	*/
976	if (lowest_vcn && (u8*)next_al_entry >= al_start &&
977	(u8*)next_al_entry + 6 < al_end &&
978	(u8*)next_al_entry + le16_to_cpu(
979	next_al_entry->length) <= al_end &&
980	sle64_to_cpu(x: next_al_entry->lowest_vcn) <=
981	lowest_vcn &&
982	next_al_entry->type == al_entry->type &&
983	next_al_entry->name_length == al_name_len &&
984	ntfs_are_names_equal(s1: (ntfschar*)((u8*)
985	next_al_entry +
986	next_al_entry->name_offset),
987	s1_len: next_al_entry->name_length,
988	s2: al_name, s2_len: al_name_len, ic: CASE_SENSITIVE,
989	upcase: vol->upcase, upcase_size: vol->upcase_len))
990	continue;
991	if (MREF_LE(al_entry->mft_reference) == ni->mft_no) {
992	if (MSEQNO_LE(al_entry->mft_reference) != ni->seq_no) {
993	ntfs_error(vol->sb, "Found stale mft "
994	"reference in attribute list "
995	"of base inode 0x%lx.%s",
996	base_ni->mft_no, es);
997	err = -EIO;
998	break;
999	}
1000	} else { /* Mft references do not match. */
1001	/* If there is a mapped record unmap it first. */
1002	if (ni != base_ni)
1003	unmap_extent_mft_record(ni);
1004	/* Do we want the base record back? */
1005	if (MREF_LE(al_entry->mft_reference) ==
1006	base_ni->mft_no) {
1007	ni = ctx->ntfs_ino = base_ni;
1008	ctx->mrec = ctx->base_mrec;
1009	} else {
1010	/* We want an extent record. */
1011	ctx->mrec = map_extent_mft_record(base_ni,
1012	le64_to_cpu(
1013	al_entry->mft_reference), ntfs_ino: &ni);
1014	if (IS_ERR(ptr: ctx->mrec)) {
1015	ntfs_error(vol->sb, "Failed to map "
1016	"extent mft record "
1017	"0x%lx of base inode "
1018	"0x%lx.%s",
1019	MREF_LE(al_entry->
1020	mft_reference),
1021	base_ni->mft_no, es);
1022	err = PTR_ERR(ptr: ctx->mrec);
1023	if (err == -ENOENT)
1024	err = -EIO;
1025	/* Cause @ctx to be sanitized below. */
1026	ni = NULL;
1027	break;
1028	}
1029	ctx->ntfs_ino = ni;
1030	}
1031	ctx->attr = (ATTR_RECORD*)((u8*)ctx->mrec +
1032	le16_to_cpu(ctx->mrec->attrs_offset));
1033	}
1034	/*
1035	* ctx->vfs_ino, ctx->mrec, and ctx->attr now point to the
1036	* mft record containing the attribute represented by the
1037	* current al_entry.
1038	*/
1039	/*
1040	* We could call into ntfs_attr_find() to find the right
1041	* attribute in this mft record but this would be less
1042	* efficient and not quite accurate as ntfs_attr_find() ignores
1043	* the attribute instance numbers for example which become
1044	* important when one plays with attribute lists. Also,
1045	* because a proper match has been found in the attribute list
1046	* entry above, the comparison can now be optimized. So it is
1047	* worth re-implementing a simplified ntfs_attr_find() here.
1048	*/
1049	a = ctx->attr;
1050	/*
1051	* Use a manual loop so we can still use break and continue
1052	* with the same meanings as above.
1053	*/
1054	do_next_attr_loop:
1055	if ((u8*)a < (u8*)ctx->mrec || (u8*)a > (u8*)ctx->mrec +
1056	le32_to_cpu(ctx->mrec->bytes_allocated))
1057	break;
1058	if (a->type == AT_END)
1059	break;
1060	if (!a->length)
1061	break;
1062	if (al_entry->instance != a->instance)
1063	goto do_next_attr;
1064	/*
1065	* If the type and/or the name are mismatched between the
1066	* attribute list entry and the attribute record, there is
1067	* corruption so we break and return error EIO.
1068	*/
1069	if (al_entry->type != a->type)
1070	break;
1071	if (!ntfs_are_names_equal(s1: (ntfschar*)((u8*)a +
1072	le16_to_cpu(a->name_offset)), s1_len: a->name_length,
1073	s2: al_name, s2_len: al_name_len, ic: CASE_SENSITIVE,
1074	upcase: vol->upcase, upcase_size: vol->upcase_len))
1075	break;
1076	ctx->attr = a;
1077	/*
1078	* If no @val specified or @val specified and it matches, we
1079	* have found it!
1080	*/
1081	if (!val || (!a->non_resident && le32_to_cpu(
1082	a->data.resident.value_length) == val_len &&
1083	!memcmp(p: (u8*)a +
1084	le16_to_cpu(a->data.resident.value_offset),
1085	q: val, size: val_len))) {
1086	ntfs_debug("Done, found.");
1087	return 0;
1088	}
1089	do_next_attr:
1090	/* Proceed to the next attribute in the current mft record. */
1091	a = (ATTR_RECORD*)((u8*)a + le32_to_cpu(a->length));
1092	goto do_next_attr_loop;
1093	}
1094	if (!err) {
1095	ntfs_error(vol->sb, "Base inode 0x%lx contains corrupt "
1096	"attribute list attribute.%s", base_ni->mft_no,
1097	es);
1098	err = -EIO;
1099	}
1100	if (ni != base_ni) {
1101	if (ni)
1102	unmap_extent_mft_record(ni);
1103	ctx->ntfs_ino = base_ni;
1104	ctx->mrec = ctx->base_mrec;
1105	ctx->attr = ctx->base_attr;
1106	}
1107	if (err != -ENOMEM)
1108	NVolSetErrors(vol);
1109	return err;
1110	not_found:
1111	/*
1112	* If we were looking for AT_END, we reset the search context @ctx and
1113	* use ntfs_attr_find() to seek to the end of the base mft record.
1114	*/
1115	if (type == AT_END) {
1116	ntfs_attr_reinit_search_ctx(ctx);
1117	return ntfs_attr_find(type: AT_END, name, name_len, ic, val, val_len,
1118	ctx);
1119	}
1120	/*
1121	* The attribute was not found. Before we return, we want to ensure
1122	* @ctx->mrec and @ctx->attr indicate the position at which the
1123	* attribute should be inserted in the base mft record. Since we also
1124	* want to preserve @ctx->al_entry we cannot reinitialize the search
1125	* context using ntfs_attr_reinit_search_ctx() as this would set
1126	* @ctx->al_entry to NULL. Thus we do the necessary bits manually (see
1127	* ntfs_attr_init_search_ctx() below). Note, we _only_ preserve
1128	* @ctx->al_entry as the remaining fields (base_*) are identical to
1129	* their non base_ counterparts and we cannot set @ctx->base_attr
1130	* correctly yet as we do not know what @ctx->attr will be set to by
1131	* the call to ntfs_attr_find() below.
1132	*/
1133	if (ni != base_ni)
1134	unmap_extent_mft_record(ni);
1135	ctx->mrec = ctx->base_mrec;
1136	ctx->attr = (ATTR_RECORD*)((u8*)ctx->mrec +
1137	le16_to_cpu(ctx->mrec->attrs_offset));
1138	ctx->is_first = true;
1139	ctx->ntfs_ino = base_ni;
1140	ctx->base_ntfs_ino = NULL;
1141	ctx->base_mrec = NULL;
1142	ctx->base_attr = NULL;
1143	/*
1144	* In case there are multiple matches in the base mft record, need to
1145	* keep enumerating until we get an attribute not found response (or
1146	* another error), otherwise we would keep returning the same attribute
1147	* over and over again and all programs using us for enumeration would
1148	* lock up in a tight loop.
1149	*/
1150	do {
1151	err = ntfs_attr_find(type, name, name_len, ic, val, val_len,
1152	ctx);
1153	} while (!err);
1154	ntfs_debug("Done, not found.");
1155	return err;
1156	}
1157
1158	/**
1159	* ntfs_attr_lookup - find an attribute in an ntfs inode
1160	* @type: attribute type to find
1161	* @name: attribute name to find (optional, i.e. NULL means don't care)
1162	* @name_len: attribute name length (only needed if @name present)
1163	* @ic: IGNORE_CASE or CASE_SENSITIVE (ignored if @name not present)
1164	* @lowest_vcn: lowest vcn to find (optional, non-resident attributes only)
1165	* @val: attribute value to find (optional, resident attributes only)
1166	* @val_len: attribute value length
1167	* @ctx: search context with mft record and attribute to search from
1168	*
1169	* Find an attribute in an ntfs inode. On first search @ctx->ntfs_ino must
1170	* be the base mft record and @ctx must have been obtained from a call to
1171	* ntfs_attr_get_search_ctx().
1172	*
1173	* This function transparently handles attribute lists and @ctx is used to
1174	* continue searches where they were left off at.
1175	*
1176	* After finishing with the attribute/mft record you need to call
1177	* ntfs_attr_put_search_ctx() to cleanup the search context (unmapping any
1178	* mapped inodes, etc).
1179	*
1180	* Return 0 if the search was successful and -errno if not.
1181	*
1182	* When 0, @ctx->attr is the found attribute and it is in mft record
1183	* @ctx->mrec. If an attribute list attribute is present, @ctx->al_entry is
1184	* the attribute list entry of the found attribute.
1185	*
1186	* When -ENOENT, @ctx->attr is the attribute which collates just after the
1187	* attribute being searched for, i.e. if one wants to add the attribute to the
1188	* mft record this is the correct place to insert it into. If an attribute
1189	* list attribute is present, @ctx->al_entry is the attribute list entry which
1190	* collates just after the attribute list entry of the attribute being searched
1191	* for, i.e. if one wants to add the attribute to the mft record this is the
1192	* correct place to insert its attribute list entry into.
1193	*
1194	* When -errno != -ENOENT, an error occurred during the lookup. @ctx->attr is
1195	* then undefined and in particular you should not rely on it not changing.
1196	*/
1197	int ntfs_attr_lookup(const ATTR_TYPE type, const ntfschar *name,
1198	const u32 name_len, const IGNORE_CASE_BOOL ic,
1199	const VCN lowest_vcn, const u8 *val, const u32 val_len,
1200	ntfs_attr_search_ctx *ctx)
1201	{
1202	ntfs_inode *base_ni;
1203
1204	ntfs_debug("Entering.");
1205	BUG_ON(IS_ERR(ctx->mrec));
1206	if (ctx->base_ntfs_ino)
1207	base_ni = ctx->base_ntfs_ino;
1208	else
1209	base_ni = ctx->ntfs_ino;
1210	/* Sanity check, just for debugging really. */
1211	BUG_ON(!base_ni);
1212	if (!NInoAttrList(ni: base_ni) || type == AT_ATTRIBUTE_LIST)
1213	return ntfs_attr_find(type, name, name_len, ic, val, val_len,
1214	ctx);
1215	return ntfs_external_attr_find(type, name, name_len, ic, lowest_vcn,
1216	val, val_len, ctx);
1217	}
1218
1219	/**
1220	* ntfs_attr_init_search_ctx - initialize an attribute search context
1221	* @ctx: attribute search context to initialize
1222	* @ni: ntfs inode with which to initialize the search context
1223	* @mrec: mft record with which to initialize the search context
1224	*
1225	* Initialize the attribute search context @ctx with @ni and @mrec.
1226	*/
1227	static inline void ntfs_attr_init_search_ctx(ntfs_attr_search_ctx *ctx,
1228	ntfs_inode *ni, MFT_RECORD *mrec)
1229	{
1230	*ctx = (ntfs_attr_search_ctx) {
1231	.mrec = mrec,
1232	/* Sanity checks are performed elsewhere. */
1233	.attr = (ATTR_RECORD*)((u8*)mrec +
1234	le16_to_cpu(mrec->attrs_offset)),
1235	.is_first = true,
1236	.ntfs_ino = ni,
1237	};
1238	}
1239
1240	/**
1241	* ntfs_attr_reinit_search_ctx - reinitialize an attribute search context
1242	* @ctx: attribute search context to reinitialize
1243	*
1244	* Reinitialize the attribute search context @ctx, unmapping an associated
1245	* extent mft record if present, and initialize the search context again.
1246	*
1247	* This is used when a search for a new attribute is being started to reset
1248	* the search context to the beginning.
1249	*/
1250	void ntfs_attr_reinit_search_ctx(ntfs_attr_search_ctx *ctx)
1251	{
1252	if (likely(!ctx->base_ntfs_ino)) {
1253	/* No attribute list. */
1254	ctx->is_first = true;
1255	/* Sanity checks are performed elsewhere. */
1256	ctx->attr = (ATTR_RECORD*)((u8*)ctx->mrec +
1257	le16_to_cpu(ctx->mrec->attrs_offset));
1258	/*
1259	* This needs resetting due to ntfs_external_attr_find() which
1260	* can leave it set despite having zeroed ctx->base_ntfs_ino.
1261	*/
1262	ctx->al_entry = NULL;
1263	return;
1264	} /* Attribute list. */
1265	if (ctx->ntfs_ino != ctx->base_ntfs_ino)
1266	unmap_extent_mft_record(ni: ctx->ntfs_ino);
1267	ntfs_attr_init_search_ctx(ctx, ni: ctx->base_ntfs_ino, mrec: ctx->base_mrec);
1268	return;
1269	}
1270
1271	/**
1272	* ntfs_attr_get_search_ctx - allocate/initialize a new attribute search context
1273	* @ni: ntfs inode with which to initialize the search context
1274	* @mrec: mft record with which to initialize the search context
1275	*
1276	* Allocate a new attribute search context, initialize it with @ni and @mrec,
1277	* and return it. Return NULL if allocation failed.
1278	*/
1279	ntfs_attr_search_ctx *ntfs_attr_get_search_ctx(ntfs_inode *ni, MFT_RECORD *mrec)
1280	{
1281	ntfs_attr_search_ctx *ctx;
1282
1283	ctx = kmem_cache_alloc(cachep: ntfs_attr_ctx_cache, GFP_NOFS);
1284	if (ctx)
1285	ntfs_attr_init_search_ctx(ctx, ni, mrec);
1286	return ctx;
1287	}
1288
1289	/**
1290	* ntfs_attr_put_search_ctx - release an attribute search context
1291	* @ctx: attribute search context to free
1292	*
1293	* Release the attribute search context @ctx, unmapping an associated extent
1294	* mft record if present.
1295	*/
1296	void ntfs_attr_put_search_ctx(ntfs_attr_search_ctx *ctx)
1297	{
1298	if (ctx->base_ntfs_ino && ctx->ntfs_ino != ctx->base_ntfs_ino)
1299	unmap_extent_mft_record(ni: ctx->ntfs_ino);
1300	kmem_cache_free(s: ntfs_attr_ctx_cache, objp: ctx);
1301	return;
1302	}
1303
1304	#ifdef NTFS_RW
1305
1306	/**
1307	* ntfs_attr_find_in_attrdef - find an attribute in the $AttrDef system file
1308	* @vol: ntfs volume to which the attribute belongs
1309	* @type: attribute type which to find
1310	*
1311	* Search for the attribute definition record corresponding to the attribute
1312	* @type in the $AttrDef system file.
1313	*
1314	* Return the attribute type definition record if found and NULL if not found.
1315	*/
1316	static ATTR_DEF *ntfs_attr_find_in_attrdef(const ntfs_volume *vol,
1317	const ATTR_TYPE type)
1318	{
1319	ATTR_DEF *ad;
1320
1321	BUG_ON(!vol->attrdef);
1322	BUG_ON(!type);
1323	for (ad = vol->attrdef; (u8*)ad - (u8*)vol->attrdef <
1324	vol->attrdef_size && ad->type; ++ad) {
1325	/* We have not found it yet, carry on searching. */
1326	if (likely(le32_to_cpu(ad->type) < le32_to_cpu(type)))
1327	continue;
1328	/* We found the attribute; return it. */
1329	if (likely(ad->type == type))
1330	return ad;
1331	/* We have gone too far already. No point in continuing. */
1332	break;
1333	}
1334	/* Attribute not found. */
1335	ntfs_debug("Attribute type 0x%x not found in $AttrDef.",
1336	le32_to_cpu(type));
1337	return NULL;
1338	}
1339
1340	/**
1341	* ntfs_attr_size_bounds_check - check a size of an attribute type for validity
1342	* @vol: ntfs volume to which the attribute belongs
1343	* @type: attribute type which to check
1344	* @size: size which to check
1345	*
1346	* Check whether the @size in bytes is valid for an attribute of @type on the
1347	* ntfs volume @vol. This information is obtained from $AttrDef system file.
1348	*
1349	* Return 0 if valid, -ERANGE if not valid, or -ENOENT if the attribute is not
1350	* listed in $AttrDef.
1351	*/
1352	int ntfs_attr_size_bounds_check(const ntfs_volume *vol, const ATTR_TYPE type,
1353	const s64 size)
1354	{
1355	ATTR_DEF *ad;
1356
1357	BUG_ON(size < 0);
1358	/*
1359	* $ATTRIBUTE_LIST has a maximum size of 256kiB, but this is not
1360	* listed in $AttrDef.
1361	*/
1362	if (unlikely(type == AT_ATTRIBUTE_LIST && size > 256 * 1024))
1363	return -ERANGE;
1364	/* Get the $AttrDef entry for the attribute @type. */
1365	ad = ntfs_attr_find_in_attrdef(vol, type);
1366	if (unlikely(!ad))
1367	return -ENOENT;
1368	/* Do the bounds check. */
1369	if (((sle64_to_cpu(x: ad->min_size) > 0) &&
1370	size < sle64_to_cpu(x: ad->min_size)) ||
1371	((sle64_to_cpu(x: ad->max_size) > 0) && size >
1372	sle64_to_cpu(x: ad->max_size)))
1373	return -ERANGE;
1374	return 0;
1375	}
1376
1377	/**
1378	* ntfs_attr_can_be_non_resident - check if an attribute can be non-resident
1379	* @vol: ntfs volume to which the attribute belongs
1380	* @type: attribute type which to check
1381	*
1382	* Check whether the attribute of @type on the ntfs volume @vol is allowed to
1383	* be non-resident. This information is obtained from $AttrDef system file.
1384	*
1385	* Return 0 if the attribute is allowed to be non-resident, -EPERM if not, and
1386	* -ENOENT if the attribute is not listed in $AttrDef.
1387	*/
1388	int ntfs_attr_can_be_non_resident(const ntfs_volume *vol, const ATTR_TYPE type)
1389	{
1390	ATTR_DEF *ad;
1391
1392	/* Find the attribute definition record in $AttrDef. */
1393	ad = ntfs_attr_find_in_attrdef(vol, type);
1394	if (unlikely(!ad))
1395	return -ENOENT;
1396	/* Check the flags and return the result. */
1397	if (ad->flags & ATTR_DEF_RESIDENT)
1398	return -EPERM;
1399	return 0;
1400	}
1401
1402	/**
1403	* ntfs_attr_can_be_resident - check if an attribute can be resident
1404	* @vol: ntfs volume to which the attribute belongs
1405	* @type: attribute type which to check
1406	*
1407	* Check whether the attribute of @type on the ntfs volume @vol is allowed to
1408	* be resident. This information is derived from our ntfs knowledge and may
1409	* not be completely accurate, especially when user defined attributes are
1410	* present. Basically we allow everything to be resident except for index
1411	* allocation and $EA attributes.
1412	*
1413	* Return 0 if the attribute is allowed to be non-resident and -EPERM if not.
1414	*
1415	* Warning: In the system file $MFT the attribute $Bitmap must be non-resident
1416	* otherwise windows will not boot (blue screen of death)! We cannot
1417	* check for this here as we do not know which inode's $Bitmap is
1418	* being asked about so the caller needs to special case this.
1419	*/
1420	int ntfs_attr_can_be_resident(const ntfs_volume *vol, const ATTR_TYPE type)
1421	{
1422	if (type == AT_INDEX_ALLOCATION)
1423	return -EPERM;
1424	return 0;
1425	}
1426
1427	/**
1428	* ntfs_attr_record_resize - resize an attribute record
1429	* @m: mft record containing attribute record
1430	* @a: attribute record to resize
1431	* @new_size: new size in bytes to which to resize the attribute record @a
1432	*
1433	* Resize the attribute record @a, i.e. the resident part of the attribute, in
1434	* the mft record @m to @new_size bytes.
1435	*
1436	* Return 0 on success and -errno on error. The following error codes are
1437	* defined:
1438	* -ENOSPC - Not enough space in the mft record @m to perform the resize.
1439	*
1440	* Note: On error, no modifications have been performed whatsoever.
1441	*
1442	* Warning: If you make a record smaller without having copied all the data you
1443	* are interested in the data may be overwritten.
1444	*/
1445	int ntfs_attr_record_resize(MFT_RECORD *m, ATTR_RECORD *a, u32 new_size)
1446	{
1447	ntfs_debug("Entering for new_size %u.", new_size);
1448	/* Align to 8 bytes if it is not already done. */
1449	if (new_size & 7)
1450	new_size = (new_size + 7) & ~7;
1451	/* If the actual attribute length has changed, move things around. */
1452	if (new_size != le32_to_cpu(a->length)) {
1453	u32 new_muse = le32_to_cpu(m->bytes_in_use) -
1454	le32_to_cpu(a->length) + new_size;
1455	/* Not enough space in this mft record. */
1456	if (new_muse > le32_to_cpu(m->bytes_allocated))
1457	return -ENOSPC;
1458	/* Move attributes following @a to their new location. */
1459	memmove((u8*)a + new_size, (u8*)a + le32_to_cpu(a->length),
1460	le32_to_cpu(m->bytes_in_use) - ((u8*)a -
1461	(u8*)m) - le32_to_cpu(a->length));
1462	/* Adjust @m to reflect the change in used space. */
1463	m->bytes_in_use = cpu_to_le32(new_muse);
1464	/* Adjust @a to reflect the new size. */
1465	if (new_size >= offsetof(ATTR_REC, length) + sizeof(a->length))
1466	a->length = cpu_to_le32(new_size);
1467	}
1468	return 0;
1469	}
1470
1471	/**
1472	* ntfs_resident_attr_value_resize - resize the value of a resident attribute
1473	* @m: mft record containing attribute record
1474	* @a: attribute record whose value to resize
1475	* @new_size: new size in bytes to which to resize the attribute value of @a
1476	*
1477	* Resize the value of the attribute @a in the mft record @m to @new_size bytes.
1478	* If the value is made bigger, the newly allocated space is cleared.
1479	*
1480	* Return 0 on success and -errno on error. The following error codes are
1481	* defined:
1482	* -ENOSPC - Not enough space in the mft record @m to perform the resize.
1483	*
1484	* Note: On error, no modifications have been performed whatsoever.
1485	*
1486	* Warning: If you make a record smaller without having copied all the data you
1487	* are interested in the data may be overwritten.
1488	*/
1489	int ntfs_resident_attr_value_resize(MFT_RECORD *m, ATTR_RECORD *a,
1490	const u32 new_size)
1491	{
1492	u32 old_size;
1493
1494	/* Resize the resident part of the attribute record. */
1495	if (ntfs_attr_record_resize(m, a,
1496	le16_to_cpu(a->data.resident.value_offset) + new_size))
1497	return -ENOSPC;
1498	/*
1499	* The resize succeeded! If we made the attribute value bigger, clear
1500	* the area between the old size and @new_size.
1501	*/
1502	old_size = le32_to_cpu(a->data.resident.value_length);
1503	if (new_size > old_size)
1504	memset((u8*)a + le16_to_cpu(a->data.resident.value_offset) +
1505	old_size, 0, new_size - old_size);
1506	/* Finally update the length of the attribute value. */
1507	a->data.resident.value_length = cpu_to_le32(new_size);
1508	return 0;
1509	}
1510
1511	/**
1512	* ntfs_attr_make_non_resident - convert a resident to a non-resident attribute
1513	* @ni: ntfs inode describing the attribute to convert
1514	* @data_size: size of the resident data to copy to the non-resident attribute
1515	*
1516	* Convert the resident ntfs attribute described by the ntfs inode @ni to a
1517	* non-resident one.
1518	*
1519	* @data_size must be equal to the attribute value size. This is needed since
1520	* we need to know the size before we can map the mft record and our callers
1521	* always know it. The reason we cannot simply read the size from the vfs
1522	* inode i_size is that this is not necessarily uptodate. This happens when
1523	* ntfs_attr_make_non_resident() is called in the ->truncate call path(s).
1524	*
1525	* Return 0 on success and -errno on error. The following error return codes
1526	* are defined:
1527	* -EPERM - The attribute is not allowed to be non-resident.
1528	* -ENOMEM - Not enough memory.
1529	* -ENOSPC - Not enough disk space.
1530	* -EINVAL - Attribute not defined on the volume.
1531	* -EIO - I/o error or other error.
1532	* Note that -ENOSPC is also returned in the case that there is not enough
1533	* space in the mft record to do the conversion. This can happen when the mft
1534	* record is already very full. The caller is responsible for trying to make
1535	* space in the mft record and trying again. FIXME: Do we need a separate
1536	* error return code for this kind of -ENOSPC or is it always worth trying
1537	* again in case the attribute may then fit in a resident state so no need to
1538	* make it non-resident at all? Ho-hum... (AIA)
1539	*
1540	* NOTE to self: No changes in the attribute list are required to move from
1541	* a resident to a non-resident attribute.
1542	*
1543	* Locking: - The caller must hold i_mutex on the inode.
1544	*/
1545	int ntfs_attr_make_non_resident(ntfs_inode *ni, const u32 data_size)
1546	{
1547	s64 new_size;
1548	struct inode *vi = VFS_I(ni);
1549	ntfs_volume *vol = ni->vol;
1550	ntfs_inode *base_ni;
1551	MFT_RECORD *m;
1552	ATTR_RECORD *a;
1553	ntfs_attr_search_ctx *ctx;
1554	struct page *page;
1555	runlist_element *rl;
1556	u8 *kaddr;
1557	unsigned long flags;
1558	int mp_size, mp_ofs, name_ofs, arec_size, err, err2;
1559	u32 attr_size;
1560	u8 old_res_attr_flags;
1561
1562	/* Check that the attribute is allowed to be non-resident. */
1563	err = ntfs_attr_can_be_non_resident(vol, type: ni->type);
1564	if (unlikely(err)) {
1565	if (err == -EPERM)
1566	ntfs_debug("Attribute is not allowed to be "
1567	"non-resident.");
1568	else
1569	ntfs_debug("Attribute not defined on the NTFS "
1570	"volume!");
1571	return err;
1572	}
1573	/*
1574	* FIXME: Compressed and encrypted attributes are not supported when
1575	* writing and we should never have gotten here for them.
1576	*/
1577	BUG_ON(NInoCompressed(ni));
1578	BUG_ON(NInoEncrypted(ni));
1579	/*
1580	* The size needs to be aligned to a cluster boundary for allocation
1581	* purposes.
1582	*/
1583	new_size = (data_size + vol->cluster_size - 1) &
1584	~(vol->cluster_size - 1);
1585	if (new_size > 0) {
1586	/*
1587	* Will need the page later and since the page lock nests
1588	* outside all ntfs locks, we need to get the page now.
1589	*/
1590	page = find_or_create_page(mapping: vi->i_mapping, index: 0,
1591	gfp_mask: mapping_gfp_mask(mapping: vi->i_mapping));
1592	if (unlikely(!page))
1593	return -ENOMEM;
1594	/* Start by allocating clusters to hold the attribute value. */
1595	rl = ntfs_cluster_alloc(vol, start_vcn: 0, count: new_size >>
1596	vol->cluster_size_bits, start_lcn: -1, zone: DATA_ZONE, is_extension: true);
1597	if (IS_ERR(ptr: rl)) {
1598	err = PTR_ERR(ptr: rl);
1599	ntfs_debug("Failed to allocate cluster%s, error code "
1600	"%i.", (new_size >>
1601	vol->cluster_size_bits) > 1 ? "s" : "",
1602	err);
1603	goto page_err_out;
1604	}
1605	} else {
1606	rl = NULL;
1607	page = NULL;
1608	}
1609	/* Determine the size of the mapping pairs array. */
1610	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl, first_vcn: 0, last_vcn: -1);
1611	if (unlikely(mp_size < 0)) {
1612	err = mp_size;
1613	ntfs_debug("Failed to get size for mapping pairs array, error "
1614	"code %i.", err);
1615	goto rl_err_out;
1616	}
1617	down_write(sem: &ni->runlist.lock);
1618	if (!NInoAttr(ni))
1619	base_ni = ni;
1620	else
1621	base_ni = ni->ext.base_ntfs_ino;
1622	m = map_mft_record(ni: base_ni);
1623	if (IS_ERR(ptr: m)) {
1624	err = PTR_ERR(ptr: m);
1625	m = NULL;
1626	ctx = NULL;
1627	goto err_out;
1628	}
1629	ctx = ntfs_attr_get_search_ctx(ni: base_ni, mrec: m);
1630	if (unlikely(!ctx)) {
1631	err = -ENOMEM;
1632	goto err_out;
1633	}
1634	err = ntfs_attr_lookup(type: ni->type, name: ni->name, name_len: ni->name_len,
1635	ic: CASE_SENSITIVE, lowest_vcn: 0, NULL, val_len: 0, ctx);
1636	if (unlikely(err)) {
1637	if (err == -ENOENT)
1638	err = -EIO;
1639	goto err_out;
1640	}
1641	m = ctx->mrec;
1642	a = ctx->attr;
1643	BUG_ON(NInoNonResident(ni));
1644	BUG_ON(a->non_resident);
1645	/*
1646	* Calculate new offsets for the name and the mapping pairs array.
1647	*/
1648	if (NInoSparse(ni) || NInoCompressed(ni))
1649	name_ofs = (offsetof(ATTR_REC,
1650	data.non_resident.compressed_size) +
1651	sizeof(a->data.non_resident.compressed_size) +
1652	7) & ~7;
1653	else
1654	name_ofs = (offsetof(ATTR_REC,
1655	data.non_resident.compressed_size) + 7) & ~7;
1656	mp_ofs = (name_ofs + a->name_length * sizeof(ntfschar) + 7) & ~7;
1657	/*
1658	* Determine the size of the resident part of the now non-resident
1659	* attribute record.
1660	*/
1661	arec_size = (mp_ofs + mp_size + 7) & ~7;
1662	/*
1663	* If the page is not uptodate bring it uptodate by copying from the
1664	* attribute value.
1665	*/
1666	attr_size = le32_to_cpu(a->data.resident.value_length);
1667	BUG_ON(attr_size != data_size);
1668	if (page && !PageUptodate(page)) {
1669	kaddr = kmap_atomic(page);
1670	memcpy(kaddr, (u8*)a +
1671	le16_to_cpu(a->data.resident.value_offset),
1672	attr_size);
1673	memset(kaddr + attr_size, 0, PAGE_SIZE - attr_size);
1674	kunmap_atomic(kaddr);
1675	flush_dcache_page(page);
1676	SetPageUptodate(page);
1677	}
1678	/* Backup the attribute flag. */
1679	old_res_attr_flags = a->data.resident.flags;
1680	/* Resize the resident part of the attribute record. */
1681	err = ntfs_attr_record_resize(m, a, new_size: arec_size);
1682	if (unlikely(err))
1683	goto err_out;
1684	/*
1685	* Convert the resident part of the attribute record to describe a
1686	* non-resident attribute.
1687	*/
1688	a->non_resident = 1;
1689	/* Move the attribute name if it exists and update the offset. */
1690	if (a->name_length)
1691	memmove((u8*)a + name_ofs, (u8*)a + le16_to_cpu(a->name_offset),
1692	a->name_length * sizeof(ntfschar));
1693	a->name_offset = cpu_to_le16(name_ofs);
1694	/* Setup the fields specific to non-resident attributes. */
1695	a->data.non_resident.lowest_vcn = 0;
1696	a->data.non_resident.highest_vcn = cpu_to_sle64(x: (new_size - 1) >>
1697	vol->cluster_size_bits);
1698	a->data.non_resident.mapping_pairs_offset = cpu_to_le16(mp_ofs);
1699	memset(&a->data.non_resident.reserved, 0,
1700	sizeof(a->data.non_resident.reserved));
1701	a->data.non_resident.allocated_size = cpu_to_sle64(x: new_size);
1702	a->data.non_resident.data_size =
1703	a->data.non_resident.initialized_size =
1704	cpu_to_sle64(x: attr_size);
1705	if (NInoSparse(ni) || NInoCompressed(ni)) {
1706	a->data.non_resident.compression_unit = 0;
1707	if (NInoCompressed(ni) || vol->major_ver < 3)
1708	a->data.non_resident.compression_unit = 4;
1709	a->data.non_resident.compressed_size =
1710	a->data.non_resident.allocated_size;
1711	} else
1712	a->data.non_resident.compression_unit = 0;
1713	/* Generate the mapping pairs array into the attribute record. */
1714	err = ntfs_mapping_pairs_build(vol, dst: (u8*)a + mp_ofs,
1715	dst_len: arec_size - mp_ofs, rl, first_vcn: 0, last_vcn: -1, NULL);
1716	if (unlikely(err)) {
1717	ntfs_debug("Failed to build mapping pairs, error code %i.",
1718	err);
1719	goto undo_err_out;
1720	}
1721	/* Setup the in-memory attribute structure to be non-resident. */
1722	ni->runlist.rl = rl;
1723	write_lock_irqsave(&ni->size_lock, flags);
1724	ni->allocated_size = new_size;
1725	if (NInoSparse(ni) || NInoCompressed(ni)) {
1726	ni->itype.compressed.size = ni->allocated_size;
1727	if (a->data.non_resident.compression_unit) {
1728	ni->itype.compressed.block_size = 1U << (a->data.
1729	non_resident.compression_unit +
1730	vol->cluster_size_bits);
1731	ni->itype.compressed.block_size_bits =
1732	ffs(ni->itype.compressed.block_size) -
1733	1;
1734	ni->itype.compressed.block_clusters = 1U <<
1735	a->data.non_resident.compression_unit;
1736	} else {
1737	ni->itype.compressed.block_size = 0;
1738	ni->itype.compressed.block_size_bits = 0;
1739	ni->itype.compressed.block_clusters = 0;
1740	}
1741	vi->i_blocks = ni->itype.compressed.size >> 9;
1742	} else
1743	vi->i_blocks = ni->allocated_size >> 9;
1744	write_unlock_irqrestore(&ni->size_lock, flags);
1745	/*
1746	* This needs to be last since the address space operations ->read_folio
1747	* and ->writepage can run concurrently with us as they are not
1748	* serialized on i_mutex. Note, we are not allowed to fail once we flip
1749	* this switch, which is another reason to do this last.
1750	*/
1751	NInoSetNonResident(ni);
1752	/* Mark the mft record dirty, so it gets written back. */
1753	flush_dcache_mft_record_page(ni: ctx->ntfs_ino);
1754	mark_mft_record_dirty(ni: ctx->ntfs_ino);
1755	ntfs_attr_put_search_ctx(ctx);
1756	unmap_mft_record(ni: base_ni);
1757	up_write(sem: &ni->runlist.lock);
1758	if (page) {
1759	set_page_dirty(page);
1760	unlock_page(page);
1761	put_page(page);
1762	}
1763	ntfs_debug("Done.");
1764	return 0;
1765	undo_err_out:
1766	/* Convert the attribute back into a resident attribute. */
1767	a->non_resident = 0;
1768	/* Move the attribute name if it exists and update the offset. */
1769	name_ofs = (offsetof(ATTR_RECORD, data.resident.reserved) +
1770	sizeof(a->data.resident.reserved) + 7) & ~7;
1771	if (a->name_length)
1772	memmove((u8*)a + name_ofs, (u8*)a + le16_to_cpu(a->name_offset),
1773	a->name_length * sizeof(ntfschar));
1774	mp_ofs = (name_ofs + a->name_length * sizeof(ntfschar) + 7) & ~7;
1775	a->name_offset = cpu_to_le16(name_ofs);
1776	arec_size = (mp_ofs + attr_size + 7) & ~7;
1777	/* Resize the resident part of the attribute record. */
1778	err2 = ntfs_attr_record_resize(m, a, new_size: arec_size);
1779	if (unlikely(err2)) {
1780	/*
1781	* This cannot happen (well if memory corruption is at work it
1782	* could happen in theory), but deal with it as well as we can.
1783	* If the old size is too small, truncate the attribute,
1784	* otherwise simply give it a larger allocated size.
1785	* FIXME: Should check whether chkdsk complains when the
1786	* allocated size is much bigger than the resident value size.
1787	*/
1788	arec_size = le32_to_cpu(a->length);
1789	if ((mp_ofs + attr_size) > arec_size) {
1790	err2 = attr_size;
1791	attr_size = arec_size - mp_ofs;
1792	ntfs_error(vol->sb, "Failed to undo partial resident "
1793	"to non-resident attribute "
1794	"conversion. Truncating inode 0x%lx, "
1795	"attribute type 0x%x from %i bytes to "
1796	"%i bytes to maintain metadata "
1797	"consistency. THIS MEANS YOU ARE "
1798	"LOSING %i BYTES DATA FROM THIS %s.",
1799	vi->i_ino,
1800	(unsigned)le32_to_cpu(ni->type),
1801	err2, attr_size, err2 - attr_size,
1802	((ni->type == AT_DATA) &&
1803	!ni->name_len) ? "FILE": "ATTRIBUTE");
1804	write_lock_irqsave(&ni->size_lock, flags);
1805	ni->initialized_size = attr_size;
1806	i_size_write(inode: vi, i_size: attr_size);
1807	write_unlock_irqrestore(&ni->size_lock, flags);
1808	}
1809	}
1810	/* Setup the fields specific to resident attributes. */
1811	a->data.resident.value_length = cpu_to_le32(attr_size);
1812	a->data.resident.value_offset = cpu_to_le16(mp_ofs);
1813	a->data.resident.flags = old_res_attr_flags;
1814	memset(&a->data.resident.reserved, 0,
1815	sizeof(a->data.resident.reserved));
1816	/* Copy the data from the page back to the attribute value. */
1817	if (page) {
1818	kaddr = kmap_atomic(page);
1819	memcpy((u8*)a + mp_ofs, kaddr, attr_size);
1820	kunmap_atomic(kaddr);
1821	}
1822	/* Setup the allocated size in the ntfs inode in case it changed. */
1823	write_lock_irqsave(&ni->size_lock, flags);
1824	ni->allocated_size = arec_size - mp_ofs;
1825	write_unlock_irqrestore(&ni->size_lock, flags);
1826	/* Mark the mft record dirty, so it gets written back. */
1827	flush_dcache_mft_record_page(ni: ctx->ntfs_ino);
1828	mark_mft_record_dirty(ni: ctx->ntfs_ino);
1829	err_out:
1830	if (ctx)
1831	ntfs_attr_put_search_ctx(ctx);
1832	if (m)
1833	unmap_mft_record(ni: base_ni);
1834	ni->runlist.rl = NULL;
1835	up_write(sem: &ni->runlist.lock);
1836	rl_err_out:
1837	if (rl) {
1838	if (ntfs_cluster_free_from_rl(vol, rl) < 0) {
1839	ntfs_error(vol->sb, "Failed to release allocated "
1840	"cluster(s) in error code path. Run "
1841	"chkdsk to recover the lost "
1842	"cluster(s).");
1843	NVolSetErrors(vol);
1844	}
1845	ntfs_free(addr: rl);
1846	page_err_out:
1847	unlock_page(page);
1848	put_page(page);
1849	}
1850	if (err == -EINVAL)
1851	err = -EIO;
1852	return err;
1853	}
1854
1855	/**
1856	* ntfs_attr_extend_allocation - extend the allocated space of an attribute
1857	* @ni: ntfs inode of the attribute whose allocation to extend
1858	* @new_alloc_size: new size in bytes to which to extend the allocation to
1859	* @new_data_size: new size in bytes to which to extend the data to
1860	* @data_start: beginning of region which is required to be non-sparse
1861	*
1862	* Extend the allocated space of an attribute described by the ntfs inode @ni
1863	* to @new_alloc_size bytes. If @data_start is -1, the whole extension may be
1864	* implemented as a hole in the file (as long as both the volume and the ntfs
1865	* inode @ni have sparse support enabled). If @data_start is >= 0, then the
1866	* region between the old allocated size and @data_start - 1 may be made sparse
1867	* but the regions between @data_start and @new_alloc_size must be backed by
1868	* actual clusters.
1869	*
1870	* If @new_data_size is -1, it is ignored. If it is >= 0, then the data size
1871	* of the attribute is extended to @new_data_size. Note that the i_size of the
1872	* vfs inode is not updated. Only the data size in the base attribute record
1873	* is updated. The caller has to update i_size separately if this is required.
1874	* WARNING: It is a BUG() for @new_data_size to be smaller than the old data
1875	* size as well as for @new_data_size to be greater than @new_alloc_size.
1876	*
1877	* For resident attributes this involves resizing the attribute record and if
1878	* necessary moving it and/or other attributes into extent mft records and/or
1879	* converting the attribute to a non-resident attribute which in turn involves
1880	* extending the allocation of a non-resident attribute as described below.
1881	*
1882	* For non-resident attributes this involves allocating clusters in the data
1883	* zone on the volume (except for regions that are being made sparse) and
1884	* extending the run list to describe the allocated clusters as well as
1885	* updating the mapping pairs array of the attribute. This in turn involves
1886	* resizing the attribute record and if necessary moving it and/or other
1887	* attributes into extent mft records and/or splitting the attribute record
1888	* into multiple extent attribute records.
1889	*
1890	* Also, the attribute list attribute is updated if present and in some of the
1891	* above cases (the ones where extent mft records/attributes come into play),
1892	* an attribute list attribute is created if not already present.
1893	*
1894	* Return the new allocated size on success and -errno on error. In the case
1895	* that an error is encountered but a partial extension at least up to
1896	* @data_start (if present) is possible, the allocation is partially extended
1897	* and this is returned. This means the caller must check the returned size to
1898	* determine if the extension was partial. If @data_start is -1 then partial
1899	* allocations are not performed.
1900	*
1901	* WARNING: Do not call ntfs_attr_extend_allocation() for $MFT/$DATA.
1902	*
1903	* Locking: This function takes the runlist lock of @ni for writing as well as
1904	* locking the mft record of the base ntfs inode. These locks are maintained
1905	* throughout execution of the function. These locks are required so that the
1906	* attribute can be resized safely and so that it can for example be converted
1907	* from resident to non-resident safely.
1908	*
1909	* TODO: At present attribute list attribute handling is not implemented.
1910	*
1911	* TODO: At present it is not safe to call this function for anything other
1912	* than the $DATA attribute(s) of an uncompressed and unencrypted file.
1913	*/
1914	s64 ntfs_attr_extend_allocation(ntfs_inode *ni, s64 new_alloc_size,
1915	const s64 new_data_size, const s64 data_start)
1916	{
1917	VCN vcn;
1918	s64 ll, allocated_size, start = data_start;
1919	struct inode *vi = VFS_I(ni);
1920	ntfs_volume *vol = ni->vol;
1921	ntfs_inode *base_ni;
1922	MFT_RECORD *m;
1923	ATTR_RECORD *a;
1924	ntfs_attr_search_ctx *ctx;
1925	runlist_element *rl, *rl2;
1926	unsigned long flags;
1927	int err, mp_size;
1928	u32 attr_len = 0; /* Silence stupid gcc warning. */
1929	bool mp_rebuilt;
1930
1931	#ifdef DEBUG
1932	read_lock_irqsave(&ni->size_lock, flags);
1933	allocated_size = ni->allocated_size;
1934	read_unlock_irqrestore(&ni->size_lock, flags);
1935	ntfs_debug("Entering for i_ino 0x%lx, attribute type 0x%x, "
1936	"old_allocated_size 0x%llx, "
1937	"new_allocated_size 0x%llx, new_data_size 0x%llx, "
1938	"data_start 0x%llx.", vi->i_ino,
1939	(unsigned)le32_to_cpu(ni->type),
1940	(unsigned long long)allocated_size,
1941	(unsigned long long)new_alloc_size,
1942	(unsigned long long)new_data_size,
1943	(unsigned long long)start);
1944	#endif
1945	retry_extend:
1946	/*
1947	* For non-resident attributes, @start and @new_size need to be aligned
1948	* to cluster boundaries for allocation purposes.
1949	*/
1950	if (NInoNonResident(ni)) {
1951	if (start > 0)
1952	start &= ~(s64)vol->cluster_size_mask;
1953	new_alloc_size = (new_alloc_size + vol->cluster_size - 1) &
1954	~(s64)vol->cluster_size_mask;
1955	}
1956	BUG_ON(new_data_size >= 0 && new_data_size > new_alloc_size);
1957	/* Check if new size is allowed in $AttrDef. */
1958	err = ntfs_attr_size_bounds_check(vol, type: ni->type, size: new_alloc_size);
1959	if (unlikely(err)) {
1960	/* Only emit errors when the write will fail completely. */
1961	read_lock_irqsave(&ni->size_lock, flags);
1962	allocated_size = ni->allocated_size;
1963	read_unlock_irqrestore(&ni->size_lock, flags);
1964	if (start < 0 || start >= allocated_size) {
1965	if (err == -ERANGE) {
1966	ntfs_error(vol->sb, "Cannot extend allocation "
1967	"of inode 0x%lx, attribute "
1968	"type 0x%x, because the new "
1969	"allocation would exceed the "
1970	"maximum allowed size for "
1971	"this attribute type.",
1972	vi->i_ino, (unsigned)
1973	le32_to_cpu(ni->type));
1974	} else {
1975	ntfs_error(vol->sb, "Cannot extend allocation "
1976	"of inode 0x%lx, attribute "
1977	"type 0x%x, because this "
1978	"attribute type is not "
1979	"defined on the NTFS volume. "
1980	"Possible corruption! You "
1981	"should run chkdsk!",
1982	vi->i_ino, (unsigned)
1983	le32_to_cpu(ni->type));
1984	}
1985	}
1986	/* Translate error code to be POSIX conformant for write(2). */
1987	if (err == -ERANGE)
1988	err = -EFBIG;
1989	else
1990	err = -EIO;
1991	return err;
1992	}
1993	if (!NInoAttr(ni))
1994	base_ni = ni;
1995	else
1996	base_ni = ni->ext.base_ntfs_ino;
1997	/*
1998	* We will be modifying both the runlist (if non-resident) and the mft
1999	* record so lock them both down.
2000	*/
2001	down_write(sem: &ni->runlist.lock);
2002	m = map_mft_record(ni: base_ni);
2003	if (IS_ERR(ptr: m)) {
2004	err = PTR_ERR(ptr: m);
2005	m = NULL;
2006	ctx = NULL;
2007	goto err_out;
2008	}
2009	ctx = ntfs_attr_get_search_ctx(ni: base_ni, mrec: m);
2010	if (unlikely(!ctx)) {
2011	err = -ENOMEM;
2012	goto err_out;
2013	}
2014	read_lock_irqsave(&ni->size_lock, flags);
2015	allocated_size = ni->allocated_size;
2016	read_unlock_irqrestore(&ni->size_lock, flags);
2017	/*
2018	* If non-resident, seek to the last extent. If resident, there is
2019	* only one extent, so seek to that.
2020	*/
2021	vcn = NInoNonResident(ni) ? allocated_size >> vol->cluster_size_bits :
2022	0;
2023	/*
2024	* Abort if someone did the work whilst we waited for the locks. If we
2025	* just converted the attribute from resident to non-resident it is
2026	* likely that exactly this has happened already. We cannot quite
2027	* abort if we need to update the data size.
2028	*/
2029	if (unlikely(new_alloc_size <= allocated_size)) {
2030	ntfs_debug("Allocated size already exceeds requested size.");
2031	new_alloc_size = allocated_size;
2032	if (new_data_size < 0)
2033	goto done;
2034	/*
2035	* We want the first attribute extent so that we can update the
2036	* data size.
2037	*/
2038	vcn = 0;
2039	}
2040	err = ntfs_attr_lookup(type: ni->type, name: ni->name, name_len: ni->name_len,
2041	ic: CASE_SENSITIVE, lowest_vcn: vcn, NULL, val_len: 0, ctx);
2042	if (unlikely(err)) {
2043	if (err == -ENOENT)
2044	err = -EIO;
2045	goto err_out;
2046	}
2047	m = ctx->mrec;
2048	a = ctx->attr;
2049	/* Use goto to reduce indentation. */
2050	if (a->non_resident)
2051	goto do_non_resident_extend;
2052	BUG_ON(NInoNonResident(ni));
2053	/* The total length of the attribute value. */
2054	attr_len = le32_to_cpu(a->data.resident.value_length);
2055	/*
2056	* Extend the attribute record to be able to store the new attribute
2057	* size. ntfs_attr_record_resize() will not do anything if the size is
2058	* not changing.
2059	*/
2060	if (new_alloc_size < vol->mft_record_size &&
2061	!ntfs_attr_record_resize(m, a,
2062	le16_to_cpu(a->data.resident.value_offset) +
2063	new_alloc_size)) {
2064	/* The resize succeeded! */
2065	write_lock_irqsave(&ni->size_lock, flags);
2066	ni->allocated_size = le32_to_cpu(a->length) -
2067	le16_to_cpu(a->data.resident.value_offset);
2068	write_unlock_irqrestore(&ni->size_lock, flags);
2069	if (new_data_size >= 0) {
2070	BUG_ON(new_data_size < attr_len);
2071	a->data.resident.value_length =
2072	cpu_to_le32((u32)new_data_size);
2073	}
2074	goto flush_done;
2075	}
2076	/*
2077	* We have to drop all the locks so we can call
2078	* ntfs_attr_make_non_resident(). This could be optimised by try-
2079	* locking the first page cache page and only if that fails dropping
2080	* the locks, locking the page, and redoing all the locking and
2081	* lookups. While this would be a huge optimisation, it is not worth
2082	* it as this is definitely a slow code path.
2083	*/
2084	ntfs_attr_put_search_ctx(ctx);
2085	unmap_mft_record(ni: base_ni);
2086	up_write(sem: &ni->runlist.lock);
2087	/*
2088	* Not enough space in the mft record, try to make the attribute
2089	* non-resident and if successful restart the extension process.
2090	*/
2091	err = ntfs_attr_make_non_resident(ni, data_size: attr_len);
2092	if (likely(!err))
2093	goto retry_extend;
2094	/*
2095	* Could not make non-resident. If this is due to this not being
2096	* permitted for this attribute type or there not being enough space,
2097	* try to make other attributes non-resident. Otherwise fail.
2098	*/
2099	if (unlikely(err != -EPERM && err != -ENOSPC)) {
2100	/* Only emit errors when the write will fail completely. */
2101	read_lock_irqsave(&ni->size_lock, flags);
2102	allocated_size = ni->allocated_size;
2103	read_unlock_irqrestore(&ni->size_lock, flags);
2104	if (start < 0 || start >= allocated_size)
2105	ntfs_error(vol->sb, "Cannot extend allocation of "
2106	"inode 0x%lx, attribute type 0x%x, "
2107	"because the conversion from resident "
2108	"to non-resident attribute failed "
2109	"with error code %i.", vi->i_ino,
2110	(unsigned)le32_to_cpu(ni->type), err);
2111	if (err != -ENOMEM)
2112	err = -EIO;
2113	goto conv_err_out;
2114	}
2115	/* TODO: Not implemented from here, abort. */
2116	read_lock_irqsave(&ni->size_lock, flags);
2117	allocated_size = ni->allocated_size;
2118	read_unlock_irqrestore(&ni->size_lock, flags);
2119	if (start < 0 || start >= allocated_size) {
2120	if (err == -ENOSPC)
2121	ntfs_error(vol->sb, "Not enough space in the mft "
2122	"record/on disk for the non-resident "
2123	"attribute value. This case is not "
2124	"implemented yet.");
2125	else /* if (err == -EPERM) */
2126	ntfs_error(vol->sb, "This attribute type may not be "
2127	"non-resident. This case is not "
2128	"implemented yet.");
2129	}
2130	err = -EOPNOTSUPP;
2131	goto conv_err_out;
2132	#if 0
2133	// TODO: Attempt to make other attributes non-resident.
2134	if (!err)
2135	goto do_resident_extend;
2136	/*
2137	* Both the attribute list attribute and the standard information
2138	* attribute must remain in the base inode. Thus, if this is one of
2139	* these attributes, we have to try to move other attributes out into
2140	* extent mft records instead.
2141	*/
2142	if (ni->type == AT_ATTRIBUTE_LIST ||
2143	ni->type == AT_STANDARD_INFORMATION) {
2144	// TODO: Attempt to move other attributes into extent mft
2145	// records.
2146	err = -EOPNOTSUPP;
2147	if (!err)
2148	goto do_resident_extend;
2149	goto err_out;
2150	}
2151	// TODO: Attempt to move this attribute to an extent mft record, but
2152	// only if it is not already the only attribute in an mft record in
2153	// which case there would be nothing to gain.
2154	err = -EOPNOTSUPP;
2155	if (!err)
2156	goto do_resident_extend;
2157	/* There is nothing we can do to make enough space. )-: */
2158	goto err_out;
2159	#endif
2160	do_non_resident_extend:
2161	BUG_ON(!NInoNonResident(ni));
2162	if (new_alloc_size == allocated_size) {
2163	BUG_ON(vcn);
2164	goto alloc_done;
2165	}
2166	/*
2167	* If the data starts after the end of the old allocation, this is a
2168	* $DATA attribute and sparse attributes are enabled on the volume and
2169	* for this inode, then create a sparse region between the old
2170	* allocated size and the start of the data. Otherwise simply proceed
2171	* with filling the whole space between the old allocated size and the
2172	* new allocated size with clusters.
2173	*/
2174	if ((start >= 0 && start <= allocated_size) || ni->type != AT_DATA ||
2175	!NVolSparseEnabled(vol) || NInoSparseDisabled(ni))
2176	goto skip_sparse;
2177	// TODO: This is not implemented yet. We just fill in with real
2178	// clusters for now...
2179	ntfs_debug("Inserting holes is not-implemented yet. Falling back to "
2180	"allocating real clusters instead.");
2181	skip_sparse:
2182	rl = ni->runlist.rl;
2183	if (likely(rl)) {
2184	/* Seek to the end of the runlist. */
2185	while (rl->length)
2186	rl++;
2187	}
2188	/* If this attribute extent is not mapped, map it now. */
2189	if (unlikely(!rl || rl->lcn == LCN_RL_NOT_MAPPED ||
2190	(rl->lcn == LCN_ENOENT && rl > ni->runlist.rl &&
2191	(rl-1)->lcn == LCN_RL_NOT_MAPPED))) {
2192	if (!rl && !allocated_size)
2193	goto first_alloc;
2194	rl = ntfs_mapping_pairs_decompress(vol, attr: a, old_rl: ni->runlist.rl);
2195	if (IS_ERR(ptr: rl)) {
2196	err = PTR_ERR(ptr: rl);
2197	if (start < 0 || start >= allocated_size)
2198	ntfs_error(vol->sb, "Cannot extend allocation "
2199	"of inode 0x%lx, attribute "
2200	"type 0x%x, because the "
2201	"mapping of a runlist "
2202	"fragment failed with error "
2203	"code %i.", vi->i_ino,
2204	(unsigned)le32_to_cpu(ni->type),
2205	err);
2206	if (err != -ENOMEM)
2207	err = -EIO;
2208	goto err_out;
2209	}
2210	ni->runlist.rl = rl;
2211	/* Seek to the end of the runlist. */
2212	while (rl->length)
2213	rl++;
2214	}
2215	/*
2216	* We now know the runlist of the last extent is mapped and @rl is at
2217	* the end of the runlist. We want to begin allocating clusters
2218	* starting at the last allocated cluster to reduce fragmentation. If
2219	* there are no valid LCNs in the attribute we let the cluster
2220	* allocator choose the starting cluster.
2221	*/
2222	/* If the last LCN is a hole or simillar seek back to last real LCN. */
2223	while (rl->lcn < 0 && rl > ni->runlist.rl)
2224	rl--;
2225	first_alloc:
2226	// FIXME: Need to implement partial allocations so at least part of the
2227	// write can be performed when start >= 0. (Needed for POSIX write(2)
2228	// conformance.)
2229	rl2 = ntfs_cluster_alloc(vol, start_vcn: allocated_size >> vol->cluster_size_bits,
2230	count: (new_alloc_size - allocated_size) >>
2231	vol->cluster_size_bits, start_lcn: (rl && (rl->lcn >= 0)) ?
2232	rl->lcn + rl->length : -1, zone: DATA_ZONE, is_extension: true);
2233	if (IS_ERR(ptr: rl2)) {
2234	err = PTR_ERR(ptr: rl2);
2235	if (start < 0 || start >= allocated_size)
2236	ntfs_error(vol->sb, "Cannot extend allocation of "
2237	"inode 0x%lx, attribute type 0x%x, "
2238	"because the allocation of clusters "
2239	"failed with error code %i.", vi->i_ino,
2240	(unsigned)le32_to_cpu(ni->type), err);
2241	if (err != -ENOMEM && err != -ENOSPC)
2242	err = -EIO;
2243	goto err_out;
2244	}
2245	rl = ntfs_runlists_merge(drl: ni->runlist.rl, srl: rl2);
2246	if (IS_ERR(ptr: rl)) {
2247	err = PTR_ERR(ptr: rl);
2248	if (start < 0 || start >= allocated_size)
2249	ntfs_error(vol->sb, "Cannot extend allocation of "
2250	"inode 0x%lx, attribute type 0x%x, "
2251	"because the runlist merge failed "
2252	"with error code %i.", vi->i_ino,
2253	(unsigned)le32_to_cpu(ni->type), err);
2254	if (err != -ENOMEM)
2255	err = -EIO;
2256	if (ntfs_cluster_free_from_rl(vol, rl: rl2)) {
2257	ntfs_error(vol->sb, "Failed to release allocated "
2258	"cluster(s) in error code path. Run "
2259	"chkdsk to recover the lost "
2260	"cluster(s).");
2261	NVolSetErrors(vol);
2262	}
2263	ntfs_free(addr: rl2);
2264	goto err_out;
2265	}
2266	ni->runlist.rl = rl;
2267	ntfs_debug("Allocated 0x%llx clusters.", (long long)(new_alloc_size -
2268	allocated_size) >> vol->cluster_size_bits);
2269	/* Find the runlist element with which the attribute extent starts. */
2270	ll = sle64_to_cpu(x: a->data.non_resident.lowest_vcn);
2271	rl2 = ntfs_rl_find_vcn_nolock(rl, vcn: ll);
2272	BUG_ON(!rl2);
2273	BUG_ON(!rl2->length);
2274	BUG_ON(rl2->lcn < LCN_HOLE);
2275	mp_rebuilt = false;
2276	/* Get the size for the new mapping pairs array for this extent. */
2277	mp_size = ntfs_get_size_for_mapping_pairs(vol, rl: rl2, first_vcn: ll, last_vcn: -1);
2278	if (unlikely(mp_size <= 0)) {
2279	err = mp_size;
2280	if (start < 0 || start >= allocated_size)
2281	ntfs_error(vol->sb, "Cannot extend allocation of "
2282	"inode 0x%lx, attribute type 0x%x, "
2283	"because determining the size for the "
2284	"mapping pairs failed with error code "
2285	"%i.", vi->i_ino,
2286	(unsigned)le32_to_cpu(ni->type), err);
2287	err = -EIO;
2288	goto undo_alloc;
2289	}
2290	/* Extend the attribute record to fit the bigger mapping pairs array. */
2291	attr_len = le32_to_cpu(a->length);
2292	err = ntfs_attr_record_resize(m, a, new_size: mp_size +
2293	le16_to_cpu(a->data.non_resident.mapping_pairs_offset));
2294	if (unlikely(err)) {
2295	BUG_ON(err != -ENOSPC);
2296	// TODO: Deal with this by moving this extent to a new mft
2297	// record or by starting a new extent in a new mft record,
2298	// possibly by extending this extent partially and filling it
2299	// and creating a new extent for the remainder, or by making
2300	// other attributes non-resident and/or by moving other
2301	// attributes out of this mft record.
2302	if (start < 0 || start >= allocated_size)
2303	ntfs_error(vol->sb, "Not enough space in the mft "
2304	"record for the extended attribute "
2305	"record. This case is not "
2306	"implemented yet.");
2307	err = -EOPNOTSUPP;
2308	goto undo_alloc;
2309	}
2310	mp_rebuilt = true;
2311	/* Generate the mapping pairs array directly into the attr record. */
2312	err = ntfs_mapping_pairs_build(vol, dst: (u8*)a +
2313	le16_to_cpu(a->data.non_resident.mapping_pairs_offset),
2314	dst_len: mp_size, rl: rl2, first_vcn: ll, last_vcn: -1, NULL);
2315	if (unlikely(err)) {
2316	if (start < 0 || start >= allocated_size)
2317	ntfs_error(vol->sb, "Cannot extend allocation of "
2318	"inode 0x%lx, attribute type 0x%x, "
2319	"because building the mapping pairs "
2320	"failed with error code %i.", vi->i_ino,
2321	(unsigned)le32_to_cpu(ni->type), err);
2322	err = -EIO;
2323	goto undo_alloc;
2324	}
2325	/* Update the highest_vcn. */
2326	a->data.non_resident.highest_vcn = cpu_to_sle64(x: (new_alloc_size >>
2327	vol->cluster_size_bits) - 1);
2328	/*
2329	* We now have extended the allocated size of the attribute. Reflect
2330	* this in the ntfs_inode structure and the attribute record.
2331	*/
2332	if (a->data.non_resident.lowest_vcn) {
2333	/*
2334	* We are not in the first attribute extent, switch to it, but
2335	* first ensure the changes will make it to disk later.
2336	*/
2337	flush_dcache_mft_record_page(ni: ctx->ntfs_ino);
2338	mark_mft_record_dirty(ni: ctx->ntfs_ino);
2339	ntfs_attr_reinit_search_ctx(ctx);
2340	err = ntfs_attr_lookup(type: ni->type, name: ni->name, name_len: ni->name_len,
2341	ic: CASE_SENSITIVE, lowest_vcn: 0, NULL, val_len: 0, ctx);
2342	if (unlikely(err))
2343	goto restore_undo_alloc;
2344	/* @m is not used any more so no need to set it. */
2345	a = ctx->attr;
2346	}
2347	write_lock_irqsave(&ni->size_lock, flags);
2348	ni->allocated_size = new_alloc_size;
2349	a->data.non_resident.allocated_size = cpu_to_sle64(x: new_alloc_size);
2350	/*
2351	* FIXME: This would fail if @ni is a directory, $MFT, or an index,
2352	* since those can have sparse/compressed set. For example can be
2353	* set compressed even though it is not compressed itself and in that
2354	* case the bit means that files are to be created compressed in the
2355	* directory... At present this is ok as this code is only called for
2356	* regular files, and only for their $DATA attribute(s).
2357	* FIXME: The calculation is wrong if we created a hole above. For now
2358	* it does not matter as we never create holes.
2359	*/
2360	if (NInoSparse(ni) || NInoCompressed(ni)) {
2361	ni->itype.compressed.size += new_alloc_size - allocated_size;
2362	a->data.non_resident.compressed_size =
2363	cpu_to_sle64(x: ni->itype.compressed.size);
2364	vi->i_blocks = ni->itype.compressed.size >> 9;
2365	} else
2366	vi->i_blocks = new_alloc_size >> 9;
2367	write_unlock_irqrestore(&ni->size_lock, flags);
2368	alloc_done:
2369	if (new_data_size >= 0) {
2370	BUG_ON(new_data_size <
2371	sle64_to_cpu(a->data.non_resident.data_size));
2372	a->data.non_resident.data_size = cpu_to_sle64(x: new_data_size);
2373	}
2374	flush_done:
2375	/* Ensure the changes make it to disk. */
2376	flush_dcache_mft_record_page(ni: ctx->ntfs_ino);
2377	mark_mft_record_dirty(ni: ctx->ntfs_ino);
2378	done:
2379	ntfs_attr_put_search_ctx(ctx);
2380	unmap_mft_record(ni: base_ni);
2381	up_write(sem: &ni->runlist.lock);
2382	ntfs_debug("Done, new_allocated_size 0x%llx.",
2383	(unsigned long long)new_alloc_size);
2384	return new_alloc_size;
2385	restore_undo_alloc:
2386	if (start < 0 || start >= allocated_size)
2387	ntfs_error(vol->sb, "Cannot complete extension of allocation "
2388	"of inode 0x%lx, attribute type 0x%x, because "
2389	"lookup of first attribute extent failed with "
2390	"error code %i.", vi->i_ino,
2391	(unsigned)le32_to_cpu(ni->type), err);
2392	if (err == -ENOENT)
2393	err = -EIO;
2394	ntfs_attr_reinit_search_ctx(ctx);
2395	if (ntfs_attr_lookup(type: ni->type, name: ni->name, name_len: ni->name_len, ic: CASE_SENSITIVE,
2396	lowest_vcn: allocated_size >> vol->cluster_size_bits, NULL, val_len: 0,
2397	ctx)) {
2398	ntfs_error(vol->sb, "Failed to find last attribute extent of "
2399	"attribute in error code path. Run chkdsk to "
2400	"recover.");
2401	write_lock_irqsave(&ni->size_lock, flags);
2402	ni->allocated_size = new_alloc_size;
2403	/*
2404	* FIXME: This would fail if @ni is a directory... See above.
2405	* FIXME: The calculation is wrong if we created a hole above.
2406	* For now it does not matter as we never create holes.
2407	*/
2408	if (NInoSparse(ni) || NInoCompressed(ni)) {
2409	ni->itype.compressed.size += new_alloc_size -
2410	allocated_size;
2411	vi->i_blocks = ni->itype.compressed.size >> 9;
2412	} else
2413	vi->i_blocks = new_alloc_size >> 9;
2414	write_unlock_irqrestore(&ni->size_lock, flags);
2415	ntfs_attr_put_search_ctx(ctx);
2416	unmap_mft_record(ni: base_ni);
2417	up_write(sem: &ni->runlist.lock);
2418	/*
2419	* The only thing that is now wrong is the allocated size of the
2420	* base attribute extent which chkdsk should be able to fix.
2421	*/
2422	NVolSetErrors(vol);
2423	return err;
2424	}
2425	ctx->attr->data.non_resident.highest_vcn = cpu_to_sle64(
2426	x: (allocated_size >> vol->cluster_size_bits) - 1);
2427	undo_alloc:
2428	ll = allocated_size >> vol->cluster_size_bits;
2429	if (ntfs_cluster_free(ni, start_vcn: ll, count: -1, ctx) < 0) {
2430	ntfs_error(vol->sb, "Failed to release allocated cluster(s) "
2431	"in error code path. Run chkdsk to recover "
2432	"the lost cluster(s).");
2433	NVolSetErrors(vol);
2434	}
2435	m = ctx->mrec;
2436	a = ctx->attr;
2437	/*
2438	* If the runlist truncation fails and/or the search context is no
2439	* longer valid, we cannot resize the attribute record or build the
2440	* mapping pairs array thus we mark the inode bad so that no access to
2441	* the freed clusters can happen.
2442	*/
2443	if (ntfs_rl_truncate_nolock(vol, runlist: &ni->runlist, new_length: ll) || IS_ERR(ptr: m)) {
2444	ntfs_error(vol->sb, "Failed to %s in error code path. Run "
2445	"chkdsk to recover.", IS_ERR(m) ?
2446	"restore attribute search context" :
2447	"truncate attribute runlist");
2448	NVolSetErrors(vol);
2449	} else if (mp_rebuilt) {
2450	if (ntfs_attr_record_resize(m, a, new_size: attr_len)) {
2451	ntfs_error(vol->sb, "Failed to restore attribute "
2452	"record in error code path. Run "
2453	"chkdsk to recover.");
2454	NVolSetErrors(vol);
2455	} else /* if (success) */ {
2456	if (ntfs_mapping_pairs_build(vol, dst: (u8*)a + le16_to_cpu(
2457	a->data.non_resident.
2458	mapping_pairs_offset), dst_len: attr_len -
2459	le16_to_cpu(a->data.non_resident.
2460	mapping_pairs_offset), rl: rl2, first_vcn: ll, last_vcn: -1,
2461	NULL)) {
2462	ntfs_error(vol->sb, "Failed to restore "
2463	"mapping pairs array in error "
2464	"code path. Run chkdsk to "
2465	"recover.");
2466	NVolSetErrors(vol);
2467	}
2468	flush_dcache_mft_record_page(ni: ctx->ntfs_ino);
2469	mark_mft_record_dirty(ni: ctx->ntfs_ino);
2470	}
2471	}
2472	err_out:
2473	if (ctx)
2474	ntfs_attr_put_search_ctx(ctx);
2475	if (m)
2476	unmap_mft_record(ni: base_ni);
2477	up_write(sem: &ni->runlist.lock);
2478	conv_err_out:
2479	ntfs_debug("Failed. Returning error code %i.", err);
2480	return err;
2481	}
2482
2483	/**
2484	* ntfs_attr_set - fill (a part of) an attribute with a byte
2485	* @ni: ntfs inode describing the attribute to fill
2486	* @ofs: offset inside the attribute at which to start to fill
2487	* @cnt: number of bytes to fill
2488	* @val: the unsigned 8-bit value with which to fill the attribute
2489	*
2490	* Fill @cnt bytes of the attribute described by the ntfs inode @ni starting at
2491	* byte offset @ofs inside the attribute with the constant byte @val.
2492	*
2493	* This function is effectively like memset() applied to an ntfs attribute.
2494	* Note this function actually only operates on the page cache pages belonging
2495	* to the ntfs attribute and it marks them dirty after doing the memset().
2496	* Thus it relies on the vm dirty page write code paths to cause the modified
2497	* pages to be written to the mft record/disk.
2498	*
2499	* Return 0 on success and -errno on error. An error code of -ESPIPE means
2500	* that @ofs + @cnt were outside the end of the attribute and no write was
2501	* performed.
2502	*/
2503	int ntfs_attr_set(ntfs_inode *ni, const s64 ofs, const s64 cnt, const u8 val)
2504	{
2505	ntfs_volume *vol = ni->vol;
2506	struct address_space *mapping;
2507	struct page *page;
2508	u8 *kaddr;
2509	pgoff_t idx, end;
2510	unsigned start_ofs, end_ofs, size;
2511
2512	ntfs_debug("Entering for ofs 0x%llx, cnt 0x%llx, val 0x%hx.",
2513	(long long)ofs, (long long)cnt, val);
2514	BUG_ON(ofs < 0);
2515	BUG_ON(cnt < 0);
2516	if (!cnt)
2517	goto done;
2518	/*
2519	* FIXME: Compressed and encrypted attributes are not supported when
2520	* writing and we should never have gotten here for them.
2521	*/
2522	BUG_ON(NInoCompressed(ni));
2523	BUG_ON(NInoEncrypted(ni));
2524	mapping = VFS_I(ni)->i_mapping;
2525	/* Work out the starting index and page offset. */
2526	idx = ofs >> PAGE_SHIFT;
2527	start_ofs = ofs & ~PAGE_MASK;
2528	/* Work out the ending index and page offset. */
2529	end = ofs + cnt;
2530	end_ofs = end & ~PAGE_MASK;
2531	/* If the end is outside the inode size return -ESPIPE. */
2532	if (unlikely(end > i_size_read(VFS_I(ni)))) {
2533	ntfs_error(vol->sb, "Request exceeds end of attribute.");
2534	return -ESPIPE;
2535	}
2536	end >>= PAGE_SHIFT;
2537	/* If there is a first partial page, need to do it the slow way. */
2538	if (start_ofs) {
2539	page = read_mapping_page(mapping, index: idx, NULL);
2540	if (IS_ERR(ptr: page)) {
2541	ntfs_error(vol->sb, "Failed to read first partial "
2542	"page (error, index 0x%lx).", idx);
2543	return PTR_ERR(ptr: page);
2544	}
2545	/*
2546	* If the last page is the same as the first page, need to
2547	* limit the write to the end offset.
2548	*/
2549	size = PAGE_SIZE;
2550	if (idx == end)
2551	size = end_ofs;
2552	kaddr = kmap_atomic(page);
2553	memset(kaddr + start_ofs, val, size - start_ofs);
2554	flush_dcache_page(page);
2555	kunmap_atomic(kaddr);
2556	set_page_dirty(page);
2557	put_page(page);
2558	balance_dirty_pages_ratelimited(mapping);
2559	cond_resched();
2560	if (idx == end)
2561	goto done;
2562	idx++;
2563	}
2564	/* Do the whole pages the fast way. */
2565	for (; idx < end; idx++) {
2566	/* Find or create the current page. (The page is locked.) */
2567	page = grab_cache_page(mapping, index: idx);
2568	if (unlikely(!page)) {
2569	ntfs_error(vol->sb, "Insufficient memory to grab "
2570	"page (index 0x%lx).", idx);
2571	return -ENOMEM;
2572	}
2573	kaddr = kmap_atomic(page);
2574	memset(kaddr, val, PAGE_SIZE);
2575	flush_dcache_page(page);
2576	kunmap_atomic(kaddr);
2577	/*
2578	* If the page has buffers, mark them uptodate since buffer
2579	* state and not page state is definitive in 2.6 kernels.
2580	*/
2581	if (page_has_buffers(page)) {
2582	struct buffer_head *bh, *head;
2583
2584	bh = head = page_buffers(page);
2585	do {
2586	set_buffer_uptodate(bh);
2587	} while ((bh = bh->b_this_page) != head);
2588	}
2589	/* Now that buffers are uptodate, set the page uptodate, too. */
2590	SetPageUptodate(page);
2591	/*
2592	* Set the page and all its buffers dirty and mark the inode
2593	* dirty, too. The VM will write the page later on.
2594	*/
2595	set_page_dirty(page);
2596	/* Finally unlock and release the page. */
2597	unlock_page(page);
2598	put_page(page);
2599	balance_dirty_pages_ratelimited(mapping);
2600	cond_resched();
2601	}
2602	/* If there is a last partial page, need to do it the slow way. */
2603	if (end_ofs) {
2604	page = read_mapping_page(mapping, index: idx, NULL);
2605	if (IS_ERR(ptr: page)) {
2606	ntfs_error(vol->sb, "Failed to read last partial page "
2607	"(error, index 0x%lx).", idx);
2608	return PTR_ERR(ptr: page);
2609	}
2610	kaddr = kmap_atomic(page);
2611	memset(kaddr, val, end_ofs);
2612	flush_dcache_page(page);
2613	kunmap_atomic(kaddr);
2614	set_page_dirty(page);
2615	put_page(page);
2616	balance_dirty_pages_ratelimited(mapping);
2617	cond_resched();
2618	}
2619	done:
2620	ntfs_debug("Done.");
2621	return 0;
2622	}
2623
2624	#endif /* NTFS_RW */
2625