From da9ec461e22915ccabb06785bf39ec34577ada12 Mon Sep 17 00:00:00 2001 From: Hugo van der Sanden Date: Sat, 11 Apr 2020 14:10:24 +0100 Subject: [PATCH] study_chunk: avoid mutating regexp program within GOSUB gh16947 and gh17743: studying GOSUB may restudy in an inner call (via a mix of recursion and enframing) something that an outer call is in the middle of looking at. Let the outer frame deal with it. (CVE-2020-12723) (cherry picked from commit c4033e740bd18d9fbe3456a9db2ec2053cdc5271) Upstream-Status: Backport [https://github.com/perl/perl5/commit/66bbb51b93253a3f87d11c2695cfb7bdb782184a] CVE: CVE-2020-12723 Signed-off-by: Chee Yang Lee --- embed.fnc | 2 +- embed.h | 2 +- proto.h | 2 +- regcomp.c | 54 +++++++++++++++++++++++++++++++++++------------------- t/re/pat.t | 26 +++++++++++++++++++++++++- 5 files changed, 63 insertions(+), 23 deletions(-) diff --git a/embed.fnc b/embed.fnc index f45c127..eff4a50 100644 --- a/embed.fnc +++ b/embed.fnc @@ -2480,7 +2480,7 @@ Es |SSize_t|study_chunk |NN RExC_state_t *pRExC_state \ |NULLOK struct scan_data_t *data \ |I32 stopparen|U32 recursed_depth \ |NULLOK regnode_ssc *and_withp \ - |U32 flags|U32 depth + |U32 flags|U32 depth|bool was_mutate_ok Es |void |rck_elide_nothing|NN regnode *node EsR |SV * |get_ANYOFM_contents|NN const regnode * n EsRn |U32 |add_data |NN RExC_state_t* const pRExC_state \ diff --git a/embed.h b/embed.h index 356a8b9..5346ec5 100644 --- a/embed.h +++ b/embed.h @@ -1239,7 +1239,7 @@ #define ssc_is_cp_posixl_init S_ssc_is_cp_posixl_init #define ssc_or(a,b,c) S_ssc_or(aTHX_ a,b,c) #define ssc_union(a,b,c) S_ssc_union(aTHX_ a,b,c) -#define study_chunk(a,b,c,d,e,f,g,h,i,j,k) S_study_chunk(aTHX_ a,b,c,d,e,f,g,h,i,j,k) +#define study_chunk(a,b,c,d,e,f,g,h,i,j,k,l) S_study_chunk(aTHX_ a,b,c,d,e,f,g,h,i,j,k,l) # endif # if defined(PERL_IN_REGCOMP_C) || defined (PERL_IN_DUMP_C) #define _invlist_dump(a,b,c,d) Perl__invlist_dump(aTHX_ a,b,c,d) diff --git a/proto.h b/proto.h index 91530b1..1bda01f 100644 --- a/proto.h +++ b/proto.h @@ -5655,7 +5655,7 @@ PERL_STATIC_INLINE void S_ssc_union(pTHX_ regnode_ssc *ssc, SV* const invlist, c #define PERL_ARGS_ASSERT_SSC_UNION \ assert(ssc); assert(invlist) #endif -STATIC SSize_t S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, SSize_t *minlenp, SSize_t *deltap, regnode *last, struct scan_data_t *data, I32 stopparen, U32 recursed_depth, regnode_ssc *and_withp, U32 flags, U32 depth); +STATIC SSize_t S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, SSize_t *minlenp, SSize_t *deltap, regnode *last, struct scan_data_t *data, I32 stopparen, U32 recursed_depth, regnode_ssc *and_withp, U32 flags, U32 depth, bool was_mutate_ok); #define PERL_ARGS_ASSERT_STUDY_CHUNK \ assert(pRExC_state); assert(scanp); assert(minlenp); assert(deltap); assert(last) #endif diff --git a/regcomp.c b/regcomp.c index 5a9adee..8d7df1f 100644 --- a/regcomp.c +++ b/regcomp.c @@ -106,6 +106,7 @@ typedef struct scan_frame { regnode *next_regnode; /* next node to process when last is reached */ U32 prev_recursed_depth; I32 stopparen; /* what stopparen do we use */ + bool in_gosub; /* this or an outer frame is for GOSUB */ struct scan_frame *this_prev_frame; /* this previous frame */ struct scan_frame *prev_frame; /* previous frame */ @@ -4466,7 +4467,7 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, I32 stopparen, U32 recursed_depth, regnode_ssc *and_withp, - U32 flags, U32 depth) + U32 flags, U32 depth, bool was_mutate_ok) /* scanp: Start here (read-write). */ /* deltap: Write maxlen-minlen here. */ /* last: Stop before this one. */ @@ -4545,6 +4546,10 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, node length to get a real minimum (because the folded version may be shorter) */ bool unfolded_multi_char = FALSE; + /* avoid mutating ops if we are anywhere within the recursed or + * enframed handling for a GOSUB: the outermost level will handle it. + */ + bool mutate_ok = was_mutate_ok && !(frame && frame->in_gosub); /* Peephole optimizer: */ DEBUG_STUDYDATA("Peep", data, depth, is_inf); DEBUG_PEEP("Peep", scan, depth, flags); @@ -4555,7 +4560,8 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, * parsing code, as each (?:..) is handled by a different invocation of * reg() -- Yves */ - JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0); + if (mutate_ok) + JOIN_EXACT(scan,&min_subtract, &unfolded_multi_char, 0); /* Follow the next-chain of the current node and optimize away all the NOTHINGs from it. @@ -4587,7 +4593,7 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, /* DEFINEP study_chunk() recursion */ (void)study_chunk(pRExC_state, &scan, &minlen, &deltanext, next, &data_fake, stopparen, - recursed_depth, NULL, f, depth+1); + recursed_depth, NULL, f, depth+1, mutate_ok); scan = next; } else @@ -4655,7 +4661,8 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, /* recurse study_chunk() for each BRANCH in an alternation */ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, next, &data_fake, stopparen, - recursed_depth, NULL, f, depth+1); + recursed_depth, NULL, f, depth+1, + mutate_ok); if (min1 > minnext) min1 = minnext; @@ -4722,9 +4729,10 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, } } - if (PERL_ENABLE_TRIE_OPTIMISATION && - OP( startbranch ) == BRANCH ) - { + if (PERL_ENABLE_TRIE_OPTIMISATION + && OP(startbranch) == BRANCH + && mutate_ok + ) { /* demq. Assuming this was/is a branch we are dealing with: 'scan' @@ -5179,6 +5187,9 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, newframe->stopparen = stopparen; newframe->prev_recursed_depth = recursed_depth; newframe->this_prev_frame= frame; + newframe->in_gosub = ( + (frame && frame->in_gosub) || OP(scan) == GOSUB + ); DEBUG_STUDYDATA("frame-new", data, depth, is_inf); DEBUG_PEEP("fnew", scan, depth, flags); @@ -5336,7 +5347,7 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, /* This temporary node can now be turned into EXACTFU, and * must, as regexec.c doesn't handle it */ - if (OP(next) == EXACTFU_S_EDGE) { + if (OP(next) == EXACTFU_S_EDGE && mutate_ok) { OP(next) = EXACTFU; } @@ -5344,8 +5355,9 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, && isALPHA_A(* STRING(next)) && ( OP(next) == EXACTFAA || ( OP(next) == EXACTFU - && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(* STRING(next))))) - { + && ! HAS_NONLATIN1_SIMPLE_FOLD_CLOSURE(* STRING(next)))) + && mutate_ok + ) { /* These differ in just one bit */ U8 mask = ~ ('A' ^ 'a'); @@ -5432,7 +5444,7 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, (mincount == 0 ? (f & ~SCF_DO_SUBSTR) : f) - ,depth+1); + , depth+1, mutate_ok); if (flags & SCF_DO_STCLASS) data->start_class = oclass; @@ -5498,7 +5510,9 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, if ( OP(oscan) == CURLYX && data && data->flags & SF_IN_PAR && !(data->flags & SF_HAS_EVAL) - && !deltanext && minnext == 1 ) { + && !deltanext && minnext == 1 + && mutate_ok + ) { /* Try to optimize to CURLYN. */ regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; regnode * const nxt1 = nxt; @@ -5548,10 +5562,10 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, && !(data->flags & SF_HAS_EVAL) && !deltanext /* atom is fixed width */ && minnext != 0 /* CURLYM can't handle zero width */ - /* Nor characters whose fold at run-time may be * multi-character */ && ! (RExC_seen & REG_UNFOLDED_MULTI_SEEN) + && mutate_ok ) { /* XXXX How to optimize if data == 0? */ /* Optimize to a simpler form. */ @@ -5604,7 +5618,7 @@ S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, /* recurse study_chunk() on optimised CURLYX => CURLYM */ study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt, NULL, stopparen, recursed_depth, NULL, 0, - depth+1); + depth+1, mutate_ok); } else oscan->flags = 0; @@ -6009,7 +6023,8 @@ Perl_re_printf( aTHX_ "LHS=%" UVuf " RHS=%" UVuf "\n", /* recurse study_chunk() for lookahead body */ minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext, last, &data_fake, stopparen, - recursed_depth, NULL, f, depth+1); + recursed_depth, NULL, f, depth+1, + mutate_ok); if (scan->flags) { if ( deltanext < 0 || deltanext > (I32) U8_MAX @@ -6114,7 +6129,7 @@ Perl_re_printf( aTHX_ "LHS=%" UVuf " RHS=%" UVuf "\n", *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext, last, &data_fake, stopparen, recursed_depth, NULL, - f, depth+1); + f, depth+1, mutate_ok); if (scan->flags) { assert(0); /* This code has never been tested since this is normally not compiled */ @@ -6282,7 +6297,8 @@ Perl_re_printf( aTHX_ "LHS=%" UVuf " RHS=%" UVuf "\n", /* optimise study_chunk() for TRIE */ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext, (regnode *)nextbranch, &data_fake, - stopparen, recursed_depth, NULL, f, depth+1); + stopparen, recursed_depth, NULL, f, depth+1, + mutate_ok); } if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH) nextbranch= regnext((regnode*)nextbranch); @@ -8075,7 +8091,7 @@ Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count, &data, -1, 0, NULL, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag | (restudied ? SCF_TRIE_DOING_RESTUDY : 0), - 0); + 0, TRUE); CHECK_RESTUDY_GOTO_butfirst(LEAVE_with_name("study_chunk")); @@ -8204,7 +8220,7 @@ Perl_re_op_compile(pTHX_ SV ** const patternp, int pat_count, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS|(restudied ? SCF_TRIE_DOING_RESTUDY : 0), - 0); + 0, TRUE); CHECK_RESTUDY_GOTO_butfirst(NOOP); diff --git a/t/re/pat.t b/t/re/pat.t index 6a868f4..2869b58 100644 --- a/t/re/pat.t +++ b/t/re/pat.t @@ -25,7 +25,7 @@ BEGIN { skip_all('no re module') unless defined &DynaLoader::boot_DynaLoader; skip_all_without_unicode_tables(); -plan tests => 864; # Update this when adding/deleting tests. +plan tests => 868; # Update this when adding/deleting tests. run_tests() unless caller; @@ -2115,6 +2115,30 @@ x{0c!}\;\;îçÿ /0f/!F/;îçÿù\Qxÿÿÿÿù`x{0c!};ù\Q like(runperl(prog => "$s", stderr => 1), qr/Unmatched \(/); } + # gh16947: test regexp corruption (GOSUB) + { + fresh_perl_is(q{ + 'xy' =~ /x(?0)|x(?|y|y)/ && print 'ok' + }, 'ok', {}, 'gh16947: test regexp corruption (GOSUB)'); + } + # gh16947: test fix doesn't break SUSPEND + { + fresh_perl_is(q{ 'sx' =~ m{ss++}i; print 'ok' }, + 'ok', {}, "gh16947: test fix doesn't break SUSPEND"); + } + + # gh17743: more regexp corruption via GOSUB + { + fresh_perl_is(q{ + "0" =~ /((0(?0)|000(?|0000|0000)(?0))|)/; print "ok" + }, 'ok', {}, 'gh17743: test regexp corruption (1)'); + + fresh_perl_is(q{ + "000000000000" =~ /(0(())(0((?0)())|000(?|\x{ef}\x{bf}\x{bd}|\x{ef}\x{bf}\x{bd}))|)/; + print "ok" + }, 'ok', {}, 'gh17743: test regexp corruption (2)'); + } + } # End of sub run_tests 1;