rpm/glibc
SHA256
1
0
Fork 0
forked from pool/glibc
Code Pull Requests Activity
f765de7465
glibc/strcoll-overflow.patch
Andreas Schwab 66993bce55 Accepting request 198529 from home:Andreas_Schwab:Factory 
- fcntl-o-tmpfile.patch: Add O_TMPFILE to <fcntl.h> (bnc#837965)
- strcoll-overflow.patch: Update to latest version

OBS-URL: https://build.opensuse.org/request/show/198529
OBS-URL: https://build.opensuse.org/package/show/Base:System/glibc?expand=0&rev=307
2013-09-11 13:17:37 +00:00

1107 lines
29 KiB
Diff
Raw Blame History

Siddhesh Poyarekar <siddhesh@redhat.com>
[BZ #14547]
* string/tst-strcoll-overflow.c: New test case.
* string/Makefile (xtests): Add tst-strcoll-overflow.
* string/strcoll_l.c (STRCOLL): Skip allocating memory for
cache if string sizes may cause integer overflow.
[BZ #14547]
* string/strcoll_l.c (coll_seq): New members rule, idx,
save_idx and back_us.
(get_next_seq_nocache): New function.
(do_compare_nocache): New function.
(STRCOLL): Use get_next_seq_nocache and do_compare_nocache
when malloc fails.
* string/strcoll_l.c (coll_seq): New structure.
(get_next_seq_cached): New function.
(get_next_seq): New function.
(do_compare): New function.
(STRCOLL): Use GNU style definition. Simplify implementation
by using get_next_seq, get_next_seq_cached and do_compare.
Index: glibc-2.18/string/Makefile
===================================================================
--- glibc-2.18.orig/string/Makefile
+++ glibc-2.18/string/Makefile
@@ -57,6 +57,8 @@ tests := tester inl-tester noinl-tester
tests-ifunc := $(strop-tests:%=test-%-ifunc)
tests += $(tests-ifunc)
+xtests = tst-strcoll-overflow
+
include ../Rules
tester-ENV = LANGUAGE=C
Index: glibc-2.18/string/strcoll_l.c
===================================================================
--- glibc-2.18.orig/string/strcoll_l.c
+++ glibc-2.18/string/strcoll_l.c
@@ -41,11 +41,434 @@
#include "../locale/localeinfo.h"
+/* Track status while looking for sequences in a string. */
+typedef struct
+{
+ int len; /* Length of the current sequence. */
+ size_t val; /* Position of the sequence relative to the
+ previous non-ignored sequence. */
+ size_t idxnow; /* Current index in sequences. */
+ size_t idxmax; /* Maximum index in sequences. */
+ size_t idxcnt; /* Current count of indices. */
+ size_t backw; /* Current Backward sequence index. */
+ size_t backw_stop; /* Index where the backward sequences stop. */
+ const USTRING_TYPE *us; /* The string. */
+ int32_t *idxarr; /* Array to cache weight indices. */
+ unsigned char *rulearr; /* Array to cache rules. */
+ unsigned char rule; /* Saved rule for the first sequence. */
+ int32_t idx; /* Index to weight of the current sequence. */
+ int32_t save_idx; /* Save looked up index of a forward
+ sequence after the last backward
+ sequence. */
+ const USTRING_TYPE *back_us; /* Beginning of the backward sequence. */
+} coll_seq;
+
+/* Get next sequence. The weight indices are cached, so we don't need to
+ traverse the string. */
+static void
+get_next_seq_cached (coll_seq *seq, int nrules, int pass,
+ const unsigned char *rulesets,
+ const USTRING_TYPE *weights)
+{
+ size_t val = seq->val = 0;
+ int len = seq->len;
+ size_t backw_stop = seq->backw_stop;
+ size_t backw = seq->backw;
+ size_t idxcnt = seq->idxcnt;
+ size_t idxmax = seq->idxmax;
+ size_t idxnow = seq->idxnow;
+ unsigned char *rulearr = seq->rulearr;
+ int32_t *idxarr = seq->idxarr;
+
+ while (len == 0)
+ {
+ ++val;
+ if (backw_stop != ~0ul)
+ {
+ /* There is something pushed. */
+ if (backw == backw_stop)
+ {
+ /* The last pushed character was handled. Continue
+ with forward characters. */
+ if (idxcnt < idxmax)
+ {
+ idxnow = idxcnt;
+ backw_stop = ~0ul;
+ }
+ else
+ {
+ /* Nothing any more. The backward sequence
+ ended with the last sequence in the string. */
+ idxnow = ~0ul;
+ break;
+ }
+ }
+ else
+ idxnow = --backw;
+ }
+ else
+ {
+ backw_stop = idxcnt;
+
+ while (idxcnt < idxmax)
+ {
+ if ((rulesets[rulearr[idxcnt] * nrules + pass]
+ & sort_backward) == 0)
+ /* No more backward characters to push. */
+ break;
+ ++idxcnt;
+ }
+
+ if (backw_stop == idxcnt)
+ {
+ /* No sequence at all or just one. */
+ if (idxcnt == idxmax)
+ /* Note that LEN is still zero. */
+ break;
+
+ backw_stop = ~0ul;
+ idxnow = idxcnt++;
+ }
+ else
+ /* We pushed backward sequences. */
+ idxnow = backw = idxcnt - 1;
+ }
+ len = weights[idxarr[idxnow]++];
+ }
+
+ /* Update the structure. */
+ seq->val = val;
+ seq->len = len;
+ seq->backw_stop = backw_stop;
+ seq->backw = backw;
+ seq->idxcnt = idxcnt;
+ seq->idxnow = idxnow;
+}
+
+/* Get next sequence. Traverse the string as required. */
+static void
+get_next_seq (coll_seq *seq, int nrules, const unsigned char *rulesets,
+ const USTRING_TYPE *weights, const int32_t *table,
+ const USTRING_TYPE *extra, const int32_t *indirect)
+{
+#include WEIGHT_H
+ size_t val = seq->val = 0;
+ int len = seq->len;
+ size_t backw_stop = seq->backw_stop;
+ size_t backw = seq->backw;
+ size_t idxcnt = seq->idxcnt;
+ size_t idxmax = seq->idxmax;
+ size_t idxnow = seq->idxnow;
+ unsigned char *rulearr = seq->rulearr;
+ int32_t *idxarr = seq->idxarr;
+ const USTRING_TYPE *us = seq->us;
+
+ while (len == 0)
+ {
+ ++val;
+ if (backw_stop != ~0ul)
+ {
+ /* There is something pushed. */
+ if (backw == backw_stop)
+ {
+ /* The last pushed character was handled. Continue
+ with forward characters. */
+ if (idxcnt < idxmax)
+ {
+ idxnow = idxcnt;
+ backw_stop = ~0ul;
+ }
+ else
+ /* Nothing any more. The backward sequence ended with
+ the last sequence in the string. Note that LEN
+ is still zero. */
+ break;
+ }
+ else
+ idxnow = --backw;
+ }
+ else
+ {
+ backw_stop = idxmax;
+
+ while (*us != L('\0'))
+ {
+ int32_t tmp = findidx (&us, -1);
+ rulearr[idxmax] = tmp >> 24;
+ idxarr[idxmax] = tmp & 0xffffff;
+ idxcnt = idxmax++;
+
+ if ((rulesets[rulearr[idxcnt] * nrules]
+ & sort_backward) == 0)
+ /* No more backward characters to push. */
+ break;
+ ++idxcnt;
+ }
+
+ if (backw_stop >= idxcnt)
+ {
+ /* No sequence at all or just one. */
+ if (idxcnt == idxmax || backw_stop > idxcnt)
+ /* Note that LEN is still zero. */
+ break;
+
+ backw_stop = ~0ul;
+ idxnow = idxcnt;
+ }
+ else
+ /* We pushed backward sequences. */
+ idxnow = backw = idxcnt - 1;
+ }
+ len = weights[idxarr[idxnow]++];
+ }
+
+ /* Update the structure. */
+ seq->val = val;
+ seq->len = len;
+ seq->backw_stop = backw_stop;
+ seq->backw = backw;
+ seq->idxcnt = idxcnt;
+ seq->idxmax = idxmax;
+ seq->idxnow = idxnow;
+ seq->us = us;
+}
+
+/* Get next sequence. Traverse the string as required. This function does not
+ set or use any index or rule cache. */
+static void
+get_next_seq_nocache (coll_seq *seq, int nrules, const unsigned char *rulesets,
+ const USTRING_TYPE *weights, const int32_t *table,
+ const USTRING_TYPE *extra, const int32_t *indirect,
+ int pass)
+{
+#include WEIGHT_H
+ size_t val = seq->val = 0;
+ int len = seq->len;
+ size_t backw_stop = seq->backw_stop;
+ size_t backw = seq->backw;
+ size_t idxcnt = seq->idxcnt;
+ size_t idxmax = seq->idxmax;
+ int32_t idx = seq->idx;
+ const USTRING_TYPE *us = seq->us;
+
+ while (len == 0)
+ {
+ ++val;
+ if (backw_stop != ~0ul)
+ {
+ /* There is something pushed. */
+ if (backw == backw_stop)
+ {
+ /* The last pushed character was handled. Continue
+ with forward characters. */
+ if (idxcnt < idxmax)
+ {
+ idx = seq->save_idx;
+ backw_stop = ~0ul;
+ }
+ else
+ {
+ /* Nothing anymore. The backward sequence ended with
+ the last sequence in the string. Note that len is
+ still zero. */
+ idx = 0;
+ break;
+ }
+ }
+ else
+ {
+ /* XXX Traverse BACKW sequences from the beginning of
+ BACKW_STOP to get the next sequence. Is ther a quicker way
+ to do this? */
+ size_t i = backw_stop;
+ us = seq->back_us;
+ while (i < backw)
+ {
+ int32_t tmp = findidx (&us, -1);
+ idx = tmp & 0xffffff;
+ i++;
+ }
+ --backw;
+ us = seq->us;
+ }
+ }
+ else
+ {
+ backw_stop = idxmax;
+ int32_t prev_idx = idx;
+
+ while (*us != L('\0'))
+ {
+ int32_t tmp = findidx (&us, -1);
+ unsigned char rule = tmp >> 24;
+ prev_idx = idx;
+ idx = tmp & 0xffffff;
+ idxcnt = idxmax++;
+
+ /* Save the rule for the first sequence. */
+ if (__glibc_unlikely (idxcnt == 0))
+ seq->rule = rule;
+
+ if ((rulesets[rule * nrules + pass]
+ & sort_backward) == 0)
+ /* No more backward characters to push. */
+ break;
+ ++idxcnt;
+ }
+
+ if (backw_stop >= idxcnt)
+ {
+ /* No sequence at all or just one. */
+ if (idxcnt == idxmax || backw_stop > idxcnt)
+ /* Note that len is still zero. */
+ break;
+
+ backw_stop = ~0ul;
+ }
+ else
+ {
+ /* We pushed backward sequences. If the stream ended with the
+ backward sequence, then we process the last sequence we
+ found. Otherwise we process the sequence before the last
+ one since the last one was a forward sequence. */
+ seq->back_us = seq->us;
+ seq->us = us;
+ backw = idxcnt;
+ if (idxmax > idxcnt)
+ {
+ backw--;
+ seq->save_idx = idx;
+ idx = prev_idx;
+ }
+ if (backw > backw_stop)
+ backw--;
+ }
+ }
+
+ len = weights[idx++];
+ /* Skip over indices of previous levels. */
+ for (int i = 0; i < pass; i++)
+ {
+ idx += len;
+ len = weights[idx];
+ idx++;
+ }
+ }
+
+ /* Update the structure. */
+ seq->val = val;
+ seq->len = len;
+ seq->backw_stop = backw_stop;
+ seq->backw = backw;
+ seq->idxcnt = idxcnt;
+ seq->idxmax = idxmax;
+ seq->us = us;
+ seq->idx = idx;
+}
+
+/* Compare two sequences. This version does not use the index and rules
+ cache. */
+static int
+do_compare_nocache (coll_seq *seq1, coll_seq *seq2, int position,
+ const USTRING_TYPE *weights)
+{
+ int seq1len = seq1->len;
+ int seq2len = seq2->len;
+ size_t val1 = seq1->val;
+ size_t val2 = seq2->val;
+ int idx1 = seq1->idx;
+ int idx2 = seq2->idx;
+ int result = 0;
+
+ /* Test for position if necessary. */
+ if (position && val1 != val2)
+ {
+ result = val1 > val2 ? 1 : -1;
+ goto out;
+ }
+
+ /* Compare the two sequences. */
+ do
+ {
+ if (weights[idx1] != weights[idx2])
+ {
+ /* The sequences differ. */
+ result = weights[idx1] - weights[idx2];
+ goto out;
+ }
+
+ /* Increment the offsets. */
+ ++idx1;
+ ++idx2;
+
+ --seq1len;
+ --seq2len;
+ }
+ while (seq1len > 0 && seq2len > 0);
+
+ if (position && seq1len != seq2len)
+ result = seq1len - seq2len;
+
+out:
+ seq1->len = seq1len;
+ seq2->len = seq2len;
+ seq1->idx = idx1;
+ seq2->idx = idx2;
+ return result;
+}
+
+/* Compare two sequences using the index cache. */
+static int
+do_compare (coll_seq *seq1, coll_seq *seq2, int position,
+ const USTRING_TYPE *weights)
+{
+ int seq1len = seq1->len;
+ int seq2len = seq2->len;
+ size_t val1 = seq1->val;
+ size_t val2 = seq2->val;
+ int32_t *idx1arr = seq1->idxarr;
+ int32_t *idx2arr = seq2->idxarr;
+ int idx1now = seq1->idxnow;
+ int idx2now = seq2->idxnow;
+ int result = 0;
+
+ /* Test for position if necessary. */
+ if (position && val1 != val2)
+ {
+ result = val1 > val2 ? 1 : -1;
+ goto out;
+ }
+
+ /* Compare the two sequences. */
+ do
+ {
+ if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
+ {
+ /* The sequences differ. */
+ result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
+ goto out;
+ }
+
+ /* Increment the offsets. */
+ ++idx1arr[idx1now];
+ ++idx2arr[idx2now];
+
+ --seq1len;
+ --seq2len;
+ }
+ while (seq1len > 0 && seq2len > 0);
+
+ if (position && seq1len != seq2len)
+ result = seq1len - seq2len;
+
+out:
+ seq1->len = seq1len;
+ seq2->len = seq2len;
+ return result;
+}
+
int
-STRCOLL (s1, s2, l)
- const STRING_TYPE *s1;
- const STRING_TYPE *s2;
- __locale_t l;
+STRCOLL (const STRING_TYPE *s1, const STRING_TYPE *s2, __locale_t l)
{
struct __locale_data *current = l->__locales[LC_COLLATE];
uint_fast32_t nrules = current->values[_NL_ITEM_INDEX (_NL_COLLATE_NRULES)].word;
@@ -56,34 +479,6 @@ STRCOLL (s1, s2, l)
const USTRING_TYPE *weights;
const USTRING_TYPE *extra;
const int32_t *indirect;
- uint_fast32_t pass;
- int result = 0;
- const USTRING_TYPE *us1;
- const USTRING_TYPE *us2;
- size_t s1len;
- size_t s2len;
- int32_t *idx1arr;
- int32_t *idx2arr;
- unsigned char *rule1arr;
- unsigned char *rule2arr;
- size_t idx1max;
- size_t idx2max;
- size_t idx1cnt;
- size_t idx2cnt;
- size_t idx1now;
- size_t idx2now;
- size_t backw1_stop;
- size_t backw2_stop;
- size_t backw1;
- size_t backw2;
- int val1;
- int val2;
- int position;
- int seq1len;
- int seq2len;
- int use_malloc;
-
-#include WEIGHT_H
if (nrules == 0)
return STRCMP (s1, s2);
@@ -98,7 +493,6 @@ STRCOLL (s1, s2, l)
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_EXTRA,SUFFIX))].string;
indirect = (const int32_t *)
current->values[_NL_ITEM_INDEX (CONCAT(_NL_COLLATE_INDIRECT,SUFFIX))].string;
- use_malloc = 0;
assert (((uintptr_t) table) % __alignof__ (table[0]) == 0);
assert (((uintptr_t) weights) % __alignof__ (weights[0]) == 0);
@@ -106,18 +500,13 @@ STRCOLL (s1, s2, l)
assert (((uintptr_t) indirect) % __alignof__ (indirect[0]) == 0);
/* We need this a few times. */
- s1len = STRLEN (s1);
- s2len = STRLEN (s2);
+ size_t s1len = STRLEN (s1);
+ size_t s2len = STRLEN (s2);
/* Catch empty strings. */
- if (__builtin_expect (s1len == 0, 0) || __builtin_expect (s2len == 0, 0))
+ if (__glibc_unlikely (s1len == 0) || __glibc_unlikely (s2len == 0))
return (s1len != 0) - (s2len != 0);
- /* We need the elements of the strings as unsigned values since they
- are used as indeces. */
- us1 = (const USTRING_TYPE *) s1;
- us2 = (const USTRING_TYPE *) s2;
-
/* Perform the first pass over the string and while doing this find
and store the weights for each character. Since we want this to
be as fast as possible we are using `alloca' to store the temporary
@@ -127,411 +516,122 @@ STRCOLL (s1, s2, l)
Please note that the localedef programs makes sure that `position'
is not used at the first level. */
- if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
- {
- idx1arr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
- idx2arr = &idx1arr[s1len];
- rule1arr = (unsigned char *) &idx2arr[s2len];
- rule2arr = &rule1arr[s1len];
-
- if (idx1arr == NULL)
- /* No memory. Well, go with the stack then.
-
- XXX Once this implementation is stable we will handle this
- differently. Instead of precomputing the indeces we will
- do this in time. This means, though, that this happens for
- every pass again. */
- goto try_stack;
- use_malloc = 1;
- }
- else
- {
- try_stack:
- idx1arr = (int32_t *) alloca (s1len * sizeof (int32_t));
- idx2arr = (int32_t *) alloca (s2len * sizeof (int32_t));
- rule1arr = (unsigned char *) alloca (s1len);
- rule2arr = (unsigned char *) alloca (s2len);
- }
-
- idx1cnt = 0;
- idx2cnt = 0;
- idx1max = 0;
- idx2max = 0;
- idx1now = 0;
- idx2now = 0;
- backw1_stop = ~0ul;
- backw2_stop = ~0ul;
- backw1 = ~0ul;
- backw2 = ~0ul;
- seq1len = 0;
- seq2len = 0;
- position = rulesets[0] & sort_position;
- while (1)
- {
- val1 = 0;
- val2 = 0;
-
- /* Get the next non-IGNOREd element for string `s1'. */
- if (seq1len == 0)
- do
- {
- ++val1;
-
- if (backw1_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw1 == backw1_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx1cnt < idx1max)
- {
- idx1now = idx1cnt;
- backw1_stop = ~0ul;
- }
- else
- /* Nothing anymore. The backward sequence ended with
- the last sequence in the string. Note that seq1len
- is still zero. */
- break;
- }
- else
- idx1now = --backw1;
- }
- else
- {
- backw1_stop = idx1max;
-
- while (*us1 != L('\0'))
- {
- int32_t tmp = findidx (&us1, -1);
- rule1arr[idx1max] = tmp >> 24;
- idx1arr[idx1max] = tmp & 0xffffff;
- idx1cnt = idx1max++;
-
- if ((rulesets[rule1arr[idx1cnt] * nrules]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx1cnt;
- }
-
- if (backw1_stop >= idx1cnt)
- {
- /* No sequence at all or just one. */
- if (idx1cnt == idx1max || backw1_stop > idx1cnt)
- /* Note that seq1len is still zero. */
- break;
-
- backw1_stop = ~0ul;
- idx1now = idx1cnt;
- }
- else
- /* We pushed backward sequences. */
- idx1now = backw1 = idx1cnt - 1;
- }
- }
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
- /* And the same for string `s2'. */
- if (seq2len == 0)
- do
- {
- ++val2;
-
- if (backw2_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw2 == backw2_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx2cnt < idx2max)
- {
- idx2now = idx2cnt;
- backw2_stop = ~0ul;
- }
- else
- /* Nothing anymore. The backward sequence ended with
- the last sequence in the string. Note that seq2len
- is still zero. */
- break;
- }
- else
- idx2now = --backw2;
- }
- else
- {
- backw2_stop = idx2max;
-
- while (*us2 != L('\0'))
- {
- int32_t tmp = findidx (&us2, -1);
- rule2arr[idx2max] = tmp >> 24;
- idx2arr[idx2max] = tmp & 0xffffff;
- idx2cnt = idx2max++;
-
- if ((rulesets[rule2arr[idx2cnt] * nrules]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx2cnt;
- }
-
- if (backw2_stop >= idx2cnt)
- {
- /* No sequence at all or just one. */
- if (idx2cnt == idx2max || backw2_stop > idx2cnt)
- /* Note that seq1len is still zero. */
- break;
-
- backw2_stop = ~0ul;
- idx2now = idx2cnt;
- }
- else
- /* We pushed backward sequences. */
- idx2now = backw2 = idx2cnt - 1;
- }
- }
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
-
- /* See whether any or both strings are empty. */
- if (seq1len == 0 || seq2len == 0)
- {
- if (seq1len == seq2len)
- /* Both ended. So far so good, both strings are equal at the
- first level. */
- break;
-
- /* This means one string is shorter than the other. Find out
- which one and return an appropriate value. */
- result = seq1len == 0 ? -1 : 1;
- goto free_and_return;
- }
-
- /* Test for position if necessary. */
- if (position && val1 != val2)
- {
- result = val1 - val2;
- goto free_and_return;
- }
- /* Compare the two sequences. */
- do
- {
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
- {
- /* The sequences differ. */
- result = weights[idx1arr[idx1now]] - weights[idx2arr[idx2now]];
- goto free_and_return;
- }
+ coll_seq seq1, seq2;
+ bool use_malloc = false;
+ int result = 0;
- /* Increment the offsets. */
- ++idx1arr[idx1now];
- ++idx2arr[idx2now];
+ memset (&seq1, 0, sizeof (seq1));
+ seq2 = seq1;
- --seq1len;
- --seq2len;
- }
- while (seq1len > 0 && seq2len > 0);
+ size_t size_max = SIZE_MAX / (sizeof (int32_t) + 1);
- if (position && seq1len != seq2len)
+ if (MIN (s1len, s2len) > size_max
+ || MAX (s1len, s2len) > size_max - MIN (s1len, s2len))
+ {
+ /* If the strings are long enough to cause overflow in the size request,
+ then skip the allocation and proceed with the non-cached routines. */
+ }
+ else if (! __libc_use_alloca ((s1len + s2len) * (sizeof (int32_t) + 1)))
+ {
+ seq1.idxarr = (int32_t *) malloc ((s1len + s2len) * (sizeof (int32_t) + 1));
+
+ /* If we failed to allocate memory, we leave everything as NULL so that
+ we use the nocache version of traversal and comparison functions. */
+ if (seq1.idxarr != NULL)
{
- result = seq1len - seq2len;
- goto free_and_return;
+ seq2.idxarr = &seq1.idxarr[s1len];
+ seq1.rulearr = (unsigned char *) &seq2.idxarr[s2len];
+ seq2.rulearr = &seq1.rulearr[s1len];
+ use_malloc = true;
}
}
+ else
+ {
+ seq1.idxarr = (int32_t *) alloca (s1len * sizeof (int32_t));
+ seq2.idxarr = (int32_t *) alloca (s2len * sizeof (int32_t));
+ seq1.rulearr = (unsigned char *) alloca (s1len);
+ seq2.rulearr = (unsigned char *) alloca (s2len);
+ }
+
+ int rule = 0;
- /* Now the remaining passes over the weights. We now use the
- indeces we found before. */
- for (pass = 1; pass < nrules; ++pass)
+ /* Cache values in the first pass and if needed, use them in subsequent
+ passes. */
+ for (int pass = 0; pass < nrules; ++pass)
{
+ seq1.idxcnt = 0;
+ seq1.idx = 0;
+ seq2.idx = 0;
+ seq1.backw_stop = ~0ul;
+ seq1.backw = ~0ul;
+ seq2.idxcnt = 0;
+ seq2.backw_stop = ~0ul;
+ seq2.backw = ~0ul;
+
+ /* We need the elements of the strings as unsigned values since they
+ are used as indices. */
+ seq1.us = (const USTRING_TYPE *) s1;
+ seq2.us = (const USTRING_TYPE *) s2;
+
/* We assume that if a rule has defined `position' in one section
this is true for all of them. */
- idx1cnt = 0;
- idx2cnt = 0;
- backw1_stop = ~0ul;
- backw2_stop = ~0ul;
- backw1 = ~0ul;
- backw2 = ~0ul;
- position = rulesets[rule1arr[0] * nrules + pass] & sort_position;
+ int position = rulesets[rule * nrules + pass] & sort_position;
while (1)
{
- val1 = 0;
- val2 = 0;
-
- /* Get the next non-IGNOREd element for string `s1'. */
- if (seq1len == 0)
- do
- {
- ++val1;
-
- if (backw1_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw1 == backw1_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx1cnt < idx1max)
- {
- idx1now = idx1cnt;
- backw1_stop = ~0ul;
- }
- else
- {
- /* Nothing anymore. The backward sequence
- ended with the last sequence in the string. */
- idx1now = ~0ul;
- break;
- }
- }
- else
- idx1now = --backw1;
- }
- else
- {
- backw1_stop = idx1cnt;
-
- while (idx1cnt < idx1max)
- {
- if ((rulesets[rule1arr[idx1cnt] * nrules + pass]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx1cnt;
- }
-
- if (backw1_stop == idx1cnt)
- {
- /* No sequence at all or just one. */
- if (idx1cnt == idx1max)
- /* Note that seq1len is still zero. */
- break;
-
- backw1_stop = ~0ul;
- idx1now = idx1cnt++;
- }
- else
- /* We pushed backward sequences. */
- idx1now = backw1 = idx1cnt - 1;
- }
- }
- while ((seq1len = weights[idx1arr[idx1now]++]) == 0);
-
- /* And the same for string `s2'. */
- if (seq2len == 0)
- do
- {
- ++val2;
-
- if (backw2_stop != ~0ul)
- {
- /* The is something pushed. */
- if (backw2 == backw2_stop)
- {
- /* The last pushed character was handled. Continue
- with forward characters. */
- if (idx2cnt < idx2max)
- {
- idx2now = idx2cnt;
- backw2_stop = ~0ul;
- }
- else
- {
- /* Nothing anymore. The backward sequence
- ended with the last sequence in the string. */
- idx2now = ~0ul;
- break;
- }
- }
- else
- idx2now = --backw2;
- }
- else
- {
- backw2_stop = idx2cnt;
-
- while (idx2cnt < idx2max)
- {
- if ((rulesets[rule2arr[idx2cnt] * nrules + pass]
- & sort_backward) == 0)
- /* No more backward characters to push. */
- break;
- ++idx2cnt;
- }
-
- if (backw2_stop == idx2cnt)
- {
- /* No sequence at all or just one. */
- if (idx2cnt == idx2max)
- /* Note that seq2len is still zero. */
- break;
-
- backw2_stop = ~0ul;
- idx2now = idx2cnt++;
- }
- else
- /* We pushed backward sequences. */
- idx2now = backw2 = idx2cnt - 1;
- }
- }
- while ((seq2len = weights[idx2arr[idx2now]++]) == 0);
+ if (__glibc_unlikely (seq1.idxarr == NULL))
+ {
+ get_next_seq_nocache (&seq1, nrules, rulesets, weights, table,
+ extra, indirect, pass);
+ get_next_seq_nocache (&seq2, nrules, rulesets, weights, table,
+ extra, indirect, pass);
+ }
+ else if (pass == 0)
+ {
+ get_next_seq (&seq1, nrules, rulesets, weights, table, extra,
+ indirect);
+ get_next_seq (&seq2, nrules, rulesets, weights, table, extra,
+ indirect);
+ }
+ else
+ {
+ get_next_seq_cached (&seq1, nrules, pass, rulesets, weights);
+ get_next_seq_cached (&seq2, nrules, pass, rulesets, weights);
+ }
/* See whether any or both strings are empty. */
- if (seq1len == 0 || seq2len == 0)
+ if (seq1.len == 0 || seq2.len == 0)
{
- if (seq1len == seq2len)
+ if (seq1.len == seq2.len)
/* Both ended. So far so good, both strings are equal
at this level. */
break;
/* This means one string is shorter than the other. Find out
which one and return an appropriate value. */
- result = seq1len == 0 ? -1 : 1;
+ result = seq1.len == 0 ? -1 : 1;
goto free_and_return;
}
- /* Test for position if necessary. */
- if (position && val1 != val2)
- {
- result = val1 - val2;
- goto free_and_return;
- }
-
- /* Compare the two sequences. */
- do
- {
- if (weights[idx1arr[idx1now]] != weights[idx2arr[idx2now]])
- {
- /* The sequences differ. */
- result = (weights[idx1arr[idx1now]]
- - weights[idx2arr[idx2now]]);
- goto free_and_return;
- }
-
- /* Increment the offsets. */
- ++idx1arr[idx1now];
- ++idx2arr[idx2now];
-
- --seq1len;
- --seq2len;
- }
- while (seq1len > 0 && seq2len > 0);
-
- if (position && seq1len != seq2len)
- {
- result = seq1len - seq2len;
- goto free_and_return;
- }
+ if (__glibc_unlikely (seq1.idxarr == NULL))
+ result = do_compare_nocache (&seq1, &seq2, position, weights);
+ else
+ result = do_compare (&seq1, &seq2, position, weights);
+ if (result != 0)
+ goto free_and_return;
}
+
+ if (__glibc_likely (seq1.rulearr != NULL))
+ rule = seq1.rulearr[0];
+ else
+ rule = seq1.rule;
}
/* Free the memory if needed. */
free_and_return:
if (use_malloc)
- free (idx1arr);
+ free (seq1.idxarr);
return result;
}
Index: glibc-2.18/string/tst-strcoll-overflow.c
===================================================================
--- /dev/null
+++ glibc-2.18/string/tst-strcoll-overflow.c
@@ -0,0 +1,61 @@
+/* Copyright (C) 2013 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <locale.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+
+/* Verify that strcoll does not crash for large strings for which it cannot
+ cache weight lookup results. The size is large enough to cause integer
+ overflows on 32-bit as well as buffer overflows on 64-bit. The test should
+ work reasonably reliably when overcommit is disabled, but it obviously
+ depends on how much memory the system has. There's a limitation to this
+ test in that it does not run to completion. Actually collating such a
+ large string can take days and we can't have xcheck running that long. For
+ that reason, we run the test for about 5 minutes and then assume that
+ everything is fine if there are no crashes. */
+#define SIZE 0x40000000ul
+
+int
+do_test (void)
+{
+ if (setlocale (LC_COLLATE, "en_GB.UTF-8") == NULL)
+ {
+ puts ("setlocale failed, cannot test for overflow");
+ return 0;
+ }
+
+ char *p = malloc (SIZE);
+
+ if (p == NULL)
+ {
+ puts ("could not allocate memory");
+ return 1;
+ }
+
+ memset (p, 'x', SIZE - 1);
+ p[SIZE - 1] = 0;
+ printf ("%d\n", strcoll (p, p));
+ return 0;
+}
+
+#define TIMEOUT 300
+#define EXPECTED_SIGNAL SIGALRM
+#define TEST_FUNCTION do_test ()
+#include "../test-skeleton.c"
View Git Blame Copy Permalink