[PATCH] Simpler signal-exit concurrency handling
[linux-2.6/verdex.git] / fs / ntfs / runlist.c
blob061b5ff6b73cec10f317f566cd3dc6fa8512e9d9
1 /**
2 * runlist.c - NTFS runlist handling code. Part of the Linux-NTFS project.
4 * Copyright (c) 2001-2005 Anton Altaparmakov
5 * Copyright (c) 2002-2005 Richard Russon
7 * This program/include file is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License as published
9 * by the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program/include file is distributed in the hope that it will be
13 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
14 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program (in the main directory of the Linux-NTFS
19 * distribution in the file COPYING); if not, write to the Free Software
20 * Foundation,Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 #include "debug.h"
24 #include "dir.h"
25 #include "endian.h"
26 #include "malloc.h"
27 #include "ntfs.h"
29 /**
30 * ntfs_rl_mm - runlist memmove
32 * It is up to the caller to serialize access to the runlist @base.
34 static inline void ntfs_rl_mm(runlist_element *base, int dst, int src,
35 int size)
37 if (likely((dst != src) && (size > 0)))
38 memmove(base + dst, base + src, size * sizeof(*base));
41 /**
42 * ntfs_rl_mc - runlist memory copy
44 * It is up to the caller to serialize access to the runlists @dstbase and
45 * @srcbase.
47 static inline void ntfs_rl_mc(runlist_element *dstbase, int dst,
48 runlist_element *srcbase, int src, int size)
50 if (likely(size > 0))
51 memcpy(dstbase + dst, srcbase + src, size * sizeof(*dstbase));
54 /**
55 * ntfs_rl_realloc - Reallocate memory for runlists
56 * @rl: original runlist
57 * @old_size: number of runlist elements in the original runlist @rl
58 * @new_size: number of runlist elements we need space for
60 * As the runlists grow, more memory will be required. To prevent the
61 * kernel having to allocate and reallocate large numbers of small bits of
62 * memory, this function returns an entire page of memory.
64 * It is up to the caller to serialize access to the runlist @rl.
66 * N.B. If the new allocation doesn't require a different number of pages in
67 * memory, the function will return the original pointer.
69 * On success, return a pointer to the newly allocated, or recycled, memory.
70 * On error, return -errno. The following error codes are defined:
71 * -ENOMEM - Not enough memory to allocate runlist array.
72 * -EINVAL - Invalid parameters were passed in.
74 static inline runlist_element *ntfs_rl_realloc(runlist_element *rl,
75 int old_size, int new_size)
77 runlist_element *new_rl;
79 old_size = PAGE_ALIGN(old_size * sizeof(*rl));
80 new_size = PAGE_ALIGN(new_size * sizeof(*rl));
81 if (old_size == new_size)
82 return rl;
84 new_rl = ntfs_malloc_nofs(new_size);
85 if (unlikely(!new_rl))
86 return ERR_PTR(-ENOMEM);
88 if (likely(rl != NULL)) {
89 if (unlikely(old_size > new_size))
90 old_size = new_size;
91 memcpy(new_rl, rl, old_size);
92 ntfs_free(rl);
94 return new_rl;
97 /**
98 * ntfs_rl_realloc_nofail - Reallocate memory for runlists
99 * @rl: original runlist
100 * @old_size: number of runlist elements in the original runlist @rl
101 * @new_size: number of runlist elements we need space for
103 * As the runlists grow, more memory will be required. To prevent the
104 * kernel having to allocate and reallocate large numbers of small bits of
105 * memory, this function returns an entire page of memory.
107 * This function guarantees that the allocation will succeed. It will sleep
108 * for as long as it takes to complete the allocation.
110 * It is up to the caller to serialize access to the runlist @rl.
112 * N.B. If the new allocation doesn't require a different number of pages in
113 * memory, the function will return the original pointer.
115 * On success, return a pointer to the newly allocated, or recycled, memory.
116 * On error, return -errno. The following error codes are defined:
117 * -ENOMEM - Not enough memory to allocate runlist array.
118 * -EINVAL - Invalid parameters were passed in.
120 static inline runlist_element *ntfs_rl_realloc_nofail(runlist_element *rl,
121 int old_size, int new_size)
123 runlist_element *new_rl;
125 old_size = PAGE_ALIGN(old_size * sizeof(*rl));
126 new_size = PAGE_ALIGN(new_size * sizeof(*rl));
127 if (old_size == new_size)
128 return rl;
130 new_rl = ntfs_malloc_nofs_nofail(new_size);
131 BUG_ON(!new_rl);
133 if (likely(rl != NULL)) {
134 if (unlikely(old_size > new_size))
135 old_size = new_size;
136 memcpy(new_rl, rl, old_size);
137 ntfs_free(rl);
139 return new_rl;
143 * ntfs_are_rl_mergeable - test if two runlists can be joined together
144 * @dst: original runlist
145 * @src: new runlist to test for mergeability with @dst
147 * Test if two runlists can be joined together. For this, their VCNs and LCNs
148 * must be adjacent.
150 * It is up to the caller to serialize access to the runlists @dst and @src.
152 * Return: TRUE Success, the runlists can be merged.
153 * FALSE Failure, the runlists cannot be merged.
155 static inline BOOL ntfs_are_rl_mergeable(runlist_element *dst,
156 runlist_element *src)
158 BUG_ON(!dst);
159 BUG_ON(!src);
161 /* We can merge unmapped regions even if they are misaligned. */
162 if ((dst->lcn == LCN_RL_NOT_MAPPED) && (src->lcn == LCN_RL_NOT_MAPPED))
163 return TRUE;
164 /* If the runs are misaligned, we cannot merge them. */
165 if ((dst->vcn + dst->length) != src->vcn)
166 return FALSE;
167 /* If both runs are non-sparse and contiguous, we can merge them. */
168 if ((dst->lcn >= 0) && (src->lcn >= 0) &&
169 ((dst->lcn + dst->length) == src->lcn))
170 return TRUE;
171 /* If we are merging two holes, we can merge them. */
172 if ((dst->lcn == LCN_HOLE) && (src->lcn == LCN_HOLE))
173 return TRUE;
174 /* Cannot merge. */
175 return FALSE;
179 * __ntfs_rl_merge - merge two runlists without testing if they can be merged
180 * @dst: original, destination runlist
181 * @src: new runlist to merge with @dst
183 * Merge the two runlists, writing into the destination runlist @dst. The
184 * caller must make sure the runlists can be merged or this will corrupt the
185 * destination runlist.
187 * It is up to the caller to serialize access to the runlists @dst and @src.
189 static inline void __ntfs_rl_merge(runlist_element *dst, runlist_element *src)
191 dst->length += src->length;
195 * ntfs_rl_append - append a runlist after a given element
196 * @dst: original runlist to be worked on
197 * @dsize: number of elements in @dst (including end marker)
198 * @src: runlist to be inserted into @dst
199 * @ssize: number of elements in @src (excluding end marker)
200 * @loc: append the new runlist @src after this element in @dst
202 * Append the runlist @src after element @loc in @dst. Merge the right end of
203 * the new runlist, if necessary. Adjust the size of the hole before the
204 * appended runlist.
206 * It is up to the caller to serialize access to the runlists @dst and @src.
208 * On success, return a pointer to the new, combined, runlist. Note, both
209 * runlists @dst and @src are deallocated before returning so you cannot use
210 * the pointers for anything any more. (Strictly speaking the returned runlist
211 * may be the same as @dst but this is irrelevant.)
213 * On error, return -errno. Both runlists are left unmodified. The following
214 * error codes are defined:
215 * -ENOMEM - Not enough memory to allocate runlist array.
216 * -EINVAL - Invalid parameters were passed in.
218 static inline runlist_element *ntfs_rl_append(runlist_element *dst,
219 int dsize, runlist_element *src, int ssize, int loc)
221 BOOL right = FALSE; /* Right end of @src needs merging. */
222 int marker; /* End of the inserted runs. */
224 BUG_ON(!dst);
225 BUG_ON(!src);
227 /* First, check if the right hand end needs merging. */
228 if ((loc + 1) < dsize)
229 right = ntfs_are_rl_mergeable(src + ssize - 1, dst + loc + 1);
231 /* Space required: @dst size + @src size, less one if we merged. */
232 dst = ntfs_rl_realloc(dst, dsize, dsize + ssize - right);
233 if (IS_ERR(dst))
234 return dst;
236 * We are guaranteed to succeed from here so can start modifying the
237 * original runlists.
240 /* First, merge the right hand end, if necessary. */
241 if (right)
242 __ntfs_rl_merge(src + ssize - 1, dst + loc + 1);
244 /* First run after the @src runs that have been inserted. */
245 marker = loc + ssize + 1;
247 /* Move the tail of @dst out of the way, then copy in @src. */
248 ntfs_rl_mm(dst, marker, loc + 1 + right, dsize - (loc + 1 + right));
249 ntfs_rl_mc(dst, loc + 1, src, 0, ssize);
251 /* Adjust the size of the preceding hole. */
252 dst[loc].length = dst[loc + 1].vcn - dst[loc].vcn;
254 /* We may have changed the length of the file, so fix the end marker */
255 if (dst[marker].lcn == LCN_ENOENT)
256 dst[marker].vcn = dst[marker - 1].vcn + dst[marker - 1].length;
258 return dst;
262 * ntfs_rl_insert - insert a runlist into another
263 * @dst: original runlist to be worked on
264 * @dsize: number of elements in @dst (including end marker)
265 * @src: new runlist to be inserted
266 * @ssize: number of elements in @src (excluding end marker)
267 * @loc: insert the new runlist @src before this element in @dst
269 * Insert the runlist @src before element @loc in the runlist @dst. Merge the
270 * left end of the new runlist, if necessary. Adjust the size of the hole
271 * after the inserted runlist.
273 * It is up to the caller to serialize access to the runlists @dst and @src.
275 * On success, return a pointer to the new, combined, runlist. Note, both
276 * runlists @dst and @src are deallocated before returning so you cannot use
277 * the pointers for anything any more. (Strictly speaking the returned runlist
278 * may be the same as @dst but this is irrelevant.)
280 * On error, return -errno. Both runlists are left unmodified. The following
281 * error codes are defined:
282 * -ENOMEM - Not enough memory to allocate runlist array.
283 * -EINVAL - Invalid parameters were passed in.
285 static inline runlist_element *ntfs_rl_insert(runlist_element *dst,
286 int dsize, runlist_element *src, int ssize, int loc)
288 BOOL left = FALSE; /* Left end of @src needs merging. */
289 BOOL disc = FALSE; /* Discontinuity between @dst and @src. */
290 int marker; /* End of the inserted runs. */
292 BUG_ON(!dst);
293 BUG_ON(!src);
296 * disc => Discontinuity between the end of @dst and the start of @src.
297 * This means we might need to insert a "not mapped" run.
299 if (loc == 0)
300 disc = (src[0].vcn > 0);
301 else {
302 s64 merged_length;
304 left = ntfs_are_rl_mergeable(dst + loc - 1, src);
306 merged_length = dst[loc - 1].length;
307 if (left)
308 merged_length += src->length;
310 disc = (src[0].vcn > dst[loc - 1].vcn + merged_length);
313 * Space required: @dst size + @src size, less one if we merged, plus
314 * one if there was a discontinuity.
316 dst = ntfs_rl_realloc(dst, dsize, dsize + ssize - left + disc);
317 if (IS_ERR(dst))
318 return dst;
320 * We are guaranteed to succeed from here so can start modifying the
321 * original runlist.
323 if (left)
324 __ntfs_rl_merge(dst + loc - 1, src);
326 * First run after the @src runs that have been inserted.
327 * Nominally, @marker equals @loc + @ssize, i.e. location + number of
328 * runs in @src. However, if @left, then the first run in @src has
329 * been merged with one in @dst. And if @disc, then @dst and @src do
330 * not meet and we need an extra run to fill the gap.
332 marker = loc + ssize - left + disc;
334 /* Move the tail of @dst out of the way, then copy in @src. */
335 ntfs_rl_mm(dst, marker, loc, dsize - loc);
336 ntfs_rl_mc(dst, loc + disc, src, left, ssize - left);
338 /* Adjust the VCN of the first run after the insertion... */
339 dst[marker].vcn = dst[marker - 1].vcn + dst[marker - 1].length;
340 /* ... and the length. */
341 if (dst[marker].lcn == LCN_HOLE || dst[marker].lcn == LCN_RL_NOT_MAPPED)
342 dst[marker].length = dst[marker + 1].vcn - dst[marker].vcn;
344 /* Writing beyond the end of the file and there is a discontinuity. */
345 if (disc) {
346 if (loc > 0) {
347 dst[loc].vcn = dst[loc - 1].vcn + dst[loc - 1].length;
348 dst[loc].length = dst[loc + 1].vcn - dst[loc].vcn;
349 } else {
350 dst[loc].vcn = 0;
351 dst[loc].length = dst[loc + 1].vcn;
353 dst[loc].lcn = LCN_RL_NOT_MAPPED;
355 return dst;
359 * ntfs_rl_replace - overwrite a runlist element with another runlist
360 * @dst: original runlist to be worked on
361 * @dsize: number of elements in @dst (including end marker)
362 * @src: new runlist to be inserted
363 * @ssize: number of elements in @src (excluding end marker)
364 * @loc: index in runlist @dst to overwrite with @src
366 * Replace the runlist element @dst at @loc with @src. Merge the left and
367 * right ends of the inserted runlist, if necessary.
369 * It is up to the caller to serialize access to the runlists @dst and @src.
371 * On success, return a pointer to the new, combined, runlist. Note, both
372 * runlists @dst and @src are deallocated before returning so you cannot use
373 * the pointers for anything any more. (Strictly speaking the returned runlist
374 * may be the same as @dst but this is irrelevant.)
376 * On error, return -errno. Both runlists are left unmodified. The following
377 * error codes are defined:
378 * -ENOMEM - Not enough memory to allocate runlist array.
379 * -EINVAL - Invalid parameters were passed in.
381 static inline runlist_element *ntfs_rl_replace(runlist_element *dst,
382 int dsize, runlist_element *src, int ssize, int loc)
384 BOOL left = FALSE; /* Left end of @src needs merging. */
385 BOOL right = FALSE; /* Right end of @src needs merging. */
386 int tail; /* Start of tail of @dst. */
387 int marker; /* End of the inserted runs. */
389 BUG_ON(!dst);
390 BUG_ON(!src);
392 /* First, see if the left and right ends need merging. */
393 if ((loc + 1) < dsize)
394 right = ntfs_are_rl_mergeable(src + ssize - 1, dst + loc + 1);
395 if (loc > 0)
396 left = ntfs_are_rl_mergeable(dst + loc - 1, src);
398 * Allocate some space. We will need less if the left, right, or both
399 * ends get merged.
401 dst = ntfs_rl_realloc(dst, dsize, dsize + ssize - left - right);
402 if (IS_ERR(dst))
403 return dst;
405 * We are guaranteed to succeed from here so can start modifying the
406 * original runlists.
409 /* First, merge the left and right ends, if necessary. */
410 if (right)
411 __ntfs_rl_merge(src + ssize - 1, dst + loc + 1);
412 if (left)
413 __ntfs_rl_merge(dst + loc - 1, src);
415 * Offset of the tail of @dst. This needs to be moved out of the way
416 * to make space for the runs to be copied from @src, i.e. the first
417 * run of the tail of @dst.
418 * Nominally, @tail equals @loc + 1, i.e. location, skipping the
419 * replaced run. However, if @right, then one of @dst's runs is
420 * already merged into @src.
422 tail = loc + right + 1;
424 * First run after the @src runs that have been inserted, i.e. where
425 * the tail of @dst needs to be moved to.
426 * Nominally, @marker equals @loc + @ssize, i.e. location + number of
427 * runs in @src. However, if @left, then the first run in @src has
428 * been merged with one in @dst.
430 marker = loc + ssize - left;
432 /* Move the tail of @dst out of the way, then copy in @src. */
433 ntfs_rl_mm(dst, marker, tail, dsize - tail);
434 ntfs_rl_mc(dst, loc, src, left, ssize - left);
436 /* We may have changed the length of the file, so fix the end marker. */
437 if (dsize - tail > 0 && dst[marker].lcn == LCN_ENOENT)
438 dst[marker].vcn = dst[marker - 1].vcn + dst[marker - 1].length;
439 return dst;
443 * ntfs_rl_split - insert a runlist into the centre of a hole
444 * @dst: original runlist to be worked on
445 * @dsize: number of elements in @dst (including end marker)
446 * @src: new runlist to be inserted
447 * @ssize: number of elements in @src (excluding end marker)
448 * @loc: index in runlist @dst at which to split and insert @src
450 * Split the runlist @dst at @loc into two and insert @new in between the two
451 * fragments. No merging of runlists is necessary. Adjust the size of the
452 * holes either side.
454 * It is up to the caller to serialize access to the runlists @dst and @src.
456 * On success, return a pointer to the new, combined, runlist. Note, both
457 * runlists @dst and @src are deallocated before returning so you cannot use
458 * the pointers for anything any more. (Strictly speaking the returned runlist
459 * may be the same as @dst but this is irrelevant.)
461 * On error, return -errno. Both runlists are left unmodified. The following
462 * error codes are defined:
463 * -ENOMEM - Not enough memory to allocate runlist array.
464 * -EINVAL - Invalid parameters were passed in.
466 static inline runlist_element *ntfs_rl_split(runlist_element *dst, int dsize,
467 runlist_element *src, int ssize, int loc)
469 BUG_ON(!dst);
470 BUG_ON(!src);
472 /* Space required: @dst size + @src size + one new hole. */
473 dst = ntfs_rl_realloc(dst, dsize, dsize + ssize + 1);
474 if (IS_ERR(dst))
475 return dst;
477 * We are guaranteed to succeed from here so can start modifying the
478 * original runlists.
481 /* Move the tail of @dst out of the way, then copy in @src. */
482 ntfs_rl_mm(dst, loc + 1 + ssize, loc, dsize - loc);
483 ntfs_rl_mc(dst, loc + 1, src, 0, ssize);
485 /* Adjust the size of the holes either size of @src. */
486 dst[loc].length = dst[loc+1].vcn - dst[loc].vcn;
487 dst[loc+ssize+1].vcn = dst[loc+ssize].vcn + dst[loc+ssize].length;
488 dst[loc+ssize+1].length = dst[loc+ssize+2].vcn - dst[loc+ssize+1].vcn;
490 return dst;
494 * ntfs_runlists_merge - merge two runlists into one
495 * @drl: original runlist to be worked on
496 * @srl: new runlist to be merged into @drl
498 * First we sanity check the two runlists @srl and @drl to make sure that they
499 * are sensible and can be merged. The runlist @srl must be either after the
500 * runlist @drl or completely within a hole (or unmapped region) in @drl.
502 * It is up to the caller to serialize access to the runlists @drl and @srl.
504 * Merging of runlists is necessary in two cases:
505 * 1. When attribute lists are used and a further extent is being mapped.
506 * 2. When new clusters are allocated to fill a hole or extend a file.
508 * There are four possible ways @srl can be merged. It can:
509 * - be inserted at the beginning of a hole,
510 * - split the hole in two and be inserted between the two fragments,
511 * - be appended at the end of a hole, or it can
512 * - replace the whole hole.
513 * It can also be appended to the end of the runlist, which is just a variant
514 * of the insert case.
516 * On success, return a pointer to the new, combined, runlist. Note, both
517 * runlists @drl and @srl are deallocated before returning so you cannot use
518 * the pointers for anything any more. (Strictly speaking the returned runlist
519 * may be the same as @dst but this is irrelevant.)
521 * On error, return -errno. Both runlists are left unmodified. The following
522 * error codes are defined:
523 * -ENOMEM - Not enough memory to allocate runlist array.
524 * -EINVAL - Invalid parameters were passed in.
525 * -ERANGE - The runlists overlap and cannot be merged.
527 runlist_element *ntfs_runlists_merge(runlist_element *drl,
528 runlist_element *srl)
530 int di, si; /* Current index into @[ds]rl. */
531 int sstart; /* First index with lcn > LCN_RL_NOT_MAPPED. */
532 int dins; /* Index into @drl at which to insert @srl. */
533 int dend, send; /* Last index into @[ds]rl. */
534 int dfinal, sfinal; /* The last index into @[ds]rl with
535 lcn >= LCN_HOLE. */
536 int marker = 0;
537 VCN marker_vcn = 0;
539 #ifdef DEBUG
540 ntfs_debug("dst:");
541 ntfs_debug_dump_runlist(drl);
542 ntfs_debug("src:");
543 ntfs_debug_dump_runlist(srl);
544 #endif
546 /* Check for silly calling... */
547 if (unlikely(!srl))
548 return drl;
549 if (IS_ERR(srl) || IS_ERR(drl))
550 return ERR_PTR(-EINVAL);
552 /* Check for the case where the first mapping is being done now. */
553 if (unlikely(!drl)) {
554 drl = srl;
555 /* Complete the source runlist if necessary. */
556 if (unlikely(drl[0].vcn)) {
557 /* Scan to the end of the source runlist. */
558 for (dend = 0; likely(drl[dend].length); dend++)
560 dend++;
561 drl = ntfs_rl_realloc(drl, dend, dend + 1);
562 if (IS_ERR(drl))
563 return drl;
564 /* Insert start element at the front of the runlist. */
565 ntfs_rl_mm(drl, 1, 0, dend);
566 drl[0].vcn = 0;
567 drl[0].lcn = LCN_RL_NOT_MAPPED;
568 drl[0].length = drl[1].vcn;
570 goto finished;
573 si = di = 0;
575 /* Skip any unmapped start element(s) in the source runlist. */
576 while (srl[si].length && srl[si].lcn < LCN_HOLE)
577 si++;
579 /* Can't have an entirely unmapped source runlist. */
580 BUG_ON(!srl[si].length);
582 /* Record the starting points. */
583 sstart = si;
586 * Skip forward in @drl until we reach the position where @srl needs to
587 * be inserted. If we reach the end of @drl, @srl just needs to be
588 * appended to @drl.
590 for (; drl[di].length; di++) {
591 if (drl[di].vcn + drl[di].length > srl[sstart].vcn)
592 break;
594 dins = di;
596 /* Sanity check for illegal overlaps. */
597 if ((drl[di].vcn == srl[si].vcn) && (drl[di].lcn >= 0) &&
598 (srl[si].lcn >= 0)) {
599 ntfs_error(NULL, "Run lists overlap. Cannot merge!");
600 return ERR_PTR(-ERANGE);
603 /* Scan to the end of both runlists in order to know their sizes. */
604 for (send = si; srl[send].length; send++)
606 for (dend = di; drl[dend].length; dend++)
609 if (srl[send].lcn == LCN_ENOENT)
610 marker_vcn = srl[marker = send].vcn;
612 /* Scan to the last element with lcn >= LCN_HOLE. */
613 for (sfinal = send; sfinal >= 0 && srl[sfinal].lcn < LCN_HOLE; sfinal--)
615 for (dfinal = dend; dfinal >= 0 && drl[dfinal].lcn < LCN_HOLE; dfinal--)
619 BOOL start;
620 BOOL finish;
621 int ds = dend + 1; /* Number of elements in drl & srl */
622 int ss = sfinal - sstart + 1;
624 start = ((drl[dins].lcn < LCN_RL_NOT_MAPPED) || /* End of file */
625 (drl[dins].vcn == srl[sstart].vcn)); /* Start of hole */
626 finish = ((drl[dins].lcn >= LCN_RL_NOT_MAPPED) && /* End of file */
627 ((drl[dins].vcn + drl[dins].length) <= /* End of hole */
628 (srl[send - 1].vcn + srl[send - 1].length)));
630 /* Or we will lose an end marker. */
631 if (finish && !drl[dins].length)
632 ss++;
633 if (marker && (drl[dins].vcn + drl[dins].length > srl[send - 1].vcn))
634 finish = FALSE;
635 #if 0
636 ntfs_debug("dfinal = %i, dend = %i", dfinal, dend);
637 ntfs_debug("sstart = %i, sfinal = %i, send = %i", sstart, sfinal, send);
638 ntfs_debug("start = %i, finish = %i", start, finish);
639 ntfs_debug("ds = %i, ss = %i, dins = %i", ds, ss, dins);
640 #endif
641 if (start) {
642 if (finish)
643 drl = ntfs_rl_replace(drl, ds, srl + sstart, ss, dins);
644 else
645 drl = ntfs_rl_insert(drl, ds, srl + sstart, ss, dins);
646 } else {
647 if (finish)
648 drl = ntfs_rl_append(drl, ds, srl + sstart, ss, dins);
649 else
650 drl = ntfs_rl_split(drl, ds, srl + sstart, ss, dins);
652 if (IS_ERR(drl)) {
653 ntfs_error(NULL, "Merge failed.");
654 return drl;
656 ntfs_free(srl);
657 if (marker) {
658 ntfs_debug("Triggering marker code.");
659 for (ds = dend; drl[ds].length; ds++)
661 /* We only need to care if @srl ended after @drl. */
662 if (drl[ds].vcn <= marker_vcn) {
663 int slots = 0;
665 if (drl[ds].vcn == marker_vcn) {
666 ntfs_debug("Old marker = 0x%llx, replacing "
667 "with LCN_ENOENT.",
668 (unsigned long long)
669 drl[ds].lcn);
670 drl[ds].lcn = LCN_ENOENT;
671 goto finished;
674 * We need to create an unmapped runlist element in
675 * @drl or extend an existing one before adding the
676 * ENOENT terminator.
678 if (drl[ds].lcn == LCN_ENOENT) {
679 ds--;
680 slots = 1;
682 if (drl[ds].lcn != LCN_RL_NOT_MAPPED) {
683 /* Add an unmapped runlist element. */
684 if (!slots) {
685 drl = ntfs_rl_realloc_nofail(drl, ds,
686 ds + 2);
687 slots = 2;
689 ds++;
690 /* Need to set vcn if it isn't set already. */
691 if (slots != 1)
692 drl[ds].vcn = drl[ds - 1].vcn +
693 drl[ds - 1].length;
694 drl[ds].lcn = LCN_RL_NOT_MAPPED;
695 /* We now used up a slot. */
696 slots--;
698 drl[ds].length = marker_vcn - drl[ds].vcn;
699 /* Finally add the ENOENT terminator. */
700 ds++;
701 if (!slots)
702 drl = ntfs_rl_realloc_nofail(drl, ds, ds + 1);
703 drl[ds].vcn = marker_vcn;
704 drl[ds].lcn = LCN_ENOENT;
705 drl[ds].length = (s64)0;
710 finished:
711 /* The merge was completed successfully. */
712 ntfs_debug("Merged runlist:");
713 ntfs_debug_dump_runlist(drl);
714 return drl;
718 * ntfs_mapping_pairs_decompress - convert mapping pairs array to runlist
719 * @vol: ntfs volume on which the attribute resides
720 * @attr: attribute record whose mapping pairs array to decompress
721 * @old_rl: optional runlist in which to insert @attr's runlist
723 * It is up to the caller to serialize access to the runlist @old_rl.
725 * Decompress the attribute @attr's mapping pairs array into a runlist. On
726 * success, return the decompressed runlist.
728 * If @old_rl is not NULL, decompressed runlist is inserted into the
729 * appropriate place in @old_rl and the resultant, combined runlist is
730 * returned. The original @old_rl is deallocated.
732 * On error, return -errno. @old_rl is left unmodified in that case.
734 * The following error codes are defined:
735 * -ENOMEM - Not enough memory to allocate runlist array.
736 * -EIO - Corrupt runlist.
737 * -EINVAL - Invalid parameters were passed in.
738 * -ERANGE - The two runlists overlap.
740 * FIXME: For now we take the conceptionally simplest approach of creating the
741 * new runlist disregarding the already existing one and then splicing the
742 * two into one, if that is possible (we check for overlap and discard the new
743 * runlist if overlap present before returning ERR_PTR(-ERANGE)).
745 runlist_element *ntfs_mapping_pairs_decompress(const ntfs_volume *vol,
746 const ATTR_RECORD *attr, runlist_element *old_rl)
748 VCN vcn; /* Current vcn. */
749 LCN lcn; /* Current lcn. */
750 s64 deltaxcn; /* Change in [vl]cn. */
751 runlist_element *rl; /* The output runlist. */
752 u8 *buf; /* Current position in mapping pairs array. */
753 u8 *attr_end; /* End of attribute. */
754 int rlsize; /* Size of runlist buffer. */
755 u16 rlpos; /* Current runlist position in units of
756 runlist_elements. */
757 u8 b; /* Current byte offset in buf. */
759 #ifdef DEBUG
760 /* Make sure attr exists and is non-resident. */
761 if (!attr || !attr->non_resident || sle64_to_cpu(
762 attr->data.non_resident.lowest_vcn) < (VCN)0) {
763 ntfs_error(vol->sb, "Invalid arguments.");
764 return ERR_PTR(-EINVAL);
766 #endif
767 /* Start at vcn = lowest_vcn and lcn 0. */
768 vcn = sle64_to_cpu(attr->data.non_resident.lowest_vcn);
769 lcn = 0;
770 /* Get start of the mapping pairs array. */
771 buf = (u8*)attr + le16_to_cpu(
772 attr->data.non_resident.mapping_pairs_offset);
773 attr_end = (u8*)attr + le32_to_cpu(attr->length);
774 if (unlikely(buf < (u8*)attr || buf > attr_end)) {
775 ntfs_error(vol->sb, "Corrupt attribute.");
776 return ERR_PTR(-EIO);
778 /* If the mapping pairs array is valid but empty, nothing to do. */
779 if (!vcn && !*buf)
780 return old_rl;
781 /* Current position in runlist array. */
782 rlpos = 0;
783 /* Allocate first page and set current runlist size to one page. */
784 rl = ntfs_malloc_nofs(rlsize = PAGE_SIZE);
785 if (unlikely(!rl))
786 return ERR_PTR(-ENOMEM);
787 /* Insert unmapped starting element if necessary. */
788 if (vcn) {
789 rl->vcn = 0;
790 rl->lcn = LCN_RL_NOT_MAPPED;
791 rl->length = vcn;
792 rlpos++;
794 while (buf < attr_end && *buf) {
796 * Allocate more memory if needed, including space for the
797 * not-mapped and terminator elements. ntfs_malloc_nofs()
798 * operates on whole pages only.
800 if (((rlpos + 3) * sizeof(*old_rl)) > rlsize) {
801 runlist_element *rl2;
803 rl2 = ntfs_malloc_nofs(rlsize + (int)PAGE_SIZE);
804 if (unlikely(!rl2)) {
805 ntfs_free(rl);
806 return ERR_PTR(-ENOMEM);
808 memcpy(rl2, rl, rlsize);
809 ntfs_free(rl);
810 rl = rl2;
811 rlsize += PAGE_SIZE;
813 /* Enter the current vcn into the current runlist element. */
814 rl[rlpos].vcn = vcn;
816 * Get the change in vcn, i.e. the run length in clusters.
817 * Doing it this way ensures that we signextend negative values.
818 * A negative run length doesn't make any sense, but hey, I
819 * didn't make up the NTFS specs and Windows NT4 treats the run
820 * length as a signed value so that's how it is...
822 b = *buf & 0xf;
823 if (b) {
824 if (unlikely(buf + b > attr_end))
825 goto io_error;
826 for (deltaxcn = (s8)buf[b--]; b; b--)
827 deltaxcn = (deltaxcn << 8) + buf[b];
828 } else { /* The length entry is compulsory. */
829 ntfs_error(vol->sb, "Missing length entry in mapping "
830 "pairs array.");
831 deltaxcn = (s64)-1;
834 * Assume a negative length to indicate data corruption and
835 * hence clean-up and return NULL.
837 if (unlikely(deltaxcn < 0)) {
838 ntfs_error(vol->sb, "Invalid length in mapping pairs "
839 "array.");
840 goto err_out;
843 * Enter the current run length into the current runlist
844 * element.
846 rl[rlpos].length = deltaxcn;
847 /* Increment the current vcn by the current run length. */
848 vcn += deltaxcn;
850 * There might be no lcn change at all, as is the case for
851 * sparse clusters on NTFS 3.0+, in which case we set the lcn
852 * to LCN_HOLE.
854 if (!(*buf & 0xf0))
855 rl[rlpos].lcn = LCN_HOLE;
856 else {
857 /* Get the lcn change which really can be negative. */
858 u8 b2 = *buf & 0xf;
859 b = b2 + ((*buf >> 4) & 0xf);
860 if (buf + b > attr_end)
861 goto io_error;
862 for (deltaxcn = (s8)buf[b--]; b > b2; b--)
863 deltaxcn = (deltaxcn << 8) + buf[b];
864 /* Change the current lcn to its new value. */
865 lcn += deltaxcn;
866 #ifdef DEBUG
868 * On NTFS 1.2-, apparently can have lcn == -1 to
869 * indicate a hole. But we haven't verified ourselves
870 * whether it is really the lcn or the deltaxcn that is
871 * -1. So if either is found give us a message so we
872 * can investigate it further!
874 if (vol->major_ver < 3) {
875 if (unlikely(deltaxcn == (LCN)-1))
876 ntfs_error(vol->sb, "lcn delta == -1");
877 if (unlikely(lcn == (LCN)-1))
878 ntfs_error(vol->sb, "lcn == -1");
880 #endif
881 /* Check lcn is not below -1. */
882 if (unlikely(lcn < (LCN)-1)) {
883 ntfs_error(vol->sb, "Invalid LCN < -1 in "
884 "mapping pairs array.");
885 goto err_out;
887 /* Enter the current lcn into the runlist element. */
888 rl[rlpos].lcn = lcn;
890 /* Get to the next runlist element. */
891 rlpos++;
892 /* Increment the buffer position to the next mapping pair. */
893 buf += (*buf & 0xf) + ((*buf >> 4) & 0xf) + 1;
895 if (unlikely(buf >= attr_end))
896 goto io_error;
898 * If there is a highest_vcn specified, it must be equal to the final
899 * vcn in the runlist - 1, or something has gone badly wrong.
901 deltaxcn = sle64_to_cpu(attr->data.non_resident.highest_vcn);
902 if (unlikely(deltaxcn && vcn - 1 != deltaxcn)) {
903 mpa_err:
904 ntfs_error(vol->sb, "Corrupt mapping pairs array in "
905 "non-resident attribute.");
906 goto err_out;
908 /* Setup not mapped runlist element if this is the base extent. */
909 if (!attr->data.non_resident.lowest_vcn) {
910 VCN max_cluster;
912 max_cluster = ((sle64_to_cpu(
913 attr->data.non_resident.allocated_size) +
914 vol->cluster_size - 1) >>
915 vol->cluster_size_bits) - 1;
917 * A highest_vcn of zero means this is a single extent
918 * attribute so simply terminate the runlist with LCN_ENOENT).
920 if (deltaxcn) {
922 * If there is a difference between the highest_vcn and
923 * the highest cluster, the runlist is either corrupt
924 * or, more likely, there are more extents following
925 * this one.
927 if (deltaxcn < max_cluster) {
928 ntfs_debug("More extents to follow; deltaxcn "
929 "= 0x%llx, max_cluster = "
930 "0x%llx",
931 (unsigned long long)deltaxcn,
932 (unsigned long long)
933 max_cluster);
934 rl[rlpos].vcn = vcn;
935 vcn += rl[rlpos].length = max_cluster -
936 deltaxcn;
937 rl[rlpos].lcn = LCN_RL_NOT_MAPPED;
938 rlpos++;
939 } else if (unlikely(deltaxcn > max_cluster)) {
940 ntfs_error(vol->sb, "Corrupt attribute. "
941 "deltaxcn = 0x%llx, "
942 "max_cluster = 0x%llx",
943 (unsigned long long)deltaxcn,
944 (unsigned long long)
945 max_cluster);
946 goto mpa_err;
949 rl[rlpos].lcn = LCN_ENOENT;
950 } else /* Not the base extent. There may be more extents to follow. */
951 rl[rlpos].lcn = LCN_RL_NOT_MAPPED;
953 /* Setup terminating runlist element. */
954 rl[rlpos].vcn = vcn;
955 rl[rlpos].length = (s64)0;
956 /* If no existing runlist was specified, we are done. */
957 if (!old_rl) {
958 ntfs_debug("Mapping pairs array successfully decompressed:");
959 ntfs_debug_dump_runlist(rl);
960 return rl;
962 /* Now combine the new and old runlists checking for overlaps. */
963 old_rl = ntfs_runlists_merge(old_rl, rl);
964 if (likely(!IS_ERR(old_rl)))
965 return old_rl;
966 ntfs_free(rl);
967 ntfs_error(vol->sb, "Failed to merge runlists.");
968 return old_rl;
969 io_error:
970 ntfs_error(vol->sb, "Corrupt attribute.");
971 err_out:
972 ntfs_free(rl);
973 return ERR_PTR(-EIO);
977 * ntfs_rl_vcn_to_lcn - convert a vcn into a lcn given a runlist
978 * @rl: runlist to use for conversion
979 * @vcn: vcn to convert
981 * Convert the virtual cluster number @vcn of an attribute into a logical
982 * cluster number (lcn) of a device using the runlist @rl to map vcns to their
983 * corresponding lcns.
985 * It is up to the caller to serialize access to the runlist @rl.
987 * Since lcns must be >= 0, we use negative return codes with special meaning:
989 * Return code Meaning / Description
990 * ==================================================
991 * LCN_HOLE Hole / not allocated on disk.
992 * LCN_RL_NOT_MAPPED This is part of the runlist which has not been
993 * inserted into the runlist yet.
994 * LCN_ENOENT There is no such vcn in the attribute.
996 * Locking: - The caller must have locked the runlist (for reading or writing).
997 * - This function does not touch the lock, nor does it modify the
998 * runlist.
1000 LCN ntfs_rl_vcn_to_lcn(const runlist_element *rl, const VCN vcn)
1002 int i;
1004 BUG_ON(vcn < 0);
1006 * If rl is NULL, assume that we have found an unmapped runlist. The
1007 * caller can then attempt to map it and fail appropriately if
1008 * necessary.
1010 if (unlikely(!rl))
1011 return LCN_RL_NOT_MAPPED;
1013 /* Catch out of lower bounds vcn. */
1014 if (unlikely(vcn < rl[0].vcn))
1015 return LCN_ENOENT;
1017 for (i = 0; likely(rl[i].length); i++) {
1018 if (unlikely(vcn < rl[i+1].vcn)) {
1019 if (likely(rl[i].lcn >= (LCN)0))
1020 return rl[i].lcn + (vcn - rl[i].vcn);
1021 return rl[i].lcn;
1025 * The terminator element is setup to the correct value, i.e. one of
1026 * LCN_HOLE, LCN_RL_NOT_MAPPED, or LCN_ENOENT.
1028 if (likely(rl[i].lcn < (LCN)0))
1029 return rl[i].lcn;
1030 /* Just in case... We could replace this with BUG() some day. */
1031 return LCN_ENOENT;
1034 #ifdef NTFS_RW
1037 * ntfs_rl_find_vcn_nolock - find a vcn in a runlist
1038 * @rl: runlist to search
1039 * @vcn: vcn to find
1041 * Find the virtual cluster number @vcn in the runlist @rl and return the
1042 * address of the runlist element containing the @vcn on success.
1044 * Return NULL if @rl is NULL or @vcn is in an unmapped part/out of bounds of
1045 * the runlist.
1047 * Locking: The runlist must be locked on entry.
1049 runlist_element *ntfs_rl_find_vcn_nolock(runlist_element *rl, const VCN vcn)
1051 BUG_ON(vcn < 0);
1052 if (unlikely(!rl || vcn < rl[0].vcn))
1053 return NULL;
1054 while (likely(rl->length)) {
1055 if (unlikely(vcn < rl[1].vcn)) {
1056 if (likely(rl->lcn >= LCN_HOLE))
1057 return rl;
1058 return NULL;
1060 rl++;
1062 if (likely(rl->lcn == LCN_ENOENT))
1063 return rl;
1064 return NULL;
1068 * ntfs_get_nr_significant_bytes - get number of bytes needed to store a number
1069 * @n: number for which to get the number of bytes for
1071 * Return the number of bytes required to store @n unambiguously as
1072 * a signed number.
1074 * This is used in the context of the mapping pairs array to determine how
1075 * many bytes will be needed in the array to store a given logical cluster
1076 * number (lcn) or a specific run length.
1078 * Return the number of bytes written. This function cannot fail.
1080 static inline int ntfs_get_nr_significant_bytes(const s64 n)
1082 s64 l = n;
1083 int i;
1084 s8 j;
1086 i = 0;
1087 do {
1088 l >>= 8;
1089 i++;
1090 } while (l != 0 && l != -1);
1091 j = (n >> 8 * (i - 1)) & 0xff;
1092 /* If the sign bit is wrong, we need an extra byte. */
1093 if ((n < 0 && j >= 0) || (n > 0 && j < 0))
1094 i++;
1095 return i;
1099 * ntfs_get_size_for_mapping_pairs - get bytes needed for mapping pairs array
1100 * @vol: ntfs volume (needed for the ntfs version)
1101 * @rl: locked runlist to determine the size of the mapping pairs of
1102 * @first_vcn: first vcn which to include in the mapping pairs array
1103 * @last_vcn: last vcn which to include in the mapping pairs array
1105 * Walk the locked runlist @rl and calculate the size in bytes of the mapping
1106 * pairs array corresponding to the runlist @rl, starting at vcn @first_vcn and
1107 * finishing with vcn @last_vcn.
1109 * A @last_vcn of -1 means end of runlist and in that case the size of the
1110 * mapping pairs array corresponding to the runlist starting at vcn @first_vcn
1111 * and finishing at the end of the runlist is determined.
1113 * This for example allows us to allocate a buffer of the right size when
1114 * building the mapping pairs array.
1116 * If @rl is NULL, just return 1 (for the single terminator byte).
1118 * Return the calculated size in bytes on success. On error, return -errno.
1119 * The following error codes are defined:
1120 * -EINVAL - Run list contains unmapped elements. Make sure to only pass
1121 * fully mapped runlists to this function.
1122 * -EIO - The runlist is corrupt.
1124 * Locking: @rl must be locked on entry (either for reading or writing), it
1125 * remains locked throughout, and is left locked upon return.
1127 int ntfs_get_size_for_mapping_pairs(const ntfs_volume *vol,
1128 const runlist_element *rl, const VCN first_vcn,
1129 const VCN last_vcn)
1131 LCN prev_lcn;
1132 int rls;
1133 BOOL the_end = FALSE;
1135 BUG_ON(first_vcn < 0);
1136 BUG_ON(last_vcn < -1);
1137 BUG_ON(last_vcn >= 0 && first_vcn > last_vcn);
1138 if (!rl) {
1139 BUG_ON(first_vcn);
1140 BUG_ON(last_vcn > 0);
1141 return 1;
1143 /* Skip to runlist element containing @first_vcn. */
1144 while (rl->length && first_vcn >= rl[1].vcn)
1145 rl++;
1146 if (unlikely((!rl->length && first_vcn > rl->vcn) ||
1147 first_vcn < rl->vcn))
1148 return -EINVAL;
1149 prev_lcn = 0;
1150 /* Always need the termining zero byte. */
1151 rls = 1;
1152 /* Do the first partial run if present. */
1153 if (first_vcn > rl->vcn) {
1154 s64 delta, length = rl->length;
1156 /* We know rl->length != 0 already. */
1157 if (unlikely(length < 0 || rl->lcn < LCN_HOLE))
1158 goto err_out;
1160 * If @stop_vcn is given and finishes inside this run, cap the
1161 * run length.
1163 if (unlikely(last_vcn >= 0 && rl[1].vcn > last_vcn)) {
1164 s64 s1 = last_vcn + 1;
1165 if (unlikely(rl[1].vcn > s1))
1166 length = s1 - rl->vcn;
1167 the_end = TRUE;
1169 delta = first_vcn - rl->vcn;
1170 /* Header byte + length. */
1171 rls += 1 + ntfs_get_nr_significant_bytes(length - delta);
1173 * If the logical cluster number (lcn) denotes a hole and we
1174 * are on NTFS 3.0+, we don't store it at all, i.e. we need
1175 * zero space. On earlier NTFS versions we just store the lcn.
1176 * Note: this assumes that on NTFS 1.2-, holes are stored with
1177 * an lcn of -1 and not a delta_lcn of -1 (unless both are -1).
1179 if (likely(rl->lcn >= 0 || vol->major_ver < 3)) {
1180 prev_lcn = rl->lcn;
1181 if (likely(rl->lcn >= 0))
1182 prev_lcn += delta;
1183 /* Change in lcn. */
1184 rls += ntfs_get_nr_significant_bytes(prev_lcn);
1186 /* Go to next runlist element. */
1187 rl++;
1189 /* Do the full runs. */
1190 for (; rl->length && !the_end; rl++) {
1191 s64 length = rl->length;
1193 if (unlikely(length < 0 || rl->lcn < LCN_HOLE))
1194 goto err_out;
1196 * If @stop_vcn is given and finishes inside this run, cap the
1197 * run length.
1199 if (unlikely(last_vcn >= 0 && rl[1].vcn > last_vcn)) {
1200 s64 s1 = last_vcn + 1;
1201 if (unlikely(rl[1].vcn > s1))
1202 length = s1 - rl->vcn;
1203 the_end = TRUE;
1205 /* Header byte + length. */
1206 rls += 1 + ntfs_get_nr_significant_bytes(length);
1208 * If the logical cluster number (lcn) denotes a hole and we
1209 * are on NTFS 3.0+, we don't store it at all, i.e. we need
1210 * zero space. On earlier NTFS versions we just store the lcn.
1211 * Note: this assumes that on NTFS 1.2-, holes are stored with
1212 * an lcn of -1 and not a delta_lcn of -1 (unless both are -1).
1214 if (likely(rl->lcn >= 0 || vol->major_ver < 3)) {
1215 /* Change in lcn. */
1216 rls += ntfs_get_nr_significant_bytes(rl->lcn -
1217 prev_lcn);
1218 prev_lcn = rl->lcn;
1221 return rls;
1222 err_out:
1223 if (rl->lcn == LCN_RL_NOT_MAPPED)
1224 rls = -EINVAL;
1225 else
1226 rls = -EIO;
1227 return rls;
1231 * ntfs_write_significant_bytes - write the significant bytes of a number
1232 * @dst: destination buffer to write to
1233 * @dst_max: pointer to last byte of destination buffer for bounds checking
1234 * @n: number whose significant bytes to write
1236 * Store in @dst, the minimum bytes of the number @n which are required to
1237 * identify @n unambiguously as a signed number, taking care not to exceed
1238 * @dest_max, the maximum position within @dst to which we are allowed to
1239 * write.
1241 * This is used when building the mapping pairs array of a runlist to compress
1242 * a given logical cluster number (lcn) or a specific run length to the minumum
1243 * size possible.
1245 * Return the number of bytes written on success. On error, i.e. the
1246 * destination buffer @dst is too small, return -ENOSPC.
1248 static inline int ntfs_write_significant_bytes(s8 *dst, const s8 *dst_max,
1249 const s64 n)
1251 s64 l = n;
1252 int i;
1253 s8 j;
1255 i = 0;
1256 do {
1257 if (unlikely(dst > dst_max))
1258 goto err_out;
1259 *dst++ = l & 0xffll;
1260 l >>= 8;
1261 i++;
1262 } while (l != 0 && l != -1);
1263 j = (n >> 8 * (i - 1)) & 0xff;
1264 /* If the sign bit is wrong, we need an extra byte. */
1265 if (n < 0 && j >= 0) {
1266 if (unlikely(dst > dst_max))
1267 goto err_out;
1268 i++;
1269 *dst = (s8)-1;
1270 } else if (n > 0 && j < 0) {
1271 if (unlikely(dst > dst_max))
1272 goto err_out;
1273 i++;
1274 *dst = (s8)0;
1276 return i;
1277 err_out:
1278 return -ENOSPC;
1282 * ntfs_mapping_pairs_build - build the mapping pairs array from a runlist
1283 * @vol: ntfs volume (needed for the ntfs version)
1284 * @dst: destination buffer to which to write the mapping pairs array
1285 * @dst_len: size of destination buffer @dst in bytes
1286 * @rl: locked runlist for which to build the mapping pairs array
1287 * @first_vcn: first vcn which to include in the mapping pairs array
1288 * @last_vcn: last vcn which to include in the mapping pairs array
1289 * @stop_vcn: first vcn outside destination buffer on success or -ENOSPC
1291 * Create the mapping pairs array from the locked runlist @rl, starting at vcn
1292 * @first_vcn and finishing with vcn @last_vcn and save the array in @dst.
1293 * @dst_len is the size of @dst in bytes and it should be at least equal to the
1294 * value obtained by calling ntfs_get_size_for_mapping_pairs().
1296 * A @last_vcn of -1 means end of runlist and in that case the mapping pairs
1297 * array corresponding to the runlist starting at vcn @first_vcn and finishing
1298 * at the end of the runlist is created.
1300 * If @rl is NULL, just write a single terminator byte to @dst.
1302 * On success or -ENOSPC error, if @stop_vcn is not NULL, *@stop_vcn is set to
1303 * the first vcn outside the destination buffer. Note that on error, @dst has
1304 * been filled with all the mapping pairs that will fit, thus it can be treated
1305 * as partial success, in that a new attribute extent needs to be created or
1306 * the next extent has to be used and the mapping pairs build has to be
1307 * continued with @first_vcn set to *@stop_vcn.
1309 * Return 0 on success and -errno on error. The following error codes are
1310 * defined:
1311 * -EINVAL - Run list contains unmapped elements. Make sure to only pass
1312 * fully mapped runlists to this function.
1313 * -EIO - The runlist is corrupt.
1314 * -ENOSPC - The destination buffer is too small.
1316 * Locking: @rl must be locked on entry (either for reading or writing), it
1317 * remains locked throughout, and is left locked upon return.
1319 int ntfs_mapping_pairs_build(const ntfs_volume *vol, s8 *dst,
1320 const int dst_len, const runlist_element *rl,
1321 const VCN first_vcn, const VCN last_vcn, VCN *const stop_vcn)
1323 LCN prev_lcn;
1324 s8 *dst_max, *dst_next;
1325 int err = -ENOSPC;
1326 BOOL the_end = FALSE;
1327 s8 len_len, lcn_len;
1329 BUG_ON(first_vcn < 0);
1330 BUG_ON(last_vcn < -1);
1331 BUG_ON(last_vcn >= 0 && first_vcn > last_vcn);
1332 BUG_ON(dst_len < 1);
1333 if (!rl) {
1334 BUG_ON(first_vcn);
1335 BUG_ON(last_vcn > 0);
1336 if (stop_vcn)
1337 *stop_vcn = 0;
1338 /* Terminator byte. */
1339 *dst = 0;
1340 return 0;
1342 /* Skip to runlist element containing @first_vcn. */
1343 while (rl->length && first_vcn >= rl[1].vcn)
1344 rl++;
1345 if (unlikely((!rl->length && first_vcn > rl->vcn) ||
1346 first_vcn < rl->vcn))
1347 return -EINVAL;
1349 * @dst_max is used for bounds checking in
1350 * ntfs_write_significant_bytes().
1352 dst_max = dst + dst_len - 1;
1353 prev_lcn = 0;
1354 /* Do the first partial run if present. */
1355 if (first_vcn > rl->vcn) {
1356 s64 delta, length = rl->length;
1358 /* We know rl->length != 0 already. */
1359 if (unlikely(length < 0 || rl->lcn < LCN_HOLE))
1360 goto err_out;
1362 * If @stop_vcn is given and finishes inside this run, cap the
1363 * run length.
1365 if (unlikely(last_vcn >= 0 && rl[1].vcn > last_vcn)) {
1366 s64 s1 = last_vcn + 1;
1367 if (unlikely(rl[1].vcn > s1))
1368 length = s1 - rl->vcn;
1369 the_end = TRUE;
1371 delta = first_vcn - rl->vcn;
1372 /* Write length. */
1373 len_len = ntfs_write_significant_bytes(dst + 1, dst_max,
1374 length - delta);
1375 if (unlikely(len_len < 0))
1376 goto size_err;
1378 * If the logical cluster number (lcn) denotes a hole and we
1379 * are on NTFS 3.0+, we don't store it at all, i.e. we need
1380 * zero space. On earlier NTFS versions we just write the lcn
1381 * change. FIXME: Do we need to write the lcn change or just
1382 * the lcn in that case? Not sure as I have never seen this
1383 * case on NT4. - We assume that we just need to write the lcn
1384 * change until someone tells us otherwise... (AIA)
1386 if (likely(rl->lcn >= 0 || vol->major_ver < 3)) {
1387 prev_lcn = rl->lcn;
1388 if (likely(rl->lcn >= 0))
1389 prev_lcn += delta;
1390 /* Write change in lcn. */
1391 lcn_len = ntfs_write_significant_bytes(dst + 1 +
1392 len_len, dst_max, prev_lcn);
1393 if (unlikely(lcn_len < 0))
1394 goto size_err;
1395 } else
1396 lcn_len = 0;
1397 dst_next = dst + len_len + lcn_len + 1;
1398 if (unlikely(dst_next > dst_max))
1399 goto size_err;
1400 /* Update header byte. */
1401 *dst = lcn_len << 4 | len_len;
1402 /* Position at next mapping pairs array element. */
1403 dst = dst_next;
1404 /* Go to next runlist element. */
1405 rl++;
1407 /* Do the full runs. */
1408 for (; rl->length && !the_end; rl++) {
1409 s64 length = rl->length;
1411 if (unlikely(length < 0 || rl->lcn < LCN_HOLE))
1412 goto err_out;
1414 * If @stop_vcn is given and finishes inside this run, cap the
1415 * run length.
1417 if (unlikely(last_vcn >= 0 && rl[1].vcn > last_vcn)) {
1418 s64 s1 = last_vcn + 1;
1419 if (unlikely(rl[1].vcn > s1))
1420 length = s1 - rl->vcn;
1421 the_end = TRUE;
1423 /* Write length. */
1424 len_len = ntfs_write_significant_bytes(dst + 1, dst_max,
1425 length);
1426 if (unlikely(len_len < 0))
1427 goto size_err;
1429 * If the logical cluster number (lcn) denotes a hole and we
1430 * are on NTFS 3.0+, we don't store it at all, i.e. we need
1431 * zero space. On earlier NTFS versions we just write the lcn
1432 * change. FIXME: Do we need to write the lcn change or just
1433 * the lcn in that case? Not sure as I have never seen this
1434 * case on NT4. - We assume that we just need to write the lcn
1435 * change until someone tells us otherwise... (AIA)
1437 if (likely(rl->lcn >= 0 || vol->major_ver < 3)) {
1438 /* Write change in lcn. */
1439 lcn_len = ntfs_write_significant_bytes(dst + 1 +
1440 len_len, dst_max, rl->lcn - prev_lcn);
1441 if (unlikely(lcn_len < 0))
1442 goto size_err;
1443 prev_lcn = rl->lcn;
1444 } else
1445 lcn_len = 0;
1446 dst_next = dst + len_len + lcn_len + 1;
1447 if (unlikely(dst_next > dst_max))
1448 goto size_err;
1449 /* Update header byte. */
1450 *dst = lcn_len << 4 | len_len;
1451 /* Position at next mapping pairs array element. */
1452 dst = dst_next;
1454 /* Success. */
1455 err = 0;
1456 size_err:
1457 /* Set stop vcn. */
1458 if (stop_vcn)
1459 *stop_vcn = rl->vcn;
1460 /* Add terminator byte. */
1461 *dst = 0;
1462 return err;
1463 err_out:
1464 if (rl->lcn == LCN_RL_NOT_MAPPED)
1465 err = -EINVAL;
1466 else
1467 err = -EIO;
1468 return err;
1472 * ntfs_rl_truncate_nolock - truncate a runlist starting at a specified vcn
1473 * @vol: ntfs volume (needed for error output)
1474 * @runlist: runlist to truncate
1475 * @new_length: the new length of the runlist in VCNs
1477 * Truncate the runlist described by @runlist as well as the memory buffer
1478 * holding the runlist elements to a length of @new_length VCNs.
1480 * If @new_length lies within the runlist, the runlist elements with VCNs of
1481 * @new_length and above are discarded. As a special case if @new_length is
1482 * zero, the runlist is discarded and set to NULL.
1484 * If @new_length lies beyond the runlist, a sparse runlist element is added to
1485 * the end of the runlist @runlist or if the last runlist element is a sparse
1486 * one already, this is extended.
1488 * Note, no checking is done for unmapped runlist elements. It is assumed that
1489 * the caller has mapped any elements that need to be mapped already.
1491 * Return 0 on success and -errno on error.
1493 * Locking: The caller must hold @runlist->lock for writing.
1495 int ntfs_rl_truncate_nolock(const ntfs_volume *vol, runlist *const runlist,
1496 const s64 new_length)
1498 runlist_element *rl;
1499 int old_size;
1501 ntfs_debug("Entering for new_length 0x%llx.", (long long)new_length);
1502 BUG_ON(!runlist);
1503 BUG_ON(new_length < 0);
1504 rl = runlist->rl;
1505 if (!new_length) {
1506 ntfs_debug("Freeing runlist.");
1507 runlist->rl = NULL;
1508 if (rl)
1509 ntfs_free(rl);
1510 return 0;
1512 if (unlikely(!rl)) {
1514 * Create a runlist consisting of a sparse runlist element of
1515 * length @new_length followed by a terminator runlist element.
1517 rl = ntfs_malloc_nofs(PAGE_SIZE);
1518 if (unlikely(!rl)) {
1519 ntfs_error(vol->sb, "Not enough memory to allocate "
1520 "runlist element buffer.");
1521 return -ENOMEM;
1523 runlist->rl = rl;
1524 rl[1].length = rl->vcn = 0;
1525 rl->lcn = LCN_HOLE;
1526 rl[1].vcn = rl->length = new_length;
1527 rl[1].lcn = LCN_ENOENT;
1528 return 0;
1530 BUG_ON(new_length < rl->vcn);
1531 /* Find @new_length in the runlist. */
1532 while (likely(rl->length && new_length >= rl[1].vcn))
1533 rl++;
1535 * If not at the end of the runlist we need to shrink it.
1536 * If at the end of the runlist we need to expand it.
1538 if (rl->length) {
1539 runlist_element *trl;
1540 BOOL is_end;
1542 ntfs_debug("Shrinking runlist.");
1543 /* Determine the runlist size. */
1544 trl = rl + 1;
1545 while (likely(trl->length))
1546 trl++;
1547 old_size = trl - runlist->rl + 1;
1548 /* Truncate the run. */
1549 rl->length = new_length - rl->vcn;
1551 * If a run was partially truncated, make the following runlist
1552 * element a terminator.
1554 is_end = FALSE;
1555 if (rl->length) {
1556 rl++;
1557 if (!rl->length)
1558 is_end = TRUE;
1559 rl->vcn = new_length;
1560 rl->length = 0;
1562 rl->lcn = LCN_ENOENT;
1563 /* Reallocate memory if necessary. */
1564 if (!is_end) {
1565 int new_size = rl - runlist->rl + 1;
1566 rl = ntfs_rl_realloc(runlist->rl, old_size, new_size);
1567 if (IS_ERR(rl))
1568 ntfs_warning(vol->sb, "Failed to shrink "
1569 "runlist buffer. This just "
1570 "wastes a bit of memory "
1571 "temporarily so we ignore it "
1572 "and return success.");
1573 else
1574 runlist->rl = rl;
1576 } else if (likely(/* !rl->length && */ new_length > rl->vcn)) {
1577 ntfs_debug("Expanding runlist.");
1579 * If there is a previous runlist element and it is a sparse
1580 * one, extend it. Otherwise need to add a new, sparse runlist
1581 * element.
1583 if ((rl > runlist->rl) && ((rl - 1)->lcn == LCN_HOLE))
1584 (rl - 1)->length = new_length - (rl - 1)->vcn;
1585 else {
1586 /* Determine the runlist size. */
1587 old_size = rl - runlist->rl + 1;
1588 /* Reallocate memory if necessary. */
1589 rl = ntfs_rl_realloc(runlist->rl, old_size,
1590 old_size + 1);
1591 if (IS_ERR(rl)) {
1592 ntfs_error(vol->sb, "Failed to expand runlist "
1593 "buffer, aborting.");
1594 return PTR_ERR(rl);
1596 runlist->rl = rl;
1598 * Set @rl to the same runlist element in the new
1599 * runlist as before in the old runlist.
1601 rl += old_size - 1;
1602 /* Add a new, sparse runlist element. */
1603 rl->lcn = LCN_HOLE;
1604 rl->length = new_length - rl->vcn;
1605 /* Add a new terminator runlist element. */
1606 rl++;
1607 rl->length = 0;
1609 rl->vcn = new_length;
1610 rl->lcn = LCN_ENOENT;
1611 } else /* if (unlikely(!rl->length && new_length == rl->vcn)) */ {
1612 /* Runlist already has same size as requested. */
1613 rl->lcn = LCN_ENOENT;
1615 ntfs_debug("Done.");
1616 return 0;
1620 * ntfs_rl_punch_nolock - punch a hole into a runlist
1621 * @vol: ntfs volume (needed for error output)
1622 * @runlist: runlist to punch a hole into
1623 * @start: starting VCN of the hole to be created
1624 * @length: size of the hole to be created in units of clusters
1626 * Punch a hole into the runlist @runlist starting at VCN @start and of size
1627 * @length clusters.
1629 * Return 0 on success and -errno on error, in which case @runlist has not been
1630 * modified.
1632 * If @start and/or @start + @length are outside the runlist return error code
1633 * -ENOENT.
1635 * If the runlist contains unmapped or error elements between @start and @start
1636 * + @length return error code -EINVAL.
1638 * Locking: The caller must hold @runlist->lock for writing.
1640 int ntfs_rl_punch_nolock(const ntfs_volume *vol, runlist *const runlist,
1641 const VCN start, const s64 length)
1643 const VCN end = start + length;
1644 s64 delta;
1645 runlist_element *rl, *rl_end, *rl_real_end, *trl;
1646 int old_size;
1647 BOOL lcn_fixup = FALSE;
1649 ntfs_debug("Entering for start 0x%llx, length 0x%llx.",
1650 (long long)start, (long long)length);
1651 BUG_ON(!runlist);
1652 BUG_ON(start < 0);
1653 BUG_ON(length < 0);
1654 BUG_ON(end < 0);
1655 rl = runlist->rl;
1656 if (unlikely(!rl)) {
1657 if (likely(!start && !length))
1658 return 0;
1659 return -ENOENT;
1661 /* Find @start in the runlist. */
1662 while (likely(rl->length && start >= rl[1].vcn))
1663 rl++;
1664 rl_end = rl;
1665 /* Find @end in the runlist. */
1666 while (likely(rl_end->length && end >= rl_end[1].vcn)) {
1667 /* Verify there are no unmapped or error elements. */
1668 if (unlikely(rl_end->lcn < LCN_HOLE))
1669 return -EINVAL;
1670 rl_end++;
1672 /* Check the last element. */
1673 if (unlikely(rl_end->length && rl_end->lcn < LCN_HOLE))
1674 return -EINVAL;
1675 /* This covers @start being out of bounds, too. */
1676 if (!rl_end->length && end > rl_end->vcn)
1677 return -ENOENT;
1678 if (!length)
1679 return 0;
1680 if (!rl->length)
1681 return -ENOENT;
1682 rl_real_end = rl_end;
1683 /* Determine the runlist size. */
1684 while (likely(rl_real_end->length))
1685 rl_real_end++;
1686 old_size = rl_real_end - runlist->rl + 1;
1687 /* If @start is in a hole simply extend the hole. */
1688 if (rl->lcn == LCN_HOLE) {
1690 * If both @start and @end are in the same sparse run, we are
1691 * done.
1693 if (end <= rl[1].vcn) {
1694 ntfs_debug("Done (requested hole is already sparse).");
1695 return 0;
1697 extend_hole:
1698 /* Extend the hole. */
1699 rl->length = end - rl->vcn;
1700 /* If @end is in a hole, merge it with the current one. */
1701 if (rl_end->lcn == LCN_HOLE) {
1702 rl_end++;
1703 rl->length = rl_end->vcn - rl->vcn;
1705 /* We have done the hole. Now deal with the remaining tail. */
1706 rl++;
1707 /* Cut out all runlist elements up to @end. */
1708 if (rl < rl_end)
1709 memmove(rl, rl_end, (rl_real_end - rl_end + 1) *
1710 sizeof(*rl));
1711 /* Adjust the beginning of the tail if necessary. */
1712 if (end > rl->vcn) {
1713 s64 delta = end - rl->vcn;
1714 rl->vcn = end;
1715 rl->length -= delta;
1716 /* Only adjust the lcn if it is real. */
1717 if (rl->lcn >= 0)
1718 rl->lcn += delta;
1720 shrink_allocation:
1721 /* Reallocate memory if the allocation changed. */
1722 if (rl < rl_end) {
1723 rl = ntfs_rl_realloc(runlist->rl, old_size,
1724 old_size - (rl_end - rl));
1725 if (IS_ERR(rl))
1726 ntfs_warning(vol->sb, "Failed to shrink "
1727 "runlist buffer. This just "
1728 "wastes a bit of memory "
1729 "temporarily so we ignore it "
1730 "and return success.");
1731 else
1732 runlist->rl = rl;
1734 ntfs_debug("Done (extend hole).");
1735 return 0;
1738 * If @start is at the beginning of a run things are easier as there is
1739 * no need to split the first run.
1741 if (start == rl->vcn) {
1743 * @start is at the beginning of a run.
1745 * If the previous run is sparse, extend its hole.
1747 * If @end is not in the same run, switch the run to be sparse
1748 * and extend the newly created hole.
1750 * Thus both of these cases reduce the problem to the above
1751 * case of "@start is in a hole".
1753 if (rl > runlist->rl && (rl - 1)->lcn == LCN_HOLE) {
1754 rl--;
1755 goto extend_hole;
1757 if (end >= rl[1].vcn) {
1758 rl->lcn = LCN_HOLE;
1759 goto extend_hole;
1762 * The final case is when @end is in the same run as @start.
1763 * For this need to split the run into two. One run for the
1764 * sparse region between the beginning of the old run, i.e.
1765 * @start, and @end and one for the remaining non-sparse
1766 * region, i.e. between @end and the end of the old run.
1768 trl = ntfs_rl_realloc(runlist->rl, old_size, old_size + 1);
1769 if (IS_ERR(trl))
1770 goto enomem_out;
1771 old_size++;
1772 if (runlist->rl != trl) {
1773 rl = trl + (rl - runlist->rl);
1774 rl_end = trl + (rl_end - runlist->rl);
1775 rl_real_end = trl + (rl_real_end - runlist->rl);
1776 runlist->rl = trl;
1778 split_end:
1779 /* Shift all the runs up by one. */
1780 memmove(rl + 1, rl, (rl_real_end - rl + 1) * sizeof(*rl));
1781 /* Finally, setup the two split runs. */
1782 rl->lcn = LCN_HOLE;
1783 rl->length = length;
1784 rl++;
1785 rl->vcn += length;
1786 /* Only adjust the lcn if it is real. */
1787 if (rl->lcn >= 0 || lcn_fixup)
1788 rl->lcn += length;
1789 rl->length -= length;
1790 ntfs_debug("Done (split one).");
1791 return 0;
1794 * @start is neither in a hole nor at the beginning of a run.
1796 * If @end is in a hole, things are easier as simply truncating the run
1797 * @start is in to end at @start - 1, deleting all runs after that up
1798 * to @end, and finally extending the beginning of the run @end is in
1799 * to be @start is all that is needed.
1801 if (rl_end->lcn == LCN_HOLE) {
1802 /* Truncate the run containing @start. */
1803 rl->length = start - rl->vcn;
1804 rl++;
1805 /* Cut out all runlist elements up to @end. */
1806 if (rl < rl_end)
1807 memmove(rl, rl_end, (rl_real_end - rl_end + 1) *
1808 sizeof(*rl));
1809 /* Extend the beginning of the run @end is in to be @start. */
1810 rl->vcn = start;
1811 rl->length = rl[1].vcn - start;
1812 goto shrink_allocation;
1815 * If @end is not in a hole there are still two cases to distinguish.
1816 * Either @end is or is not in the same run as @start.
1818 * The second case is easier as it can be reduced to an already solved
1819 * problem by truncating the run @start is in to end at @start - 1.
1820 * Then, if @end is in the next run need to split the run into a sparse
1821 * run followed by a non-sparse run (already covered above) and if @end
1822 * is not in the next run switching it to be sparse, again reduces the
1823 * problem to the already covered case of "@start is in a hole".
1825 if (end >= rl[1].vcn) {
1827 * If @end is not in the next run, reduce the problem to the
1828 * case of "@start is in a hole".
1830 if (rl[1].length && end >= rl[2].vcn) {
1831 /* Truncate the run containing @start. */
1832 rl->length = start - rl->vcn;
1833 rl++;
1834 rl->vcn = start;
1835 rl->lcn = LCN_HOLE;
1836 goto extend_hole;
1838 trl = ntfs_rl_realloc(runlist->rl, old_size, old_size + 1);
1839 if (IS_ERR(trl))
1840 goto enomem_out;
1841 old_size++;
1842 if (runlist->rl != trl) {
1843 rl = trl + (rl - runlist->rl);
1844 rl_end = trl + (rl_end - runlist->rl);
1845 rl_real_end = trl + (rl_real_end - runlist->rl);
1846 runlist->rl = trl;
1848 /* Truncate the run containing @start. */
1849 rl->length = start - rl->vcn;
1850 rl++;
1852 * @end is in the next run, reduce the problem to the case
1853 * where "@start is at the beginning of a run and @end is in
1854 * the same run as @start".
1856 delta = rl->vcn - start;
1857 rl->vcn = start;
1858 if (rl->lcn >= 0) {
1859 rl->lcn -= delta;
1860 /* Need this in case the lcn just became negative. */
1861 lcn_fixup = TRUE;
1863 rl->length += delta;
1864 goto split_end;
1867 * The first case from above, i.e. @end is in the same run as @start.
1868 * We need to split the run into three. One run for the non-sparse
1869 * region between the beginning of the old run and @start, one for the
1870 * sparse region between @start and @end, and one for the remaining
1871 * non-sparse region, i.e. between @end and the end of the old run.
1873 trl = ntfs_rl_realloc(runlist->rl, old_size, old_size + 2);
1874 if (IS_ERR(trl))
1875 goto enomem_out;
1876 old_size += 2;
1877 if (runlist->rl != trl) {
1878 rl = trl + (rl - runlist->rl);
1879 rl_end = trl + (rl_end - runlist->rl);
1880 rl_real_end = trl + (rl_real_end - runlist->rl);
1881 runlist->rl = trl;
1883 /* Shift all the runs up by two. */
1884 memmove(rl + 2, rl, (rl_real_end - rl + 1) * sizeof(*rl));
1885 /* Finally, setup the three split runs. */
1886 rl->length = start - rl->vcn;
1887 rl++;
1888 rl->vcn = start;
1889 rl->lcn = LCN_HOLE;
1890 rl->length = length;
1891 rl++;
1892 delta = end - rl->vcn;
1893 rl->vcn = end;
1894 rl->lcn += delta;
1895 rl->length -= delta;
1896 ntfs_debug("Done (split both).");
1897 return 0;
1898 enomem_out:
1899 ntfs_error(vol->sb, "Not enough memory to extend runlist buffer.");
1900 return -ENOMEM;
1903 #endif /* NTFS_RW */