# [PATCH] name-hash: fix buffer overrun 11 messages from 2017-03-31 to 2017-04-02. Participants: git@jeffhostetler.com, Junio C Hamano, Johannes Schindelin, Jakub Narębski. Thread: https://gitlist.dev/t/45570 ## git@jeffhostetler.com, 2017-03-31 17:32 Subject: [PATCH] name-hash: fix buffer overrun Message-ID: <20170331173214.47514-1-git@jeffhostetler.com> URL: https://gitlist.dev/e/20170331173214.47514-1-git%40jeffhostetler.com ``` From: Jeff Hostetler Fix buffer overrun in handle_range_dir() when the final entry in the index was the only file in the last directory, such as "a/b/foo.txt". The look ahead (k_start + 1) was invalid since (k_start + 1) == k_end. This bug was introduced by Jeff in "jh/memihash-opt" which was recently merged into master. Kevin Willford (1): name-hash: fix buffer overrun name-hash.c | 4 +++- t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ 2 files changed, 22 insertions(+), 1 deletion(-) create mode 100644 t/t3008-ls-files-lazy-init-name-hash.sh -- 2.9.3 ``` ## git@jeffhostetler.com, 2017-03-31 17:32 Subject: [PATCH] name-hash: fix buffer overrun Message-ID: <20170331173214.47514-2-git@jeffhostetler.com> URL: https://gitlist.dev/e/20170331173214.47514-2-git%40jeffhostetler.com In-Reply-To: <20170331173214.47514-1-git@jeffhostetler.com> ``` From: Kevin Willford Add check for the end of the entries for the thread partition. Add test for lazy init name hash with specific directory structure The lazy init hash name was causing a buffer overflow when the last entry in the index was multiple folder deep with parent folders that did not have any files in them. This adds a test for the boundary condition of the thread partitions with the folder structure that was triggering the buffer overflow. The fix was to check if it is the last entry for the thread partition in the handle_range_dir and not try to use the next entry in the cache. Signed-off-by: Kevin Willford Signed-off-by: Johannes Schindelin Signed-off-by: Jeff Hostetler --- name-hash.c | 4 +++- t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ 2 files changed, 22 insertions(+), 1 deletion(-) create mode 100644 t/t3008-ls-files-lazy-init-name-hash.sh diff --git a/name-hash.c b/name-hash.c index cac313c..39309ef 100644 --- a/name-hash.c +++ b/name-hash.c @@ -342,7 +342,9 @@ static int handle_range_dir( * Scan forward in the index array for index entries having the same * path prefix (that are also in this directory). */ - if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0) + if (k_start + 1 >= k_end) + k = k_end; + else if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0) k = k_start + 1; else if (strncmp(istate->cache[k_end - 1]->name, prefix->buf, prefix->len) == 0) k = k_end; diff --git a/t/t3008-ls-files-lazy-init-name-hash.sh b/t/t3008-ls-files-lazy-init-name-hash.sh new file mode 100644 index 0000000..e46a11b --- /dev/null +++ b/t/t3008-ls-files-lazy-init-name-hash.sh @@ -0,0 +1,19 @@ +#!/bin/sh + +test_description='Test the lazy init name hash with various folder structures' + +. ./test-lib.sh + +test_expect_success 'no buffer overflow in lazy_init_name_hash' ' + ( + test_seq 2000 | sed "s/^/a_/" + echo b/b/b + test_seq 2000 | sed "s/^/c_/" + test_seq 50 | sed "s/^/d_/" | tr "\n" "/"; echo d + ) | + sed "s/^/100644 $EMPTY_BLOB\t/" | + git update-index --index-info && + test-lazy-init-name-hash -m +' + +test_done -- 2.9.3 ``` ## Junio C Hamano, 2017-03-31 19:35 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/xmqqo9whxmrq.fsf%40gitster.mtv.corp.google.com In-Reply-To: <20170331173214.47514-2-git@jeffhostetler.com> ``` git@jeffhostetler.com writes: > From: Kevin Willford > > Add check for the end of the entries for the thread partition. > Add test for lazy init name hash with specific directory structure > > The lazy init hash name was causing a buffer overflow when the last > entry in the index was multiple folder deep with parent folders that > did not have any files in them. > > This adds a test for the boundary condition of the thread partitions > with the folder structure that was triggering the buffer overflow. > > The fix was to check if it is the last entry for the thread partition > in the handle_range_dir and not try to use the next entry in the cache. > > Signed-off-by: Kevin Willford > Signed-off-by: Johannes Schindelin > Signed-off-by: Jeff Hostetler > > --- Will queue with ... > name-hash.c | 4 +++- > t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ > 2 files changed, 22 insertions(+), 1 deletion(-) > create mode 100644 t/t3008-ls-files-lazy-init-name-hash.sh ... this thing fixed by "chmod +x" (otherwise the tests won't start). Thanks. ``` ## Junio C Hamano, 2017-03-31 21:18 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/xmqqy3vlw3f9.fsf%40gitster.mtv.corp.google.com In-Reply-To: ``` Junio C Hamano writes: > Will queue with ... > >> name-hash.c | 4 +++- >> t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ >> 2 files changed, 22 insertions(+), 1 deletion(-) >> create mode 100644 t/t3008-ls-files-lazy-init-name-hash.sh > > ... this thing fixed by "chmod +x" (otherwise the tests won't start). > > Thanks. Also, https://travis-ci.org/git/git/jobs/217303927 seems to say that MacOS is not happy with this change. ``` ## Junio C Hamano, 2017-03-31 23:18 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/CAPc5daWU5XDVNiYk_pTFk_qziuDr6W2XDHXXH-0oR49_KiCUYA%40mail.gmail.com In-Reply-To: ``` On Fri, Mar 31, 2017 at 2:18 PM, Junio C Hamano wrote: > Junio C Hamano writes: > >> Will queue with ... >> >>> name-hash.c | 4 +++- >>> t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ >>> 2 files changed, 22 insertions(+), 1 deletion(-) >>> create mode 100644 t/t3008-ls-files-lazy-init-name-hash.sh >> >> ... this thing fixed by "chmod +x" (otherwise the tests won't start). >> >> Thanks. > > Also, https://travis-ci.org/git/git/jobs/217303927 seems to say that > MacOS is not happy with this change. Ah, of course. Avoid GNUism to spell HT as "\t" in a sed script. ``` ## Junio C Hamano, 2017-04-01 04:12 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/xmqqh928wyu7.fsf%40gitster.mtv.corp.google.com In-Reply-To: ``` Junio C Hamano writes: > Ah, of course. Avoid GNUism to spell HT as "\t" in a sed script. Here is what I replaced the original patch with. Let's see how well it fares with Travis tonight. -- >8 -- From: Kevin Willford Date: Fri, 31 Mar 2017 17:32:14 +0000 Subject: [PATCH] name-hash: fix buffer overrun Add check for the end of the entries for the thread partition. Add test for lazy init name hash with specific directory structure The lazy init hash name was causing a buffer overflow when the last entry in the index was multiple folder deep with parent folders that did not have any files in them. This adds a test for the boundary condition of the thread partitions with the folder structure that was triggering the buffer overflow. The fix was to check if it is the last entry for the thread partition in the handle_range_dir and not try to use the next entry in the cache. Signed-off-by: Kevin Willford Signed-off-by: Johannes Schindelin Signed-off-by: Jeff Hostetler Signed-off-by: Junio C Hamano --- name-hash.c | 4 +++- t/t3008-ls-files-lazy-init-name-hash.sh | 19 +++++++++++++++++++ 2 files changed, 22 insertions(+), 1 deletion(-) create mode 100755 t/t3008-ls-files-lazy-init-name-hash.sh diff --git a/name-hash.c b/name-hash.c index cac313c78d..39309efb7f 100644 --- a/name-hash.c +++ b/name-hash.c @@ -342,7 +342,9 @@ static int handle_range_dir( * Scan forward in the index array for index entries having the same * path prefix (that are also in this directory). */ - if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0) + if (k_start + 1 >= k_end) + k = k_end; + else if (strncmp(istate->cache[k_start + 1]->name, prefix->buf, prefix->len) > 0) k = k_start + 1; else if (strncmp(istate->cache[k_end - 1]->name, prefix->buf, prefix->len) == 0) k = k_end; diff --git a/t/t3008-ls-files-lazy-init-name-hash.sh b/t/t3008-ls-files-lazy-init-name-hash.sh new file mode 100755 index 0000000000..971975bff4 --- /dev/null +++ b/t/t3008-ls-files-lazy-init-name-hash.sh @@ -0,0 +1,19 @@ +#!/bin/sh + +test_description='Test the lazy init name hash with various folder structures' + +. ./test-lib.sh + +test_expect_success 'no buffer overflow in lazy_init_name_hash' ' + ( + test_seq 2000 | sed "s/^/a_/" + echo b/b/b + test_seq 2000 | sed "s/^/c_/" + test_seq 50 | sed "s/^/d_/" | tr "\n" "/"; echo d + ) | + sed -e "s/^/100644 $EMPTY_BLOB /" | + git update-index --index-info && + test-lazy-init-name-hash -m +' + +test_done -- 2.12.2-752-g2215051a9e ``` ## Junio C Hamano, 2017-04-01 04:14 Subject: [PATCH] contrib/git-resurrect.sh: do not write \t for HT in sed scripts Message-ID: URL: https://gitlist.dev/e/xmqqd1cwwyr5.fsf_-_%40gitster.mtv.corp.google.com In-Reply-To: ``` Just like we did in 0d1d6e50 ("t/t7003: replace \t with literal tab in sed expression", 2010-08-12), avoid writing "\t" for HT in sed scripts, which is not portable. Signed-off-by: Junio C Hamano --- * This should hopefully kill the last instances of the same issue to make sure people do not copy-and-paste this unportable construct. contrib/git-resurrect.sh | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/contrib/git-resurrect.sh b/contrib/git-resurrect.sh index c364dda696..3b78ffd079 100755 --- a/contrib/git-resurrect.sh +++ b/contrib/git-resurrect.sh @@ -24,13 +24,13 @@ n,dry-run don't recreate the branch" . git-sh-setup search_reflog () { - sed -ne 's~^\([^ ]*\) .*\tcheckout: moving from '"$1"' .*~\1~p' \ + sed -ne 's~^\([^ ]*\) .* checkout: moving from '"$1"' .*~\1~p' \ < "$GIT_DIR"/logs/HEAD } search_reflog_merges () { git rev-parse $( - sed -ne 's~^[^ ]* \([^ ]*\) .*\tmerge '"$1"':.*~\1^2~p' \ + sed -ne 's~^[^ ]* \([^ ]*\) .* merge '"$1"':.*~\1^2~p' \ < "$GIT_DIR"/logs/HEAD ) } -- 2.12.2-752-g2215051a9e ``` ## Johannes Schindelin, 2017-04-01 12:12 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/alpine.DEB.2.20.1704011411460.3742%40virtualbox In-Reply-To: ``` Hi Junio, On Fri, 31 Mar 2017, Junio C Hamano wrote: > Junio C Hamano writes: > > > Ah, of course. Avoid GNUism to spell HT as "\t" in a sed script. Sorry about that, I suggested this snippet to Kevin, it is my fault for not remembering BSD sed's idiosynchracies. Ciao, Johannes P.S.: enjoy your time off! ``` ## Junio C Hamano, 2017-04-01 18:16 Subject: Re: [PATCH] name-hash: fix buffer overrun Message-ID: URL: https://gitlist.dev/e/xmqqk274uh6w.fsf%40gitster.mtv.corp.google.com In-Reply-To: ``` Johannes Schindelin writes: > On Fri, 31 Mar 2017, Junio C Hamano wrote: > >> Junio C Hamano writes: >> >> > Ah, of course. Avoid GNUism to spell HT as "\t" in a sed script. > > Sorry about that, I suggested this snippet to Kevin, it is my fault for > not remembering BSD sed's idiosynchracies. Heh, don't fret about that. We all make mistakes and that is why we review on the list so that patches get exposure to more sets of eyes. We may be interested to learn from our common mistakes, but for that purpose, "whose fault is it?" is far less interesting than "how it came about?", i.e. what made that mistake common and if/how we can help people avoid it (removing cuttable-and-pastable bad examples is one way to do so). Thanks. ``` ## Jakub Narębski, 2017-04-02 09:58 Subject: Re: [PATCH] contrib/git-resurrect.sh: do not write \t for HT in sed scripts Message-ID: <7028b4eb-aa2b-aaf6-764d-cffab24a9968@gmail.com> URL: https://gitlist.dev/e/7028b4eb-aa2b-aaf6-764d-cffab24a9968%40gmail.com In-Reply-To: ``` W dniu 01.04.2017 o 06:14, Junio C Hamano pisze: > Just like we did in 0d1d6e50 ("t/t7003: replace \t with literal tab > in sed expression", 2010-08-12), avoid writing "\t" for HT in sed > scripts, which is not portable. I guess it is not possible to use HT variable (similar to LF that we have in t/test-lib.sh, and which is defined by some scripts) set to literal tab HT=" " wouldn't work there, is it? Using this variable would make literal tab character more visible. -- Jakub Narębski ``` ## Junio C Hamano, 2017-04-02 16:46 Subject: Re: [PATCH] contrib/git-resurrect.sh: do not write \t for HT in sed scripts Message-ID: URL: https://gitlist.dev/e/xmqqzifysqol.fsf%40gitster.mtv.corp.google.com In-Reply-To: <7028b4eb-aa2b-aaf6-764d-cffab24a9968@gmail.com> ``` Jakub Narębski writes: > W dniu 01.04.2017 o 06:14, Junio C Hamano pisze: > >> Just like we did in 0d1d6e50 ("t/t7003: replace \t with literal tab >> in sed expression", 2010-08-12), avoid writing "\t" for HT in sed >> scripts, which is not portable. > > I guess it is not possible to use HT variable (similar to LF that we > have in t/test-lib.sh, and which is defined by some scripts) set to > literal tab > > HT=" " > > wouldn't work there, is it? > > Using this variable would make literal tab character more visible. Patches welcome. ```