Linux 5.18
[linux-block.git] / fs / ntfs3 / inode.c
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/buffer_head.h>
9 #include <linux/fs.h>
10 #include <linux/mpage.h>
11 #include <linux/namei.h>
12 #include <linux/nls.h>
13 #include <linux/uio.h>
14 #include <linux/writeback.h>
15
16 #include "debug.h"
17 #include "ntfs.h"
18 #include "ntfs_fs.h"
19
20 /*
21  * ntfs_read_mft - Read record and parses MFT.
22  */
23 static struct inode *ntfs_read_mft(struct inode *inode,
24                                    const struct cpu_str *name,
25                                    const struct MFT_REF *ref)
26 {
27         int err = 0;
28         struct ntfs_inode *ni = ntfs_i(inode);
29         struct super_block *sb = inode->i_sb;
30         struct ntfs_sb_info *sbi = sb->s_fs_info;
31         mode_t mode = 0;
32         struct ATTR_STD_INFO5 *std5 = NULL;
33         struct ATTR_LIST_ENTRY *le;
34         struct ATTRIB *attr;
35         bool is_match = false;
36         bool is_root = false;
37         bool is_dir;
38         unsigned long ino = inode->i_ino;
39         u32 rp_fa = 0, asize, t32;
40         u16 roff, rsize, names = 0;
41         const struct ATTR_FILE_NAME *fname = NULL;
42         const struct INDEX_ROOT *root;
43         struct REPARSE_DATA_BUFFER rp; // 0x18 bytes
44         u64 t64;
45         struct MFT_REC *rec;
46         struct runs_tree *run;
47
48         inode->i_op = NULL;
49         /* Setup 'uid' and 'gid' */
50         inode->i_uid = sbi->options->fs_uid;
51         inode->i_gid = sbi->options->fs_gid;
52
53         err = mi_init(&ni->mi, sbi, ino);
54         if (err)
55                 goto out;
56
57         if (!sbi->mft.ni && ino == MFT_REC_MFT && !sb->s_root) {
58                 t64 = sbi->mft.lbo >> sbi->cluster_bits;
59                 t32 = bytes_to_cluster(sbi, MFT_REC_VOL * sbi->record_size);
60                 sbi->mft.ni = ni;
61                 init_rwsem(&ni->file.run_lock);
62
63                 if (!run_add_entry(&ni->file.run, 0, t64, t32, true)) {
64                         err = -ENOMEM;
65                         goto out;
66                 }
67         }
68
69         err = mi_read(&ni->mi, ino == MFT_REC_MFT);
70
71         if (err)
72                 goto out;
73
74         rec = ni->mi.mrec;
75
76         if (sbi->flags & NTFS_FLAGS_LOG_REPLAYING) {
77                 ;
78         } else if (ref->seq != rec->seq) {
79                 err = -EINVAL;
80                 ntfs_err(sb, "MFT: r=%lx, expect seq=%x instead of %x!", ino,
81                          le16_to_cpu(ref->seq), le16_to_cpu(rec->seq));
82                 goto out;
83         } else if (!is_rec_inuse(rec)) {
84                 err = -EINVAL;
85                 ntfs_err(sb, "Inode r=%x is not in use!", (u32)ino);
86                 goto out;
87         }
88
89         if (le32_to_cpu(rec->total) != sbi->record_size) {
90                 /* Bad inode? */
91                 err = -EINVAL;
92                 goto out;
93         }
94
95         if (!is_rec_base(rec))
96                 goto Ok;
97
98         /* Record should contain $I30 root. */
99         is_dir = rec->flags & RECORD_FLAG_DIR;
100
101         inode->i_generation = le16_to_cpu(rec->seq);
102
103         /* Enumerate all struct Attributes MFT. */
104         le = NULL;
105         attr = NULL;
106
107         /*
108          * To reduce tab pressure use goto instead of
109          * while( (attr = ni_enum_attr_ex(ni, attr, &le, NULL) ))
110          */
111 next_attr:
112         run = NULL;
113         err = -EINVAL;
114         attr = ni_enum_attr_ex(ni, attr, &le, NULL);
115         if (!attr)
116                 goto end_enum;
117
118         if (le && le->vcn) {
119                 /* This is non primary attribute segment. Ignore if not MFT. */
120                 if (ino != MFT_REC_MFT || attr->type != ATTR_DATA)
121                         goto next_attr;
122
123                 run = &ni->file.run;
124                 asize = le32_to_cpu(attr->size);
125                 goto attr_unpack_run;
126         }
127
128         roff = attr->non_res ? 0 : le16_to_cpu(attr->res.data_off);
129         rsize = attr->non_res ? 0 : le32_to_cpu(attr->res.data_size);
130         asize = le32_to_cpu(attr->size);
131
132         switch (attr->type) {
133         case ATTR_STD:
134                 if (attr->non_res ||
135                     asize < sizeof(struct ATTR_STD_INFO) + roff ||
136                     rsize < sizeof(struct ATTR_STD_INFO))
137                         goto out;
138
139                 if (std5)
140                         goto next_attr;
141
142                 std5 = Add2Ptr(attr, roff);
143
144 #ifdef STATX_BTIME
145                 nt2kernel(std5->cr_time, &ni->i_crtime);
146 #endif
147                 nt2kernel(std5->a_time, &inode->i_atime);
148                 nt2kernel(std5->c_time, &inode->i_ctime);
149                 nt2kernel(std5->m_time, &inode->i_mtime);
150
151                 ni->std_fa = std5->fa;
152
153                 if (asize >= sizeof(struct ATTR_STD_INFO5) + roff &&
154                     rsize >= sizeof(struct ATTR_STD_INFO5))
155                         ni->std_security_id = std5->security_id;
156                 goto next_attr;
157
158         case ATTR_LIST:
159                 if (attr->name_len || le || ino == MFT_REC_LOG)
160                         goto out;
161
162                 err = ntfs_load_attr_list(ni, attr);
163                 if (err)
164                         goto out;
165
166                 le = NULL;
167                 attr = NULL;
168                 goto next_attr;
169
170         case ATTR_NAME:
171                 if (attr->non_res || asize < SIZEOF_ATTRIBUTE_FILENAME + roff ||
172                     rsize < SIZEOF_ATTRIBUTE_FILENAME)
173                         goto out;
174
175                 fname = Add2Ptr(attr, roff);
176                 if (fname->type == FILE_NAME_DOS)
177                         goto next_attr;
178
179                 names += 1;
180                 if (name && name->len == fname->name_len &&
181                     !ntfs_cmp_names_cpu(name, (struct le_str *)&fname->name_len,
182                                         NULL, false))
183                         is_match = true;
184
185                 goto next_attr;
186
187         case ATTR_DATA:
188                 if (is_dir) {
189                         /* Ignore data attribute in dir record. */
190                         goto next_attr;
191                 }
192
193                 if (ino == MFT_REC_BADCLUST && !attr->non_res)
194                         goto next_attr;
195
196                 if (attr->name_len &&
197                     ((ino != MFT_REC_BADCLUST || !attr->non_res ||
198                       attr->name_len != ARRAY_SIZE(BAD_NAME) ||
199                       memcmp(attr_name(attr), BAD_NAME, sizeof(BAD_NAME))) &&
200                      (ino != MFT_REC_SECURE || !attr->non_res ||
201                       attr->name_len != ARRAY_SIZE(SDS_NAME) ||
202                       memcmp(attr_name(attr), SDS_NAME, sizeof(SDS_NAME))))) {
203                         /* File contains stream attribute. Ignore it. */
204                         goto next_attr;
205                 }
206
207                 if (is_attr_sparsed(attr))
208                         ni->std_fa |= FILE_ATTRIBUTE_SPARSE_FILE;
209                 else
210                         ni->std_fa &= ~FILE_ATTRIBUTE_SPARSE_FILE;
211
212                 if (is_attr_compressed(attr))
213                         ni->std_fa |= FILE_ATTRIBUTE_COMPRESSED;
214                 else
215                         ni->std_fa &= ~FILE_ATTRIBUTE_COMPRESSED;
216
217                 if (is_attr_encrypted(attr))
218                         ni->std_fa |= FILE_ATTRIBUTE_ENCRYPTED;
219                 else
220                         ni->std_fa &= ~FILE_ATTRIBUTE_ENCRYPTED;
221
222                 if (!attr->non_res) {
223                         ni->i_valid = inode->i_size = rsize;
224                         inode_set_bytes(inode, rsize);
225                 }
226
227                 mode = S_IFREG | (0777 & sbi->options->fs_fmask_inv);
228
229                 if (!attr->non_res) {
230                         ni->ni_flags |= NI_FLAG_RESIDENT;
231                         goto next_attr;
232                 }
233
234                 inode_set_bytes(inode, attr_ondisk_size(attr));
235
236                 ni->i_valid = le64_to_cpu(attr->nres.valid_size);
237                 inode->i_size = le64_to_cpu(attr->nres.data_size);
238                 if (!attr->nres.alloc_size)
239                         goto next_attr;
240
241                 run = ino == MFT_REC_BITMAP ? &sbi->used.bitmap.run
242                                             : &ni->file.run;
243                 break;
244
245         case ATTR_ROOT:
246                 if (attr->non_res)
247                         goto out;
248
249                 root = Add2Ptr(attr, roff);
250                 is_root = true;
251
252                 if (attr->name_len != ARRAY_SIZE(I30_NAME) ||
253                     memcmp(attr_name(attr), I30_NAME, sizeof(I30_NAME)))
254                         goto next_attr;
255
256                 if (root->type != ATTR_NAME ||
257                     root->rule != NTFS_COLLATION_TYPE_FILENAME)
258                         goto out;
259
260                 if (!is_dir)
261                         goto next_attr;
262
263                 ni->ni_flags |= NI_FLAG_DIR;
264
265                 err = indx_init(&ni->dir, sbi, attr, INDEX_MUTEX_I30);
266                 if (err)
267                         goto out;
268
269                 mode = sb->s_root
270                                ? (S_IFDIR | (0777 & sbi->options->fs_dmask_inv))
271                                : (S_IFDIR | 0777);
272                 goto next_attr;
273
274         case ATTR_ALLOC:
275                 if (!is_root || attr->name_len != ARRAY_SIZE(I30_NAME) ||
276                     memcmp(attr_name(attr), I30_NAME, sizeof(I30_NAME)))
277                         goto next_attr;
278
279                 inode->i_size = le64_to_cpu(attr->nres.data_size);
280                 ni->i_valid = le64_to_cpu(attr->nres.valid_size);
281                 inode_set_bytes(inode, le64_to_cpu(attr->nres.alloc_size));
282
283                 run = &ni->dir.alloc_run;
284                 break;
285
286         case ATTR_BITMAP:
287                 if (ino == MFT_REC_MFT) {
288                         if (!attr->non_res)
289                                 goto out;
290 #ifndef CONFIG_NTFS3_64BIT_CLUSTER
291                         /* 0x20000000 = 2^32 / 8 */
292                         if (le64_to_cpu(attr->nres.alloc_size) >= 0x20000000)
293                                 goto out;
294 #endif
295                         run = &sbi->mft.bitmap.run;
296                         break;
297                 } else if (is_dir && attr->name_len == ARRAY_SIZE(I30_NAME) &&
298                            !memcmp(attr_name(attr), I30_NAME,
299                                    sizeof(I30_NAME)) &&
300                            attr->non_res) {
301                         run = &ni->dir.bitmap_run;
302                         break;
303                 }
304                 goto next_attr;
305
306         case ATTR_REPARSE:
307                 if (attr->name_len)
308                         goto next_attr;
309
310                 rp_fa = ni_parse_reparse(ni, attr, &rp);
311                 switch (rp_fa) {
312                 case REPARSE_LINK:
313                         /*
314                          * Normal symlink.
315                          * Assume one unicode symbol == one utf8.
316                          */
317                         inode->i_size = le16_to_cpu(rp.SymbolicLinkReparseBuffer
318                                                             .PrintNameLength) /
319                                         sizeof(u16);
320
321                         ni->i_valid = inode->i_size;
322
323                         /* Clear directory bit. */
324                         if (ni->ni_flags & NI_FLAG_DIR) {
325                                 indx_clear(&ni->dir);
326                                 memset(&ni->dir, 0, sizeof(ni->dir));
327                                 ni->ni_flags &= ~NI_FLAG_DIR;
328                         } else {
329                                 run_close(&ni->file.run);
330                         }
331                         mode = S_IFLNK | 0777;
332                         is_dir = false;
333                         if (attr->non_res) {
334                                 run = &ni->file.run;
335                                 goto attr_unpack_run; // Double break.
336                         }
337                         break;
338
339                 case REPARSE_COMPRESSED:
340                         break;
341
342                 case REPARSE_DEDUPLICATED:
343                         break;
344                 }
345                 goto next_attr;
346
347         case ATTR_EA_INFO:
348                 if (!attr->name_len &&
349                     resident_data_ex(attr, sizeof(struct EA_INFO))) {
350                         ni->ni_flags |= NI_FLAG_EA;
351                         /*
352                          * ntfs_get_wsl_perm updates inode->i_uid, inode->i_gid, inode->i_mode
353                          */
354                         inode->i_mode = mode;
355                         ntfs_get_wsl_perm(inode);
356                         mode = inode->i_mode;
357                 }
358                 goto next_attr;
359
360         default:
361                 goto next_attr;
362         }
363
364 attr_unpack_run:
365         roff = le16_to_cpu(attr->nres.run_off);
366
367         t64 = le64_to_cpu(attr->nres.svcn);
368         err = run_unpack_ex(run, sbi, ino, t64, le64_to_cpu(attr->nres.evcn),
369                             t64, Add2Ptr(attr, roff), asize - roff);
370         if (err < 0)
371                 goto out;
372         err = 0;
373         goto next_attr;
374
375 end_enum:
376
377         if (!std5)
378                 goto out;
379
380         if (!is_match && name) {
381                 /* Reuse rec as buffer for ascii name. */
382                 err = -ENOENT;
383                 goto out;
384         }
385
386         if (std5->fa & FILE_ATTRIBUTE_READONLY)
387                 mode &= ~0222;
388
389         if (!names) {
390                 err = -EINVAL;
391                 goto out;
392         }
393
394         if (names != le16_to_cpu(rec->hard_links)) {
395                 /* Correct minor error on the fly. Do not mark inode as dirty. */
396                 rec->hard_links = cpu_to_le16(names);
397                 ni->mi.dirty = true;
398         }
399
400         set_nlink(inode, names);
401
402         if (S_ISDIR(mode)) {
403                 ni->std_fa |= FILE_ATTRIBUTE_DIRECTORY;
404
405                 /*
406                  * Dot and dot-dot should be included in count but was not
407                  * included in enumeration.
408                  * Usually a hard links to directories are disabled.
409                  */
410                 inode->i_op = &ntfs_dir_inode_operations;
411                 inode->i_fop = &ntfs_dir_operations;
412                 ni->i_valid = 0;
413         } else if (S_ISLNK(mode)) {
414                 ni->std_fa &= ~FILE_ATTRIBUTE_DIRECTORY;
415                 inode->i_op = &ntfs_link_inode_operations;
416                 inode->i_fop = NULL;
417                 inode_nohighmem(inode);
418         } else if (S_ISREG(mode)) {
419                 ni->std_fa &= ~FILE_ATTRIBUTE_DIRECTORY;
420                 inode->i_op = &ntfs_file_inode_operations;
421                 inode->i_fop = &ntfs_file_operations;
422                 inode->i_mapping->a_ops =
423                         is_compressed(ni) ? &ntfs_aops_cmpr : &ntfs_aops;
424                 if (ino != MFT_REC_MFT)
425                         init_rwsem(&ni->file.run_lock);
426         } else if (S_ISCHR(mode) || S_ISBLK(mode) || S_ISFIFO(mode) ||
427                    S_ISSOCK(mode)) {
428                 inode->i_op = &ntfs_special_inode_operations;
429                 init_special_inode(inode, mode, inode->i_rdev);
430         } else if (fname && fname->home.low == cpu_to_le32(MFT_REC_EXTEND) &&
431                    fname->home.seq == cpu_to_le16(MFT_REC_EXTEND)) {
432                 /* Records in $Extend are not a files or general directories. */
433         } else {
434                 err = -EINVAL;
435                 goto out;
436         }
437
438         if ((sbi->options->sys_immutable &&
439              (std5->fa & FILE_ATTRIBUTE_SYSTEM)) &&
440             !S_ISFIFO(mode) && !S_ISSOCK(mode) && !S_ISLNK(mode)) {
441                 inode->i_flags |= S_IMMUTABLE;
442         } else {
443                 inode->i_flags &= ~S_IMMUTABLE;
444         }
445
446         inode->i_mode = mode;
447         if (!(ni->ni_flags & NI_FLAG_EA)) {
448                 /* If no xattr then no security (stored in xattr). */
449                 inode->i_flags |= S_NOSEC;
450         }
451
452 Ok:
453         if (ino == MFT_REC_MFT && !sb->s_root)
454                 sbi->mft.ni = NULL;
455
456         unlock_new_inode(inode);
457
458         return inode;
459
460 out:
461         if (ino == MFT_REC_MFT && !sb->s_root)
462                 sbi->mft.ni = NULL;
463
464         iget_failed(inode);
465         return ERR_PTR(err);
466 }
467
468 /*
469  * ntfs_test_inode
470  *
471  * Return: 1 if match.
472  */
473 static int ntfs_test_inode(struct inode *inode, void *data)
474 {
475         struct MFT_REF *ref = data;
476
477         return ino_get(ref) == inode->i_ino;
478 }
479
480 static int ntfs_set_inode(struct inode *inode, void *data)
481 {
482         const struct MFT_REF *ref = data;
483
484         inode->i_ino = ino_get(ref);
485         return 0;
486 }
487
488 struct inode *ntfs_iget5(struct super_block *sb, const struct MFT_REF *ref,
489                          const struct cpu_str *name)
490 {
491         struct inode *inode;
492
493         inode = iget5_locked(sb, ino_get(ref), ntfs_test_inode, ntfs_set_inode,
494                              (void *)ref);
495         if (unlikely(!inode))
496                 return ERR_PTR(-ENOMEM);
497
498         /* If this is a freshly allocated inode, need to read it now. */
499         if (inode->i_state & I_NEW)
500                 inode = ntfs_read_mft(inode, name, ref);
501         else if (ref->seq != ntfs_i(inode)->mi.mrec->seq) {
502                 /* Inode overlaps? */
503                 make_bad_inode(inode);
504         }
505
506         return inode;
507 }
508
509 enum get_block_ctx {
510         GET_BLOCK_GENERAL = 0,
511         GET_BLOCK_WRITE_BEGIN = 1,
512         GET_BLOCK_DIRECT_IO_R = 2,
513         GET_BLOCK_DIRECT_IO_W = 3,
514         GET_BLOCK_BMAP = 4,
515 };
516
517 static noinline int ntfs_get_block_vbo(struct inode *inode, u64 vbo,
518                                        struct buffer_head *bh, int create,
519                                        enum get_block_ctx ctx)
520 {
521         struct super_block *sb = inode->i_sb;
522         struct ntfs_sb_info *sbi = sb->s_fs_info;
523         struct ntfs_inode *ni = ntfs_i(inode);
524         struct page *page = bh->b_page;
525         u8 cluster_bits = sbi->cluster_bits;
526         u32 block_size = sb->s_blocksize;
527         u64 bytes, lbo, valid;
528         u32 off;
529         int err;
530         CLST vcn, lcn, len;
531         bool new;
532
533         /* Clear previous state. */
534         clear_buffer_new(bh);
535         clear_buffer_uptodate(bh);
536
537         /* Direct write uses 'create=0'. */
538         if (!create && vbo >= ni->i_valid) {
539                 /* Out of valid. */
540                 return 0;
541         }
542
543         if (vbo >= inode->i_size) {
544                 /* Out of size. */
545                 return 0;
546         }
547
548         if (is_resident(ni)) {
549                 ni_lock(ni);
550                 err = attr_data_read_resident(ni, page);
551                 ni_unlock(ni);
552
553                 if (!err)
554                         set_buffer_uptodate(bh);
555                 bh->b_size = block_size;
556                 return err;
557         }
558
559         vcn = vbo >> cluster_bits;
560         off = vbo & sbi->cluster_mask;
561         new = false;
562
563         err = attr_data_get_block(ni, vcn, 1, &lcn, &len, create ? &new : NULL);
564         if (err)
565                 goto out;
566
567         if (!len)
568                 return 0;
569
570         bytes = ((u64)len << cluster_bits) - off;
571
572         if (lcn == SPARSE_LCN) {
573                 if (!create) {
574                         if (bh->b_size > bytes)
575                                 bh->b_size = bytes;
576                         return 0;
577                 }
578                 WARN_ON(1);
579         }
580
581         if (new) {
582                 set_buffer_new(bh);
583                 if ((len << cluster_bits) > block_size)
584                         ntfs_sparse_cluster(inode, page, vcn, len);
585         }
586
587         lbo = ((u64)lcn << cluster_bits) + off;
588
589         set_buffer_mapped(bh);
590         bh->b_bdev = sb->s_bdev;
591         bh->b_blocknr = lbo >> sb->s_blocksize_bits;
592
593         valid = ni->i_valid;
594
595         if (ctx == GET_BLOCK_DIRECT_IO_W) {
596                 /* ntfs_direct_IO will update ni->i_valid. */
597                 if (vbo >= valid)
598                         set_buffer_new(bh);
599         } else if (create) {
600                 /* Normal write. */
601                 if (bytes > bh->b_size)
602                         bytes = bh->b_size;
603
604                 if (vbo >= valid)
605                         set_buffer_new(bh);
606
607                 if (vbo + bytes > valid) {
608                         ni->i_valid = vbo + bytes;
609                         mark_inode_dirty(inode);
610                 }
611         } else if (vbo >= valid) {
612                 /* Read out of valid data. */
613                 /* Should never be here 'cause already checked. */
614                 clear_buffer_mapped(bh);
615         } else if (vbo + bytes <= valid) {
616                 /* Normal read. */
617         } else if (vbo + block_size <= valid) {
618                 /* Normal short read. */
619                 bytes = block_size;
620         } else {
621                 /*
622                  * Read across valid size: vbo < valid && valid < vbo + block_size
623                  */
624                 bytes = block_size;
625
626                 if (page) {
627                         u32 voff = valid - vbo;
628
629                         bh->b_size = block_size;
630                         off = vbo & (PAGE_SIZE - 1);
631                         set_bh_page(bh, page, off);
632                         ll_rw_block(REQ_OP_READ, 0, 1, &bh);
633                         wait_on_buffer(bh);
634                         if (!buffer_uptodate(bh)) {
635                                 err = -EIO;
636                                 goto out;
637                         }
638                         zero_user_segment(page, off + voff, off + block_size);
639                 }
640         }
641
642         if (bh->b_size > bytes)
643                 bh->b_size = bytes;
644
645 #ifndef __LP64__
646         if (ctx == GET_BLOCK_DIRECT_IO_W || ctx == GET_BLOCK_DIRECT_IO_R) {
647                 static_assert(sizeof(size_t) < sizeof(loff_t));
648                 if (bytes > 0x40000000u)
649                         bh->b_size = 0x40000000u;
650         }
651 #endif
652
653         return 0;
654
655 out:
656         return err;
657 }
658
659 int ntfs_get_block(struct inode *inode, sector_t vbn,
660                    struct buffer_head *bh_result, int create)
661 {
662         return ntfs_get_block_vbo(inode, (u64)vbn << inode->i_blkbits,
663                                   bh_result, create, GET_BLOCK_GENERAL);
664 }
665
666 static int ntfs_get_block_bmap(struct inode *inode, sector_t vsn,
667                                struct buffer_head *bh_result, int create)
668 {
669         return ntfs_get_block_vbo(inode,
670                                   (u64)vsn << inode->i_sb->s_blocksize_bits,
671                                   bh_result, create, GET_BLOCK_BMAP);
672 }
673
674 static sector_t ntfs_bmap(struct address_space *mapping, sector_t block)
675 {
676         return generic_block_bmap(mapping, block, ntfs_get_block_bmap);
677 }
678
679 static int ntfs_readpage(struct file *file, struct page *page)
680 {
681         int err;
682         struct address_space *mapping = page->mapping;
683         struct inode *inode = mapping->host;
684         struct ntfs_inode *ni = ntfs_i(inode);
685
686         if (is_resident(ni)) {
687                 ni_lock(ni);
688                 err = attr_data_read_resident(ni, page);
689                 ni_unlock(ni);
690                 if (err != E_NTFS_NONRESIDENT) {
691                         unlock_page(page);
692                         return err;
693                 }
694         }
695
696         if (is_compressed(ni)) {
697                 ni_lock(ni);
698                 err = ni_readpage_cmpr(ni, page);
699                 ni_unlock(ni);
700                 return err;
701         }
702
703         /* Normal + sparse files. */
704         return mpage_readpage(page, ntfs_get_block);
705 }
706
707 static void ntfs_readahead(struct readahead_control *rac)
708 {
709         struct address_space *mapping = rac->mapping;
710         struct inode *inode = mapping->host;
711         struct ntfs_inode *ni = ntfs_i(inode);
712         u64 valid;
713         loff_t pos;
714
715         if (is_resident(ni)) {
716                 /* No readahead for resident. */
717                 return;
718         }
719
720         if (is_compressed(ni)) {
721                 /* No readahead for compressed. */
722                 return;
723         }
724
725         valid = ni->i_valid;
726         pos = readahead_pos(rac);
727
728         if (valid < i_size_read(inode) && pos <= valid &&
729             valid < pos + readahead_length(rac)) {
730                 /* Range cross 'valid'. Read it page by page. */
731                 return;
732         }
733
734         mpage_readahead(rac, ntfs_get_block);
735 }
736
737 static int ntfs_get_block_direct_IO_R(struct inode *inode, sector_t iblock,
738                                       struct buffer_head *bh_result, int create)
739 {
740         return ntfs_get_block_vbo(inode, (u64)iblock << inode->i_blkbits,
741                                   bh_result, create, GET_BLOCK_DIRECT_IO_R);
742 }
743
744 static int ntfs_get_block_direct_IO_W(struct inode *inode, sector_t iblock,
745                                       struct buffer_head *bh_result, int create)
746 {
747         return ntfs_get_block_vbo(inode, (u64)iblock << inode->i_blkbits,
748                                   bh_result, create, GET_BLOCK_DIRECT_IO_W);
749 }
750
751 static ssize_t ntfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
752 {
753         struct file *file = iocb->ki_filp;
754         struct address_space *mapping = file->f_mapping;
755         struct inode *inode = mapping->host;
756         struct ntfs_inode *ni = ntfs_i(inode);
757         loff_t vbo = iocb->ki_pos;
758         loff_t end;
759         int wr = iov_iter_rw(iter) & WRITE;
760         loff_t valid;
761         ssize_t ret;
762
763         if (is_resident(ni)) {
764                 /* Switch to buffered write. */
765                 ret = 0;
766                 goto out;
767         }
768
769         ret = blockdev_direct_IO(iocb, inode, iter,
770                                  wr ? ntfs_get_block_direct_IO_W
771                                     : ntfs_get_block_direct_IO_R);
772
773         if (ret <= 0)
774                 goto out;
775
776         end = vbo + ret;
777         valid = ni->i_valid;
778         if (wr) {
779                 if (end > valid && !S_ISBLK(inode->i_mode)) {
780                         ni->i_valid = end;
781                         mark_inode_dirty(inode);
782                 }
783         } else if (vbo < valid && valid < end) {
784                 /* Fix page. */
785                 iov_iter_revert(iter, end - valid);
786                 iov_iter_zero(end - valid, iter);
787         }
788
789 out:
790         return ret;
791 }
792
793 int ntfs_set_size(struct inode *inode, u64 new_size)
794 {
795         struct super_block *sb = inode->i_sb;
796         struct ntfs_sb_info *sbi = sb->s_fs_info;
797         struct ntfs_inode *ni = ntfs_i(inode);
798         int err;
799
800         /* Check for maximum file size. */
801         if (is_sparsed(ni) || is_compressed(ni)) {
802                 if (new_size > sbi->maxbytes_sparse) {
803                         err = -EFBIG;
804                         goto out;
805                 }
806         } else if (new_size > sbi->maxbytes) {
807                 err = -EFBIG;
808                 goto out;
809         }
810
811         ni_lock(ni);
812         down_write(&ni->file.run_lock);
813
814         err = attr_set_size(ni, ATTR_DATA, NULL, 0, &ni->file.run, new_size,
815                             &ni->i_valid, true, NULL);
816
817         up_write(&ni->file.run_lock);
818         ni_unlock(ni);
819
820         mark_inode_dirty(inode);
821
822 out:
823         return err;
824 }
825
826 static int ntfs_writepage(struct page *page, struct writeback_control *wbc)
827 {
828         struct address_space *mapping = page->mapping;
829         struct inode *inode = mapping->host;
830         struct ntfs_inode *ni = ntfs_i(inode);
831         int err;
832
833         if (is_resident(ni)) {
834                 ni_lock(ni);
835                 err = attr_data_write_resident(ni, page);
836                 ni_unlock(ni);
837                 if (err != E_NTFS_NONRESIDENT) {
838                         unlock_page(page);
839                         return err;
840                 }
841         }
842
843         return block_write_full_page(page, ntfs_get_block, wbc);
844 }
845
846 static int ntfs_writepages(struct address_space *mapping,
847                            struct writeback_control *wbc)
848 {
849         struct inode *inode = mapping->host;
850         struct ntfs_inode *ni = ntfs_i(inode);
851         /* Redirect call to 'ntfs_writepage' for resident files. */
852         get_block_t *get_block = is_resident(ni) ? NULL : &ntfs_get_block;
853
854         return mpage_writepages(mapping, wbc, get_block);
855 }
856
857 static int ntfs_get_block_write_begin(struct inode *inode, sector_t vbn,
858                                       struct buffer_head *bh_result, int create)
859 {
860         return ntfs_get_block_vbo(inode, (u64)vbn << inode->i_blkbits,
861                                   bh_result, create, GET_BLOCK_WRITE_BEGIN);
862 }
863
864 static int ntfs_write_begin(struct file *file, struct address_space *mapping,
865                             loff_t pos, u32 len, u32 flags, struct page **pagep,
866                             void **fsdata)
867 {
868         int err;
869         struct inode *inode = mapping->host;
870         struct ntfs_inode *ni = ntfs_i(inode);
871
872         *pagep = NULL;
873         if (is_resident(ni)) {
874                 struct page *page = grab_cache_page_write_begin(
875                         mapping, pos >> PAGE_SHIFT, flags);
876
877                 if (!page) {
878                         err = -ENOMEM;
879                         goto out;
880                 }
881
882                 ni_lock(ni);
883                 err = attr_data_read_resident(ni, page);
884                 ni_unlock(ni);
885
886                 if (!err) {
887                         *pagep = page;
888                         goto out;
889                 }
890                 unlock_page(page);
891                 put_page(page);
892
893                 if (err != E_NTFS_NONRESIDENT)
894                         goto out;
895         }
896
897         err = block_write_begin(mapping, pos, len, flags, pagep,
898                                 ntfs_get_block_write_begin);
899
900 out:
901         return err;
902 }
903
904 /*
905  * ntfs_write_end - Address_space_operations::write_end.
906  */
907 static int ntfs_write_end(struct file *file, struct address_space *mapping,
908                           loff_t pos, u32 len, u32 copied, struct page *page,
909                           void *fsdata)
910
911 {
912         struct inode *inode = mapping->host;
913         struct ntfs_inode *ni = ntfs_i(inode);
914         u64 valid = ni->i_valid;
915         bool dirty = false;
916         int err;
917
918         if (is_resident(ni)) {
919                 ni_lock(ni);
920                 err = attr_data_write_resident(ni, page);
921                 ni_unlock(ni);
922                 if (!err) {
923                         dirty = true;
924                         /* Clear any buffers in page. */
925                         if (page_has_buffers(page)) {
926                                 struct buffer_head *head, *bh;
927
928                                 bh = head = page_buffers(page);
929                                 do {
930                                         clear_buffer_dirty(bh);
931                                         clear_buffer_mapped(bh);
932                                         set_buffer_uptodate(bh);
933                                 } while (head != (bh = bh->b_this_page));
934                         }
935                         SetPageUptodate(page);
936                         err = copied;
937                 }
938                 unlock_page(page);
939                 put_page(page);
940         } else {
941                 err = generic_write_end(file, mapping, pos, len, copied, page,
942                                         fsdata);
943         }
944
945         if (err >= 0) {
946                 if (!(ni->std_fa & FILE_ATTRIBUTE_ARCHIVE)) {
947                         inode->i_ctime = inode->i_mtime = current_time(inode);
948                         ni->std_fa |= FILE_ATTRIBUTE_ARCHIVE;
949                         dirty = true;
950                 }
951
952                 if (valid != ni->i_valid) {
953                         /* ni->i_valid is changed in ntfs_get_block_vbo. */
954                         dirty = true;
955                 }
956
957                 if (dirty)
958                         mark_inode_dirty(inode);
959         }
960
961         return err;
962 }
963
964 int reset_log_file(struct inode *inode)
965 {
966         int err;
967         loff_t pos = 0;
968         u32 log_size = inode->i_size;
969         struct address_space *mapping = inode->i_mapping;
970
971         for (;;) {
972                 u32 len;
973                 void *kaddr;
974                 struct page *page;
975
976                 len = pos + PAGE_SIZE > log_size ? (log_size - pos) : PAGE_SIZE;
977
978                 err = block_write_begin(mapping, pos, len, 0, &page,
979                                         ntfs_get_block_write_begin);
980                 if (err)
981                         goto out;
982
983                 kaddr = kmap_atomic(page);
984                 memset(kaddr, -1, len);
985                 kunmap_atomic(kaddr);
986                 flush_dcache_page(page);
987
988                 err = block_write_end(NULL, mapping, pos, len, len, page, NULL);
989                 if (err < 0)
990                         goto out;
991                 pos += len;
992
993                 if (pos >= log_size)
994                         break;
995                 balance_dirty_pages_ratelimited(mapping);
996         }
997 out:
998         mark_inode_dirty_sync(inode);
999
1000         return err;
1001 }
1002
1003 int ntfs3_write_inode(struct inode *inode, struct writeback_control *wbc)
1004 {
1005         return _ni_write_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1006 }
1007
1008 int ntfs_sync_inode(struct inode *inode)
1009 {
1010         return _ni_write_inode(inode, 1);
1011 }
1012
1013 /*
1014  * writeback_inode - Helper function for ntfs_flush_inodes().
1015  *
1016  * This writes both the inode and the file data blocks, waiting
1017  * for in flight data blocks before the start of the call.  It
1018  * does not wait for any io started during the call.
1019  */
1020 static int writeback_inode(struct inode *inode)
1021 {
1022         int ret = sync_inode_metadata(inode, 0);
1023
1024         if (!ret)
1025                 ret = filemap_fdatawrite(inode->i_mapping);
1026         return ret;
1027 }
1028
1029 /*
1030  * ntfs_flush_inodes
1031  *
1032  * Write data and metadata corresponding to i1 and i2.  The io is
1033  * started but we do not wait for any of it to finish.
1034  *
1035  * filemap_flush() is used for the block device, so if there is a dirty
1036  * page for a block already in flight, we will not wait and start the
1037  * io over again.
1038  */
1039 int ntfs_flush_inodes(struct super_block *sb, struct inode *i1,
1040                       struct inode *i2)
1041 {
1042         int ret = 0;
1043
1044         if (i1)
1045                 ret = writeback_inode(i1);
1046         if (!ret && i2)
1047                 ret = writeback_inode(i2);
1048         if (!ret)
1049                 ret = sync_blockdev_nowait(sb->s_bdev);
1050         return ret;
1051 }
1052
1053 int inode_write_data(struct inode *inode, const void *data, size_t bytes)
1054 {
1055         pgoff_t idx;
1056
1057         /* Write non resident data. */
1058         for (idx = 0; bytes; idx++) {
1059                 size_t op = bytes > PAGE_SIZE ? PAGE_SIZE : bytes;
1060                 struct page *page = ntfs_map_page(inode->i_mapping, idx);
1061
1062                 if (IS_ERR(page))
1063                         return PTR_ERR(page);
1064
1065                 lock_page(page);
1066                 WARN_ON(!PageUptodate(page));
1067                 ClearPageUptodate(page);
1068
1069                 memcpy(page_address(page), data, op);
1070
1071                 flush_dcache_page(page);
1072                 SetPageUptodate(page);
1073                 unlock_page(page);
1074
1075                 ntfs_unmap_page(page);
1076
1077                 bytes -= op;
1078                 data = Add2Ptr(data, PAGE_SIZE);
1079         }
1080         return 0;
1081 }
1082
1083 /*
1084  * ntfs_reparse_bytes
1085  *
1086  * Number of bytes for REPARSE_DATA_BUFFER(IO_REPARSE_TAG_SYMLINK)
1087  * for unicode string of @uni_len length.
1088  */
1089 static inline u32 ntfs_reparse_bytes(u32 uni_len)
1090 {
1091         /* Header + unicode string + decorated unicode string. */
1092         return sizeof(short) * (2 * uni_len + 4) +
1093                offsetof(struct REPARSE_DATA_BUFFER,
1094                         SymbolicLinkReparseBuffer.PathBuffer);
1095 }
1096
1097 static struct REPARSE_DATA_BUFFER *
1098 ntfs_create_reparse_buffer(struct ntfs_sb_info *sbi, const char *symname,
1099                            u32 size, u16 *nsize)
1100 {
1101         int i, err;
1102         struct REPARSE_DATA_BUFFER *rp;
1103         __le16 *rp_name;
1104         typeof(rp->SymbolicLinkReparseBuffer) *rs;
1105
1106         rp = kzalloc(ntfs_reparse_bytes(2 * size + 2), GFP_NOFS);
1107         if (!rp)
1108                 return ERR_PTR(-ENOMEM);
1109
1110         rs = &rp->SymbolicLinkReparseBuffer;
1111         rp_name = rs->PathBuffer;
1112
1113         /* Convert link name to UTF-16. */
1114         err = ntfs_nls_to_utf16(sbi, symname, size,
1115                                 (struct cpu_str *)(rp_name - 1), 2 * size,
1116                                 UTF16_LITTLE_ENDIAN);
1117         if (err < 0)
1118                 goto out;
1119
1120         /* err = the length of unicode name of symlink. */
1121         *nsize = ntfs_reparse_bytes(err);
1122
1123         if (*nsize > sbi->reparse.max_size) {
1124                 err = -EFBIG;
1125                 goto out;
1126         }
1127
1128         /* Translate Linux '/' into Windows '\'. */
1129         for (i = 0; i < err; i++) {
1130                 if (rp_name[i] == cpu_to_le16('/'))
1131                         rp_name[i] = cpu_to_le16('\\');
1132         }
1133
1134         rp->ReparseTag = IO_REPARSE_TAG_SYMLINK;
1135         rp->ReparseDataLength =
1136                 cpu_to_le16(*nsize - offsetof(struct REPARSE_DATA_BUFFER,
1137                                               SymbolicLinkReparseBuffer));
1138
1139         /* PrintName + SubstituteName. */
1140         rs->SubstituteNameOffset = cpu_to_le16(sizeof(short) * err);
1141         rs->SubstituteNameLength = cpu_to_le16(sizeof(short) * err + 8);
1142         rs->PrintNameLength = rs->SubstituteNameOffset;
1143
1144         /*
1145          * TODO: Use relative path if possible to allow Windows to
1146          * parse this path.
1147          * 0-absolute path 1- relative path (SYMLINK_FLAG_RELATIVE).
1148          */
1149         rs->Flags = 0;
1150
1151         memmove(rp_name + err + 4, rp_name, sizeof(short) * err);
1152
1153         /* Decorate SubstituteName. */
1154         rp_name += err;
1155         rp_name[0] = cpu_to_le16('\\');
1156         rp_name[1] = cpu_to_le16('?');
1157         rp_name[2] = cpu_to_le16('?');
1158         rp_name[3] = cpu_to_le16('\\');
1159
1160         return rp;
1161 out:
1162         kfree(rp);
1163         return ERR_PTR(err);
1164 }
1165
1166 struct inode *ntfs_create_inode(struct user_namespace *mnt_userns,
1167                                 struct inode *dir, struct dentry *dentry,
1168                                 const struct cpu_str *uni, umode_t mode,
1169                                 dev_t dev, const char *symname, u32 size,
1170                                 struct ntfs_fnd *fnd)
1171 {
1172         int err;
1173         struct super_block *sb = dir->i_sb;
1174         struct ntfs_sb_info *sbi = sb->s_fs_info;
1175         const struct qstr *name = &dentry->d_name;
1176         CLST ino = 0;
1177         struct ntfs_inode *dir_ni = ntfs_i(dir);
1178         struct ntfs_inode *ni = NULL;
1179         struct inode *inode = NULL;
1180         struct ATTRIB *attr;
1181         struct ATTR_STD_INFO5 *std5;
1182         struct ATTR_FILE_NAME *fname;
1183         struct MFT_REC *rec;
1184         u32 asize, dsize, sd_size;
1185         enum FILE_ATTRIBUTE fa;
1186         __le32 security_id = SECURITY_ID_INVALID;
1187         CLST vcn;
1188         const void *sd;
1189         u16 t16, nsize = 0, aid = 0;
1190         struct INDEX_ROOT *root, *dir_root;
1191         struct NTFS_DE *e, *new_de = NULL;
1192         struct REPARSE_DATA_BUFFER *rp = NULL;
1193         bool rp_inserted = false;
1194
1195         ni_lock_dir(dir_ni);
1196
1197         dir_root = indx_get_root(&dir_ni->dir, dir_ni, NULL, NULL);
1198         if (!dir_root) {
1199                 err = -EINVAL;
1200                 goto out1;
1201         }
1202
1203         if (S_ISDIR(mode)) {
1204                 /* Use parent's directory attributes. */
1205                 fa = dir_ni->std_fa | FILE_ATTRIBUTE_DIRECTORY |
1206                      FILE_ATTRIBUTE_ARCHIVE;
1207                 /*
1208                  * By default child directory inherits parent attributes.
1209                  * Root directory is hidden + system.
1210                  * Make an exception for children in root.
1211                  */
1212                 if (dir->i_ino == MFT_REC_ROOT)
1213                         fa &= ~(FILE_ATTRIBUTE_HIDDEN | FILE_ATTRIBUTE_SYSTEM);
1214         } else if (S_ISLNK(mode)) {
1215                 /* It is good idea that link should be the same type (file/dir) as target */
1216                 fa = FILE_ATTRIBUTE_REPARSE_POINT;
1217
1218                 /*
1219                  * Linux: there are dir/file/symlink and so on.
1220                  * NTFS: symlinks are "dir + reparse" or "file + reparse"
1221                  * It is good idea to create:
1222                  * dir + reparse if 'symname' points to directory
1223                  * or
1224                  * file + reparse if 'symname' points to file
1225                  * Unfortunately kern_path hangs if symname contains 'dir'.
1226                  */
1227
1228                 /*
1229                  *      struct path path;
1230                  *
1231                  *      if (!kern_path(symname, LOOKUP_FOLLOW, &path)){
1232                  *              struct inode *target = d_inode(path.dentry);
1233                  *
1234                  *              if (S_ISDIR(target->i_mode))
1235                  *                      fa |= FILE_ATTRIBUTE_DIRECTORY;
1236                  *              // if ( target->i_sb == sb ){
1237                  *              //      use relative path?
1238                  *              // }
1239                  *              path_put(&path);
1240                  *      }
1241                  */
1242         } else if (S_ISREG(mode)) {
1243                 if (sbi->options->sparse) {
1244                         /* Sparsed regular file, cause option 'sparse'. */
1245                         fa = FILE_ATTRIBUTE_SPARSE_FILE |
1246                              FILE_ATTRIBUTE_ARCHIVE;
1247                 } else if (dir_ni->std_fa & FILE_ATTRIBUTE_COMPRESSED) {
1248                         /* Compressed regular file, if parent is compressed. */
1249                         fa = FILE_ATTRIBUTE_COMPRESSED | FILE_ATTRIBUTE_ARCHIVE;
1250                 } else {
1251                         /* Regular file, default attributes. */
1252                         fa = FILE_ATTRIBUTE_ARCHIVE;
1253                 }
1254         } else {
1255                 fa = FILE_ATTRIBUTE_ARCHIVE;
1256         }
1257
1258         if (!(mode & 0222))
1259                 fa |= FILE_ATTRIBUTE_READONLY;
1260
1261         /* Allocate PATH_MAX bytes. */
1262         new_de = __getname();
1263         if (!new_de) {
1264                 err = -ENOMEM;
1265                 goto out1;
1266         }
1267
1268         /* Mark rw ntfs as dirty. it will be cleared at umount. */
1269         ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
1270
1271         /* Step 1: allocate and fill new mft record. */
1272         err = ntfs_look_free_mft(sbi, &ino, false, NULL, NULL);
1273         if (err)
1274                 goto out2;
1275
1276         ni = ntfs_new_inode(sbi, ino, fa & FILE_ATTRIBUTE_DIRECTORY);
1277         if (IS_ERR(ni)) {
1278                 err = PTR_ERR(ni);
1279                 ni = NULL;
1280                 goto out3;
1281         }
1282         inode = &ni->vfs_inode;
1283         inode_init_owner(mnt_userns, inode, dir, mode);
1284         mode = inode->i_mode;
1285
1286         inode->i_atime = inode->i_mtime = inode->i_ctime = ni->i_crtime =
1287                 current_time(inode);
1288
1289         rec = ni->mi.mrec;
1290         rec->hard_links = cpu_to_le16(1);
1291         attr = Add2Ptr(rec, le16_to_cpu(rec->attr_off));
1292
1293         /* Get default security id. */
1294         sd = s_default_security;
1295         sd_size = sizeof(s_default_security);
1296
1297         if (is_ntfs3(sbi)) {
1298                 security_id = dir_ni->std_security_id;
1299                 if (le32_to_cpu(security_id) < SECURITY_ID_FIRST) {
1300                         security_id = sbi->security.def_security_id;
1301
1302                         if (security_id == SECURITY_ID_INVALID &&
1303                             !ntfs_insert_security(sbi, sd, sd_size,
1304                                                   &security_id, NULL))
1305                                 sbi->security.def_security_id = security_id;
1306                 }
1307         }
1308
1309         /* Insert standard info. */
1310         std5 = Add2Ptr(attr, SIZEOF_RESIDENT);
1311
1312         if (security_id == SECURITY_ID_INVALID) {
1313                 dsize = sizeof(struct ATTR_STD_INFO);
1314         } else {
1315                 dsize = sizeof(struct ATTR_STD_INFO5);
1316                 std5->security_id = security_id;
1317                 ni->std_security_id = security_id;
1318         }
1319         asize = SIZEOF_RESIDENT + dsize;
1320
1321         attr->type = ATTR_STD;
1322         attr->size = cpu_to_le32(asize);
1323         attr->id = cpu_to_le16(aid++);
1324         attr->res.data_off = SIZEOF_RESIDENT_LE;
1325         attr->res.data_size = cpu_to_le32(dsize);
1326
1327         std5->cr_time = std5->m_time = std5->c_time = std5->a_time =
1328                 kernel2nt(&inode->i_atime);
1329
1330         ni->std_fa = fa;
1331         std5->fa = fa;
1332
1333         attr = Add2Ptr(attr, asize);
1334
1335         /* Insert file name. */
1336         err = fill_name_de(sbi, new_de, name, uni);
1337         if (err)
1338                 goto out4;
1339
1340         mi_get_ref(&ni->mi, &new_de->ref);
1341
1342         fname = (struct ATTR_FILE_NAME *)(new_de + 1);
1343         mi_get_ref(&dir_ni->mi, &fname->home);
1344         fname->dup.cr_time = fname->dup.m_time = fname->dup.c_time =
1345                 fname->dup.a_time = std5->cr_time;
1346         fname->dup.alloc_size = fname->dup.data_size = 0;
1347         fname->dup.fa = std5->fa;
1348         fname->dup.ea_size = fname->dup.reparse = 0;
1349
1350         dsize = le16_to_cpu(new_de->key_size);
1351         asize = ALIGN(SIZEOF_RESIDENT + dsize, 8);
1352
1353         attr->type = ATTR_NAME;
1354         attr->size = cpu_to_le32(asize);
1355         attr->res.data_off = SIZEOF_RESIDENT_LE;
1356         attr->res.flags = RESIDENT_FLAG_INDEXED;
1357         attr->id = cpu_to_le16(aid++);
1358         attr->res.data_size = cpu_to_le32(dsize);
1359         memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), fname, dsize);
1360
1361         attr = Add2Ptr(attr, asize);
1362
1363         if (security_id == SECURITY_ID_INVALID) {
1364                 /* Insert security attribute. */
1365                 asize = SIZEOF_RESIDENT + ALIGN(sd_size, 8);
1366
1367                 attr->type = ATTR_SECURE;
1368                 attr->size = cpu_to_le32(asize);
1369                 attr->id = cpu_to_le16(aid++);
1370                 attr->res.data_off = SIZEOF_RESIDENT_LE;
1371                 attr->res.data_size = cpu_to_le32(sd_size);
1372                 memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), sd, sd_size);
1373
1374                 attr = Add2Ptr(attr, asize);
1375         }
1376
1377         attr->id = cpu_to_le16(aid++);
1378         if (fa & FILE_ATTRIBUTE_DIRECTORY) {
1379                 /*
1380                  * Regular directory or symlink to directory.
1381                  * Create root attribute.
1382                  */
1383                 dsize = sizeof(struct INDEX_ROOT) + sizeof(struct NTFS_DE);
1384                 asize = sizeof(I30_NAME) + SIZEOF_RESIDENT + dsize;
1385
1386                 attr->type = ATTR_ROOT;
1387                 attr->size = cpu_to_le32(asize);
1388
1389                 attr->name_len = ARRAY_SIZE(I30_NAME);
1390                 attr->name_off = SIZEOF_RESIDENT_LE;
1391                 attr->res.data_off =
1392                         cpu_to_le16(sizeof(I30_NAME) + SIZEOF_RESIDENT);
1393                 attr->res.data_size = cpu_to_le32(dsize);
1394                 memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), I30_NAME,
1395                        sizeof(I30_NAME));
1396
1397                 root = Add2Ptr(attr, sizeof(I30_NAME) + SIZEOF_RESIDENT);
1398                 memcpy(root, dir_root, offsetof(struct INDEX_ROOT, ihdr));
1399                 root->ihdr.de_off =
1400                         cpu_to_le32(sizeof(struct INDEX_HDR)); // 0x10
1401                 root->ihdr.used = cpu_to_le32(sizeof(struct INDEX_HDR) +
1402                                               sizeof(struct NTFS_DE));
1403                 root->ihdr.total = root->ihdr.used;
1404
1405                 e = Add2Ptr(root, sizeof(struct INDEX_ROOT));
1406                 e->size = cpu_to_le16(sizeof(struct NTFS_DE));
1407                 e->flags = NTFS_IE_LAST;
1408         } else if (S_ISLNK(mode)) {
1409                 /*
1410                  * Symlink to file.
1411                  * Create empty resident data attribute.
1412                  */
1413                 asize = SIZEOF_RESIDENT;
1414
1415                 /* Insert empty ATTR_DATA */
1416                 attr->type = ATTR_DATA;
1417                 attr->size = cpu_to_le32(SIZEOF_RESIDENT);
1418                 attr->name_off = SIZEOF_RESIDENT_LE;
1419                 attr->res.data_off = SIZEOF_RESIDENT_LE;
1420         } else if (S_ISREG(mode)) {
1421                 /*
1422                  * Regular file. Create empty non resident data attribute.
1423                  */
1424                 attr->type = ATTR_DATA;
1425                 attr->non_res = 1;
1426                 attr->nres.evcn = cpu_to_le64(-1ll);
1427                 if (fa & FILE_ATTRIBUTE_SPARSE_FILE) {
1428                         attr->size = cpu_to_le32(SIZEOF_NONRESIDENT_EX + 8);
1429                         attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
1430                         attr->flags = ATTR_FLAG_SPARSED;
1431                         asize = SIZEOF_NONRESIDENT_EX + 8;
1432                 } else if (fa & FILE_ATTRIBUTE_COMPRESSED) {
1433                         attr->size = cpu_to_le32(SIZEOF_NONRESIDENT_EX + 8);
1434                         attr->name_off = SIZEOF_NONRESIDENT_EX_LE;
1435                         attr->flags = ATTR_FLAG_COMPRESSED;
1436                         attr->nres.c_unit = COMPRESSION_UNIT;
1437                         asize = SIZEOF_NONRESIDENT_EX + 8;
1438                 } else {
1439                         attr->size = cpu_to_le32(SIZEOF_NONRESIDENT + 8);
1440                         attr->name_off = SIZEOF_NONRESIDENT_LE;
1441                         asize = SIZEOF_NONRESIDENT + 8;
1442                 }
1443                 attr->nres.run_off = attr->name_off;
1444         } else {
1445                 /*
1446                  * Node. Create empty resident data attribute.
1447                  */
1448                 attr->type = ATTR_DATA;
1449                 attr->size = cpu_to_le32(SIZEOF_RESIDENT);
1450                 attr->name_off = SIZEOF_RESIDENT_LE;
1451                 if (fa & FILE_ATTRIBUTE_SPARSE_FILE)
1452                         attr->flags = ATTR_FLAG_SPARSED;
1453                 else if (fa & FILE_ATTRIBUTE_COMPRESSED)
1454                         attr->flags = ATTR_FLAG_COMPRESSED;
1455                 attr->res.data_off = SIZEOF_RESIDENT_LE;
1456                 asize = SIZEOF_RESIDENT;
1457                 ni->ni_flags |= NI_FLAG_RESIDENT;
1458         }
1459
1460         if (S_ISDIR(mode)) {
1461                 ni->ni_flags |= NI_FLAG_DIR;
1462                 err = indx_init(&ni->dir, sbi, attr, INDEX_MUTEX_I30);
1463                 if (err)
1464                         goto out4;
1465         } else if (S_ISLNK(mode)) {
1466                 rp = ntfs_create_reparse_buffer(sbi, symname, size, &nsize);
1467
1468                 if (IS_ERR(rp)) {
1469                         err = PTR_ERR(rp);
1470                         rp = NULL;
1471                         goto out4;
1472                 }
1473
1474                 /*
1475                  * Insert ATTR_REPARSE.
1476                  */
1477                 attr = Add2Ptr(attr, asize);
1478                 attr->type = ATTR_REPARSE;
1479                 attr->id = cpu_to_le16(aid++);
1480
1481                 /* Resident or non resident? */
1482                 asize = ALIGN(SIZEOF_RESIDENT + nsize, 8);
1483                 t16 = PtrOffset(rec, attr);
1484
1485                 /*
1486                  * Below function 'ntfs_save_wsl_perm' requires 0x78 bytes.
1487                  * It is good idea to keep extened attributes resident.
1488                  */
1489                 if (asize + t16 + 0x78 + 8 > sbi->record_size) {
1490                         CLST alen;
1491                         CLST clst = bytes_to_cluster(sbi, nsize);
1492
1493                         /* Bytes per runs. */
1494                         t16 = sbi->record_size - t16 - SIZEOF_NONRESIDENT;
1495
1496                         attr->non_res = 1;
1497                         attr->nres.evcn = cpu_to_le64(clst - 1);
1498                         attr->name_off = SIZEOF_NONRESIDENT_LE;
1499                         attr->nres.run_off = attr->name_off;
1500                         attr->nres.data_size = cpu_to_le64(nsize);
1501                         attr->nres.valid_size = attr->nres.data_size;
1502                         attr->nres.alloc_size =
1503                                 cpu_to_le64(ntfs_up_cluster(sbi, nsize));
1504
1505                         err = attr_allocate_clusters(sbi, &ni->file.run, 0, 0,
1506                                                      clst, NULL, 0, &alen, 0,
1507                                                      NULL);
1508                         if (err)
1509                                 goto out5;
1510
1511                         err = run_pack(&ni->file.run, 0, clst,
1512                                        Add2Ptr(attr, SIZEOF_NONRESIDENT), t16,
1513                                        &vcn);
1514                         if (err < 0)
1515                                 goto out5;
1516
1517                         if (vcn != clst) {
1518                                 err = -EINVAL;
1519                                 goto out5;
1520                         }
1521
1522                         asize = SIZEOF_NONRESIDENT + ALIGN(err, 8);
1523                 } else {
1524                         attr->res.data_off = SIZEOF_RESIDENT_LE;
1525                         attr->res.data_size = cpu_to_le32(nsize);
1526                         memcpy(Add2Ptr(attr, SIZEOF_RESIDENT), rp, nsize);
1527                         nsize = 0;
1528                 }
1529                 /* Size of symlink equals the length of input string. */
1530                 inode->i_size = size;
1531
1532                 attr->size = cpu_to_le32(asize);
1533
1534                 err = ntfs_insert_reparse(sbi, IO_REPARSE_TAG_SYMLINK,
1535                                           &new_de->ref);
1536                 if (err)
1537                         goto out5;
1538
1539                 rp_inserted = true;
1540         }
1541
1542         attr = Add2Ptr(attr, asize);
1543         attr->type = ATTR_END;
1544
1545         rec->used = cpu_to_le32(PtrOffset(rec, attr) + 8);
1546         rec->next_attr_id = cpu_to_le16(aid);
1547
1548         /* Step 2: Add new name in index. */
1549         err = indx_insert_entry(&dir_ni->dir, dir_ni, new_de, sbi, fnd, 0);
1550         if (err)
1551                 goto out6;
1552
1553         /* Unlock parent directory before ntfs_init_acl. */
1554         ni_unlock(dir_ni);
1555
1556         inode->i_generation = le16_to_cpu(rec->seq);
1557
1558         dir->i_mtime = dir->i_ctime = inode->i_atime;
1559
1560         if (S_ISDIR(mode)) {
1561                 inode->i_op = &ntfs_dir_inode_operations;
1562                 inode->i_fop = &ntfs_dir_operations;
1563         } else if (S_ISLNK(mode)) {
1564                 inode->i_op = &ntfs_link_inode_operations;
1565                 inode->i_fop = NULL;
1566                 inode->i_mapping->a_ops = &ntfs_aops;
1567                 inode->i_size = size;
1568                 inode_nohighmem(inode);
1569         } else if (S_ISREG(mode)) {
1570                 inode->i_op = &ntfs_file_inode_operations;
1571                 inode->i_fop = &ntfs_file_operations;
1572                 inode->i_mapping->a_ops =
1573                         is_compressed(ni) ? &ntfs_aops_cmpr : &ntfs_aops;
1574                 init_rwsem(&ni->file.run_lock);
1575         } else {
1576                 inode->i_op = &ntfs_special_inode_operations;
1577                 init_special_inode(inode, mode, dev);
1578         }
1579
1580 #ifdef CONFIG_NTFS3_FS_POSIX_ACL
1581         if (!S_ISLNK(mode) && (sb->s_flags & SB_POSIXACL)) {
1582                 err = ntfs_init_acl(mnt_userns, inode, dir);
1583                 if (err)
1584                         goto out7;
1585         } else
1586 #endif
1587         {
1588                 inode->i_flags |= S_NOSEC;
1589         }
1590
1591         /* Write non resident data. */
1592         if (nsize) {
1593                 err = ntfs_sb_write_run(sbi, &ni->file.run, 0, rp, nsize, 0);
1594                 if (err)
1595                         goto out7;
1596         }
1597
1598         /*
1599          * Call 'd_instantiate' after inode->i_op is set
1600          * but before finish_open.
1601          */
1602         d_instantiate(dentry, inode);
1603
1604         ntfs_save_wsl_perm(inode);
1605         mark_inode_dirty(dir);
1606         mark_inode_dirty(inode);
1607
1608         /* Normal exit. */
1609         goto out2;
1610
1611 out7:
1612
1613         /* Undo 'indx_insert_entry'. */
1614         ni_lock_dir(dir_ni);
1615         indx_delete_entry(&dir_ni->dir, dir_ni, new_de + 1,
1616                           le16_to_cpu(new_de->key_size), sbi);
1617         /* ni_unlock(dir_ni); will be called later. */
1618 out6:
1619         if (rp_inserted)
1620                 ntfs_remove_reparse(sbi, IO_REPARSE_TAG_SYMLINK, &new_de->ref);
1621
1622 out5:
1623         if (S_ISDIR(mode) || run_is_empty(&ni->file.run))
1624                 goto out4;
1625
1626         run_deallocate(sbi, &ni->file.run, false);
1627
1628 out4:
1629         clear_rec_inuse(rec);
1630         clear_nlink(inode);
1631         ni->mi.dirty = false;
1632         discard_new_inode(inode);
1633 out3:
1634         ntfs_mark_rec_free(sbi, ino);
1635
1636 out2:
1637         __putname(new_de);
1638         kfree(rp);
1639
1640 out1:
1641         if (err) {
1642                 ni_unlock(dir_ni);
1643                 return ERR_PTR(err);
1644         }
1645
1646         unlock_new_inode(inode);
1647
1648         return inode;
1649 }
1650
1651 int ntfs_link_inode(struct inode *inode, struct dentry *dentry)
1652 {
1653         int err;
1654         struct ntfs_inode *ni = ntfs_i(inode);
1655         struct ntfs_sb_info *sbi = inode->i_sb->s_fs_info;
1656         struct NTFS_DE *de;
1657         struct ATTR_FILE_NAME *de_name;
1658
1659         /* Allocate PATH_MAX bytes. */
1660         de = __getname();
1661         if (!de)
1662                 return -ENOMEM;
1663
1664         /* Mark rw ntfs as dirty. It will be cleared at umount. */
1665         ntfs_set_state(sbi, NTFS_DIRTY_DIRTY);
1666
1667         /* Construct 'de'. */
1668         err = fill_name_de(sbi, de, &dentry->d_name, NULL);
1669         if (err)
1670                 goto out;
1671
1672         de_name = (struct ATTR_FILE_NAME *)(de + 1);
1673         /* Fill duplicate info. */
1674         de_name->dup.cr_time = de_name->dup.m_time = de_name->dup.c_time =
1675                 de_name->dup.a_time = kernel2nt(&inode->i_ctime);
1676         de_name->dup.alloc_size = de_name->dup.data_size =
1677                 cpu_to_le64(inode->i_size);
1678         de_name->dup.fa = ni->std_fa;
1679         de_name->dup.ea_size = de_name->dup.reparse = 0;
1680
1681         err = ni_add_name(ntfs_i(d_inode(dentry->d_parent)), ni, de);
1682 out:
1683         __putname(de);
1684         return err;
1685 }
1686
1687 /*
1688  * ntfs_unlink_inode
1689  *
1690  * inode_operations::unlink
1691  * inode_operations::rmdir
1692  */
1693 int ntfs_unlink_inode(struct inode *dir, const struct dentry *dentry)
1694 {
1695         int err;
1696         struct ntfs_sb_info *sbi = dir->i_sb->s_fs_info;
1697         struct inode *inode = d_inode(dentry);
1698         struct ntfs_inode *ni = ntfs_i(inode);
1699         struct ntfs_inode *dir_ni = ntfs_i(dir);
1700         struct NTFS_DE *de, *de2 = NULL;
1701         int undo_remove;
1702
1703         if (ntfs_is_meta_file(sbi, ni->mi.rno))
1704                 return -EINVAL;
1705
1706         /* Allocate PATH_MAX bytes. */
1707         de = __getname();
1708         if (!de)
1709                 return -ENOMEM;
1710
1711         ni_lock(ni);
1712
1713         if (S_ISDIR(inode->i_mode) && !dir_is_empty(inode)) {
1714                 err = -ENOTEMPTY;
1715                 goto out;
1716         }
1717
1718         err = fill_name_de(sbi, de, &dentry->d_name, NULL);
1719         if (err < 0)
1720                 goto out;
1721
1722         undo_remove = 0;
1723         err = ni_remove_name(dir_ni, ni, de, &de2, &undo_remove);
1724
1725         if (!err) {
1726                 drop_nlink(inode);
1727                 dir->i_mtime = dir->i_ctime = current_time(dir);
1728                 mark_inode_dirty(dir);
1729                 inode->i_ctime = dir->i_ctime;
1730                 if (inode->i_nlink)
1731                         mark_inode_dirty(inode);
1732         } else if (!ni_remove_name_undo(dir_ni, ni, de, de2, undo_remove)) {
1733                 make_bad_inode(inode);
1734                 ntfs_inode_err(inode, "failed to undo unlink");
1735                 ntfs_set_state(sbi, NTFS_DIRTY_ERROR);
1736         } else {
1737                 if (ni_is_dirty(dir))
1738                         mark_inode_dirty(dir);
1739                 if (ni_is_dirty(inode))
1740                         mark_inode_dirty(inode);
1741         }
1742
1743 out:
1744         ni_unlock(ni);
1745         __putname(de);
1746         return err;
1747 }
1748
1749 void ntfs_evict_inode(struct inode *inode)
1750 {
1751         truncate_inode_pages_final(&inode->i_data);
1752
1753         if (inode->i_nlink)
1754                 _ni_write_inode(inode, inode_needs_sync(inode));
1755
1756         invalidate_inode_buffers(inode);
1757         clear_inode(inode);
1758
1759         ni_clear(ntfs_i(inode));
1760 }
1761
1762 static noinline int ntfs_readlink_hlp(struct inode *inode, char *buffer,
1763                                       int buflen)
1764 {
1765         int i, err = -EINVAL;
1766         struct ntfs_inode *ni = ntfs_i(inode);
1767         struct super_block *sb = inode->i_sb;
1768         struct ntfs_sb_info *sbi = sb->s_fs_info;
1769         u64 size;
1770         u16 ulen = 0;
1771         void *to_free = NULL;
1772         struct REPARSE_DATA_BUFFER *rp;
1773         const __le16 *uname;
1774         struct ATTRIB *attr;
1775
1776         /* Reparse data present. Try to parse it. */
1777         static_assert(!offsetof(struct REPARSE_DATA_BUFFER, ReparseTag));
1778         static_assert(sizeof(u32) == sizeof(rp->ReparseTag));
1779
1780         *buffer = 0;
1781
1782         attr = ni_find_attr(ni, NULL, NULL, ATTR_REPARSE, NULL, 0, NULL, NULL);
1783         if (!attr)
1784                 goto out;
1785
1786         if (!attr->non_res) {
1787                 rp = resident_data_ex(attr, sizeof(struct REPARSE_DATA_BUFFER));
1788                 if (!rp)
1789                         goto out;
1790                 size = le32_to_cpu(attr->res.data_size);
1791         } else {
1792                 size = le64_to_cpu(attr->nres.data_size);
1793                 rp = NULL;
1794         }
1795
1796         if (size > sbi->reparse.max_size || size <= sizeof(u32))
1797                 goto out;
1798
1799         if (!rp) {
1800                 rp = kmalloc(size, GFP_NOFS);
1801                 if (!rp) {
1802                         err = -ENOMEM;
1803                         goto out;
1804                 }
1805                 to_free = rp;
1806                 /* Read into temporal buffer. */
1807                 err = ntfs_read_run_nb(sbi, &ni->file.run, 0, rp, size, NULL);
1808                 if (err)
1809                         goto out;
1810         }
1811
1812         /* Microsoft Tag. */
1813         switch (rp->ReparseTag) {
1814         case IO_REPARSE_TAG_MOUNT_POINT:
1815                 /* Mount points and junctions. */
1816                 /* Can we use 'Rp->MountPointReparseBuffer.PrintNameLength'? */
1817                 if (size <= offsetof(struct REPARSE_DATA_BUFFER,
1818                                      MountPointReparseBuffer.PathBuffer))
1819                         goto out;
1820                 uname = Add2Ptr(rp,
1821                                 offsetof(struct REPARSE_DATA_BUFFER,
1822                                          MountPointReparseBuffer.PathBuffer) +
1823                                         le16_to_cpu(rp->MountPointReparseBuffer
1824                                                             .PrintNameOffset));
1825                 ulen = le16_to_cpu(rp->MountPointReparseBuffer.PrintNameLength);
1826                 break;
1827
1828         case IO_REPARSE_TAG_SYMLINK:
1829                 /* FolderSymbolicLink */
1830                 /* Can we use 'Rp->SymbolicLinkReparseBuffer.PrintNameLength'? */
1831                 if (size <= offsetof(struct REPARSE_DATA_BUFFER,
1832                                      SymbolicLinkReparseBuffer.PathBuffer))
1833                         goto out;
1834                 uname = Add2Ptr(
1835                         rp, offsetof(struct REPARSE_DATA_BUFFER,
1836                                      SymbolicLinkReparseBuffer.PathBuffer) +
1837                                     le16_to_cpu(rp->SymbolicLinkReparseBuffer
1838                                                         .PrintNameOffset));
1839                 ulen = le16_to_cpu(
1840                         rp->SymbolicLinkReparseBuffer.PrintNameLength);
1841                 break;
1842
1843         case IO_REPARSE_TAG_CLOUD:
1844         case IO_REPARSE_TAG_CLOUD_1:
1845         case IO_REPARSE_TAG_CLOUD_2:
1846         case IO_REPARSE_TAG_CLOUD_3:
1847         case IO_REPARSE_TAG_CLOUD_4:
1848         case IO_REPARSE_TAG_CLOUD_5:
1849         case IO_REPARSE_TAG_CLOUD_6:
1850         case IO_REPARSE_TAG_CLOUD_7:
1851         case IO_REPARSE_TAG_CLOUD_8:
1852         case IO_REPARSE_TAG_CLOUD_9:
1853         case IO_REPARSE_TAG_CLOUD_A:
1854         case IO_REPARSE_TAG_CLOUD_B:
1855         case IO_REPARSE_TAG_CLOUD_C:
1856         case IO_REPARSE_TAG_CLOUD_D:
1857         case IO_REPARSE_TAG_CLOUD_E:
1858         case IO_REPARSE_TAG_CLOUD_F:
1859                 err = sizeof("OneDrive") - 1;
1860                 if (err > buflen)
1861                         err = buflen;
1862                 memcpy(buffer, "OneDrive", err);
1863                 goto out;
1864
1865         default:
1866                 if (IsReparseTagMicrosoft(rp->ReparseTag)) {
1867                         /* Unknown Microsoft Tag. */
1868                         goto out;
1869                 }
1870                 if (!IsReparseTagNameSurrogate(rp->ReparseTag) ||
1871                     size <= sizeof(struct REPARSE_POINT)) {
1872                         goto out;
1873                 }
1874
1875                 /* Users tag. */
1876                 uname = Add2Ptr(rp, sizeof(struct REPARSE_POINT));
1877                 ulen = le16_to_cpu(rp->ReparseDataLength) -
1878                        sizeof(struct REPARSE_POINT);
1879         }
1880
1881         /* Convert nlen from bytes to UNICODE chars. */
1882         ulen >>= 1;
1883
1884         /* Check that name is available. */
1885         if (!ulen || uname + ulen > (__le16 *)Add2Ptr(rp, size))
1886                 goto out;
1887
1888         /* If name is already zero terminated then truncate it now. */
1889         if (!uname[ulen - 1])
1890                 ulen -= 1;
1891
1892         err = ntfs_utf16_to_nls(sbi, uname, ulen, buffer, buflen);
1893
1894         if (err < 0)
1895                 goto out;
1896
1897         /* Translate Windows '\' into Linux '/'. */
1898         for (i = 0; i < err; i++) {
1899                 if (buffer[i] == '\\')
1900                         buffer[i] = '/';
1901         }
1902
1903         /* Always set last zero. */
1904         buffer[err] = 0;
1905 out:
1906         kfree(to_free);
1907         return err;
1908 }
1909
1910 static const char *ntfs_get_link(struct dentry *de, struct inode *inode,
1911                                  struct delayed_call *done)
1912 {
1913         int err;
1914         char *ret;
1915
1916         if (!de)
1917                 return ERR_PTR(-ECHILD);
1918
1919         ret = kmalloc(PAGE_SIZE, GFP_NOFS);
1920         if (!ret)
1921                 return ERR_PTR(-ENOMEM);
1922
1923         err = ntfs_readlink_hlp(inode, ret, PAGE_SIZE);
1924         if (err < 0) {
1925                 kfree(ret);
1926                 return ERR_PTR(err);
1927         }
1928
1929         set_delayed_call(done, kfree_link, ret);
1930
1931         return ret;
1932 }
1933
1934 // clang-format off
1935 const struct inode_operations ntfs_link_inode_operations = {
1936         .get_link       = ntfs_get_link,
1937         .setattr        = ntfs3_setattr,
1938         .listxattr      = ntfs_listxattr,
1939         .permission     = ntfs_permission,
1940         .get_acl        = ntfs_get_acl,
1941         .set_acl        = ntfs_set_acl,
1942 };
1943
1944 const struct address_space_operations ntfs_aops = {
1945         .readpage       = ntfs_readpage,
1946         .readahead      = ntfs_readahead,
1947         .writepage      = ntfs_writepage,
1948         .writepages     = ntfs_writepages,
1949         .write_begin    = ntfs_write_begin,
1950         .write_end      = ntfs_write_end,
1951         .direct_IO      = ntfs_direct_IO,
1952         .bmap           = ntfs_bmap,
1953         .dirty_folio    = block_dirty_folio,
1954 };
1955
1956 const struct address_space_operations ntfs_aops_cmpr = {
1957         .readpage       = ntfs_readpage,
1958         .readahead      = ntfs_readahead,
1959 };
1960 // clang-format on