1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Cluster (de)allocation code.
4 *
5 * Copyright (c) 2004-2005 Anton Altaparmakov
6 * Copyright (c) 2025 LG Electronics Co., Ltd.
7 *
8 * Part of this file is based on code from the NTFS-3G.
9 * and is copyrighted by the respective authors below:
10 * Copyright (c) 2002-2004 Anton Altaparmakov
11 * Copyright (c) 2004 Yura Pakhuchiy
12 * Copyright (c) 2004-2008 Szabolcs Szakacsits
13 * Copyright (c) 2008-2009 Jean-Pierre Andre
14 */
15
16 #include <linux/blkdev.h>
17
18 #include "lcnalloc.h"
19 #include "bitmap.h"
20 #include "ntfs.h"
21
22 /*
23 * ntfs_cluster_free_from_rl_nolock - free clusters from runlist
24 * @vol: mounted ntfs volume on which to free the clusters
25 * @rl: runlist describing the clusters to free
26 *
27 * Free all the clusters described by the runlist @rl on the volume @vol. In
28 * the case of an error being returned, at least some of the clusters were not
29 * freed.
30 *
31 * Return 0 on success and -errno on error.
32 *
33 * Locking: - The volume lcn bitmap must be locked for writing on entry and is
34 * left locked on return.
35 */
ntfs_cluster_free_from_rl_nolock(struct ntfs_volume * vol,const struct runlist_element * rl)36 int ntfs_cluster_free_from_rl_nolock(struct ntfs_volume *vol,
37 const struct runlist_element *rl)
38 {
39 struct inode *lcnbmp_vi = vol->lcnbmp_ino;
40 int ret = 0;
41 s64 nr_freed = 0;
42
43 ntfs_debug("Entering.");
44 if (!rl)
45 return 0;
46
47 if (!NVolFreeClusterKnown(vol))
48 wait_event(vol->free_waitq, NVolFreeClusterKnown(vol));
49
50 for (; rl->length; rl++) {
51 int err;
52
53 if (rl->lcn < 0)
54 continue;
55 err = ntfs_bitmap_clear_run(lcnbmp_vi, rl->lcn, rl->length);
56 if (likely(!err))
57 nr_freed += rl->length;
58 else if (!ret || ret == -ENOMEM)
59 ret = err;
60 }
61 ntfs_inc_free_clusters(vol, nr_freed);
62 ntfs_debug("Done.");
63 return ret;
64 }
65
max_empty_bit_range(unsigned char * buf,int size)66 static s64 max_empty_bit_range(unsigned char *buf, int size)
67 {
68 int i, j, run = 0;
69 int max_range = 0;
70 s64 start_pos = -1;
71
72 ntfs_debug("Entering\n");
73
74 i = 0;
75 while (i < size) {
76 switch (*buf) {
77 case 0:
78 do {
79 buf++;
80 run += 8;
81 i++;
82 } while ((i < size) && !*buf);
83 break;
84 case 255:
85 if (run > max_range) {
86 max_range = run;
87 start_pos = (s64)i * 8 - run;
88 }
89 run = 0;
90 do {
91 buf++;
92 i++;
93 } while ((i < size) && (*buf == 255));
94 break;
95 default:
96 for (j = 0; j < 8; j++) {
97 int bit = *buf & (1 << j);
98
99 if (bit) {
100 if (run > max_range) {
101 max_range = run;
102 start_pos = (s64)i * 8 + (j - run);
103 }
104 run = 0;
105 } else
106 run++;
107 }
108 i++;
109 buf++;
110 }
111 }
112
113 if (run > max_range)
114 start_pos = (s64)i * 8 - run;
115
116 return start_pos;
117 }
118
119 /*
120 * ntfs_cluster_alloc - allocate clusters on an ntfs volume
121 * @vol: mounted ntfs volume on which to allocate clusters
122 * @start_vcn: vcn of the first allocated cluster
123 * @count: number of clusters to allocate
124 * @start_lcn: starting lcn at which to allocate the clusters or -1 if none
125 * @zone: zone from which to allocate (MFT_ZONE or DATA_ZONE)
126 * @is_extension: if true, the caller is extending an attribute
127 * @is_contig: if true, require contiguous allocation
128 * @is_dealloc: if true, the allocation is for deallocation purposes
129 *
130 * Allocate @count clusters preferably starting at cluster @start_lcn or at the
131 * current allocator position if @start_lcn is -1, on the mounted ntfs volume
132 * @vol. @zone is either DATA_ZONE for allocation of normal clusters or
133 * MFT_ZONE for allocation of clusters for the master file table, i.e. the
134 * $MFT/$DATA attribute.
135 *
136 * @start_vcn specifies the vcn of the first allocated cluster. This makes
137 * merging the resulting runlist with the old runlist easier.
138 *
139 * If @is_extension is 'true', the caller is allocating clusters to extend an
140 * attribute and if it is 'false', the caller is allocating clusters to fill a
141 * hole in an attribute. Practically the difference is that if @is_extension
142 * is 'true' the returned runlist will be terminated with LCN_ENOENT and if
143 * @is_extension is 'false' the runlist will be terminated with
144 * LCN_RL_NOT_MAPPED.
145 *
146 * You need to check the return value with IS_ERR(). If this is false, the
147 * function was successful and the return value is a runlist describing the
148 * allocated cluster(s). If IS_ERR() is true, the function failed and
149 * PTR_ERR() gives you the error code.
150 *
151 * Notes on the allocation algorithm
152 * =================================
153 *
154 * There are two data zones. First is the area between the end of the mft zone
155 * and the end of the volume, and second is the area between the start of the
156 * volume and the start of the mft zone. On unmodified/standard NTFS 1.x
157 * volumes, the second data zone does not exist due to the mft zone being
158 * expanded to cover the start of the volume in order to reserve space for the
159 * mft bitmap attribute.
160 *
161 * This is not the prettiest function but the complexity stems from the need of
162 * implementing the mft vs data zoned approach and from the fact that we have
163 * access to the lcn bitmap in portions of up to 8192 bytes at a time, so we
164 * need to cope with crossing over boundaries of two buffers. Further, the
165 * fact that the allocator allows for caller supplied hints as to the location
166 * of where allocation should begin and the fact that the allocator keeps track
167 * of where in the data zones the next natural allocation should occur,
168 * contribute to the complexity of the function. But it should all be
169 * worthwhile, because this allocator should: 1) be a full implementation of
170 * the MFT zone approach used by Windows NT, 2) cause reduction in
171 * fragmentation, and 3) be speedy in allocations (the code is not optimized
172 * for speed, but the algorithm is, so further speed improvements are probably
173 * possible).
174 *
175 * Locking: - The volume lcn bitmap must be unlocked on entry and is unlocked
176 * on return.
177 * - This function takes the volume lcn bitmap lock for writing and
178 * modifies the bitmap contents.
179 *
180 * Return: Runlist describing the allocated cluster(s) on success, error pointer
181 * on failure.
182 */
ntfs_cluster_alloc(struct ntfs_volume * vol,const s64 start_vcn,const s64 count,const s64 start_lcn,const int zone,const bool is_extension,const bool is_contig,const bool is_dealloc)183 struct runlist_element *ntfs_cluster_alloc(struct ntfs_volume *vol, const s64 start_vcn,
184 const s64 count, const s64 start_lcn,
185 const int zone,
186 const bool is_extension,
187 const bool is_contig,
188 const bool is_dealloc)
189 {
190 s64 zone_start, zone_end, bmp_pos, bmp_initial_pos, last_read_pos, lcn;
191 s64 prev_lcn = 0, prev_run_len = 0, mft_zone_size;
192 s64 clusters, free_clusters;
193 loff_t i_size;
194 struct inode *lcnbmp_vi;
195 struct runlist_element *rl = NULL;
196 struct address_space *mapping;
197 struct folio *folio = NULL;
198 u8 *buf = NULL, *byte;
199 int err = 0, rlpos, rlsize, buf_size, pg_off;
200 u8 pass, done_zones, search_zone, need_writeback = 0, bit;
201 unsigned int memalloc_flags;
202 u8 has_guess, used_zone_pos;
203 pgoff_t index;
204
205 ntfs_debug("Entering for start_vcn 0x%llx, count 0x%llx, start_lcn 0x%llx, zone %s_ZONE.",
206 start_vcn, count, start_lcn,
207 zone == MFT_ZONE ? "MFT" : "DATA");
208
209 lcnbmp_vi = vol->lcnbmp_ino;
210 if (start_vcn < 0 || start_lcn < LCN_HOLE ||
211 zone < FIRST_ZONE || zone > LAST_ZONE)
212 return ERR_PTR(-EINVAL);
213
214 /* Return NULL if @count is zero. */
215 if (count < 0 || !count)
216 return ERR_PTR(-EINVAL);
217
218 memalloc_flags = memalloc_nofs_save();
219
220 if (!NVolFreeClusterKnown(vol))
221 wait_event(vol->free_waitq, NVolFreeClusterKnown(vol));
222 free_clusters = atomic64_read(&vol->free_clusters);
223
224 /* Take the lcnbmp lock for writing. */
225 down_write(&vol->lcnbmp_lock);
226 if (is_dealloc == false)
227 free_clusters -= atomic64_read(&vol->dirty_clusters);
228
229 if (free_clusters < count) {
230 err = -ENOSPC;
231 goto out_restore;
232 }
233
234 /*
235 * If no specific @start_lcn was requested, use the current data zone
236 * position, otherwise use the requested @start_lcn but make sure it
237 * lies outside the mft zone. Also set done_zones to 0 (no zones done)
238 * and pass depending on whether we are starting inside a zone (1) or
239 * at the beginning of a zone (2). If requesting from the MFT_ZONE,
240 * we either start at the current position within the mft zone or at
241 * the specified position. If the latter is out of bounds then we start
242 * at the beginning of the MFT_ZONE.
243 */
244 done_zones = 0;
245 pass = 1;
246 /*
247 * zone_start and zone_end are the current search range. search_zone
248 * is 1 for mft zone, 2 for data zone 1 (end of mft zone till end of
249 * volume) and 4 for data zone 2 (start of volume till start of mft
250 * zone).
251 */
252 has_guess = 1;
253 zone_start = start_lcn;
254
255 if (zone_start < 0) {
256 if (zone == DATA_ZONE)
257 zone_start = vol->data1_zone_pos;
258 else
259 zone_start = vol->mft_zone_pos;
260 if (!zone_start) {
261 /*
262 * Zone starts at beginning of volume which means a
263 * single pass is sufficient.
264 */
265 pass = 2;
266 }
267 has_guess = 0;
268 }
269
270 used_zone_pos = has_guess ? 0 : 1;
271
272 if (!zone_start || zone_start == vol->mft_zone_start ||
273 zone_start == vol->mft_zone_end)
274 pass = 2;
275
276 if (zone_start < vol->mft_zone_start) {
277 zone_end = vol->mft_zone_start;
278 search_zone = 4;
279 /* Skip searching the mft zone. */
280 done_zones |= 1;
281 } else if (zone_start < vol->mft_zone_end) {
282 zone_end = vol->mft_zone_end;
283 search_zone = 1;
284 } else {
285 zone_end = vol->nr_clusters;
286 search_zone = 2;
287 /* Skip searching the mft zone. */
288 done_zones |= 1;
289 }
290
291 /*
292 * bmp_pos is the current bit position inside the bitmap. We use
293 * bmp_initial_pos to determine whether or not to do a zone switch.
294 */
295 bmp_pos = bmp_initial_pos = zone_start;
296
297 /* Loop until all clusters are allocated, i.e. clusters == 0. */
298 clusters = count;
299 rlpos = rlsize = 0;
300 mapping = lcnbmp_vi->i_mapping;
301 /*
302 * lcn_empty_bits_per_page is sized from nr_clusters, but $Bitmap can
303 * cover more clusters than that; bound the scan by the array.
304 */
305 i_size = min_t(s64, i_size_read(lcnbmp_vi),
306 ((s64)vol->nr_clusters + 7) >> 3);
307 while (1) {
308 ntfs_debug("Start of outer while loop: done_zones 0x%x, search_zone %i, pass %i, zone_start 0x%llx, zone_end 0x%llx, bmp_initial_pos 0x%llx, bmp_pos 0x%llx, rlpos %i, rlsize %i.",
309 done_zones, search_zone, pass,
310 zone_start, zone_end, bmp_initial_pos,
311 bmp_pos, rlpos, rlsize);
312 /* Loop until we run out of free clusters. */
313 last_read_pos = bmp_pos >> 3;
314 ntfs_debug("last_read_pos 0x%llx.", last_read_pos);
315 if (last_read_pos >= i_size) {
316 ntfs_debug("End of attribute reached. Skipping to zone_pass_done.");
317 goto zone_pass_done;
318 }
319 if (likely(folio)) {
320 if (need_writeback) {
321 ntfs_debug("Marking page dirty.");
322 folio_mark_dirty(folio);
323 need_writeback = 0;
324 }
325 folio_unlock(folio);
326 kunmap_local(buf);
327 folio_put(folio);
328 folio = NULL;
329 }
330
331 index = last_read_pos >> PAGE_SHIFT;
332 pg_off = last_read_pos & ~PAGE_MASK;
333 buf_size = PAGE_SIZE - pg_off;
334 if (unlikely(last_read_pos + buf_size > i_size))
335 buf_size = i_size - last_read_pos;
336 buf_size <<= 3;
337 lcn = bmp_pos & 7;
338 bmp_pos &= ~(s64)7;
339
340 if (vol->lcn_empty_bits_per_page[index] == 0)
341 goto next_bmp_pos;
342
343 folio = read_mapping_folio(mapping, index, NULL);
344 if (IS_ERR(folio)) {
345 err = PTR_ERR(folio);
346 ntfs_error(vol->sb, "Failed to map page.");
347 goto out;
348 }
349
350 folio_lock(folio);
351 buf = kmap_local_folio(folio, 0) + pg_off;
352 ntfs_debug("Before inner while loop: buf_size %i, lcn 0x%llx, bmp_pos 0x%llx, need_writeback %i.",
353 buf_size, lcn, bmp_pos, need_writeback);
354 while (lcn < buf_size && lcn + bmp_pos < zone_end) {
355 byte = buf + (lcn >> 3);
356 ntfs_debug("In inner while loop: buf_size %i, lcn 0x%llx, bmp_pos 0x%llx, need_writeback %i, byte ofs 0x%x, *byte 0x%x.",
357 buf_size, lcn, bmp_pos, need_writeback,
358 (unsigned int)(lcn >> 3),
359 (unsigned int)*byte);
360 bit = 1 << (lcn & 7);
361 ntfs_debug("bit 0x%x.", bit);
362
363 if (has_guess) {
364 if (*byte & bit) {
365 if (is_contig == true && prev_run_len > 0)
366 goto done;
367
368 has_guess = 0;
369 break;
370 }
371 } else {
372 lcn = max_empty_bit_range(buf, buf_size >> 3);
373 if (lcn < 0)
374 break;
375 has_guess = 1;
376 continue;
377 }
378 /*
379 * Allocate more memory if needed, including space for
380 * the terminator element.
381 * kvzalloc() operates on whole pages only.
382 */
383 if ((rlpos + 2) * sizeof(*rl) > rlsize) {
384 struct runlist_element *rl2;
385
386 ntfs_debug("Reallocating memory.");
387 if (!rl)
388 ntfs_debug("First free bit is at s64 0x%llx.",
389 lcn + bmp_pos);
390 rl2 = kvzalloc(rlsize + PAGE_SIZE, GFP_NOFS);
391 if (unlikely(!rl2)) {
392 err = -ENOMEM;
393 ntfs_error(vol->sb, "Failed to allocate memory.");
394 goto out;
395 }
396 memcpy(rl2, rl, rlsize);
397 kvfree(rl);
398 rl = rl2;
399 rlsize += PAGE_SIZE;
400 ntfs_debug("Reallocated memory, rlsize 0x%x.",
401 rlsize);
402 }
403 /* Allocate the bitmap bit. */
404 *byte |= bit;
405 /* We need to write this bitmap page to disk. */
406 need_writeback = 1;
407 ntfs_debug("*byte 0x%x, need_writeback is set.",
408 (unsigned int)*byte);
409 ntfs_dec_free_clusters(vol, 1);
410 ntfs_set_lcn_empty_bits(vol, index, 1, 1);
411
412 /*
413 * Coalesce with previous run if adjacent LCNs.
414 * Otherwise, append a new run.
415 */
416 ntfs_debug("Adding run (lcn 0x%llx, len 0x%llx), prev_lcn 0x%llx, lcn 0x%llx, bmp_pos 0x%llx, prev_run_len 0x%llx, rlpos %i.",
417 lcn + bmp_pos, 1ULL, prev_lcn,
418 lcn, bmp_pos, prev_run_len, rlpos);
419 if (prev_lcn == lcn + bmp_pos - prev_run_len && rlpos) {
420 ntfs_debug("Coalescing to run (lcn 0x%llx, len 0x%llx).",
421 rl[rlpos - 1].lcn,
422 rl[rlpos - 1].length);
423 rl[rlpos - 1].length = ++prev_run_len;
424 ntfs_debug("Run now (lcn 0x%llx, len 0x%llx), prev_run_len 0x%llx.",
425 rl[rlpos - 1].lcn,
426 rl[rlpos - 1].length,
427 prev_run_len);
428 } else {
429 if (likely(rlpos)) {
430 ntfs_debug("Adding new run, (previous run lcn 0x%llx, len 0x%llx).",
431 rl[rlpos - 1].lcn, rl[rlpos - 1].length);
432 rl[rlpos].vcn = rl[rlpos - 1].vcn +
433 prev_run_len;
434 } else {
435 ntfs_debug("Adding new run, is first run.");
436 rl[rlpos].vcn = start_vcn;
437 }
438 rl[rlpos].lcn = prev_lcn = lcn + bmp_pos;
439 rl[rlpos].length = prev_run_len = 1;
440 rlpos++;
441 }
442 /* Done? */
443 if (!--clusters) {
444 s64 tc;
445 done:
446 if (!used_zone_pos)
447 goto out;
448 /*
449 * Update the current zone position. Positions
450 * of already scanned zones have been updated
451 * during the respective zone switches.
452 */
453 tc = lcn + bmp_pos + 1;
454 ntfs_debug("Done. Updating current zone position, tc 0x%llx, search_zone %i.",
455 tc, search_zone);
456 switch (search_zone) {
457 case 1:
458 ntfs_debug("Before checks, vol->mft_zone_pos 0x%llx.",
459 vol->mft_zone_pos);
460 if (tc >= vol->mft_zone_end) {
461 vol->mft_zone_pos =
462 vol->mft_lcn;
463 if (!vol->mft_zone_end)
464 vol->mft_zone_pos = 0;
465 } else if ((bmp_initial_pos >=
466 vol->mft_zone_pos ||
467 tc > vol->mft_zone_pos)
468 && tc >= vol->mft_lcn)
469 vol->mft_zone_pos = tc;
470 ntfs_debug("After checks, vol->mft_zone_pos 0x%llx.",
471 vol->mft_zone_pos);
472 break;
473 case 2:
474 ntfs_debug("Before checks, vol->data1_zone_pos 0x%llx.",
475 vol->data1_zone_pos);
476 if (tc >= vol->nr_clusters)
477 vol->data1_zone_pos =
478 vol->mft_zone_end;
479 else if ((bmp_initial_pos >=
480 vol->data1_zone_pos ||
481 tc > vol->data1_zone_pos)
482 && tc >= vol->mft_zone_end)
483 vol->data1_zone_pos = tc;
484 ntfs_debug("After checks, vol->data1_zone_pos 0x%llx.",
485 vol->data1_zone_pos);
486 break;
487 case 4:
488 ntfs_debug("Before checks, vol->data2_zone_pos 0x%llx.",
489 vol->data2_zone_pos);
490 if (tc >= vol->mft_zone_start)
491 vol->data2_zone_pos = 0;
492 else if (bmp_initial_pos >=
493 vol->data2_zone_pos ||
494 tc > vol->data2_zone_pos)
495 vol->data2_zone_pos = tc;
496 ntfs_debug("After checks, vol->data2_zone_pos 0x%llx.",
497 vol->data2_zone_pos);
498 break;
499 default:
500 WARN_ON(1);
501 }
502 ntfs_debug("Finished. Going to out.");
503 goto out;
504 }
505 lcn++;
506 }
507
508 if (!used_zone_pos) {
509 used_zone_pos = 1;
510 if (search_zone == 1)
511 zone_start = vol->mft_zone_pos;
512 else if (search_zone == 2)
513 zone_start = vol->data1_zone_pos;
514 else
515 zone_start = vol->data2_zone_pos;
516
517 if (!zone_start || zone_start == vol->mft_zone_start ||
518 zone_start == vol->mft_zone_end)
519 pass = 2;
520 bmp_pos = zone_start;
521 } else {
522 next_bmp_pos:
523 bmp_pos += buf_size;
524 }
525
526 ntfs_debug("After inner while loop: buf_size 0x%x, lcn 0x%llx, bmp_pos 0x%llx, need_writeback %i.",
527 buf_size, lcn, bmp_pos, need_writeback);
528 if (bmp_pos < zone_end) {
529 ntfs_debug("Continuing outer while loop, bmp_pos 0x%llx, zone_end 0x%llx.",
530 bmp_pos, zone_end);
531 continue;
532 }
533 zone_pass_done: /* Finished with the current zone pass. */
534 ntfs_debug("At zone_pass_done, pass %i.", pass);
535 if (pass == 1) {
536 /*
537 * Now do pass 2, scanning the first part of the zone
538 * we omitted in pass 1.
539 */
540 pass = 2;
541 zone_end = zone_start;
542 switch (search_zone) {
543 case 1: /* mft_zone */
544 zone_start = vol->mft_zone_start;
545 break;
546 case 2: /* data1_zone */
547 zone_start = vol->mft_zone_end;
548 break;
549 case 4: /* data2_zone */
550 zone_start = 0;
551 break;
552 default:
553 WARN_ON(1);
554 }
555 /* Sanity check. */
556 if (zone_end < zone_start)
557 zone_end = zone_start;
558 bmp_pos = zone_start;
559 ntfs_debug("Continuing outer while loop, pass 2, zone_start 0x%llx, zone_end 0x%llx, bmp_pos 0x%llx.",
560 zone_start, zone_end, bmp_pos);
561 continue;
562 } /* pass == 2 */
563 done_zones_check:
564 ntfs_debug("At done_zones_check, search_zone %i, done_zones before 0x%x, done_zones after 0x%x.",
565 search_zone, done_zones,
566 done_zones | search_zone);
567 done_zones |= search_zone;
568 if (done_zones < 7) {
569 ntfs_debug("Switching zone.");
570 /* Now switch to the next zone we haven't done yet. */
571 pass = 1;
572 switch (search_zone) {
573 case 1:
574 ntfs_debug("Switching from mft zone to data1 zone.");
575 /* Update mft zone position. */
576 if (rlpos && used_zone_pos) {
577 s64 tc;
578
579 ntfs_debug("Before checks, vol->mft_zone_pos 0x%llx.",
580 vol->mft_zone_pos);
581 tc = rl[rlpos - 1].lcn +
582 rl[rlpos - 1].length;
583 if (tc >= vol->mft_zone_end) {
584 vol->mft_zone_pos =
585 vol->mft_lcn;
586 if (!vol->mft_zone_end)
587 vol->mft_zone_pos = 0;
588 } else if ((bmp_initial_pos >=
589 vol->mft_zone_pos ||
590 tc > vol->mft_zone_pos)
591 && tc >= vol->mft_lcn)
592 vol->mft_zone_pos = tc;
593 ntfs_debug("After checks, vol->mft_zone_pos 0x%llx.",
594 vol->mft_zone_pos);
595 }
596 /* Switch from mft zone to data1 zone. */
597 switch_to_data1_zone: search_zone = 2;
598 zone_start = bmp_initial_pos =
599 vol->data1_zone_pos;
600 zone_end = vol->nr_clusters;
601 if (zone_start == vol->mft_zone_end)
602 pass = 2;
603 if (zone_start >= zone_end) {
604 vol->data1_zone_pos = zone_start =
605 vol->mft_zone_end;
606 pass = 2;
607 }
608 break;
609 case 2:
610 ntfs_debug("Switching from data1 zone to data2 zone.");
611 /* Update data1 zone position. */
612 if (rlpos && used_zone_pos) {
613 s64 tc;
614
615 ntfs_debug("Before checks, vol->data1_zone_pos 0x%llx.",
616 vol->data1_zone_pos);
617 tc = rl[rlpos - 1].lcn +
618 rl[rlpos - 1].length;
619 if (tc >= vol->nr_clusters)
620 vol->data1_zone_pos =
621 vol->mft_zone_end;
622 else if ((bmp_initial_pos >=
623 vol->data1_zone_pos ||
624 tc > vol->data1_zone_pos)
625 && tc >= vol->mft_zone_end)
626 vol->data1_zone_pos = tc;
627 ntfs_debug("After checks, vol->data1_zone_pos 0x%llx.",
628 vol->data1_zone_pos);
629 }
630 /* Switch from data1 zone to data2 zone. */
631 search_zone = 4;
632 zone_start = bmp_initial_pos =
633 vol->data2_zone_pos;
634 zone_end = vol->mft_zone_start;
635 if (!zone_start)
636 pass = 2;
637 if (zone_start >= zone_end) {
638 vol->data2_zone_pos = zone_start =
639 bmp_initial_pos = 0;
640 pass = 2;
641 }
642 break;
643 case 4:
644 ntfs_debug("Switching from data2 zone to data1 zone.");
645 /* Update data2 zone position. */
646 if (rlpos && used_zone_pos) {
647 s64 tc;
648
649 ntfs_debug("Before checks, vol->data2_zone_pos 0x%llx.",
650 vol->data2_zone_pos);
651 tc = rl[rlpos - 1].lcn +
652 rl[rlpos - 1].length;
653 if (tc >= vol->mft_zone_start)
654 vol->data2_zone_pos = 0;
655 else if (bmp_initial_pos >=
656 vol->data2_zone_pos ||
657 tc > vol->data2_zone_pos)
658 vol->data2_zone_pos = tc;
659 ntfs_debug("After checks, vol->data2_zone_pos 0x%llx.",
660 vol->data2_zone_pos);
661 }
662 /* Switch from data2 zone to data1 zone. */
663 goto switch_to_data1_zone;
664 default:
665 WARN_ON(1);
666 }
667 ntfs_debug("After zone switch, search_zone %i, pass %i, bmp_initial_pos 0x%llx, zone_start 0x%llx, zone_end 0x%llx.",
668 search_zone, pass,
669 bmp_initial_pos,
670 zone_start,
671 zone_end);
672 bmp_pos = zone_start;
673 if (zone_start == zone_end) {
674 ntfs_debug("Empty zone, going to done_zones_check.");
675 /* Empty zone. Don't bother searching it. */
676 goto done_zones_check;
677 }
678 ntfs_debug("Continuing outer while loop.");
679 continue;
680 } /* done_zones == 7 */
681 ntfs_debug("All zones are finished.");
682 /*
683 * All zones are finished! If DATA_ZONE, shrink mft zone. If
684 * MFT_ZONE, we have really run out of space.
685 */
686 mft_zone_size = vol->mft_zone_end - vol->mft_zone_start;
687 ntfs_debug("vol->mft_zone_start 0x%llx, vol->mft_zone_end 0x%llx, mft_zone_size 0x%llx.",
688 vol->mft_zone_start, vol->mft_zone_end,
689 mft_zone_size);
690 if (zone == MFT_ZONE || mft_zone_size <= 0) {
691 ntfs_debug("No free clusters left, going to out.");
692 /* Really no more space left on device. */
693 err = -ENOSPC;
694 goto out;
695 } /* zone == DATA_ZONE && mft_zone_size > 0 */
696 ntfs_debug("Shrinking mft zone.");
697 zone_end = vol->mft_zone_end;
698 mft_zone_size >>= 1;
699 if (mft_zone_size > 0)
700 vol->mft_zone_end = vol->mft_zone_start + mft_zone_size;
701 else /* mft zone and data2 zone no longer exist. */
702 vol->data2_zone_pos = vol->mft_zone_start =
703 vol->mft_zone_end = 0;
704 if (vol->mft_zone_pos >= vol->mft_zone_end) {
705 vol->mft_zone_pos = vol->mft_lcn;
706 if (!vol->mft_zone_end)
707 vol->mft_zone_pos = 0;
708 }
709 bmp_pos = zone_start = bmp_initial_pos =
710 vol->data1_zone_pos = vol->mft_zone_end;
711 search_zone = 2;
712 pass = 2;
713 done_zones &= ~2;
714 ntfs_debug("After shrinking mft zone, mft_zone_size 0x%llx, vol->mft_zone_start 0x%llx, vol->mft_zone_end 0x%llx, vol->mft_zone_pos 0x%llx, search_zone 2, pass 2, dones_zones 0x%x, zone_start 0x%llx, zone_end 0x%llx, vol->data1_zone_pos 0x%llx, continuing outer while loop.",
715 mft_zone_size, vol->mft_zone_start,
716 vol->mft_zone_end, vol->mft_zone_pos,
717 done_zones, zone_start, zone_end,
718 vol->data1_zone_pos);
719 }
720 ntfs_debug("After outer while loop.");
721 out:
722 ntfs_debug("At out.");
723 /* Add runlist terminator element. */
724 if (likely(rl)) {
725 rl[rlpos].vcn = rl[rlpos - 1].vcn + rl[rlpos - 1].length;
726 rl[rlpos].lcn = is_extension ? LCN_ENOENT : LCN_RL_NOT_MAPPED;
727 rl[rlpos].length = 0;
728 }
729 if (!IS_ERR_OR_NULL(folio)) {
730 if (need_writeback) {
731 ntfs_debug("Marking page dirty.");
732 folio_mark_dirty(folio);
733 need_writeback = 0;
734 }
735 folio_unlock(folio);
736 kunmap_local(buf);
737 folio_put(folio);
738 }
739 if (likely(!err)) {
740 if (!rl) {
741 err = -EIO;
742 goto out_restore;
743 }
744 if (is_dealloc == true)
745 ntfs_release_dirty_clusters(vol, rl->length);
746 ntfs_debug("Done.");
747 goto out_restore;
748 }
749 if (err != -ENOSPC)
750 ntfs_error(vol->sb,
751 "Failed to allocate clusters, aborting (error %i).",
752 err);
753 if (rl) {
754 int err2;
755
756 if (err == -ENOSPC)
757 ntfs_debug("Not enough space to complete allocation, err -ENOSPC, first free lcn 0x%llx, could allocate up to 0x%llx clusters.",
758 rl[0].lcn, count - clusters);
759 /* Deallocate all allocated clusters. */
760 ntfs_debug("Attempting rollback...");
761 err2 = ntfs_cluster_free_from_rl_nolock(vol, rl);
762 if (err2) {
763 ntfs_error(vol->sb,
764 "Failed to rollback (error %i). Leaving inconsistent metadata! Unmount and run chkdsk.",
765 err2);
766 NVolSetErrors(vol);
767 }
768 /* Free the runlist. */
769 kvfree(rl);
770 } else if (err == -ENOSPC)
771 ntfs_debug("No space left at all, err = -ENOSPC, first free lcn = 0x%llx.",
772 vol->data1_zone_pos);
773 atomic64_set(&vol->dirty_clusters, 0);
774
775 out_restore:
776 up_write(&vol->lcnbmp_lock);
777 memalloc_nofs_restore(memalloc_flags);
778
779 return err < 0 ? ERR_PTR(err) : rl;
780 }
781
782 /*
783 * __ntfs_cluster_free - free clusters on an ntfs volume
784 * @ni: ntfs inode whose runlist describes the clusters to free
785 * @start_vcn: vcn in the runlist of @ni at which to start freeing clusters
786 * @count: number of clusters to free or -1 for all clusters
787 * @ctx: active attribute search context if present or NULL if not
788 * @is_rollback: true if this is a rollback operation
789 *
790 * Free @count clusters starting at the cluster @start_vcn in the runlist
791 * described by the vfs inode @ni.
792 *
793 * If @count is -1, all clusters from @start_vcn to the end of the runlist are
794 * deallocated. Thus, to completely free all clusters in a runlist, use
795 * @start_vcn = 0 and @count = -1.
796 *
797 * If @ctx is specified, it is an active search context of @ni and its base mft
798 * record. This is needed when __ntfs_cluster_free() encounters unmapped
799 * runlist fragments and allows their mapping. If you do not have the mft
800 * record mapped, you can specify @ctx as NULL and __ntfs_cluster_free() will
801 * perform the necessary mapping and unmapping.
802 *
803 * Note, __ntfs_cluster_free() saves the state of @ctx on entry and restores it
804 * before returning. Thus, @ctx will be left pointing to the same attribute on
805 * return as on entry. However, the actual pointers in @ctx may point to
806 * different memory locations on return, so you must remember to reset any
807 * cached pointers from the @ctx, i.e. after the call to __ntfs_cluster_free(),
808 * you will probably want to do:
809 * m = ctx->mrec;
810 * a = ctx->attr;
811 * Assuming you cache ctx->attr in a variable @a of type attr_record * and that
812 * you cache ctx->mrec in a variable @m of type struct mft_record *.
813 *
814 * @is_rollback should always be 'false', it is for internal use to rollback
815 * errors. You probably want to use ntfs_cluster_free() instead.
816 *
817 * Note, __ntfs_cluster_free() does not modify the runlist, so you have to
818 * remove from the runlist or mark sparse the freed runs later.
819 *
820 * Return the number of deallocated clusters (not counting sparse ones) on
821 * success and -errno on error.
822 *
823 * WARNING: If @ctx is supplied, regardless of whether success or failure is
824 * returned, you need to check IS_ERR(@ctx->mrec) and if 'true' the @ctx
825 * is no longer valid, i.e. you need to either call
826 * ntfs_attr_reinit_search_ctx() or ntfs_attr_put_search_ctx() on it.
827 * In that case PTR_ERR(@ctx->mrec) will give you the error code for
828 * why the mapping of the old inode failed.
829 *
830 * Locking: - The runlist described by @ni must be locked for writing on entry
831 * and is locked on return. Note the runlist may be modified when
832 * needed runlist fragments need to be mapped.
833 * - The volume lcn bitmap must be unlocked on entry and is unlocked
834 * on return.
835 * - This function takes the volume lcn bitmap lock for writing and
836 * modifies the bitmap contents.
837 * - If @ctx is NULL, the base mft record of @ni must not be mapped on
838 * entry and it will be left unmapped on return.
839 * - If @ctx is not NULL, the base mft record must be mapped on entry
840 * and it will be left mapped on return.
841 */
__ntfs_cluster_free(struct ntfs_inode * ni,const s64 start_vcn,s64 count,struct ntfs_attr_search_ctx * ctx,const bool is_rollback)842 s64 __ntfs_cluster_free(struct ntfs_inode *ni, const s64 start_vcn, s64 count,
843 struct ntfs_attr_search_ctx *ctx, const bool is_rollback)
844 {
845 s64 delta, to_free, total_freed, real_freed;
846 struct ntfs_volume *vol;
847 struct inode *lcnbmp_vi;
848 struct runlist_element *rl;
849 int err;
850 unsigned int memalloc_flags;
851
852 ntfs_debug("Entering for i_ino 0x%llx, start_vcn 0x%llx, count 0x%llx.%s",
853 ni->mft_no, start_vcn, count,
854 is_rollback ? " (rollback)" : "");
855 vol = ni->vol;
856 lcnbmp_vi = vol->lcnbmp_ino;
857 if (start_vcn < 0 || count < -1)
858 return -EINVAL;
859
860 if (!NVolFreeClusterKnown(vol))
861 wait_event(vol->free_waitq, NVolFreeClusterKnown(vol));
862
863 /*
864 * Lock the lcn bitmap for writing but only if not rolling back. We
865 * must hold the lock all the way including through rollback otherwise
866 * rollback is not possible because once we have cleared a bit and
867 * dropped the lock, anyone could have set the bit again, thus
868 * allocating the cluster for another use.
869 */
870 if (likely(!is_rollback)) {
871 memalloc_flags = memalloc_nofs_save();
872 down_write(&vol->lcnbmp_lock);
873 }
874
875 total_freed = real_freed = 0;
876
877 rl = ntfs_attr_find_vcn_nolock(ni, start_vcn, ctx);
878 if (IS_ERR(rl)) {
879 err = PTR_ERR(rl);
880 if (err == -ENOENT) {
881 if (likely(!is_rollback)) {
882 up_write(&vol->lcnbmp_lock);
883 memalloc_nofs_restore(memalloc_flags);
884 }
885 return 0;
886 }
887
888 if (!is_rollback)
889 ntfs_error(vol->sb,
890 "Failed to find first runlist element (error %d), aborting.",
891 err);
892 goto err_out;
893 }
894 if (unlikely(rl->lcn < LCN_HOLE)) {
895 if (!is_rollback)
896 ntfs_error(vol->sb, "First runlist element has invalid lcn, aborting.");
897 err = -EIO;
898 goto err_out;
899 }
900 /* Find the starting cluster inside the run that needs freeing. */
901 delta = start_vcn - rl->vcn;
902
903 /* The number of clusters in this run that need freeing. */
904 to_free = rl->length - delta;
905 if (count >= 0 && to_free > count)
906 to_free = count;
907
908 if (likely(rl->lcn >= 0)) {
909 /* Do the actual freeing of the clusters in this run. */
910 err = ntfs_bitmap_set_bits_in_run(lcnbmp_vi, rl->lcn + delta,
911 to_free, likely(!is_rollback) ? 0 : 1);
912 if (unlikely(err)) {
913 if (!is_rollback)
914 ntfs_error(vol->sb,
915 "Failed to clear first run (error %i), aborting.",
916 err);
917 goto err_out;
918 }
919 /* We have freed @to_free real clusters. */
920 real_freed = to_free;
921 }
922 /* Go to the next run and adjust the number of clusters left to free. */
923 ++rl;
924 if (count >= 0)
925 count -= to_free;
926
927 /* Keep track of the total "freed" clusters, including sparse ones. */
928 total_freed = to_free;
929 /*
930 * Loop over the remaining runs, using @count as a capping value, and
931 * free them.
932 */
933 for (; rl->length && count != 0; ++rl) {
934 if (unlikely(rl->lcn < LCN_HOLE)) {
935 s64 vcn;
936
937 /* Attempt to map runlist. */
938 vcn = rl->vcn;
939 rl = ntfs_attr_find_vcn_nolock(ni, vcn, ctx);
940 if (IS_ERR(rl)) {
941 err = PTR_ERR(rl);
942 if (!is_rollback)
943 ntfs_error(vol->sb,
944 "Failed to map runlist fragment or failed to find subsequent runlist element.");
945 goto err_out;
946 }
947 if (unlikely(rl->lcn < LCN_HOLE)) {
948 if (!is_rollback)
949 ntfs_error(vol->sb,
950 "Runlist element has invalid lcn (0x%llx).",
951 rl->lcn);
952 err = -EIO;
953 goto err_out;
954 }
955 }
956 /* The number of clusters in this run that need freeing. */
957 to_free = rl->length;
958 if (count >= 0 && to_free > count)
959 to_free = count;
960
961 if (likely(rl->lcn >= 0)) {
962 /* Do the actual freeing of the clusters in the run. */
963 err = ntfs_bitmap_set_bits_in_run(lcnbmp_vi, rl->lcn,
964 to_free, likely(!is_rollback) ? 0 : 1);
965 if (unlikely(err)) {
966 if (!is_rollback)
967 ntfs_error(vol->sb, "Failed to clear subsequent run.");
968 goto err_out;
969 }
970 /* We have freed @to_free real clusters. */
971 real_freed += to_free;
972 }
973 /* Adjust the number of clusters left to free. */
974 if (count >= 0)
975 count -= to_free;
976
977 /* Update the total done clusters. */
978 total_freed += to_free;
979 }
980 ntfs_inc_free_clusters(vol, real_freed);
981 if (likely(!is_rollback)) {
982 up_write(&vol->lcnbmp_lock);
983 memalloc_nofs_restore(memalloc_flags);
984 }
985
986 WARN_ON(count > 0);
987
988 if (NVolDiscard(vol) && !is_rollback) {
989 s64 total_discarded = 0, rl_off;
990 u32 gran = bdev_discard_granularity(vol->sb->s_bdev);
991
992 rl = ntfs_attr_find_vcn_nolock(ni, start_vcn, ctx);
993 if (IS_ERR(rl))
994 return real_freed;
995 rl_off = start_vcn - rl->vcn;
996 while (rl->length && total_discarded < total_freed) {
997 s64 to_discard = rl->length - rl_off;
998
999 if (to_discard + total_discarded > total_freed)
1000 to_discard = total_freed - total_discarded;
1001 if (rl->lcn >= 0) {
1002 sector_t start_sector, end_sector;
1003 int ret;
1004
1005 start_sector = ALIGN(NTFS_CLU_TO_B(vol, rl->lcn + rl_off),
1006 gran) >> SECTOR_SHIFT;
1007 end_sector = ALIGN_DOWN(NTFS_CLU_TO_B(vol,
1008 rl->lcn + rl_off + to_discard),
1009 gran) >> SECTOR_SHIFT;
1010 if (start_sector < end_sector) {
1011 ret = blkdev_issue_discard(vol->sb->s_bdev, start_sector,
1012 end_sector - start_sector,
1013 GFP_NOFS);
1014 if (ret)
1015 break;
1016 }
1017 }
1018
1019 total_discarded += to_discard;
1020 ++rl;
1021 rl_off = 0;
1022 }
1023 }
1024
1025 /* We are done. Return the number of actually freed clusters. */
1026 ntfs_debug("Done.");
1027 return real_freed;
1028 err_out:
1029 if (is_rollback)
1030 return err;
1031 /* If no real clusters were freed, no need to rollback. */
1032 if (!real_freed) {
1033 up_write(&vol->lcnbmp_lock);
1034 memalloc_nofs_restore(memalloc_flags);
1035 return err;
1036 }
1037 /*
1038 * Attempt to rollback and if that succeeds just return the error code.
1039 * If rollback fails, set the volume errors flag, emit an error
1040 * message, and return the error code.
1041 */
1042 delta = __ntfs_cluster_free(ni, start_vcn, total_freed, ctx, true);
1043 if (delta < 0) {
1044 ntfs_error(vol->sb,
1045 "Failed to rollback (error %i). Leaving inconsistent metadata! Unmount and run chkdsk.",
1046 (int)delta);
1047 NVolSetErrors(vol);
1048 } else {
1049 ntfs_dec_free_clusters(vol, delta);
1050 }
1051 up_write(&vol->lcnbmp_lock);
1052 memalloc_nofs_restore(memalloc_flags);
1053 ntfs_error(vol->sb, "Aborting (error %i).", err);
1054 return err;
1055 }
1056