1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	*
4	* Copyright (C) 2019-2021 Paragon Software GmbH, All rights reserved.
5	*
6	*/
7
8	#include <linux/fiemap.h>
9	#include <linux/fs.h>
10	#include <linux/minmax.h>
11	#include <linux/vmalloc.h>
12
13	#include "debug.h"
14	#include "ntfs.h"
15	#include "ntfs_fs.h"
16	#ifdef CONFIG_NTFS3_LZX_XPRESS
17	#include "lib/lib.h"
18	#endif
19
20	static struct mft_inode *ni_ins_mi(struct ntfs_inode *ni, struct rb_root *tree,
21	CLST ino, struct rb_node *ins)
22	{
23	struct rb_node **p = &tree->rb_node;
24	struct rb_node *pr = NULL;
25
26	while (*p) {
27	struct mft_inode *mi;
28
29	pr = *p;
30	mi = rb_entry(pr, struct mft_inode, node);
31	if (mi->rno > ino)
32	p = &pr->rb_left;
33	else if (mi->rno < ino)
34	p = &pr->rb_right;
35	else
36	return mi;
37	}
38
39	if (!ins)
40	return NULL;
41
42	rb_link_node(node: ins, parent: pr, rb_link: p);
43	rb_insert_color(ins, tree);
44	return rb_entry(ins, struct mft_inode, node);
45	}
46
47	/*
48	* ni_find_mi - Find mft_inode by record number.
49	*/
50	static struct mft_inode *ni_find_mi(struct ntfs_inode *ni, CLST rno)
51	{
52	return ni_ins_mi(ni, tree: &ni->mi_tree, ino: rno, NULL);
53	}
54
55	/*
56	* ni_add_mi - Add new mft_inode into ntfs_inode.
57	*/
58	static void ni_add_mi(struct ntfs_inode *ni, struct mft_inode *mi)
59	{
60	ni_ins_mi(ni, tree: &ni->mi_tree, ino: mi->rno, ins: &mi->node);
61	}
62
63	/*
64	* ni_remove_mi - Remove mft_inode from ntfs_inode.
65	*/
66	void ni_remove_mi(struct ntfs_inode *ni, struct mft_inode *mi)
67	{
68	rb_erase(&mi->node, &ni->mi_tree);
69	}
70
71	/*
72	* ni_std - Return: Pointer into std_info from primary record.
73	*/
74	struct ATTR_STD_INFO *ni_std(struct ntfs_inode *ni)
75	{
76	const struct ATTRIB *attr;
77
78	attr = mi_find_attr(mi: &ni->mi, NULL, type: ATTR_STD, NULL, name_len: 0, NULL);
79	return attr ? resident_data_ex(attr, datasize: sizeof(struct ATTR_STD_INFO)) :
80	NULL;
81	}
82
83	/*
84	* ni_std5
85	*
86	* Return: Pointer into std_info from primary record.
87	*/
88	struct ATTR_STD_INFO5 *ni_std5(struct ntfs_inode *ni)
89	{
90	const struct ATTRIB *attr;
91
92	attr = mi_find_attr(mi: &ni->mi, NULL, type: ATTR_STD, NULL, name_len: 0, NULL);
93
94	return attr ? resident_data_ex(attr, datasize: sizeof(struct ATTR_STD_INFO5)) :
95	NULL;
96	}
97
98	/*
99	* ni_clear - Clear resources allocated by ntfs_inode.
100	*/
101	void ni_clear(struct ntfs_inode *ni)
102	{
103	struct rb_node *node;
104
105	if (!ni->vfs_inode.i_nlink && ni->mi.mrec && is_rec_inuse(rec: ni->mi.mrec))
106	ni_delete_all(ni);
107
108	al_destroy(ni);
109
110	for (node = rb_first(&ni->mi_tree); node;) {
111	struct rb_node *next = rb_next(node);
112	struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
113
114	rb_erase(node, &ni->mi_tree);
115	mi_put(mi);
116	node = next;
117	}
118
119	/* Bad inode always has mode == S_IFREG. */
120	if (ni->ni_flags & NI_FLAG_DIR)
121	indx_clear(idx: &ni->dir);
122	else {
123	run_close(run: &ni->file.run);
124	#ifdef CONFIG_NTFS3_LZX_XPRESS
125	if (ni->file.offs_page) {
126	/* On-demand allocated page for offsets. */
127	put_page(page: ni->file.offs_page);
128	ni->file.offs_page = NULL;
129	}
130	#endif
131	}
132
133	mi_clear(mi: &ni->mi);
134	}
135
136	/*
137	* ni_load_mi_ex - Find mft_inode by record number.
138	*/
139	int ni_load_mi_ex(struct ntfs_inode *ni, CLST rno, struct mft_inode **mi)
140	{
141	int err;
142	struct mft_inode *r;
143
144	r = ni_find_mi(ni, rno);
145	if (r)
146	goto out;
147
148	err = mi_get(sbi: ni->mi.sbi, rno, mi: &r);
149	if (err)
150	return err;
151
152	ni_add_mi(ni, mi: r);
153
154	out:
155	if (mi)
156	*mi = r;
157	return 0;
158	}
159
160	/*
161	* ni_load_mi - Load mft_inode corresponded list_entry.
162	*/
163	int ni_load_mi(struct ntfs_inode *ni, const struct ATTR_LIST_ENTRY *le,
164	struct mft_inode **mi)
165	{
166	CLST rno;
167
168	if (!le) {
169	*mi = &ni->mi;
170	return 0;
171	}
172
173	rno = ino_get(ref: &le->ref);
174	if (rno == ni->mi.rno) {
175	*mi = &ni->mi;
176	return 0;
177	}
178	return ni_load_mi_ex(ni, rno, mi);
179	}
180
181	/*
182	* ni_find_attr
183	*
184	* Return: Attribute and record this attribute belongs to.
185	*/
186	struct ATTRIB *ni_find_attr(struct ntfs_inode *ni, struct ATTRIB *attr,
187	struct ATTR_LIST_ENTRY **le_o, enum ATTR_TYPE type,
188	const __le16 *name, u8 name_len, const CLST *vcn,
189	struct mft_inode **mi)
190	{
191	struct ATTR_LIST_ENTRY *le;
192	struct mft_inode *m;
193
194	if (!ni->attr_list.size ||
195	(!name_len && (type == ATTR_LIST || type == ATTR_STD))) {
196	if (le_o)
197	*le_o = NULL;
198	if (mi)
199	*mi = &ni->mi;
200
201	/* Look for required attribute in primary record. */
202	return mi_find_attr(mi: &ni->mi, attr, type, name, name_len, NULL);
203	}
204
205	/* First look for list entry of required type. */
206	le = al_find_ex(ni, le: le_o ? *le_o : NULL, type, name, name_len, vcn);
207	if (!le)
208	return NULL;
209
210	if (le_o)
211	*le_o = le;
212
213	/* Load record that contains this attribute. */
214	if (ni_load_mi(ni, le, mi: &m))
215	return NULL;
216
217	/* Look for required attribute. */
218	attr = mi_find_attr(mi: m, NULL, type, name, name_len, id: &le->id);
219
220	if (!attr)
221	goto out;
222
223	if (!attr->non_res) {
224	if (vcn && *vcn)
225	goto out;
226	} else if (!vcn) {
227	if (attr->nres.svcn)
228	goto out;
229	} else if (le64_to_cpu(attr->nres.svcn) > *vcn ||
230	*vcn > le64_to_cpu(attr->nres.evcn)) {
231	goto out;
232	}
233
234	if (mi)
235	*mi = m;
236	return attr;
237
238	out:
239	ntfs_inode_err(&ni->vfs_inode, "failed to parse mft record");
240	ntfs_set_state(sbi: ni->mi.sbi, dirty: NTFS_DIRTY_ERROR);
241	return NULL;
242	}
243
244	/*
245	* ni_enum_attr_ex - Enumerates attributes in ntfs_inode.
246	*/
247	struct ATTRIB *ni_enum_attr_ex(struct ntfs_inode *ni, struct ATTRIB *attr,
248	struct ATTR_LIST_ENTRY **le,
249	struct mft_inode **mi)
250	{
251	struct mft_inode *mi2;
252	struct ATTR_LIST_ENTRY *le2;
253
254	/* Do we have an attribute list? */
255	if (!ni->attr_list.size) {
256	*le = NULL;
257	if (mi)
258	*mi = &ni->mi;
259	/* Enum attributes in primary record. */
260	return mi_enum_attr(mi: &ni->mi, attr);
261	}
262
263	/* Get next list entry. */
264	le2 = *le = al_enumerate(ni, le: attr ? *le : NULL);
265	if (!le2)
266	return NULL;
267
268	/* Load record that contains the required attribute. */
269	if (ni_load_mi(ni, le: le2, mi: &mi2))
270	return NULL;
271
272	if (mi)
273	*mi = mi2;
274
275	/* Find attribute in loaded record. */
276	return rec_find_attr_le(rec: mi2, le: le2);
277	}
278
279	/*
280	* ni_load_attr - Load attribute that contains given VCN.
281	*/
282	struct ATTRIB *ni_load_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
283	const __le16 *name, u8 name_len, CLST vcn,
284	struct mft_inode **pmi)
285	{
286	struct ATTR_LIST_ENTRY *le;
287	struct ATTRIB *attr;
288	struct mft_inode *mi;
289	struct ATTR_LIST_ENTRY *next;
290
291	if (!ni->attr_list.size) {
292	if (pmi)
293	*pmi = &ni->mi;
294	return mi_find_attr(mi: &ni->mi, NULL, type, name, name_len, NULL);
295	}
296
297	le = al_find_ex(ni, NULL, type, name, name_len, NULL);
298	if (!le)
299	return NULL;
300
301	/*
302	* Unfortunately ATTR_LIST_ENTRY contains only start VCN.
303	* So to find the ATTRIB segment that contains 'vcn' we should
304	* enumerate some entries.
305	*/
306	if (vcn) {
307	for (;; le = next) {
308	next = al_find_ex(ni, le, type, name, name_len, NULL);
309	if (!next || le64_to_cpu(next->vcn) > vcn)
310	break;
311	}
312	}
313
314	if (ni_load_mi(ni, le, mi: &mi))
315	return NULL;
316
317	if (pmi)
318	*pmi = mi;
319
320	attr = mi_find_attr(mi, NULL, type, name, name_len, id: &le->id);
321	if (!attr)
322	return NULL;
323
324	if (!attr->non_res)
325	return attr;
326
327	if (le64_to_cpu(attr->nres.svcn) <= vcn &&
328	vcn <= le64_to_cpu(attr->nres.evcn))
329	return attr;
330
331	return NULL;
332	}
333
334	/*
335	* ni_load_all_mi - Load all subrecords.
336	*/
337	int ni_load_all_mi(struct ntfs_inode *ni)
338	{
339	int err;
340	struct ATTR_LIST_ENTRY *le;
341
342	if (!ni->attr_list.size)
343	return 0;
344
345	le = NULL;
346
347	while ((le = al_enumerate(ni, le))) {
348	CLST rno = ino_get(ref: &le->ref);
349
350	if (rno == ni->mi.rno)
351	continue;
352
353	err = ni_load_mi_ex(ni, rno, NULL);
354	if (err)
355	return err;
356	}
357
358	return 0;
359	}
360
361	/*
362	* ni_add_subrecord - Allocate + format + attach a new subrecord.
363	*/
364	bool ni_add_subrecord(struct ntfs_inode *ni, CLST rno, struct mft_inode **mi)
365	{
366	struct mft_inode *m;
367
368	m = kzalloc(size: sizeof(struct mft_inode), GFP_NOFS);
369	if (!m)
370	return false;
371
372	if (mi_format_new(mi: m, sbi: ni->mi.sbi, rno, flags: 0, is_mft: ni->mi.rno == MFT_REC_MFT)) {
373	mi_put(mi: m);
374	return false;
375	}
376
377	mi_get_ref(mi: &ni->mi, ref: &m->mrec->parent_ref);
378
379	ni_add_mi(ni, mi: m);
380	*mi = m;
381	return true;
382	}
383
384	/*
385	* ni_remove_attr - Remove all attributes for the given type/name/id.
386	*/
387	int ni_remove_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
388	const __le16 *name, u8 name_len, bool base_only,
389	const __le16 *id)
390	{
391	int err;
392	struct ATTRIB *attr;
393	struct ATTR_LIST_ENTRY *le;
394	struct mft_inode *mi;
395	u32 type_in;
396	int diff;
397
398	if (base_only || type == ATTR_LIST || !ni->attr_list.size) {
399	attr = mi_find_attr(mi: &ni->mi, NULL, type, name, name_len, id);
400	if (!attr)
401	return -ENOENT;
402
403	mi_remove_attr(ni, mi: &ni->mi, attr);
404	return 0;
405	}
406
407	type_in = le32_to_cpu(type);
408	le = NULL;
409
410	for (;;) {
411	le = al_enumerate(ni, le);
412	if (!le)
413	return 0;
414
415	next_le2:
416	diff = le32_to_cpu(le->type) - type_in;
417	if (diff < 0)
418	continue;
419
420	if (diff > 0)
421	return 0;
422
423	if (le->name_len != name_len)
424	continue;
425
426	if (name_len &&
427	memcmp(p: le_name(le), q: name, size: name_len * sizeof(short)))
428	continue;
429
430	if (id && le->id != *id)
431	continue;
432	err = ni_load_mi(ni, le, mi: &mi);
433	if (err)
434	return err;
435
436	al_remove_le(ni, le);
437
438	attr = mi_find_attr(mi, NULL, type, name, name_len, id);
439	if (!attr)
440	return -ENOENT;
441
442	mi_remove_attr(ni, mi, attr);
443
444	if (PtrOffset(ni->attr_list.le, le) >= ni->attr_list.size)
445	return 0;
446	goto next_le2;
447	}
448	}
449
450	/*
451	* ni_ins_new_attr - Insert the attribute into record.
452	*
453	* Return: Not full constructed attribute or NULL if not possible to create.
454	*/
455	static struct ATTRIB *
456	ni_ins_new_attr(struct ntfs_inode *ni, struct mft_inode *mi,
457	struct ATTR_LIST_ENTRY *le, enum ATTR_TYPE type,
458	const __le16 *name, u8 name_len, u32 asize, u16 name_off,
459	CLST svcn, struct ATTR_LIST_ENTRY **ins_le)
460	{
461	int err;
462	struct ATTRIB *attr;
463	bool le_added = false;
464	struct MFT_REF ref;
465
466	mi_get_ref(mi, ref: &ref);
467
468	if (type != ATTR_LIST && !le && ni->attr_list.size) {
469	err = al_add_le(ni, type, name, name_len, svcn, cpu_to_le16(-1),
470	ref: &ref, new_le: &le);
471	if (err) {
472	/* No memory or no space. */
473	return ERR_PTR(error: err);
474	}
475	le_added = true;
476
477	/*
478	* al_add_le -> attr_set_size (list) -> ni_expand_list
479	* which moves some attributes out of primary record
480	* this means that name may point into moved memory
481	* reinit 'name' from le.
482	*/
483	name = le->name;
484	}
485
486	attr = mi_insert_attr(mi, type, name, name_len, asize, name_off);
487	if (!attr) {
488	if (le_added)
489	al_remove_le(ni, le);
490	return NULL;
491	}
492
493	if (type == ATTR_LIST) {
494	/* Attr list is not in list entry array. */
495	goto out;
496	}
497
498	if (!le)
499	goto out;
500
501	/* Update ATTRIB Id and record reference. */
502	le->id = attr->id;
503	ni->attr_list.dirty = true;
504	le->ref = ref;
505
506	out:
507	if (ins_le)
508	*ins_le = le;
509	return attr;
510	}
511
512	/*
513	* ni_repack
514	*
515	* Random write access to sparsed or compressed file may result to
516	* not optimized packed runs.
517	* Here is the place to optimize it.
518	*/
519	static int ni_repack(struct ntfs_inode *ni)
520	{
521	#if 1
522	return 0;
523	#else
524	int err = 0;
525	struct ntfs_sb_info *sbi = ni->mi.sbi;
526	struct mft_inode *mi, *mi_p = NULL;
527	struct ATTRIB *attr = NULL, *attr_p;
528	struct ATTR_LIST_ENTRY *le = NULL, *le_p;
529	CLST alloc = 0;
530	u8 cluster_bits = sbi->cluster_bits;
531	CLST svcn, evcn = 0, svcn_p, evcn_p, next_svcn;
532	u32 roff, rs = sbi->record_size;
533	struct runs_tree run;
534
535	run_init(&run);
536
537	while ((attr = ni_enum_attr_ex(ni, attr, &le, &mi))) {
538	if (!attr->non_res)
539	continue;
540
541	svcn = le64_to_cpu(attr->nres.svcn);
542	if (svcn != le64_to_cpu(le->vcn)) {
543	err = -EINVAL;
544	break;
545	}
546
547	if (!svcn) {
548	alloc = le64_to_cpu(attr->nres.alloc_size) >>
549	cluster_bits;
550	mi_p = NULL;
551	} else if (svcn != evcn + 1) {
552	err = -EINVAL;
553	break;
554	}
555
556	evcn = le64_to_cpu(attr->nres.evcn);
557
558	if (svcn > evcn + 1) {
559	err = -EINVAL;
560	break;
561	}
562
563	if (!mi_p) {
564	/* Do not try if not enough free space. */
565	if (le32_to_cpu(mi->mrec->used) + 8 >= rs)
566	continue;
567
568	/* Do not try if last attribute segment. */
569	if (evcn + 1 == alloc)
570	continue;
571	run_close(&run);
572	}
573
574	roff = le16_to_cpu(attr->nres.run_off);
575
576	if (roff > le32_to_cpu(attr->size)) {
577	err = -EINVAL;
578	break;
579	}
580
581	err = run_unpack(&run, sbi, ni->mi.rno, svcn, evcn, svcn,
582	Add2Ptr(attr, roff),
583	le32_to_cpu(attr->size) - roff);
584	if (err < 0)
585	break;
586
587	if (!mi_p) {
588	mi_p = mi;
589	attr_p = attr;
590	svcn_p = svcn;
591	evcn_p = evcn;
592	le_p = le;
593	err = 0;
594	continue;
595	}
596
597	/*
598	* Run contains data from two records: mi_p and mi
599	* Try to pack in one.
600	*/
601	err = mi_pack_runs(mi_p, attr_p, &run, evcn + 1 - svcn_p);
602	if (err)
603	break;
604
605	next_svcn = le64_to_cpu(attr_p->nres.evcn) + 1;
606
607	if (next_svcn >= evcn + 1) {
608	/* We can remove this attribute segment. */
609	al_remove_le(ni, le);
610	mi_remove_attr(NULL, mi, attr);
611	le = le_p;
612	continue;
613	}
614
615	attr->nres.svcn = le->vcn = cpu_to_le64(next_svcn);
616	mi->dirty = true;
617	ni->attr_list.dirty = true;
618
619	if (evcn + 1 == alloc) {
620	err = mi_pack_runs(mi, attr, &run,
621	evcn + 1 - next_svcn);
622	if (err)
623	break;
624	mi_p = NULL;
625	} else {
626	mi_p = mi;
627	attr_p = attr;
628	svcn_p = next_svcn;
629	evcn_p = evcn;
630	le_p = le;
631	run_truncate_head(&run, next_svcn);
632	}
633	}
634
635	if (err) {
636	ntfs_inode_warn(&ni->vfs_inode, "repack problem");
637	ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
638
639	/* Pack loaded but not packed runs. */
640	if (mi_p)
641	mi_pack_runs(mi_p, attr_p, &run, evcn_p + 1 - svcn_p);
642	}
643
644	run_close(&run);
645	return err;
646	#endif
647	}
648
649	/*
650	* ni_try_remove_attr_list
651	*
652	* Can we remove attribute list?
653	* Check the case when primary record contains enough space for all attributes.
654	*/
655	static int ni_try_remove_attr_list(struct ntfs_inode *ni)
656	{
657	int err = 0;
658	struct ntfs_sb_info *sbi = ni->mi.sbi;
659	struct ATTRIB *attr, *attr_list, *attr_ins;
660	struct ATTR_LIST_ENTRY *le;
661	struct mft_inode *mi;
662	u32 asize, free;
663	struct MFT_REF ref;
664	struct MFT_REC *mrec;
665	__le16 id;
666
667	if (!ni->attr_list.dirty)
668	return 0;
669
670	err = ni_repack(ni);
671	if (err)
672	return err;
673
674	attr_list = mi_find_attr(mi: &ni->mi, NULL, type: ATTR_LIST, NULL, name_len: 0, NULL);
675	if (!attr_list)
676	return 0;
677
678	asize = le32_to_cpu(attr_list->size);
679
680	/* Free space in primary record without attribute list. */
681	free = sbi->record_size - le32_to_cpu(ni->mi.mrec->used) + asize;
682	mi_get_ref(mi: &ni->mi, ref: &ref);
683
684	le = NULL;
685	while ((le = al_enumerate(ni, le))) {
686	if (!memcmp(p: &le->ref, q: &ref, size: sizeof(ref)))
687	continue;
688
689	if (le->vcn)
690	return 0;
691
692	mi = ni_find_mi(ni, rno: ino_get(ref: &le->ref));
693	if (!mi)
694	return 0;
695
696	attr = mi_find_attr(mi, NULL, type: le->type, name: le_name(le),
697	name_len: le->name_len, id: &le->id);
698	if (!attr)
699	return 0;
700
701	asize = le32_to_cpu(attr->size);
702	if (asize > free)
703	return 0;
704
705	free -= asize;
706	}
707
708	/* Make a copy of primary record to restore if error. */
709	mrec = kmemdup(p: ni->mi.mrec, size: sbi->record_size, GFP_NOFS);
710	if (!mrec)
711	return 0; /* Not critical. */
712
713	/* It seems that attribute list can be removed from primary record. */
714	mi_remove_attr(NULL, mi: &ni->mi, attr: attr_list);
715
716	/*
717	* Repeat the cycle above and copy all attributes to primary record.
718	* Do not remove original attributes from subrecords!
719	* It should be success!
720	*/
721	le = NULL;
722	while ((le = al_enumerate(ni, le))) {
723	if (!memcmp(p: &le->ref, q: &ref, size: sizeof(ref)))
724	continue;
725
726	mi = ni_find_mi(ni, rno: ino_get(ref: &le->ref));
727	if (!mi) {
728	/* Should never happened, 'cause already checked. */
729	goto out;
730	}
731
732	attr = mi_find_attr(mi, NULL, type: le->type, name: le_name(le),
733	name_len: le->name_len, id: &le->id);
734	if (!attr) {
735	/* Should never happened, 'cause already checked. */
736	goto out;
737	}
738	asize = le32_to_cpu(attr->size);
739
740	/* Insert into primary record. */
741	attr_ins = mi_insert_attr(mi: &ni->mi, type: le->type, name: le_name(le),
742	name_len: le->name_len, asize,
743	le16_to_cpu(attr->name_off));
744	if (!attr_ins) {
745	/*
746	* No space in primary record (already checked).
747	*/
748	goto out;
749	}
750
751	/* Copy all except id. */
752	id = attr_ins->id;
753	memcpy(attr_ins, attr, asize);
754	attr_ins->id = id;
755	}
756
757	/*
758	* Repeat the cycle above and remove all attributes from subrecords.
759	*/
760	le = NULL;
761	while ((le = al_enumerate(ni, le))) {
762	if (!memcmp(p: &le->ref, q: &ref, size: sizeof(ref)))
763	continue;
764
765	mi = ni_find_mi(ni, rno: ino_get(ref: &le->ref));
766	if (!mi)
767	continue;
768
769	attr = mi_find_attr(mi, NULL, type: le->type, name: le_name(le),
770	name_len: le->name_len, id: &le->id);
771	if (!attr)
772	continue;
773
774	/* Remove from original record. */
775	mi_remove_attr(NULL, mi, attr);
776	}
777
778	run_deallocate(sbi, run: &ni->attr_list.run, trim: true);
779	run_close(run: &ni->attr_list.run);
780	ni->attr_list.size = 0;
781	kvfree(addr: ni->attr_list.le);
782	ni->attr_list.le = NULL;
783	ni->attr_list.dirty = false;
784
785	kfree(objp: mrec);
786	return 0;
787	out:
788	/* Restore primary record. */
789	swap(mrec, ni->mi.mrec);
790	kfree(objp: mrec);
791	return 0;
792	}
793
794	/*
795	* ni_create_attr_list - Generates an attribute list for this primary record.
796	*/
797	int ni_create_attr_list(struct ntfs_inode *ni)
798	{
799	struct ntfs_sb_info *sbi = ni->mi.sbi;
800	int err;
801	u32 lsize;
802	struct ATTRIB *attr;
803	struct ATTRIB *arr_move[7];
804	struct ATTR_LIST_ENTRY *le, *le_b[7];
805	struct MFT_REC *rec;
806	bool is_mft;
807	CLST rno = 0;
808	struct mft_inode *mi;
809	u32 free_b, nb, to_free, rs;
810	u16 sz;
811
812	is_mft = ni->mi.rno == MFT_REC_MFT;
813	rec = ni->mi.mrec;
814	rs = sbi->record_size;
815
816	/*
817	* Skip estimating exact memory requirement.
818	* Looks like one record_size is always enough.
819	*/
820	le = kmalloc(size: al_aligned(size: rs), GFP_NOFS);
821	if (!le)
822	return -ENOMEM;
823
824	mi_get_ref(mi: &ni->mi, ref: &le->ref);
825	ni->attr_list.le = le;
826
827	attr = NULL;
828	nb = 0;
829	free_b = 0;
830	attr = NULL;
831
832	for (; (attr = mi_enum_attr(mi: &ni->mi, attr)); le = Add2Ptr(le, sz)) {
833	sz = le_size(name_len: attr->name_len);
834	le->type = attr->type;
835	le->size = cpu_to_le16(sz);
836	le->name_len = attr->name_len;
837	le->name_off = offsetof(struct ATTR_LIST_ENTRY, name);
838	le->vcn = 0;
839	if (le != ni->attr_list.le)
840	le->ref = ni->attr_list.le->ref;
841	le->id = attr->id;
842
843	if (attr->name_len)
844	memcpy(le->name, attr_name(attr),
845	sizeof(short) * attr->name_len);
846	else if (attr->type == ATTR_STD)
847	continue;
848	else if (attr->type == ATTR_LIST)
849	continue;
850	else if (is_mft && attr->type == ATTR_DATA)
851	continue;
852
853	if (!nb || nb < ARRAY_SIZE(arr_move)) {
854	le_b[nb] = le;
855	arr_move[nb++] = attr;
856	free_b += le32_to_cpu(attr->size);
857	}
858	}
859
860	lsize = PtrOffset(ni->attr_list.le, le);
861	ni->attr_list.size = lsize;
862
863	to_free = le32_to_cpu(rec->used) + lsize + SIZEOF_RESIDENT;
864	if (to_free <= rs) {
865	to_free = 0;
866	} else {
867	to_free -= rs;
868
869	if (to_free > free_b) {
870	err = -EINVAL;
871	goto out;
872	}
873	}
874
875	/* Allocate child MFT. */
876	err = ntfs_look_free_mft(sbi, rno: &rno, mft: is_mft, ni, mi: &mi);
877	if (err)
878	goto out;
879
880	err = -EINVAL;
881	/* Call mi_remove_attr() in reverse order to keep pointers 'arr_move' valid. */
882	while (to_free > 0) {
883	struct ATTRIB *b = arr_move[--nb];
884	u32 asize = le32_to_cpu(b->size);
885	u16 name_off = le16_to_cpu(b->name_off);
886
887	attr = mi_insert_attr(mi, type: b->type, Add2Ptr(b, name_off),
888	name_len: b->name_len, asize, name_off);
889	if (!attr)
890	goto out;
891
892	mi_get_ref(mi, ref: &le_b[nb]->ref);
893	le_b[nb]->id = attr->id;
894
895	/* Copy all except id. */
896	memcpy(attr, b, asize);
897	attr->id = le_b[nb]->id;
898
899	/* Remove from primary record. */
900	if (!mi_remove_attr(NULL, mi: &ni->mi, attr: b))
901	goto out;
902
903	if (to_free <= asize)
904	break;
905	to_free -= asize;
906	if (!nb)
907	goto out;
908	}
909
910	attr = mi_insert_attr(mi: &ni->mi, type: ATTR_LIST, NULL, name_len: 0,
911	asize: lsize + SIZEOF_RESIDENT, SIZEOF_RESIDENT);
912	if (!attr)
913	goto out;
914
915	attr->non_res = 0;
916	attr->flags = 0;
917	attr->res.data_size = cpu_to_le32(lsize);
918	attr->res.data_off = SIZEOF_RESIDENT_LE;
919	attr->res.flags = 0;
920	attr->res.res = 0;
921
922	memcpy(resident_data_ex(attr, lsize), ni->attr_list.le, lsize);
923
924	ni->attr_list.dirty = false;
925
926	mark_inode_dirty(inode: &ni->vfs_inode);
927	return 0;
928
929	out:
930	kvfree(addr: ni->attr_list.le);
931	ni->attr_list.le = NULL;
932	ni->attr_list.size = 0;
933	return err;
934	}
935
936	/*
937	* ni_ins_attr_ext - Add an external attribute to the ntfs_inode.
938	*/
939	static int ni_ins_attr_ext(struct ntfs_inode *ni, struct ATTR_LIST_ENTRY *le,
940	enum ATTR_TYPE type, const __le16 *name, u8 name_len,
941	u32 asize, CLST svcn, u16 name_off, bool force_ext,
942	struct ATTRIB **ins_attr, struct mft_inode **ins_mi,
943	struct ATTR_LIST_ENTRY **ins_le)
944	{
945	struct ATTRIB *attr;
946	struct mft_inode *mi;
947	CLST rno;
948	u64 vbo;
949	struct rb_node *node;
950	int err;
951	bool is_mft, is_mft_data;
952	struct ntfs_sb_info *sbi = ni->mi.sbi;
953
954	is_mft = ni->mi.rno == MFT_REC_MFT;
955	is_mft_data = is_mft && type == ATTR_DATA && !name_len;
956
957	if (asize > sbi->max_bytes_per_attr) {
958	err = -EINVAL;
959	goto out;
960	}
961
962	/*
963	* Standard information and attr_list cannot be made external.
964	* The Log File cannot have any external attributes.
965	*/
966	if (type == ATTR_STD || type == ATTR_LIST ||
967	ni->mi.rno == MFT_REC_LOG) {
968	err = -EINVAL;
969	goto out;
970	}
971
972	/* Create attribute list if it is not already existed. */
973	if (!ni->attr_list.size) {
974	err = ni_create_attr_list(ni);
975	if (err)
976	goto out;
977	}
978
979	vbo = is_mft_data ? ((u64)svcn << sbi->cluster_bits) : 0;
980
981	if (force_ext)
982	goto insert_ext;
983
984	/* Load all subrecords into memory. */
985	err = ni_load_all_mi(ni);
986	if (err)
987	goto out;
988
989	/* Check each of loaded subrecord. */
990	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
991	mi = rb_entry(node, struct mft_inode, node);
992
993	if (is_mft_data &&
994	(mi_enum_attr(mi, NULL) ||
995	vbo <= ((u64)mi->rno << sbi->record_bits))) {
996	/* We can't accept this record 'cause MFT's bootstrapping. */
997	continue;
998	}
999	if (is_mft &&
1000	mi_find_attr(mi, NULL, type: ATTR_DATA, NULL, name_len: 0, NULL)) {
1001	/*
1002	* This child record already has a ATTR_DATA.
1003	* So it can't accept any other records.
1004	*/
1005	continue;
1006	}
1007
1008	if ((type != ATTR_NAME || name_len) &&
1009	mi_find_attr(mi, NULL, type, name, name_len, NULL)) {
1010	/* Only indexed attributes can share same record. */
1011	continue;
1012	}
1013
1014	/*
1015	* Do not try to insert this attribute
1016	* if there is no room in record.
1017	*/
1018	if (le32_to_cpu(mi->mrec->used) + asize > sbi->record_size)
1019	continue;
1020
1021	/* Try to insert attribute into this subrecord. */
1022	attr = ni_ins_new_attr(ni, mi, le, type, name, name_len, asize,
1023	name_off, svcn, ins_le);
1024	if (!attr)
1025	continue;
1026	if (IS_ERR(ptr: attr))
1027	return PTR_ERR(ptr: attr);
1028
1029	if (ins_attr)
1030	*ins_attr = attr;
1031	if (ins_mi)
1032	*ins_mi = mi;
1033	return 0;
1034	}
1035
1036	insert_ext:
1037	/* We have to allocate a new child subrecord. */
1038	err = ntfs_look_free_mft(sbi, rno: &rno, mft: is_mft_data, ni, mi: &mi);
1039	if (err)
1040	goto out;
1041
1042	if (is_mft_data && vbo <= ((u64)rno << sbi->record_bits)) {
1043	err = -EINVAL;
1044	goto out1;
1045	}
1046
1047	attr = ni_ins_new_attr(ni, mi, le, type, name, name_len, asize,
1048	name_off, svcn, ins_le);
1049	if (!attr) {
1050	err = -EINVAL;
1051	goto out2;
1052	}
1053
1054	if (IS_ERR(ptr: attr)) {
1055	err = PTR_ERR(ptr: attr);
1056	goto out2;
1057	}
1058
1059	if (ins_attr)
1060	*ins_attr = attr;
1061	if (ins_mi)
1062	*ins_mi = mi;
1063
1064	return 0;
1065
1066	out2:
1067	ni_remove_mi(ni, mi);
1068	mi_put(mi);
1069
1070	out1:
1071	ntfs_mark_rec_free(sbi, rno, is_mft);
1072
1073	out:
1074	return err;
1075	}
1076
1077	/*
1078	* ni_insert_attr - Insert an attribute into the file.
1079	*
1080	* If the primary record has room, it will just insert the attribute.
1081	* If not, it may make the attribute external.
1082	* For $MFT::Data it may make room for the attribute by
1083	* making other attributes external.
1084	*
1085	* NOTE:
1086	* The ATTR_LIST and ATTR_STD cannot be made external.
1087	* This function does not fill new attribute full.
1088	* It only fills 'size'/'type'/'id'/'name_len' fields.
1089	*/
1090	static int ni_insert_attr(struct ntfs_inode *ni, enum ATTR_TYPE type,
1091	const __le16 *name, u8 name_len, u32 asize,
1092	u16 name_off, CLST svcn, struct ATTRIB **ins_attr,
1093	struct mft_inode **ins_mi,
1094	struct ATTR_LIST_ENTRY **ins_le)
1095	{
1096	struct ntfs_sb_info *sbi = ni->mi.sbi;
1097	int err;
1098	struct ATTRIB *attr, *eattr;
1099	struct MFT_REC *rec;
1100	bool is_mft;
1101	struct ATTR_LIST_ENTRY *le;
1102	u32 list_reserve, max_free, free, used, t32;
1103	__le16 id;
1104	u16 t16;
1105
1106	is_mft = ni->mi.rno == MFT_REC_MFT;
1107	rec = ni->mi.mrec;
1108
1109	list_reserve = SIZEOF_NONRESIDENT + 3 * (1 + 2 * sizeof(u32));
1110	used = le32_to_cpu(rec->used);
1111	free = sbi->record_size - used;
1112
1113	if (is_mft && type != ATTR_LIST) {
1114	/* Reserve space for the ATTRIB list. */
1115	if (free < list_reserve)
1116	free = 0;
1117	else
1118	free -= list_reserve;
1119	}
1120
1121	if (asize <= free) {
1122	attr = ni_ins_new_attr(ni, mi: &ni->mi, NULL, type, name, name_len,
1123	asize, name_off, svcn, ins_le);
1124	if (IS_ERR(ptr: attr)) {
1125	err = PTR_ERR(ptr: attr);
1126	goto out;
1127	}
1128
1129	if (attr) {
1130	if (ins_attr)
1131	*ins_attr = attr;
1132	if (ins_mi)
1133	*ins_mi = &ni->mi;
1134	err = 0;
1135	goto out;
1136	}
1137	}
1138
1139	if (!is_mft || type != ATTR_DATA || svcn) {
1140	/* This ATTRIB will be external. */
1141	err = ni_ins_attr_ext(ni, NULL, type, name, name_len, asize,
1142	svcn, name_off, force_ext: false, ins_attr, ins_mi,
1143	ins_le);
1144	goto out;
1145	}
1146
1147	/*
1148	* Here we have: "is_mft && type == ATTR_DATA && !svcn"
1149	*
1150	* The first chunk of the $MFT::Data ATTRIB must be the base record.
1151	* Evict as many other attributes as possible.
1152	*/
1153	max_free = free;
1154
1155	/* Estimate the result of moving all possible attributes away. */
1156	attr = NULL;
1157
1158	while ((attr = mi_enum_attr(mi: &ni->mi, attr))) {
1159	if (attr->type == ATTR_STD)
1160	continue;
1161	if (attr->type == ATTR_LIST)
1162	continue;
1163	max_free += le32_to_cpu(attr->size);
1164	}
1165
1166	if (max_free < asize + list_reserve) {
1167	/* Impossible to insert this attribute into primary record. */
1168	err = -EINVAL;
1169	goto out;
1170	}
1171
1172	/* Start real attribute moving. */
1173	attr = NULL;
1174
1175	for (;;) {
1176	attr = mi_enum_attr(mi: &ni->mi, attr);
1177	if (!attr) {
1178	/* We should never be here 'cause we have already check this case. */
1179	err = -EINVAL;
1180	goto out;
1181	}
1182
1183	/* Skip attributes that MUST be primary record. */
1184	if (attr->type == ATTR_STD || attr->type == ATTR_LIST)
1185	continue;
1186
1187	le = NULL;
1188	if (ni->attr_list.size) {
1189	le = al_find_le(ni, NULL, attr);
1190	if (!le) {
1191	/* Really this is a serious bug. */
1192	err = -EINVAL;
1193	goto out;
1194	}
1195	}
1196
1197	t32 = le32_to_cpu(attr->size);
1198	t16 = le16_to_cpu(attr->name_off);
1199	err = ni_ins_attr_ext(ni, le, type: attr->type, Add2Ptr(attr, t16),
1200	name_len: attr->name_len, asize: t32, svcn: attr_svcn(attr), name_off: t16,
1201	force_ext: false, ins_attr: &eattr, NULL, NULL);
1202	if (err)
1203	return err;
1204
1205	id = eattr->id;
1206	memcpy(eattr, attr, t32);
1207	eattr->id = id;
1208
1209	/* Remove from primary record. */
1210	mi_remove_attr(NULL, mi: &ni->mi, attr);
1211
1212	/* attr now points to next attribute. */
1213	if (attr->type == ATTR_END)
1214	goto out;
1215	}
1216	while (asize + list_reserve > sbi->record_size - le32_to_cpu(rec->used))
1217	;
1218
1219	attr = ni_ins_new_attr(ni, mi: &ni->mi, NULL, type, name, name_len, asize,
1220	name_off, svcn, ins_le);
1221	if (!attr) {
1222	err = -EINVAL;
1223	goto out;
1224	}
1225
1226	if (IS_ERR(ptr: attr)) {
1227	err = PTR_ERR(ptr: attr);
1228	goto out;
1229	}
1230
1231	if (ins_attr)
1232	*ins_attr = attr;
1233	if (ins_mi)
1234	*ins_mi = &ni->mi;
1235
1236	out:
1237	return err;
1238	}
1239
1240	/* ni_expand_mft_list - Split ATTR_DATA of $MFT. */
1241	static int ni_expand_mft_list(struct ntfs_inode *ni)
1242	{
1243	int err = 0;
1244	struct runs_tree *run = &ni->file.run;
1245	u32 asize, run_size, done = 0;
1246	struct ATTRIB *attr;
1247	struct rb_node *node;
1248	CLST mft_min, mft_new, svcn, evcn, plen;
1249	struct mft_inode *mi, *mi_min, *mi_new;
1250	struct ntfs_sb_info *sbi = ni->mi.sbi;
1251
1252	/* Find the nearest MFT. */
1253	mft_min = 0;
1254	mft_new = 0;
1255	mi_min = NULL;
1256
1257	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
1258	mi = rb_entry(node, struct mft_inode, node);
1259
1260	attr = mi_enum_attr(mi, NULL);
1261
1262	if (!attr) {
1263	mft_min = mi->rno;
1264	mi_min = mi;
1265	break;
1266	}
1267	}
1268
1269	if (ntfs_look_free_mft(sbi, rno: &mft_new, mft: true, ni, mi: &mi_new)) {
1270	mft_new = 0;
1271	/* Really this is not critical. */
1272	} else if (mft_min > mft_new) {
1273	mft_min = mft_new;
1274	mi_min = mi_new;
1275	} else {
1276	ntfs_mark_rec_free(sbi, rno: mft_new, is_mft: true);
1277	mft_new = 0;
1278	ni_remove_mi(ni, mi: mi_new);
1279	}
1280
1281	attr = mi_find_attr(mi: &ni->mi, NULL, type: ATTR_DATA, NULL, name_len: 0, NULL);
1282	if (!attr) {
1283	err = -EINVAL;
1284	goto out;
1285	}
1286
1287	asize = le32_to_cpu(attr->size);
1288
1289	evcn = le64_to_cpu(attr->nres.evcn);
1290	svcn = bytes_to_cluster(sbi, size: (u64)(mft_min + 1) << sbi->record_bits);
1291	if (evcn + 1 >= svcn) {
1292	err = -EINVAL;
1293	goto out;
1294	}
1295
1296	/*
1297	* Split primary attribute [0 evcn] in two parts [0 svcn) + [svcn evcn].
1298	*
1299	* Update first part of ATTR_DATA in 'primary MFT.
1300	*/
1301	err = run_pack(run, svcn: 0, len: svcn, Add2Ptr(attr, SIZEOF_NONRESIDENT),
1302	run_buf_size: asize - SIZEOF_NONRESIDENT, packed_vcns: &plen);
1303	if (err < 0)
1304	goto out;
1305
1306	run_size = ALIGN(err, 8);
1307	err = 0;
1308
1309	if (plen < svcn) {
1310	err = -EINVAL;
1311	goto out;
1312	}
1313
1314	attr->nres.evcn = cpu_to_le64(svcn - 1);
1315	attr->size = cpu_to_le32(run_size + SIZEOF_NONRESIDENT);
1316	/* 'done' - How many bytes of primary MFT becomes free. */
1317	done = asize - run_size - SIZEOF_NONRESIDENT;
1318	le32_sub_cpu(var: &ni->mi.mrec->used, val: done);
1319
1320	/* Estimate packed size (run_buf=NULL). */
1321	err = run_pack(run, svcn, len: evcn + 1 - svcn, NULL, run_buf_size: sbi->record_size,
1322	packed_vcns: &plen);
1323	if (err < 0)
1324	goto out;
1325
1326	run_size = ALIGN(err, 8);
1327	err = 0;
1328
1329	if (plen < evcn + 1 - svcn) {
1330	err = -EINVAL;
1331	goto out;
1332	}
1333
1334	/*
1335	* This function may implicitly call expand attr_list.
1336	* Insert second part of ATTR_DATA in 'mi_min'.
1337	*/
1338	attr = ni_ins_new_attr(ni, mi: mi_min, NULL, type: ATTR_DATA, NULL, name_len: 0,
1339	SIZEOF_NONRESIDENT + run_size,
1340	SIZEOF_NONRESIDENT, svcn, NULL);
1341	if (!attr) {
1342	err = -EINVAL;
1343	goto out;
1344	}
1345
1346	if (IS_ERR(ptr: attr)) {
1347	err = PTR_ERR(ptr: attr);
1348	goto out;
1349	}
1350
1351	attr->non_res = 1;
1352	attr->name_off = SIZEOF_NONRESIDENT_LE;
1353	attr->flags = 0;
1354
1355	/* This function can't fail - cause already checked above. */
1356	run_pack(run, svcn, len: evcn + 1 - svcn, Add2Ptr(attr, SIZEOF_NONRESIDENT),
1357	run_buf_size: run_size, packed_vcns: &plen);
1358
1359	attr->nres.svcn = cpu_to_le64(svcn);
1360	attr->nres.evcn = cpu_to_le64(evcn);
1361	attr->nres.run_off = cpu_to_le16(SIZEOF_NONRESIDENT);
1362
1363	out:
1364	if (mft_new) {
1365	ntfs_mark_rec_free(sbi, rno: mft_new, is_mft: true);
1366	ni_remove_mi(ni, mi: mi_new);
1367	}
1368
1369	return !err && !done ? -EOPNOTSUPP : err;
1370	}
1371
1372	/*
1373	* ni_expand_list - Move all possible attributes out of primary record.
1374	*/
1375	int ni_expand_list(struct ntfs_inode *ni)
1376	{
1377	int err = 0;
1378	u32 asize, done = 0;
1379	struct ATTRIB *attr, *ins_attr;
1380	struct ATTR_LIST_ENTRY *le;
1381	bool is_mft = ni->mi.rno == MFT_REC_MFT;
1382	struct MFT_REF ref;
1383
1384	mi_get_ref(mi: &ni->mi, ref: &ref);
1385	le = NULL;
1386
1387	while ((le = al_enumerate(ni, le))) {
1388	if (le->type == ATTR_STD)
1389	continue;
1390
1391	if (memcmp(p: &ref, q: &le->ref, size: sizeof(struct MFT_REF)))
1392	continue;
1393
1394	if (is_mft && le->type == ATTR_DATA)
1395	continue;
1396
1397	/* Find attribute in primary record. */
1398	attr = rec_find_attr_le(rec: &ni->mi, le);
1399	if (!attr) {
1400	err = -EINVAL;
1401	goto out;
1402	}
1403
1404	asize = le32_to_cpu(attr->size);
1405
1406	/* Always insert into new record to avoid collisions (deep recursive). */
1407	err = ni_ins_attr_ext(ni, le, type: attr->type, name: attr_name(attr),
1408	name_len: attr->name_len, asize, svcn: attr_svcn(attr),
1409	le16_to_cpu(attr->name_off), force_ext: true,
1410	ins_attr: &ins_attr, NULL, NULL);
1411
1412	if (err)
1413	goto out;
1414
1415	memcpy(ins_attr, attr, asize);
1416	ins_attr->id = le->id;
1417	/* Remove from primary record. */
1418	mi_remove_attr(NULL, mi: &ni->mi, attr);
1419
1420	done += asize;
1421	goto out;
1422	}
1423
1424	if (!is_mft) {
1425	err = -EFBIG; /* Attr list is too big(?) */
1426	goto out;
1427	}
1428
1429	/* Split MFT data as much as possible. */
1430	err = ni_expand_mft_list(ni);
1431
1432	out:
1433	return !err && !done ? -EOPNOTSUPP : err;
1434	}
1435
1436	/*
1437	* ni_insert_nonresident - Insert new nonresident attribute.
1438	*/
1439	int ni_insert_nonresident(struct ntfs_inode *ni, enum ATTR_TYPE type,
1440	const __le16 *name, u8 name_len,
1441	const struct runs_tree *run, CLST svcn, CLST len,
1442	__le16 flags, struct ATTRIB **new_attr,
1443	struct mft_inode **mi, struct ATTR_LIST_ENTRY **le)
1444	{
1445	int err;
1446	CLST plen;
1447	struct ATTRIB *attr;
1448	bool is_ext = (flags & (ATTR_FLAG_SPARSED | ATTR_FLAG_COMPRESSED)) &&
1449	!svcn;
1450	u32 name_size = ALIGN(name_len * sizeof(short), 8);
1451	u32 name_off = is_ext ? SIZEOF_NONRESIDENT_EX : SIZEOF_NONRESIDENT;
1452	u32 run_off = name_off + name_size;
1453	u32 run_size, asize;
1454	struct ntfs_sb_info *sbi = ni->mi.sbi;
1455
1456	/* Estimate packed size (run_buf=NULL). */
1457	err = run_pack(run, svcn, len, NULL, run_buf_size: sbi->max_bytes_per_attr - run_off,
1458	packed_vcns: &plen);
1459	if (err < 0)
1460	goto out;
1461
1462	run_size = ALIGN(err, 8);
1463
1464	if (plen < len) {
1465	err = -EINVAL;
1466	goto out;
1467	}
1468
1469	asize = run_off + run_size;
1470
1471	if (asize > sbi->max_bytes_per_attr) {
1472	err = -EINVAL;
1473	goto out;
1474	}
1475
1476	err = ni_insert_attr(ni, type, name, name_len, asize, name_off, svcn,
1477	ins_attr: &attr, ins_mi: mi, ins_le: le);
1478
1479	if (err)
1480	goto out;
1481
1482	attr->non_res = 1;
1483	attr->name_off = cpu_to_le16(name_off);
1484	attr->flags = flags;
1485
1486	/* This function can't fail - cause already checked above. */
1487	run_pack(run, svcn, len, Add2Ptr(attr, run_off), run_buf_size: run_size, packed_vcns: &plen);
1488
1489	attr->nres.svcn = cpu_to_le64(svcn);
1490	attr->nres.evcn = cpu_to_le64((u64)svcn + len - 1);
1491
1492	if (new_attr)
1493	*new_attr = attr;
1494
1495	*(__le64 *)&attr->nres.run_off = cpu_to_le64(run_off);
1496
1497	attr->nres.alloc_size =
1498	svcn ? 0 : cpu_to_le64((u64)len << ni->mi.sbi->cluster_bits);
1499	attr->nres.data_size = attr->nres.alloc_size;
1500	attr->nres.valid_size = attr->nres.alloc_size;
1501
1502	if (is_ext) {
1503	if (flags & ATTR_FLAG_COMPRESSED)
1504	attr->nres.c_unit = COMPRESSION_UNIT;
1505	attr->nres.total_size = attr->nres.alloc_size;
1506	}
1507
1508	out:
1509	return err;
1510	}
1511
1512	/*
1513	* ni_insert_resident - Inserts new resident attribute.
1514	*/
1515	int ni_insert_resident(struct ntfs_inode *ni, u32 data_size,
1516	enum ATTR_TYPE type, const __le16 *name, u8 name_len,
1517	struct ATTRIB **new_attr, struct mft_inode **mi,
1518	struct ATTR_LIST_ENTRY **le)
1519	{
1520	int err;
1521	u32 name_size = ALIGN(name_len * sizeof(short), 8);
1522	u32 asize = SIZEOF_RESIDENT + name_size + ALIGN(data_size, 8);
1523	struct ATTRIB *attr;
1524
1525	err = ni_insert_attr(ni, type, name, name_len, asize, SIZEOF_RESIDENT,
1526	svcn: 0, ins_attr: &attr, ins_mi: mi, ins_le: le);
1527	if (err)
1528	return err;
1529
1530	attr->non_res = 0;
1531	attr->flags = 0;
1532
1533	attr->res.data_size = cpu_to_le32(data_size);
1534	attr->res.data_off = cpu_to_le16(SIZEOF_RESIDENT + name_size);
1535	if (type == ATTR_NAME) {
1536	attr->res.flags = RESIDENT_FLAG_INDEXED;
1537
1538	/* is_attr_indexed(attr)) == true */
1539	le16_add_cpu(var: &ni->mi.mrec->hard_links, val: 1);
1540	ni->mi.dirty = true;
1541	}
1542	attr->res.res = 0;
1543
1544	if (new_attr)
1545	*new_attr = attr;
1546
1547	return 0;
1548	}
1549
1550	/*
1551	* ni_remove_attr_le - Remove attribute from record.
1552	*/
1553	void ni_remove_attr_le(struct ntfs_inode *ni, struct ATTRIB *attr,
1554	struct mft_inode *mi, struct ATTR_LIST_ENTRY *le)
1555	{
1556	mi_remove_attr(ni, mi, attr);
1557
1558	if (le)
1559	al_remove_le(ni, le);
1560	}
1561
1562	/*
1563	* ni_delete_all - Remove all attributes and frees allocates space.
1564	*
1565	* ntfs_evict_inode->ntfs_clear_inode->ni_delete_all (if no links).
1566	*/
1567	int ni_delete_all(struct ntfs_inode *ni)
1568	{
1569	int err;
1570	struct ATTR_LIST_ENTRY *le = NULL;
1571	struct ATTRIB *attr = NULL;
1572	struct rb_node *node;
1573	u16 roff;
1574	u32 asize;
1575	CLST svcn, evcn;
1576	struct ntfs_sb_info *sbi = ni->mi.sbi;
1577	bool nt3 = is_ntfs3(sbi);
1578	struct MFT_REF ref;
1579
1580	while ((attr = ni_enum_attr_ex(ni, attr, le: &le, NULL))) {
1581	if (!nt3 || attr->name_len) {
1582	;
1583	} else if (attr->type == ATTR_REPARSE) {
1584	mi_get_ref(mi: &ni->mi, ref: &ref);
1585	ntfs_remove_reparse(sbi, rtag: 0, ref: &ref);
1586	} else if (attr->type == ATTR_ID && !attr->non_res &&
1587	le32_to_cpu(attr->res.data_size) >=
1588	sizeof(struct GUID)) {
1589	ntfs_objid_remove(sbi, guid: resident_data(attr));
1590	}
1591
1592	if (!attr->non_res)
1593	continue;
1594
1595	svcn = le64_to_cpu(attr->nres.svcn);
1596	evcn = le64_to_cpu(attr->nres.evcn);
1597
1598	if (evcn + 1 <= svcn)
1599	continue;
1600
1601	asize = le32_to_cpu(attr->size);
1602	roff = le16_to_cpu(attr->nres.run_off);
1603
1604	if (roff > asize)
1605	return -EINVAL;
1606
1607	/* run==1 means unpack and deallocate. */
1608	run_unpack_ex(RUN_DEALLOCATE, sbi, ino: ni->mi.rno, svcn, evcn, vcn: svcn,
1609	Add2Ptr(attr, roff), run_buf_size: asize - roff);
1610	}
1611
1612	if (ni->attr_list.size) {
1613	run_deallocate(sbi: ni->mi.sbi, run: &ni->attr_list.run, trim: true);
1614	al_destroy(ni);
1615	}
1616
1617	/* Free all subrecords. */
1618	for (node = rb_first(&ni->mi_tree); node;) {
1619	struct rb_node *next = rb_next(node);
1620	struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
1621
1622	clear_rec_inuse(rec: mi->mrec);
1623	mi->dirty = true;
1624	mi_write(mi, wait: 0);
1625
1626	ntfs_mark_rec_free(sbi, rno: mi->rno, is_mft: false);
1627	ni_remove_mi(ni, mi);
1628	mi_put(mi);
1629	node = next;
1630	}
1631
1632	/* Free base record. */
1633	clear_rec_inuse(rec: ni->mi.mrec);
1634	ni->mi.dirty = true;
1635	err = mi_write(mi: &ni->mi, wait: 0);
1636
1637	ntfs_mark_rec_free(sbi, rno: ni->mi.rno, is_mft: false);
1638
1639	return err;
1640	}
1641
1642	/* ni_fname_name
1643	*
1644	* Return: File name attribute by its value.
1645	*/
1646	struct ATTR_FILE_NAME *ni_fname_name(struct ntfs_inode *ni,
1647	const struct le_str *uni,
1648	const struct MFT_REF *home_dir,
1649	struct mft_inode **mi,
1650	struct ATTR_LIST_ENTRY **le)
1651	{
1652	struct ATTRIB *attr = NULL;
1653	struct ATTR_FILE_NAME *fname;
1654
1655	if (le)
1656	*le = NULL;
1657
1658	/* Enumerate all names. */
1659	next:
1660	attr = ni_find_attr(ni, attr, le_o: le, type: ATTR_NAME, NULL, name_len: 0, NULL, mi);
1661	if (!attr)
1662	return NULL;
1663
1664	fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
1665	if (!fname)
1666	goto next;
1667
1668	if (home_dir && memcmp(p: home_dir, q: &fname->home, size: sizeof(*home_dir)))
1669	goto next;
1670
1671	if (!uni)
1672	return fname;
1673
1674	if (uni->len != fname->name_len)
1675	goto next;
1676
1677	if (ntfs_cmp_names(s1: uni->name, l1: uni->len, s2: fname->name, l2: uni->len, NULL,
1678	bothcase: false))
1679	goto next;
1680	return fname;
1681	}
1682
1683	/*
1684	* ni_fname_type
1685	*
1686	* Return: File name attribute with given type.
1687	*/
1688	struct ATTR_FILE_NAME *ni_fname_type(struct ntfs_inode *ni, u8 name_type,
1689	struct mft_inode **mi,
1690	struct ATTR_LIST_ENTRY **le)
1691	{
1692	struct ATTRIB *attr = NULL;
1693	struct ATTR_FILE_NAME *fname;
1694
1695	*le = NULL;
1696
1697	if (name_type == FILE_NAME_POSIX)
1698	return NULL;
1699
1700	/* Enumerate all names. */
1701	for (;;) {
1702	attr = ni_find_attr(ni, attr, le_o: le, type: ATTR_NAME, NULL, name_len: 0, NULL, mi);
1703	if (!attr)
1704	return NULL;
1705
1706	fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
1707	if (fname && name_type == fname->type)
1708	return fname;
1709	}
1710	}
1711
1712	/*
1713	* ni_new_attr_flags
1714	*
1715	* Process compressed/sparsed in special way.
1716	* NOTE: You need to set ni->std_fa = new_fa
1717	* after this function to keep internal structures in consistency.
1718	*/
1719	int ni_new_attr_flags(struct ntfs_inode *ni, enum FILE_ATTRIBUTE new_fa)
1720	{
1721	struct ATTRIB *attr;
1722	struct mft_inode *mi;
1723	__le16 new_aflags;
1724	u32 new_asize;
1725
1726	attr = ni_find_attr(ni, NULL, NULL, type: ATTR_DATA, NULL, name_len: 0, NULL, mi: &mi);
1727	if (!attr)
1728	return -EINVAL;
1729
1730	new_aflags = attr->flags;
1731
1732	if (new_fa & FILE_ATTRIBUTE_SPARSE_FILE)
1733	new_aflags |= ATTR_FLAG_SPARSED;
1734	else
1735	new_aflags &= ~ATTR_FLAG_SPARSED;
1736
1737	if (new_fa & FILE_ATTRIBUTE_COMPRESSED)
1738	new_aflags |= ATTR_FLAG_COMPRESSED;
1739	else
1740	new_aflags &= ~ATTR_FLAG_COMPRESSED;
1741
1742	if (new_aflags == attr->flags)
1743	return 0;
1744
1745	if ((new_aflags & (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED)) ==
1746	(ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED)) {
1747	ntfs_inode_warn(&ni->vfs_inode,
1748	"file can't be sparsed and compressed");
1749	return -EOPNOTSUPP;
1750	}
1751
1752	if (!attr->non_res)
1753	goto out;
1754
1755	if (attr->nres.data_size) {
1756	ntfs_inode_warn(
1757	&ni->vfs_inode,
1758	"one can change sparsed/compressed only for empty files");
1759	return -EOPNOTSUPP;
1760	}
1761
1762	/* Resize nonresident empty attribute in-place only. */
1763	new_asize = (new_aflags & (ATTR_FLAG_COMPRESSED | ATTR_FLAG_SPARSED)) ?
1764	(SIZEOF_NONRESIDENT_EX + 8) :
1765	(SIZEOF_NONRESIDENT + 8);
1766
1767	if (!mi_resize_attr(mi, attr, bytes: new_asize - le32_to_cpu(attr->size)))
1768	return -EOPNOTSUPP;
1769
1770	if (new_aflags & ATTR_FLAG_SPARSED) {
1771	attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
1772	/* Windows uses 16 clusters per frame but supports one cluster per frame too. */
1773	attr->nres.c_unit = 0;
1774	ni->vfs_inode.i_mapping->a_ops = &ntfs_aops;
1775	} else if (new_aflags & ATTR_FLAG_COMPRESSED) {
1776	attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
1777	/* The only allowed: 16 clusters per frame. */
1778	attr->nres.c_unit = NTFS_LZNT_CUNIT;
1779	ni->vfs_inode.i_mapping->a_ops = &ntfs_aops_cmpr;
1780	} else {
1781	attr->name_off = SIZEOF_NONRESIDENT_LE;
1782	/* Normal files. */
1783	attr->nres.c_unit = 0;
1784	ni->vfs_inode.i_mapping->a_ops = &ntfs_aops;
1785	}
1786	attr->nres.run_off = attr->name_off;
1787	out:
1788	attr->flags = new_aflags;
1789	mi->dirty = true;
1790
1791	return 0;
1792	}
1793
1794	/*
1795	* ni_parse_reparse
1796	*
1797	* buffer - memory for reparse buffer header
1798	*/
1799	enum REPARSE_SIGN ni_parse_reparse(struct ntfs_inode *ni, struct ATTRIB *attr,
1800	struct REPARSE_DATA_BUFFER *buffer)
1801	{
1802	const struct REPARSE_DATA_BUFFER *rp = NULL;
1803	u8 bits;
1804	u16 len;
1805	typeof(rp->CompressReparseBuffer) *cmpr;
1806
1807	/* Try to estimate reparse point. */
1808	if (!attr->non_res) {
1809	rp = resident_data_ex(attr, datasize: sizeof(struct REPARSE_DATA_BUFFER));
1810	} else if (le64_to_cpu(attr->nres.data_size) >=
1811	sizeof(struct REPARSE_DATA_BUFFER)) {
1812	struct runs_tree run;
1813
1814	run_init(run: &run);
1815
1816	if (!attr_load_runs_vcn(ni, type: ATTR_REPARSE, NULL, name_len: 0, run: &run, vcn: 0) &&
1817	!ntfs_read_run_nb(sbi: ni->mi.sbi, run: &run, vbo: 0, buf: buffer,
1818	bytes: sizeof(struct REPARSE_DATA_BUFFER),
1819	NULL)) {
1820	rp = buffer;
1821	}
1822
1823	run_close(run: &run);
1824	}
1825
1826	if (!rp)
1827	return REPARSE_NONE;
1828
1829	len = le16_to_cpu(rp->ReparseDataLength);
1830	switch (rp->ReparseTag) {
1831	case (IO_REPARSE_TAG_MICROSOFT | IO_REPARSE_TAG_SYMBOLIC_LINK):
1832	break; /* Symbolic link. */
1833	case IO_REPARSE_TAG_MOUNT_POINT:
1834	break; /* Mount points and junctions. */
1835	case IO_REPARSE_TAG_SYMLINK:
1836	break;
1837	case IO_REPARSE_TAG_COMPRESS:
1838	/*
1839	* WOF - Windows Overlay Filter - Used to compress files with
1840	* LZX/Xpress.
1841	*
1842	* Unlike native NTFS file compression, the Windows
1843	* Overlay Filter supports only read operations. This means
1844	* that it doesn't need to sector-align each compressed chunk,
1845	* so the compressed data can be packed more tightly together.
1846	* If you open the file for writing, the WOF just decompresses
1847	* the entire file, turning it back into a plain file.
1848	*
1849	* Ntfs3 driver decompresses the entire file only on write or
1850	* change size requests.
1851	*/
1852
1853	cmpr = &rp->CompressReparseBuffer;
1854	if (len < sizeof(*cmpr) ||
1855	cmpr->WofVersion != WOF_CURRENT_VERSION ||
1856	cmpr->WofProvider != WOF_PROVIDER_SYSTEM ||
1857	cmpr->ProviderVer != WOF_PROVIDER_CURRENT_VERSION) {
1858	return REPARSE_NONE;
1859	}
1860
1861	switch (cmpr->CompressionFormat) {
1862	case WOF_COMPRESSION_XPRESS4K:
1863	bits = 0xc; // 4k
1864	break;
1865	case WOF_COMPRESSION_XPRESS8K:
1866	bits = 0xd; // 8k
1867	break;
1868	case WOF_COMPRESSION_XPRESS16K:
1869	bits = 0xe; // 16k
1870	break;
1871	case WOF_COMPRESSION_LZX32K:
1872	bits = 0xf; // 32k
1873	break;
1874	default:
1875	bits = 0x10; // 64k
1876	break;
1877	}
1878	ni_set_ext_compress_bits(ni, bits);
1879	return REPARSE_COMPRESSED;
1880
1881	case IO_REPARSE_TAG_DEDUP:
1882	ni->ni_flags |= NI_FLAG_DEDUPLICATED;
1883	return REPARSE_DEDUPLICATED;
1884
1885	default:
1886	if (rp->ReparseTag & IO_REPARSE_TAG_NAME_SURROGATE)
1887	break;
1888
1889	return REPARSE_NONE;
1890	}
1891
1892	if (buffer != rp)
1893	memcpy(buffer, rp, sizeof(struct REPARSE_DATA_BUFFER));
1894
1895	/* Looks like normal symlink. */
1896	return REPARSE_LINK;
1897	}
1898
1899	/*
1900	* ni_fiemap - Helper for file_fiemap().
1901	*
1902	* Assumed ni_lock.
1903	* TODO: Less aggressive locks.
1904	*/
1905	int ni_fiemap(struct ntfs_inode *ni, struct fiemap_extent_info *fieinfo,
1906	__u64 vbo, __u64 len)
1907	{
1908	int err = 0;
1909	struct ntfs_sb_info *sbi = ni->mi.sbi;
1910	u8 cluster_bits = sbi->cluster_bits;
1911	struct runs_tree *run;
1912	struct rw_semaphore *run_lock;
1913	struct ATTRIB *attr;
1914	CLST vcn = vbo >> cluster_bits;
1915	CLST lcn, clen;
1916	u64 valid = ni->i_valid;
1917	u64 lbo, bytes;
1918	u64 end, alloc_size;
1919	size_t idx = -1;
1920	u32 flags;
1921	bool ok;
1922
1923	if (S_ISDIR(ni->vfs_inode.i_mode)) {
1924	run = &ni->dir.alloc_run;
1925	attr = ni_find_attr(ni, NULL, NULL, type: ATTR_ALLOC, name: I30_NAME,
1926	ARRAY_SIZE(I30_NAME), NULL, NULL);
1927	run_lock = &ni->dir.run_lock;
1928	} else {
1929	run = &ni->file.run;
1930	attr = ni_find_attr(ni, NULL, NULL, type: ATTR_DATA, NULL, name_len: 0, NULL,
1931	NULL);
1932	if (!attr) {
1933	err = -EINVAL;
1934	goto out;
1935	}
1936	if (is_attr_compressed(attr)) {
1937	/* Unfortunately cp -r incorrectly treats compressed clusters. */
1938	err = -EOPNOTSUPP;
1939	ntfs_inode_warn(
1940	&ni->vfs_inode,
1941	"fiemap is not supported for compressed file (cp -r)");
1942	goto out;
1943	}
1944	run_lock = &ni->file.run_lock;
1945	}
1946
1947	if (!attr || !attr->non_res) {
1948	err = fiemap_fill_next_extent(
1949	info: fieinfo, logical: 0, phys: 0,
1950	len: attr ? le32_to_cpu(attr->res.data_size) : 0,
1951	FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_LAST |
1952	FIEMAP_EXTENT_MERGED);
1953	goto out;
1954	}
1955
1956	end = vbo + len;
1957	alloc_size = le64_to_cpu(attr->nres.alloc_size);
1958	if (end > alloc_size)
1959	end = alloc_size;
1960
1961	down_read(sem: run_lock);
1962
1963	while (vbo < end) {
1964	if (idx == -1) {
1965	ok = run_lookup_entry(run, vcn, lcn: &lcn, len: &clen, index: &idx);
1966	} else {
1967	CLST vcn_next = vcn;
1968
1969	ok = run_get_entry(run, index: ++idx, vcn: &vcn, lcn: &lcn, len: &clen) &&
1970	vcn == vcn_next;
1971	if (!ok)
1972	vcn = vcn_next;
1973	}
1974
1975	if (!ok) {
1976	up_read(sem: run_lock);
1977	down_write(sem: run_lock);
1978
1979	err = attr_load_runs_vcn(ni, type: attr->type,
1980	name: attr_name(attr),
1981	name_len: attr->name_len, run, vcn);
1982
1983	up_write(sem: run_lock);
1984	down_read(sem: run_lock);
1985
1986	if (err)
1987	break;
1988
1989	ok = run_lookup_entry(run, vcn, lcn: &lcn, len: &clen, index: &idx);
1990
1991	if (!ok) {
1992	err = -EINVAL;
1993	break;
1994	}
1995	}
1996
1997	if (!clen) {
1998	err = -EINVAL; // ?
1999	break;
2000	}
2001
2002	if (lcn == SPARSE_LCN) {
2003	vcn += clen;
2004	vbo = (u64)vcn << cluster_bits;
2005	continue;
2006	}
2007
2008	flags = FIEMAP_EXTENT_MERGED;
2009	if (S_ISDIR(ni->vfs_inode.i_mode)) {
2010	;
2011	} else if (is_attr_compressed(attr)) {
2012	CLST clst_data;
2013
2014	err = attr_is_frame_compressed(
2015	ni, attr, frame: vcn >> attr->nres.c_unit, clst_data: &clst_data);
2016	if (err)
2017	break;
2018	if (clst_data < NTFS_LZNT_CLUSTERS)
2019	flags |= FIEMAP_EXTENT_ENCODED;
2020	} else if (is_attr_encrypted(attr)) {
2021	flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2022	}
2023
2024	vbo = (u64)vcn << cluster_bits;
2025	bytes = (u64)clen << cluster_bits;
2026	lbo = (u64)lcn << cluster_bits;
2027
2028	vcn += clen;
2029
2030	if (vbo + bytes >= end)
2031	bytes = end - vbo;
2032
2033	if (vbo + bytes <= valid) {
2034	;
2035	} else if (vbo >= valid) {
2036	flags |= FIEMAP_EXTENT_UNWRITTEN;
2037	} else {
2038	/* vbo < valid && valid < vbo + bytes */
2039	u64 dlen = valid - vbo;
2040
2041	if (vbo + dlen >= end)
2042	flags |= FIEMAP_EXTENT_LAST;
2043
2044	err = fiemap_fill_next_extent(info: fieinfo, logical: vbo, phys: lbo, len: dlen,
2045	flags);
2046	if (err < 0)
2047	break;
2048	if (err == 1) {
2049	err = 0;
2050	break;
2051	}
2052
2053	vbo = valid;
2054	bytes -= dlen;
2055	if (!bytes)
2056	continue;
2057
2058	lbo += dlen;
2059	flags |= FIEMAP_EXTENT_UNWRITTEN;
2060	}
2061
2062	if (vbo + bytes >= end)
2063	flags |= FIEMAP_EXTENT_LAST;
2064
2065	err = fiemap_fill_next_extent(info: fieinfo, logical: vbo, phys: lbo, len: bytes, flags);
2066	if (err < 0)
2067	break;
2068	if (err == 1) {
2069	err = 0;
2070	break;
2071	}
2072
2073	vbo += bytes;
2074	}
2075
2076	up_read(sem: run_lock);
2077
2078	out:
2079	return err;
2080	}
2081
2082	/*
2083	* ni_readpage_cmpr
2084	*
2085	* When decompressing, we typically obtain more than one page per reference.
2086	* We inject the additional pages into the page cache.
2087	*/
2088	int ni_readpage_cmpr(struct ntfs_inode *ni, struct page *page)
2089	{
2090	int err;
2091	struct ntfs_sb_info *sbi = ni->mi.sbi;
2092	struct address_space *mapping = page->mapping;
2093	pgoff_t index = page->index;
2094	u64 frame_vbo, vbo = (u64)index << PAGE_SHIFT;
2095	struct page **pages = NULL; /* Array of at most 16 pages. stack? */
2096	u8 frame_bits;
2097	CLST frame;
2098	u32 i, idx, frame_size, pages_per_frame;
2099	gfp_t gfp_mask;
2100	struct page *pg;
2101
2102	if (vbo >= i_size_read(inode: &ni->vfs_inode)) {
2103	SetPageUptodate(page);
2104	err = 0;
2105	goto out;
2106	}
2107
2108	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
2109	/* Xpress or LZX. */
2110	frame_bits = ni_ext_compress_bits(ni);
2111	} else {
2112	/* LZNT compression. */
2113	frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
2114	}
2115	frame_size = 1u << frame_bits;
2116	frame = vbo >> frame_bits;
2117	frame_vbo = (u64)frame << frame_bits;
2118	idx = (vbo - frame_vbo) >> PAGE_SHIFT;
2119
2120	pages_per_frame = frame_size >> PAGE_SHIFT;
2121	pages = kcalloc(n: pages_per_frame, size: sizeof(struct page *), GFP_NOFS);
2122	if (!pages) {
2123	err = -ENOMEM;
2124	goto out;
2125	}
2126
2127	pages[idx] = page;
2128	index = frame_vbo >> PAGE_SHIFT;
2129	gfp_mask = mapping_gfp_mask(mapping);
2130
2131	for (i = 0; i < pages_per_frame; i++, index++) {
2132	if (i == idx)
2133	continue;
2134
2135	pg = find_or_create_page(mapping, index, gfp_mask);
2136	if (!pg) {
2137	err = -ENOMEM;
2138	goto out1;
2139	}
2140	pages[i] = pg;
2141	}
2142
2143	err = ni_read_frame(ni, frame_vbo, pages, pages_per_frame);
2144
2145	out1:
2146	if (err)
2147	SetPageError(page);
2148
2149	for (i = 0; i < pages_per_frame; i++) {
2150	pg = pages[i];
2151	if (i == idx || !pg)
2152	continue;
2153	unlock_page(page: pg);
2154	put_page(page: pg);
2155	}
2156
2157	out:
2158	/* At this point, err contains 0 or -EIO depending on the "critical" page. */
2159	kfree(objp: pages);
2160	unlock_page(page);
2161
2162	return err;
2163	}
2164
2165	#ifdef CONFIG_NTFS3_LZX_XPRESS
2166	/*
2167	* ni_decompress_file - Decompress LZX/Xpress compressed file.
2168	*
2169	* Remove ATTR_DATA::WofCompressedData.
2170	* Remove ATTR_REPARSE.
2171	*/
2172	int ni_decompress_file(struct ntfs_inode *ni)
2173	{
2174	struct ntfs_sb_info *sbi = ni->mi.sbi;
2175	struct inode *inode = &ni->vfs_inode;
2176	loff_t i_size = i_size_read(inode);
2177	struct address_space *mapping = inode->i_mapping;
2178	gfp_t gfp_mask = mapping_gfp_mask(mapping);
2179	struct page **pages = NULL;
2180	struct ATTR_LIST_ENTRY *le;
2181	struct ATTRIB *attr;
2182	CLST vcn, cend, lcn, clen, end;
2183	pgoff_t index;
2184	u64 vbo;
2185	u8 frame_bits;
2186	u32 i, frame_size, pages_per_frame, bytes;
2187	struct mft_inode *mi;
2188	int err;
2189
2190	/* Clusters for decompressed data. */
2191	cend = bytes_to_cluster(sbi, size: i_size);
2192
2193	if (!i_size)
2194	goto remove_wof;
2195
2196	/* Check in advance. */
2197	if (cend > wnd_zeroes(wnd: &sbi->used.bitmap)) {
2198	err = -ENOSPC;
2199	goto out;
2200	}
2201
2202	frame_bits = ni_ext_compress_bits(ni);
2203	frame_size = 1u << frame_bits;
2204	pages_per_frame = frame_size >> PAGE_SHIFT;
2205	pages = kcalloc(n: pages_per_frame, size: sizeof(struct page *), GFP_NOFS);
2206	if (!pages) {
2207	err = -ENOMEM;
2208	goto out;
2209	}
2210
2211	/*
2212	* Step 1: Decompress data and copy to new allocated clusters.
2213	*/
2214	index = 0;
2215	for (vbo = 0; vbo < i_size; vbo += bytes) {
2216	u32 nr_pages;
2217	bool new;
2218
2219	if (vbo + frame_size > i_size) {
2220	bytes = i_size - vbo;
2221	nr_pages = (bytes + PAGE_SIZE - 1) >> PAGE_SHIFT;
2222	} else {
2223	nr_pages = pages_per_frame;
2224	bytes = frame_size;
2225	}
2226
2227	end = bytes_to_cluster(sbi, size: vbo + bytes);
2228
2229	for (vcn = vbo >> sbi->cluster_bits; vcn < end; vcn += clen) {
2230	err = attr_data_get_block(ni, vcn, clen: cend - vcn, lcn: &lcn,
2231	len: &clen, new: &new, zero: false);
2232	if (err)
2233	goto out;
2234	}
2235
2236	for (i = 0; i < pages_per_frame; i++, index++) {
2237	struct page *pg;
2238
2239	pg = find_or_create_page(mapping, index, gfp_mask);
2240	if (!pg) {
2241	while (i--) {
2242	unlock_page(page: pages[i]);
2243	put_page(page: pages[i]);
2244	}
2245	err = -ENOMEM;
2246	goto out;
2247	}
2248	pages[i] = pg;
2249	}
2250
2251	err = ni_read_frame(ni, frame_vbo: vbo, pages, pages_per_frame);
2252
2253	if (!err) {
2254	down_read(sem: &ni->file.run_lock);
2255	err = ntfs_bio_pages(sbi, run: &ni->file.run, pages,
2256	nr_pages, vbo, bytes,
2257	op: REQ_OP_WRITE);
2258	up_read(sem: &ni->file.run_lock);
2259	}
2260
2261	for (i = 0; i < pages_per_frame; i++) {
2262	unlock_page(page: pages[i]);
2263	put_page(page: pages[i]);
2264	}
2265
2266	if (err)
2267	goto out;
2268
2269	cond_resched();
2270	}
2271
2272	remove_wof:
2273	/*
2274	* Step 2: Deallocate attributes ATTR_DATA::WofCompressedData
2275	* and ATTR_REPARSE.
2276	*/
2277	attr = NULL;
2278	le = NULL;
2279	while ((attr = ni_enum_attr_ex(ni, attr, le: &le, NULL))) {
2280	CLST svcn, evcn;
2281	u32 asize, roff;
2282
2283	if (attr->type == ATTR_REPARSE) {
2284	struct MFT_REF ref;
2285
2286	mi_get_ref(mi: &ni->mi, ref: &ref);
2287	ntfs_remove_reparse(sbi, rtag: 0, ref: &ref);
2288	}
2289
2290	if (!attr->non_res)
2291	continue;
2292
2293	if (attr->type != ATTR_REPARSE &&
2294	(attr->type != ATTR_DATA ||
2295	attr->name_len != ARRAY_SIZE(WOF_NAME) ||
2296	memcmp(p: attr_name(attr), q: WOF_NAME, size: sizeof(WOF_NAME))))
2297	continue;
2298
2299	svcn = le64_to_cpu(attr->nres.svcn);
2300	evcn = le64_to_cpu(attr->nres.evcn);
2301
2302	if (evcn + 1 <= svcn)
2303	continue;
2304
2305	asize = le32_to_cpu(attr->size);
2306	roff = le16_to_cpu(attr->nres.run_off);
2307
2308	if (roff > asize) {
2309	err = -EINVAL;
2310	goto out;
2311	}
2312
2313	/*run==1 Means unpack and deallocate. */
2314	run_unpack_ex(RUN_DEALLOCATE, sbi, ino: ni->mi.rno, svcn, evcn, vcn: svcn,
2315	Add2Ptr(attr, roff), run_buf_size: asize - roff);
2316	}
2317
2318	/*
2319	* Step 3: Remove attribute ATTR_DATA::WofCompressedData.
2320	*/
2321	err = ni_remove_attr(ni, type: ATTR_DATA, name: WOF_NAME, ARRAY_SIZE(WOF_NAME),
2322	base_only: false, NULL);
2323	if (err)
2324	goto out;
2325
2326	/*
2327	* Step 4: Remove ATTR_REPARSE.
2328	*/
2329	err = ni_remove_attr(ni, type: ATTR_REPARSE, NULL, name_len: 0, base_only: false, NULL);
2330	if (err)
2331	goto out;
2332
2333	/*
2334	* Step 5: Remove sparse flag from data attribute.
2335	*/
2336	attr = ni_find_attr(ni, NULL, NULL, type: ATTR_DATA, NULL, name_len: 0, NULL, mi: &mi);
2337	if (!attr) {
2338	err = -EINVAL;
2339	goto out;
2340	}
2341
2342	if (attr->non_res && is_attr_sparsed(attr)) {
2343	/* Sparsed attribute header is 8 bytes bigger than normal. */
2344	struct MFT_REC *rec = mi->mrec;
2345	u32 used = le32_to_cpu(rec->used);
2346	u32 asize = le32_to_cpu(attr->size);
2347	u16 roff = le16_to_cpu(attr->nres.run_off);
2348	char *rbuf = Add2Ptr(attr, roff);
2349
2350	memmove(rbuf - 8, rbuf, used - PtrOffset(rec, rbuf));
2351	attr->size = cpu_to_le32(asize - 8);
2352	attr->flags &= ~ATTR_FLAG_SPARSED;
2353	attr->nres.run_off = cpu_to_le16(roff - 8);
2354	attr->nres.c_unit = 0;
2355	rec->used = cpu_to_le32(used - 8);
2356	mi->dirty = true;
2357	ni->std_fa &= ~(FILE_ATTRIBUTE_SPARSE_FILE |
2358	FILE_ATTRIBUTE_REPARSE_POINT);
2359
2360	mark_inode_dirty(inode);
2361	}
2362
2363	/* Clear cached flag. */
2364	ni->ni_flags &= ~NI_FLAG_COMPRESSED_MASK;
2365	if (ni->file.offs_page) {
2366	put_page(page: ni->file.offs_page);
2367	ni->file.offs_page = NULL;
2368	}
2369	mapping->a_ops = &ntfs_aops;
2370
2371	out:
2372	kfree(objp: pages);
2373	if (err)
2374	_ntfs_bad_inode(inode);
2375
2376	return err;
2377	}
2378
2379	/*
2380	* decompress_lzx_xpress - External compression LZX/Xpress.
2381	*/
2382	static int decompress_lzx_xpress(struct ntfs_sb_info *sbi, const char *cmpr,
2383	size_t cmpr_size, void *unc, size_t unc_size,
2384	u32 frame_size)
2385	{
2386	int err;
2387	void *ctx;
2388
2389	if (cmpr_size == unc_size) {
2390	/* Frame not compressed. */
2391	memcpy(unc, cmpr, unc_size);
2392	return 0;
2393	}
2394
2395	err = 0;
2396	if (frame_size == 0x8000) {
2397	mutex_lock(&sbi->compress.mtx_lzx);
2398	/* LZX: Frame compressed. */
2399	ctx = sbi->compress.lzx;
2400	if (!ctx) {
2401	/* Lazy initialize LZX decompress context. */
2402	ctx = lzx_allocate_decompressor();
2403	if (!ctx) {
2404	err = -ENOMEM;
2405	goto out1;
2406	}
2407
2408	sbi->compress.lzx = ctx;
2409	}
2410
2411	if (lzx_decompress(d: ctx, compressed_data: cmpr, compressed_size: cmpr_size, uncompressed_data: unc, uncompressed_size: unc_size)) {
2412	/* Treat all errors as "invalid argument". */
2413	err = -EINVAL;
2414	}
2415	out1:
2416	mutex_unlock(lock: &sbi->compress.mtx_lzx);
2417	} else {
2418	/* XPRESS: Frame compressed. */
2419	mutex_lock(&sbi->compress.mtx_xpress);
2420	ctx = sbi->compress.xpress;
2421	if (!ctx) {
2422	/* Lazy initialize Xpress decompress context. */
2423	ctx = xpress_allocate_decompressor();
2424	if (!ctx) {
2425	err = -ENOMEM;
2426	goto out2;
2427	}
2428
2429	sbi->compress.xpress = ctx;
2430	}
2431
2432	if (xpress_decompress(d: ctx, compressed_data: cmpr, compressed_size: cmpr_size, uncompressed_data: unc, uncompressed_size: unc_size)) {
2433	/* Treat all errors as "invalid argument". */
2434	err = -EINVAL;
2435	}
2436	out2:
2437	mutex_unlock(lock: &sbi->compress.mtx_xpress);
2438	}
2439	return err;
2440	}
2441	#endif
2442
2443	/*
2444	* ni_read_frame
2445	*
2446	* Pages - Array of locked pages.
2447	*/
2448	int ni_read_frame(struct ntfs_inode *ni, u64 frame_vbo, struct page **pages,
2449	u32 pages_per_frame)
2450	{
2451	int err;
2452	struct ntfs_sb_info *sbi = ni->mi.sbi;
2453	u8 cluster_bits = sbi->cluster_bits;
2454	char *frame_ondisk = NULL;
2455	char *frame_mem = NULL;
2456	struct page **pages_disk = NULL;
2457	struct ATTR_LIST_ENTRY *le = NULL;
2458	struct runs_tree *run = &ni->file.run;
2459	u64 valid_size = ni->i_valid;
2460	u64 vbo_disk;
2461	size_t unc_size;
2462	u32 frame_size, i, npages_disk, ondisk_size;
2463	struct page *pg;
2464	struct ATTRIB *attr;
2465	CLST frame, clst_data;
2466
2467	/*
2468	* To simplify decompress algorithm do vmap for source
2469	* and target pages.
2470	*/
2471	for (i = 0; i < pages_per_frame; i++)
2472	kmap(page: pages[i]);
2473
2474	frame_size = pages_per_frame << PAGE_SHIFT;
2475	frame_mem = vmap(pages, count: pages_per_frame, VM_MAP, PAGE_KERNEL);
2476	if (!frame_mem) {
2477	err = -ENOMEM;
2478	goto out;
2479	}
2480
2481	attr = ni_find_attr(ni, NULL, le_o: &le, type: ATTR_DATA, NULL, name_len: 0, NULL, NULL);
2482	if (!attr) {
2483	err = -ENOENT;
2484	goto out1;
2485	}
2486
2487	if (!attr->non_res) {
2488	u32 data_size = le32_to_cpu(attr->res.data_size);
2489
2490	memset(frame_mem, 0, frame_size);
2491	if (frame_vbo < data_size) {
2492	ondisk_size = data_size - frame_vbo;
2493	memcpy(frame_mem, resident_data(attr) + frame_vbo,
2494	min(ondisk_size, frame_size));
2495	}
2496	err = 0;
2497	goto out1;
2498	}
2499
2500	if (frame_vbo >= valid_size) {
2501	memset(frame_mem, 0, frame_size);
2502	err = 0;
2503	goto out1;
2504	}
2505
2506	if (ni->ni_flags & NI_FLAG_COMPRESSED_MASK) {
2507	#ifndef CONFIG_NTFS3_LZX_XPRESS
2508	err = -EOPNOTSUPP;
2509	goto out1;
2510	#else
2511	loff_t i_size = i_size_read(inode: &ni->vfs_inode);
2512	u32 frame_bits = ni_ext_compress_bits(ni);
2513	u64 frame64 = frame_vbo >> frame_bits;
2514	u64 frames, vbo_data;
2515
2516	if (frame_size != (1u << frame_bits)) {
2517	err = -EINVAL;
2518	goto out1;
2519	}
2520	switch (frame_size) {
2521	case 0x1000:
2522	case 0x2000:
2523	case 0x4000:
2524	case 0x8000:
2525	break;
2526	default:
2527	/* Unknown compression. */
2528	err = -EOPNOTSUPP;
2529	goto out1;
2530	}
2531
2532	attr = ni_find_attr(ni, attr, le_o: &le, type: ATTR_DATA, name: WOF_NAME,
2533	ARRAY_SIZE(WOF_NAME), NULL, NULL);
2534	if (!attr) {
2535	ntfs_inode_err(
2536	&ni->vfs_inode,
2537	"external compressed file should contains data attribute \"WofCompressedData\"");
2538	err = -EINVAL;
2539	goto out1;
2540	}
2541
2542	if (!attr->non_res) {
2543	run = NULL;
2544	} else {
2545	run = run_alloc();
2546	if (!run) {
2547	err = -ENOMEM;
2548	goto out1;
2549	}
2550	}
2551
2552	frames = (i_size - 1) >> frame_bits;
2553
2554	err = attr_wof_frame_info(ni, attr, run, frame: frame64, frames,
2555	frame_bits, ondisk_size: &ondisk_size, vbo_data: &vbo_data);
2556	if (err)
2557	goto out2;
2558
2559	if (frame64 == frames) {
2560	unc_size = 1 + ((i_size - 1) & (frame_size - 1));
2561	ondisk_size = attr_size(attr) - vbo_data;
2562	} else {
2563	unc_size = frame_size;
2564	}
2565
2566	if (ondisk_size > frame_size) {
2567	err = -EINVAL;
2568	goto out2;
2569	}
2570
2571	if (!attr->non_res) {
2572	if (vbo_data + ondisk_size >
2573	le32_to_cpu(attr->res.data_size)) {
2574	err = -EINVAL;
2575	goto out1;
2576	}
2577
2578	err = decompress_lzx_xpress(
2579	sbi, Add2Ptr(resident_data(attr), vbo_data),
2580	cmpr_size: ondisk_size, unc: frame_mem, unc_size, frame_size);
2581	goto out1;
2582	}
2583	vbo_disk = vbo_data;
2584	/* Load all runs to read [vbo_disk-vbo_to). */
2585	err = attr_load_runs_range(ni, type: ATTR_DATA, name: WOF_NAME,
2586	ARRAY_SIZE(WOF_NAME), run, from: vbo_disk,
2587	to: vbo_data + ondisk_size);
2588	if (err)
2589	goto out2;
2590	npages_disk = (ondisk_size + (vbo_disk & (PAGE_SIZE - 1)) +
2591	PAGE_SIZE - 1) >>
2592	PAGE_SHIFT;
2593	#endif
2594	} else if (is_attr_compressed(attr)) {
2595	/* LZNT compression. */
2596	if (sbi->cluster_size > NTFS_LZNT_MAX_CLUSTER) {
2597	err = -EOPNOTSUPP;
2598	goto out1;
2599	}
2600
2601	if (attr->nres.c_unit != NTFS_LZNT_CUNIT) {
2602	err = -EOPNOTSUPP;
2603	goto out1;
2604	}
2605
2606	down_write(sem: &ni->file.run_lock);
2607	run_truncate_around(run, le64_to_cpu(attr->nres.svcn));
2608	frame = frame_vbo >> (cluster_bits + NTFS_LZNT_CUNIT);
2609	err = attr_is_frame_compressed(ni, attr, frame, clst_data: &clst_data);
2610	up_write(sem: &ni->file.run_lock);
2611	if (err)
2612	goto out1;
2613
2614	if (!clst_data) {
2615	memset(frame_mem, 0, frame_size);
2616	goto out1;
2617	}
2618
2619	frame_size = sbi->cluster_size << NTFS_LZNT_CUNIT;
2620	ondisk_size = clst_data << cluster_bits;
2621
2622	if (clst_data >= NTFS_LZNT_CLUSTERS) {
2623	/* Frame is not compressed. */
2624	down_read(sem: &ni->file.run_lock);
2625	err = ntfs_bio_pages(sbi, run, pages, nr_pages: pages_per_frame,
2626	vbo: frame_vbo, bytes: ondisk_size,
2627	op: REQ_OP_READ);
2628	up_read(sem: &ni->file.run_lock);
2629	goto out1;
2630	}
2631	vbo_disk = frame_vbo;
2632	npages_disk = (ondisk_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2633	} else {
2634	__builtin_unreachable();
2635	err = -EINVAL;
2636	goto out1;
2637	}
2638
2639	pages_disk = kzalloc(size: npages_disk * sizeof(struct page *), GFP_NOFS);
2640	if (!pages_disk) {
2641	err = -ENOMEM;
2642	goto out2;
2643	}
2644
2645	for (i = 0; i < npages_disk; i++) {
2646	pg = alloc_page(GFP_KERNEL);
2647	if (!pg) {
2648	err = -ENOMEM;
2649	goto out3;
2650	}
2651	pages_disk[i] = pg;
2652	lock_page(page: pg);
2653	kmap(page: pg);
2654	}
2655
2656	/* Read 'ondisk_size' bytes from disk. */
2657	down_read(sem: &ni->file.run_lock);
2658	err = ntfs_bio_pages(sbi, run, pages: pages_disk, nr_pages: npages_disk, vbo: vbo_disk,
2659	bytes: ondisk_size, op: REQ_OP_READ);
2660	up_read(sem: &ni->file.run_lock);
2661	if (err)
2662	goto out3;
2663
2664	/*
2665	* To simplify decompress algorithm do vmap for source and target pages.
2666	*/
2667	frame_ondisk = vmap(pages: pages_disk, count: npages_disk, VM_MAP, PAGE_KERNEL_RO);
2668	if (!frame_ondisk) {
2669	err = -ENOMEM;
2670	goto out3;
2671	}
2672
2673	/* Decompress: Frame_ondisk -> frame_mem. */
2674	#ifdef CONFIG_NTFS3_LZX_XPRESS
2675	if (run != &ni->file.run) {
2676	/* LZX or XPRESS */
2677	err = decompress_lzx_xpress(
2678	sbi, cmpr: frame_ondisk + (vbo_disk & (PAGE_SIZE - 1)),
2679	cmpr_size: ondisk_size, unc: frame_mem, unc_size, frame_size);
2680	} else
2681	#endif
2682	{
2683	/* LZNT - Native NTFS compression. */
2684	unc_size = decompress_lznt(compressed: frame_ondisk, compressed_size: ondisk_size, uncompressed: frame_mem,
2685	uncompressed_size: frame_size);
2686	if ((ssize_t)unc_size < 0)
2687	err = unc_size;
2688	else if (!unc_size || unc_size > frame_size)
2689	err = -EINVAL;
2690	}
2691	if (!err && valid_size < frame_vbo + frame_size) {
2692	size_t ok = valid_size - frame_vbo;
2693
2694	memset(frame_mem + ok, 0, frame_size - ok);
2695	}
2696
2697	vunmap(addr: frame_ondisk);
2698
2699	out3:
2700	for (i = 0; i < npages_disk; i++) {
2701	pg = pages_disk[i];
2702	if (pg) {
2703	kunmap(page: pg);
2704	unlock_page(page: pg);
2705	put_page(page: pg);
2706	}
2707	}
2708	kfree(objp: pages_disk);
2709
2710	out2:
2711	#ifdef CONFIG_NTFS3_LZX_XPRESS
2712	if (run != &ni->file.run)
2713	run_free(run);
2714	#endif
2715	out1:
2716	vunmap(addr: frame_mem);
2717	out:
2718	for (i = 0; i < pages_per_frame; i++) {
2719	pg = pages[i];
2720	kunmap(page: pg);
2721	ClearPageError(page: pg);
2722	SetPageUptodate(pg);
2723	}
2724
2725	return err;
2726	}
2727
2728	/*
2729	* ni_write_frame
2730	*
2731	* Pages - Array of locked pages.
2732	*/
2733	int ni_write_frame(struct ntfs_inode *ni, struct page **pages,
2734	u32 pages_per_frame)
2735	{
2736	int err;
2737	struct ntfs_sb_info *sbi = ni->mi.sbi;
2738	u8 frame_bits = NTFS_LZNT_CUNIT + sbi->cluster_bits;
2739	u32 frame_size = sbi->cluster_size << NTFS_LZNT_CUNIT;
2740	u64 frame_vbo = (u64)pages[0]->index << PAGE_SHIFT;
2741	CLST frame = frame_vbo >> frame_bits;
2742	char *frame_ondisk = NULL;
2743	struct page **pages_disk = NULL;
2744	struct ATTR_LIST_ENTRY *le = NULL;
2745	char *frame_mem;
2746	struct ATTRIB *attr;
2747	struct mft_inode *mi;
2748	u32 i;
2749	struct page *pg;
2750	size_t compr_size, ondisk_size;
2751	struct lznt *lznt;
2752
2753	attr = ni_find_attr(ni, NULL, le_o: &le, type: ATTR_DATA, NULL, name_len: 0, NULL, mi: &mi);
2754	if (!attr) {
2755	err = -ENOENT;
2756	goto out;
2757	}
2758
2759	if (WARN_ON(!is_attr_compressed(attr))) {
2760	err = -EINVAL;
2761	goto out;
2762	}
2763
2764	if (sbi->cluster_size > NTFS_LZNT_MAX_CLUSTER) {
2765	err = -EOPNOTSUPP;
2766	goto out;
2767	}
2768
2769	if (!attr->non_res) {
2770	down_write(sem: &ni->file.run_lock);
2771	err = attr_make_nonresident(ni, attr, le, mi,
2772	le32_to_cpu(attr->res.data_size),
2773	run: &ni->file.run, ins_attr: &attr, page: pages[0]);
2774	up_write(sem: &ni->file.run_lock);
2775	if (err)
2776	goto out;
2777	}
2778
2779	if (attr->nres.c_unit != NTFS_LZNT_CUNIT) {
2780	err = -EOPNOTSUPP;
2781	goto out;
2782	}
2783
2784	pages_disk = kcalloc(n: pages_per_frame, size: sizeof(struct page *), GFP_NOFS);
2785	if (!pages_disk) {
2786	err = -ENOMEM;
2787	goto out;
2788	}
2789
2790	for (i = 0; i < pages_per_frame; i++) {
2791	pg = alloc_page(GFP_KERNEL);
2792	if (!pg) {
2793	err = -ENOMEM;
2794	goto out1;
2795	}
2796	pages_disk[i] = pg;
2797	lock_page(page: pg);
2798	kmap(page: pg);
2799	}
2800
2801	/* To simplify compress algorithm do vmap for source and target pages. */
2802	frame_ondisk = vmap(pages: pages_disk, count: pages_per_frame, VM_MAP, PAGE_KERNEL);
2803	if (!frame_ondisk) {
2804	err = -ENOMEM;
2805	goto out1;
2806	}
2807
2808	for (i = 0; i < pages_per_frame; i++)
2809	kmap(page: pages[i]);
2810
2811	/* Map in-memory frame for read-only. */
2812	frame_mem = vmap(pages, count: pages_per_frame, VM_MAP, PAGE_KERNEL_RO);
2813	if (!frame_mem) {
2814	err = -ENOMEM;
2815	goto out2;
2816	}
2817
2818	mutex_lock(&sbi->compress.mtx_lznt);
2819	lznt = NULL;
2820	if (!sbi->compress.lznt) {
2821	/*
2822	* LZNT implements two levels of compression:
2823	* 0 - Standard compression
2824	* 1 - Best compression, requires a lot of cpu
2825	* use mount option?
2826	*/
2827	lznt = get_lznt_ctx(level: 0);
2828	if (!lznt) {
2829	mutex_unlock(lock: &sbi->compress.mtx_lznt);
2830	err = -ENOMEM;
2831	goto out3;
2832	}
2833
2834	sbi->compress.lznt = lznt;
2835	lznt = NULL;
2836	}
2837
2838	/* Compress: frame_mem -> frame_ondisk */
2839	compr_size = compress_lznt(uncompressed: frame_mem, uncompressed_size: frame_size, compressed: frame_ondisk,
2840	compressed_size: frame_size, ctx: sbi->compress.lznt);
2841	mutex_unlock(lock: &sbi->compress.mtx_lznt);
2842	kfree(objp: lznt);
2843
2844	if (compr_size + sbi->cluster_size > frame_size) {
2845	/* Frame is not compressed. */
2846	compr_size = frame_size;
2847	ondisk_size = frame_size;
2848	} else if (compr_size) {
2849	/* Frame is compressed. */
2850	ondisk_size = ntfs_up_cluster(sbi, size: compr_size);
2851	memset(frame_ondisk + compr_size, 0, ondisk_size - compr_size);
2852	} else {
2853	/* Frame is sparsed. */
2854	ondisk_size = 0;
2855	}
2856
2857	down_write(sem: &ni->file.run_lock);
2858	run_truncate_around(run: &ni->file.run, le64_to_cpu(attr->nres.svcn));
2859	err = attr_allocate_frame(ni, frame, compr_size, new_valid: ni->i_valid);
2860	up_write(sem: &ni->file.run_lock);
2861	if (err)
2862	goto out2;
2863
2864	if (!ondisk_size)
2865	goto out2;
2866
2867	down_read(sem: &ni->file.run_lock);
2868	err = ntfs_bio_pages(sbi, run: &ni->file.run,
2869	pages: ondisk_size < frame_size ? pages_disk : pages,
2870	nr_pages: pages_per_frame, vbo: frame_vbo, bytes: ondisk_size,
2871	op: REQ_OP_WRITE);
2872	up_read(sem: &ni->file.run_lock);
2873
2874	out3:
2875	vunmap(addr: frame_mem);
2876
2877	out2:
2878	for (i = 0; i < pages_per_frame; i++)
2879	kunmap(page: pages[i]);
2880
2881	vunmap(addr: frame_ondisk);
2882	out1:
2883	for (i = 0; i < pages_per_frame; i++) {
2884	pg = pages_disk[i];
2885	if (pg) {
2886	kunmap(page: pg);
2887	unlock_page(page: pg);
2888	put_page(page: pg);
2889	}
2890	}
2891	kfree(objp: pages_disk);
2892	out:
2893	return err;
2894	}
2895
2896	/*
2897	* ni_remove_name - Removes name 'de' from MFT and from directory.
2898	* 'de2' and 'undo_step' are used to restore MFT/dir, if error occurs.
2899	*/
2900	int ni_remove_name(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
2901	struct NTFS_DE *de, struct NTFS_DE **de2, int *undo_step)
2902	{
2903	int err;
2904	struct ntfs_sb_info *sbi = ni->mi.sbi;
2905	struct ATTR_FILE_NAME *de_name = (struct ATTR_FILE_NAME *)(de + 1);
2906	struct ATTR_FILE_NAME *fname;
2907	struct ATTR_LIST_ENTRY *le;
2908	struct mft_inode *mi;
2909	u16 de_key_size = le16_to_cpu(de->key_size);
2910	u8 name_type;
2911
2912	*undo_step = 0;
2913
2914	/* Find name in record. */
2915	mi_get_ref(mi: &dir_ni->mi, ref: &de_name->home);
2916
2917	fname = ni_fname_name(ni, uni: (struct le_str *)&de_name->name_len,
2918	home_dir: &de_name->home, mi: &mi, le: &le);
2919	if (!fname)
2920	return -ENOENT;
2921
2922	memcpy(&de_name->dup, &fname->dup, sizeof(struct NTFS_DUP_INFO));
2923	name_type = paired_name(type: fname->type);
2924
2925	/* Mark ntfs as dirty. It will be cleared at umount. */
2926	ntfs_set_state(sbi, dirty: NTFS_DIRTY_DIRTY);
2927
2928	/* Step 1: Remove name from directory. */
2929	err = indx_delete_entry(indx: &dir_ni->dir, ni: dir_ni, key: fname, key_len: de_key_size, param: sbi);
2930	if (err)
2931	return err;
2932
2933	/* Step 2: Remove name from MFT. */
2934	ni_remove_attr_le(ni, attr: attr_from_name(fname), mi, le);
2935
2936	*undo_step = 2;
2937
2938	/* Get paired name. */
2939	fname = ni_fname_type(ni, name_type, mi: &mi, le: &le);
2940	if (fname) {
2941	u16 de2_key_size = fname_full_size(fname);
2942
2943	*de2 = Add2Ptr(de, 1024);
2944	(*de2)->key_size = cpu_to_le16(de2_key_size);
2945
2946	memcpy(*de2 + 1, fname, de2_key_size);
2947
2948	/* Step 3: Remove paired name from directory. */
2949	err = indx_delete_entry(indx: &dir_ni->dir, ni: dir_ni, key: fname,
2950	key_len: de2_key_size, param: sbi);
2951	if (err)
2952	return err;
2953
2954	/* Step 4: Remove paired name from MFT. */
2955	ni_remove_attr_le(ni, attr: attr_from_name(fname), mi, le);
2956
2957	*undo_step = 4;
2958	}
2959	return 0;
2960	}
2961
2962	/*
2963	* ni_remove_name_undo - Paired function for ni_remove_name.
2964	*
2965	* Return: True if ok
2966	*/
2967	bool ni_remove_name_undo(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
2968	struct NTFS_DE *de, struct NTFS_DE *de2, int undo_step)
2969	{
2970	struct ntfs_sb_info *sbi = ni->mi.sbi;
2971	struct ATTRIB *attr;
2972	u16 de_key_size;
2973
2974	switch (undo_step) {
2975	case 4:
2976	de_key_size = le16_to_cpu(de2->key_size);
2977	if (ni_insert_resident(ni, data_size: de_key_size, type: ATTR_NAME, NULL, name_len: 0,
2978	new_attr: &attr, NULL, NULL))
2979	return false;
2980	memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de2 + 1, de_key_size);
2981
2982	mi_get_ref(mi: &ni->mi, ref: &de2->ref);
2983	de2->size = cpu_to_le16(ALIGN(de_key_size, 8) +
2984	sizeof(struct NTFS_DE));
2985	de2->flags = 0;
2986	de2->res = 0;
2987
2988	if (indx_insert_entry(indx: &dir_ni->dir, ni: dir_ni, new_de: de2, param: sbi, NULL, undo: 1))
2989	return false;
2990	fallthrough;
2991
2992	case 2:
2993	de_key_size = le16_to_cpu(de->key_size);
2994
2995	if (ni_insert_resident(ni, data_size: de_key_size, type: ATTR_NAME, NULL, name_len: 0,
2996	new_attr: &attr, NULL, NULL))
2997	return false;
2998
2999	memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de + 1, de_key_size);
3000	mi_get_ref(mi: &ni->mi, ref: &de->ref);
3001
3002	if (indx_insert_entry(indx: &dir_ni->dir, ni: dir_ni, new_de: de, param: sbi, NULL, undo: 1))
3003	return false;
3004	}
3005
3006	return true;
3007	}
3008
3009	/*
3010	* ni_add_name - Add new name into MFT and into directory.
3011	*/
3012	int ni_add_name(struct ntfs_inode *dir_ni, struct ntfs_inode *ni,
3013	struct NTFS_DE *de)
3014	{
3015	int err;
3016	struct ntfs_sb_info *sbi = ni->mi.sbi;
3017	struct ATTRIB *attr;
3018	struct ATTR_LIST_ENTRY *le;
3019	struct mft_inode *mi;
3020	struct ATTR_FILE_NAME *fname;
3021	struct ATTR_FILE_NAME *de_name = (struct ATTR_FILE_NAME *)(de + 1);
3022	u16 de_key_size = le16_to_cpu(de->key_size);
3023
3024	if (sbi->options->windows_names &&
3025	!valid_windows_name(sbi, name: (struct le_str *)&de_name->name_len))
3026	return -EINVAL;
3027
3028	/* If option "hide_dot_files" then set hidden attribute for dot files. */
3029	if (ni->mi.sbi->options->hide_dot_files) {
3030	if (de_name->name_len > 0 &&
3031	le16_to_cpu(de_name->name[0]) == '.')
3032	ni->std_fa |= FILE_ATTRIBUTE_HIDDEN;
3033	else
3034	ni->std_fa &= ~FILE_ATTRIBUTE_HIDDEN;
3035	}
3036
3037	mi_get_ref(mi: &ni->mi, ref: &de->ref);
3038	mi_get_ref(mi: &dir_ni->mi, ref: &de_name->home);
3039
3040	/* Fill duplicate from any ATTR_NAME. */
3041	fname = ni_fname_name(ni, NULL, NULL, NULL, NULL);
3042	if (fname)
3043	memcpy(&de_name->dup, &fname->dup, sizeof(fname->dup));
3044	de_name->dup.fa = ni->std_fa;
3045
3046	/* Insert new name into MFT. */
3047	err = ni_insert_resident(ni, data_size: de_key_size, type: ATTR_NAME, NULL, name_len: 0, new_attr: &attr,
3048	mi: &mi, le: &le);
3049	if (err)
3050	return err;
3051
3052	memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), de_name, de_key_size);
3053
3054	/* Insert new name into directory. */
3055	err = indx_insert_entry(indx: &dir_ni->dir, ni: dir_ni, new_de: de, param: sbi, NULL, undo: 0);
3056	if (err)
3057	ni_remove_attr_le(ni, attr, mi, le);
3058
3059	return err;
3060	}
3061
3062	/*
3063	* ni_rename - Remove one name and insert new name.
3064	*/
3065	int ni_rename(struct ntfs_inode *dir_ni, struct ntfs_inode *new_dir_ni,
3066	struct ntfs_inode *ni, struct NTFS_DE *de, struct NTFS_DE *new_de,
3067	bool *is_bad)
3068	{
3069	int err;
3070	struct NTFS_DE *de2 = NULL;
3071	int undo = 0;
3072
3073	/*
3074	* There are two possible ways to rename:
3075	* 1) Add new name and remove old name.
3076	* 2) Remove old name and add new name.
3077	*
3078	* In most cases (not all!) adding new name into MFT and into directory can
3079	* allocate additional cluster(s).
3080	* Second way may result to bad inode if we can't add new name
3081	* and then can't restore (add) old name.
3082	*/
3083
3084	/*
3085	* Way 1 - Add new + remove old.
3086	*/
3087	err = ni_add_name(dir_ni: new_dir_ni, ni, de: new_de);
3088	if (!err) {
3089	err = ni_remove_name(dir_ni, ni, de, de2: &de2, undo_step: &undo);
3090	if (err && ni_remove_name(dir_ni: new_dir_ni, ni, de: new_de, de2: &de2, undo_step: &undo))
3091	*is_bad = true;
3092	}
3093
3094	/*
3095	* Way 2 - Remove old + add new.
3096	*/
3097	/*
3098	* err = ni_remove_name(dir_ni, ni, de, &de2, &undo);
3099	* if (!err) {
3100	* err = ni_add_name(new_dir_ni, ni, new_de);
3101	* if (err && !ni_remove_name_undo(dir_ni, ni, de, de2, undo))
3102	* *is_bad = true;
3103	* }
3104	*/
3105
3106	return err;
3107	}
3108
3109	/*
3110	* ni_is_dirty - Return: True if 'ni' requires ni_write_inode.
3111	*/
3112	bool ni_is_dirty(struct inode *inode)
3113	{
3114	struct ntfs_inode *ni = ntfs_i(inode);
3115	struct rb_node *node;
3116
3117	if (ni->mi.dirty || ni->attr_list.dirty ||
3118	(ni->ni_flags & NI_FLAG_UPDATE_PARENT))
3119	return true;
3120
3121	for (node = rb_first(&ni->mi_tree); node; node = rb_next(node)) {
3122	if (rb_entry(node, struct mft_inode, node)->dirty)
3123	return true;
3124	}
3125
3126	return false;
3127	}
3128
3129	/*
3130	* ni_update_parent
3131	*
3132	* Update duplicate info of ATTR_FILE_NAME in MFT and in parent directories.
3133	*/
3134	static bool ni_update_parent(struct ntfs_inode *ni, struct NTFS_DUP_INFO *dup,
3135	int sync)
3136	{
3137	struct ATTRIB *attr;
3138	struct mft_inode *mi;
3139	struct ATTR_LIST_ENTRY *le = NULL;
3140	struct ntfs_sb_info *sbi = ni->mi.sbi;
3141	struct super_block *sb = sbi->sb;
3142	bool re_dirty = false;
3143
3144	if (ni->mi.mrec->flags & RECORD_FLAG_DIR) {
3145	dup->fa |= FILE_ATTRIBUTE_DIRECTORY;
3146	attr = NULL;
3147	dup->alloc_size = 0;
3148	dup->data_size = 0;
3149	} else {
3150	dup->fa &= ~FILE_ATTRIBUTE_DIRECTORY;
3151
3152	attr = ni_find_attr(ni, NULL, le_o: &le, type: ATTR_DATA, NULL, name_len: 0, NULL,
3153	mi: &mi);
3154	if (!attr) {
3155	dup->alloc_size = dup->data_size = 0;
3156	} else if (!attr->non_res) {
3157	u32 data_size = le32_to_cpu(attr->res.data_size);
3158
3159	dup->alloc_size = cpu_to_le64(ALIGN(data_size, 8));
3160	dup->data_size = cpu_to_le64(data_size);
3161	} else {
3162	u64 new_valid = ni->i_valid;
3163	u64 data_size = le64_to_cpu(attr->nres.data_size);
3164	__le64 valid_le;
3165
3166	dup->alloc_size = is_attr_ext(attr) ?
3167	attr->nres.total_size :
3168	attr->nres.alloc_size;
3169	dup->data_size = attr->nres.data_size;
3170
3171	if (new_valid > data_size)
3172	new_valid = data_size;
3173
3174	valid_le = cpu_to_le64(new_valid);
3175	if (valid_le != attr->nres.valid_size) {
3176	attr->nres.valid_size = valid_le;
3177	mi->dirty = true;
3178	}
3179	}
3180	}
3181
3182	/* TODO: Fill reparse info. */
3183	dup->reparse = 0;
3184	dup->ea_size = 0;
3185
3186	if (ni->ni_flags & NI_FLAG_EA) {
3187	attr = ni_find_attr(ni, attr, le_o: &le, type: ATTR_EA_INFO, NULL, name_len: 0, NULL,
3188	NULL);
3189	if (attr) {
3190	const struct EA_INFO *info;
3191
3192	info = resident_data_ex(attr, datasize: sizeof(struct EA_INFO));
3193	/* If ATTR_EA_INFO exists 'info' can't be NULL. */
3194	if (info)
3195	dup->ea_size = info->size_pack;
3196	}
3197	}
3198
3199	attr = NULL;
3200	le = NULL;
3201
3202	while ((attr = ni_find_attr(ni, attr, le_o: &le, type: ATTR_NAME, NULL, name_len: 0, NULL,
3203	mi: &mi))) {
3204	struct inode *dir;
3205	struct ATTR_FILE_NAME *fname;
3206
3207	fname = resident_data_ex(attr, SIZEOF_ATTRIBUTE_FILENAME);
3208	if (!fname || !memcmp(p: &fname->dup, q: dup, size: sizeof(fname->dup)))
3209	continue;
3210
3211	/* Check simple case when parent inode equals current inode. */
3212	if (ino_get(ref: &fname->home) == ni->vfs_inode.i_ino) {
3213	ntfs_set_state(sbi, dirty: NTFS_DIRTY_ERROR);
3214	continue;
3215	}
3216
3217	/* ntfs_iget5 may sleep. */
3218	dir = ntfs_iget5(sb, ref: &fname->home, NULL);
3219	if (IS_ERR(ptr: dir)) {
3220	ntfs_inode_warn(
3221	&ni->vfs_inode,
3222	"failed to open parent directory r=%lx to update",
3223	(long)ino_get(&fname->home));
3224	continue;
3225	}
3226
3227	if (!is_bad_inode(dir)) {
3228	struct ntfs_inode *dir_ni = ntfs_i(inode: dir);
3229
3230	if (!ni_trylock(ni: dir_ni)) {
3231	re_dirty = true;
3232	} else {
3233	indx_update_dup(ni: dir_ni, sbi, fname, dup, sync);
3234	ni_unlock(ni: dir_ni);
3235	memcpy(&fname->dup, dup, sizeof(fname->dup));
3236	mi->dirty = true;
3237	}
3238	}
3239	iput(dir);
3240	}
3241
3242	return re_dirty;
3243	}
3244
3245	/*
3246	* ni_write_inode - Write MFT base record and all subrecords to disk.
3247	*/
3248	int ni_write_inode(struct inode *inode, int sync, const char *hint)
3249	{
3250	int err = 0, err2;
3251	struct ntfs_inode *ni = ntfs_i(inode);
3252	struct super_block *sb = inode->i_sb;
3253	struct ntfs_sb_info *sbi = sb->s_fs_info;
3254	bool re_dirty = false;
3255	struct ATTR_STD_INFO *std;
3256	struct rb_node *node, *next;
3257	struct NTFS_DUP_INFO dup;
3258
3259	if (is_bad_inode(inode) || sb_rdonly(sb))
3260	return 0;
3261
3262	if (unlikely(ntfs3_forced_shutdown(sb)))
3263	return -EIO;
3264
3265	if (!ni_trylock(ni)) {
3266	/* 'ni' is under modification, skip for now. */
3267	mark_inode_dirty_sync(inode);
3268	return 0;
3269	}
3270
3271	if (!ni->mi.mrec)
3272	goto out;
3273
3274	if (is_rec_inuse(rec: ni->mi.mrec) &&
3275	!(sbi->flags & NTFS_FLAGS_LOG_REPLAYING) && inode->i_nlink) {
3276	bool modified = false;
3277	struct timespec64 ts;
3278
3279	/* Update times in standard attribute. */
3280	std = ni_std(ni);
3281	if (!std) {
3282	err = -EINVAL;
3283	goto out;
3284	}
3285
3286	/* Update the access times if they have changed. */
3287	ts = inode_get_mtime(inode);
3288	dup.m_time = kernel2nt(ts: &ts);
3289	if (std->m_time != dup.m_time) {
3290	std->m_time = dup.m_time;
3291	modified = true;
3292	}
3293
3294	ts = inode_get_ctime(inode);
3295	dup.c_time = kernel2nt(ts: &ts);
3296	if (std->c_time != dup.c_time) {
3297	std->c_time = dup.c_time;
3298	modified = true;
3299	}
3300
3301	ts = inode_get_atime(inode);
3302	dup.a_time = kernel2nt(ts: &ts);
3303	if (std->a_time != dup.a_time) {
3304	std->a_time = dup.a_time;
3305	modified = true;
3306	}
3307
3308	dup.fa = ni->std_fa;
3309	if (std->fa != dup.fa) {
3310	std->fa = dup.fa;
3311	modified = true;
3312	}
3313
3314	/* std attribute is always in primary MFT record. */
3315	if (modified)
3316	ni->mi.dirty = true;
3317
3318	if (!ntfs_is_meta_file(sbi, rno: inode->i_ino) &&
3319	(modified || (ni->ni_flags & NI_FLAG_UPDATE_PARENT))
3320	/* Avoid __wait_on_freeing_inode(inode). */
3321	&& (sb->s_flags & SB_ACTIVE)) {
3322	dup.cr_time = std->cr_time;
3323	/* Not critical if this function fail. */
3324	re_dirty = ni_update_parent(ni, dup: &dup, sync);
3325
3326	if (re_dirty)
3327	ni->ni_flags |= NI_FLAG_UPDATE_PARENT;
3328	else
3329	ni->ni_flags &= ~NI_FLAG_UPDATE_PARENT;
3330	}
3331
3332	/* Update attribute list. */
3333	if (ni->attr_list.size && ni->attr_list.dirty) {
3334	if (inode->i_ino != MFT_REC_MFT || sync) {
3335	err = ni_try_remove_attr_list(ni);
3336	if (err)
3337	goto out;
3338	}
3339
3340	err = al_update(ni, sync);
3341	if (err)
3342	goto out;
3343	}
3344	}
3345
3346	for (node = rb_first(&ni->mi_tree); node; node = next) {
3347	struct mft_inode *mi = rb_entry(node, struct mft_inode, node);
3348	bool is_empty;
3349
3350	next = rb_next(node);
3351
3352	if (!mi->dirty)
3353	continue;
3354
3355	is_empty = !mi_enum_attr(mi, NULL);
3356
3357	if (is_empty)
3358	clear_rec_inuse(rec: mi->mrec);
3359
3360	err2 = mi_write(mi, wait: sync);
3361	if (!err && err2)
3362	err = err2;
3363
3364	if (is_empty) {
3365	ntfs_mark_rec_free(sbi, rno: mi->rno, is_mft: false);
3366	rb_erase(node, &ni->mi_tree);
3367	mi_put(mi);
3368	}
3369	}
3370
3371	if (ni->mi.dirty) {
3372	err2 = mi_write(mi: &ni->mi, wait: sync);
3373	if (!err && err2)
3374	err = err2;
3375	}
3376	out:
3377	ni_unlock(ni);
3378
3379	if (err) {
3380	ntfs_inode_err(inode, "%s failed, %d.", hint, err);
3381	ntfs_set_state(sbi, dirty: NTFS_DIRTY_ERROR);
3382	return err;
3383	}
3384
3385	if (re_dirty)
3386	mark_inode_dirty_sync(inode);
3387
3388	return 0;
3389	}
3390