Skip to contentSkip to conversation

name-rev: avoid cutoff timestamp underflow

PATCH In review SZEDER Gábor 4 participants 11 messages Sep 22, 2019 – Sep 24, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 name-rev: avoid cutoff timestamp underflow

When 'git name-rev' is invoked with commit-ish parameters, it tries to save some work, and doesn't visit commits older than the committer date of the oldest given commit minus a one day worth of slop. Since our 'timestamp_t' is an unsigned type, this leads to a timestamp underflow when the committer date of the oldest given commit is within a day of the UNIX epoch. As a result the cutoff timestamp ends up far-far in the future, and 'git name-rev' doesn't visit any commits, and names each given commit as 'undefined'.

Check whether substacting the slop from the oldest committer date would lead to an underflow, and use a 0 as cutoff in that case. This way it will handle commits shortly after the epoch even if we were to switch to a signed 'timestamp_t' (but then we'll have to worry about signed underflow for very old commits).

Note that the type of the cutoff timestamp variable used to be signed before 5589e87fd8 (name-rev: change a "long" variable to timestamp_t, 2017-05-20). The behavior was still the same even back then, but the underflow didn't happen when substracting the slop from the oldest committer date, but when comparing the signed cutoff timestamp with unsigned committer dates in name_rev(). IOW, this underflow bug is as old as 'git name-rev' itself.

---

This patch adds a test at the end of 't6120-describe.sh', so it will conflict with my non-recursive name-rev patch series, which adds a test there as well, but the conflict should be wasy to resolve.

https://public-inbox.org/git/[email protected]/

 builtin/name-rev.c  | 15 ++++++++++++---
 t/t6120-describe.sh | 15 +++++++++++++++
 2 files changed, 27 insertions(+), 3 deletions(-)
builtin/name-rev.c +12 -3
index c785fe16ba..a4d8d312ab 100644
@@ -9,7 +9,11 @@
 #include "sha1-lookup.h"
 #include "commit-slab.h"
 
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 
 typedef struct rev_name {
 	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 		add_object_array(object, *argv, &revs);
 	}
 
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)
+			cutoff = cutoff - CUTOFF_DATE_SLOP;
+		else
+			cutoff = 0;
+	}
 	for_each_ref(name_ref, &data);
 
 	if (transform_stdin) {
t/t6120-describe.sh +15 -0
index 2b883d8174..965e633c32 100755
@@ -424,4 +424,19 @@ test_expect_success 'describe complains about missing object' '
 	test_must_fail git describe $ZERO_OID
 '
 
+test_expect_success 'name-rev a rev shortly after epoch' '
+	test_when_finished "git checkout master" &&
+
+	git checkout --orphan no-timestamp-underflow &&
+	# Any date closer to epoch than the CUTOFF_DATE_SLOP constant
+	# in builtin/name-rev.c.
+	GIT_COMMITTER_DATE="@1234 +0000" \
+	git commit -m "committer date shortly after epoch" &&
+	near_commit_oid=$(git rev-parse HEAD) &&
+
+	echo "$near_commit_oid no-timestamp-underflow" >expect &&
+	git name-rev $near_commit_oid >actual &&
+	test_cmp expect actual
+'
+
 test_done
Signature · 1 line
2.23.0.597.gb58f4577a1
#1 lore raw

Hi Gábor

Quoted 54 lines · SZEDER Gábor

When 'git name-rev' is invoked with commit-ish parameters, it tries to save some work, and doesn't visit commits older than the committer date of the oldest given commit minus a one day worth of slop. Since our 'timestamp_t' is an unsigned type, this leads to a timestamp underflow when the committer date of the oldest given commit is within a day of the UNIX epoch. As a result the cutoff timestamp ends up far-far in the future, and 'git name-rev' doesn't visit any commits, and names each given commit as 'undefined'.

Check whether substacting the slop from the oldest committer date would lead to an underflow, and use a 0 as cutoff in that case. This way it will handle commits shortly after the epoch even if we were to switch to a signed 'timestamp_t' (but then we'll have to worry about signed underflow for very old commits).

Note that the type of the cutoff timestamp variable used to be signed before 5589e87fd8 (name-rev: change a "long" variable to timestamp_t, 2017-05-20). The behavior was still the same even back then, but the underflow didn't happen when substracting the slop from the oldest committer date, but when comparing the signed cutoff timestamp with unsigned committer dates in name_rev(). IOW, this underflow bug is as old as 'git name-rev' itself.

---

This patch adds a test at the end of 't6120-describe.sh', so it will conflict with my non-recursive name-rev patch series, which adds a test there as well, but the conflict should be wasy to resolve.

   https://public-inbox.org/git/[email protected]/
  builtin/name-rev.c  | 15 ++++++++++++---
  t/t6120-describe.sh | 15 +++++++++++++++
  2 files changed, 27 insertions(+), 3 deletions(-)
builtin/name-rev.c +5 -1
index c785fe16ba..a4d8d312ab 100644
@@ -9,7 +9,11 @@
  #include "sha1-lookup.h"
  #include "commit-slab.h"
 
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 
  typedef struct rev_name {
  	const char *tip_name;

builtin/name-rev.c int cmd_name_rev(int argc, const char **argv, const char *prefix) · line 490

Quoted 8 lines
builtin/name-rev.c +3 -2
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
  		add_object_array(object, *argv, &revs);
  	}
 
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)

Nice catch but wouldn't this be clearer as
if (cutoff > CUTOFF_DATE_SLOP) ?

Best Wishes

Phillip

Quoted 32 lines
builtin/name-rev.c +4 -0
+			cutoff = cutoff - CUTOFF_DATE_SLOP;
+		else
+			cutoff = 0;
+	}
  	for_each_ref(name_ref, &data);
 
  	if (transform_stdin) {
t/t6120-describe.sh +15 -0
index 2b883d8174..965e633c32 100755
@@ -424,4 +424,19 @@ test_expect_success 'describe complains about missing object' '
  	test_must_fail git describe $ZERO_OID
  '
 
+test_expect_success 'name-rev a rev shortly after epoch' '
+	test_when_finished "git checkout master" &&
+
+	git checkout --orphan no-timestamp-underflow &&
+	# Any date closer to epoch than the CUTOFF_DATE_SLOP constant
+	# in builtin/name-rev.c.
+	GIT_COMMITTER_DATE="@1234 +0000" \
+	git commit -m "committer date shortly after epoch" &&
+	near_commit_oid=$(git rev-parse HEAD) &&
+
+	echo "$near_commit_oid no-timestamp-underflow" >expect &&
+	git name-rev $near_commit_oid >actual &&
+	test_cmp expect actual
+'
+
  test_done
#2 lore raw
Quoted 55 lines · Phillip Wood
Quoted 54 lines · SZEDER Gábor

When 'git name-rev' is invoked with commit-ish parameters, it tries to save some work, and doesn't visit commits older than the committer date of the oldest given commit minus a one day worth of slop. Since our 'timestamp_t' is an unsigned type, this leads to a timestamp underflow when the committer date of the oldest given commit is within a day of the UNIX epoch. As a result the cutoff timestamp ends up far-far in the future, and 'git name-rev' doesn't visit any commits, and names each given commit as 'undefined'.

Check whether substacting the slop from the oldest committer date would lead to an underflow, and use a 0 as cutoff in that case. This way it will handle commits shortly after the epoch even if we were to switch to a signed 'timestamp_t' (but then we'll have to worry about signed underflow for very old commits).

Note that the type of the cutoff timestamp variable used to be signed before 5589e87fd8 (name-rev: change a "long" variable to timestamp_t, 2017-05-20). The behavior was still the same even back then, but the underflow didn't happen when substracting the slop from the oldest committer date, but when comparing the signed cutoff timestamp with unsigned committer dates in name_rev(). IOW, this underflow bug is as old as 'git name-rev' itself.

---

This patch adds a test at the end of 't6120-describe.sh', so it will conflict with my non-recursive name-rev patch series, which adds a test there as well, but the conflict should be wasy to resolve.

https://public-inbox.org/git/[email protected]/

 builtin/name-rev.c  | 15 ++++++++++++---
 t/t6120-describe.sh | 15 +++++++++++++++
 2 files changed, 27 insertions(+), 3 deletions(-)
builtin/name-rev.c +5 -1
index c785fe16ba..a4d8d312ab 100644
@@ -9,7 +9,11 @@
 #include "sha1-lookup.h"
 #include "commit-slab.h"
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 typedef struct rev_name {
 	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 		add_object_array(object, *argv, &revs);
 	}
Quoted 8 lines

builtin/name-rev.c int cmd_name_rev(int argc, const char **argv, const char *prefix) · line 490

Quoted 5 lines
builtin/name-rev.c +3 -2
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)

Nice catch but wouldn't this be clearer as
if (cutoff > CUTOFF_DATE_SLOP) ?

It would only be clearer now, with an unsigned 'timestamp_t'. I tried to future-proof for a signed 'timestamp_t' and a cutoff date before the UNIX epoch.

Quoted 4 lines
Quoted 4 lines
builtin/name-rev.c +4 -0
+			cutoff = cutoff - CUTOFF_DATE_SLOP;
+		else
+			cutoff = 0;
+	}
#3 lore raw

Am 22.09.19 um 21:53 schrieb SZEDER Gábor:

Quoted 21 lines · Phillip Wood
Quoted 11 lines · SZEDER Gábor
Quoted 10 lines
(quoted diff) +5 -0
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 typedef struct rev_name {
 	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 		add_object_array(object, *argv, &revs);
 	}
Quoted 8 lines

Review int cmd_name_rev(int argc, const char **argv, const char *prefix) · line 489

Quoted 5 lines
(quoted diff) +3 -2
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)

Nice catch but wouldn't this be clearer as
if (cutoff > CUTOFF_DATE_SLOP) ?

Quoted 3 lines

It would only be clearer now, with an unsigned 'timestamp_t'. I tried to future-proof for a signed 'timestamp_t' and a cutoff date before the UNIX epoch.

Huh? For signed cutoff and positive CUTOFF_DATE_SLOP, cutoff - CUTOFF_DATE_SLOP < cutoff is ALWAYS true. Signed interger underflow is undefined behavior and signed integer arithmetic does not wrap around!

IOW, the new condition makes only sense today, because cutoff is an unsigned type, but breaks down should we switch to a signed type.

You need this on top:

builtin/name-rev.c +2 -2
index a4d8d312ab..2d83c2b172 100644
@@ -487,10 +487,10 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 
 	if (cutoff) {
 		/* check for undeflow */
-		if (cutoff - CUTOFF_DATE_SLOP < cutoff)
+		if (cutoff > TIME_MIN + CUTOFF_DATE_SLOP)
 			cutoff = cutoff - CUTOFF_DATE_SLOP;
 		else
-			cutoff = 0;
+			cutoff = TIME_MIN;
 	}
 	for_each_ref(name_ref, &data);
 
git-compat-util.h +1 -0
index c68c61d07c..1bdc21a069 100644
@@ -344,6 +344,7 @@ typedef uintmax_t timestamp_t;
 #define PRItime PRIuMAX
 #define parse_timestamp strtoumax
 #define TIME_MAX UINTMAX_MAX
+#define TIME_MIN 0
 
 #ifndef PATH_SEP
 #define PATH_SEP ':'
#8 lore raw
In reply to SZEDER Gábor

builtin/name-rev.c line 13

Quoted 11 lines · SZEDER Gábor
builtin/name-rev.c +2 -1
index c785fe16ba..a4d8d312ab 100644
@@ -9,7 +9,11 @@
 #include "sha1-lookup.h"
 #include "commit-slab.h"
 
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when

This piece of code says "close to"…

builtin/name-rev.c line 14

Quoted 1 line
builtin/name-rev.c +1 -0
+test_expect_success 'name-rev a rev shortly after epoch' '

…but this says "shortly after."

Overall, I think the idea is definitely sane, though.

Signature · 2 lines
brian m. carlson: Houston, Texas, US
OpenPGP: https://keybase.io/bk2204
#4 lore raw
In reply to Johannes Sixt
Quoted 26 lines · Johannes Sixt

Am 22.09.19 um 21:53 schrieb SZEDER Gábor:

Quoted 21 lines · Phillip Wood
Quoted 11 lines · SZEDER Gábor
Quoted 10 lines
(quoted diff) +5 -0
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 typedef struct rev_name {
 	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 		add_object_array(object, *argv, &revs);
 	}
Quoted 8 lines

Review int cmd_name_rev(int argc, const char **argv, const char *prefix) · line 489

Quoted 5 lines
(quoted diff) +3 -2
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)

Nice catch but wouldn't this be clearer as
if (cutoff > CUTOFF_DATE_SLOP) ?

Quoted 3 lines

It would only be clearer now, with an unsigned 'timestamp_t'. I tried to future-proof for a signed 'timestamp_t' and a cutoff date before the UNIX epoch.

Quoted 7 lines

Huh? For signed cutoff and positive CUTOFF_DATE_SLOP, cutoff - CUTOFF_DATE_SLOP < cutoff is ALWAYS true. Signed interger underflow is undefined behavior and signed integer arithmetic does not wrap around!

IOW, the new condition makes only sense today, because cutoff is an unsigned type, but breaks down should we switch to a signed type.

Yeah, that's what I meant with worrying about signed underflow in the commit message. As long as the cutoff is at least a day later than the minimum value of our future signed 'timestamp_t', the condition does the right thing. And considering that oldest time a signed 64 bit timestamp can represent far exceeds the age of the universe, and the oldest value of even a signed 32 bit timestamp is almost half the age of the Earth, I wasn't too worried.

Quoted 31 lines

You need this on top:

builtin/name-rev.c +2 -2
index a4d8d312ab..2d83c2b172 100644
@@ -487,10 +487,10 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 
 	if (cutoff) {
 		/* check for undeflow */
-		if (cutoff - CUTOFF_DATE_SLOP < cutoff)
+		if (cutoff > TIME_MIN + CUTOFF_DATE_SLOP)
 			cutoff = cutoff - CUTOFF_DATE_SLOP;
 		else
-			cutoff = 0;
+			cutoff = TIME_MIN;
 	}
 	for_each_ref(name_ref, &data);
 
git-compat-util.h +1 -0
index c68c61d07c..1bdc21a069 100644
@@ -344,6 +344,7 @@ typedef uintmax_t timestamp_t;
 #define PRItime PRIuMAX
 #define parse_timestamp strtoumax
 #define TIME_MAX UINTMAX_MAX
+#define TIME_MIN 0
 
 #ifndef PATH_SEP
 #define PATH_SEP ':'
#9 lore raw
In reply to brian m. carlson
Quoted 15 lines · brian m. carlson

builtin/name-rev.c line 13

Quoted 11 lines · SZEDER Gábor
builtin/name-rev.c +2 -1
index c785fe16ba..a4d8d312ab 100644
@@ -9,7 +9,11 @@
 #include "sha1-lookup.h"
 #include "commit-slab.h"
 
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when

This piece of code says "close to"…

Quoted 3 lines

builtin/name-rev.c line 14

Quoted 1 line
builtin/name-rev.c +1 -0
+test_expect_success 'name-rev a rev shortly after epoch' '

…but this says "shortly after."

Thanks. I did a last minute 'close to' -> 'shortly after' edit, but apparently not everywhere.

Quoted 4 lines

Overall, I think the idea is definitely sane, though.

Signature · 2 lines
brian m. carlson: Houston, Texas, US
OpenPGP: https://keybase.io/bk2204
#5 lore raw
In reply to SZEDER Gábor
Quoted 35 lines · SZEDER Gábor
Quoted 26 lines · Johannes Sixt

Am 22.09.19 um 21:53 schrieb SZEDER Gábor:

Quoted 21 lines · Phillip Wood
Quoted 11 lines · SZEDER Gábor
Quoted 10 lines
(quoted diff) +5 -0
+/*
+ * One day.  See the 'name a rev close to epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
  typedef struct rev_name {
  	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
  		add_object_array(object, *argv, &revs);
  	}
Quoted 8 lines

Review int cmd_name_rev(int argc, const char **argv, const char *prefix) · line 489

Quoted 5 lines
(quoted diff) +3 -2
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff - CUTOFF_DATE_SLOP < cutoff)

Nice catch but wouldn't this be clearer as
if (cutoff > CUTOFF_DATE_SLOP) ?

Quoted 3 lines

It would only be clearer now, with an unsigned 'timestamp_t'. I tried to future-proof for a signed 'timestamp_t' and a cutoff date before the UNIX epoch.

Quoted 7 lines

Huh? For signed cutoff and positive CUTOFF_DATE_SLOP, cutoff - CUTOFF_DATE_SLOP < cutoff is ALWAYS true. Signed interger underflow is undefined behavior and signed integer arithmetic does not wrap around!

IOW, the new condition makes only sense today, because cutoff is an unsigned type, but breaks down should we switch to a signed type.

Quoted 7 lines

Yeah, that's what I meant with worrying about signed underflow in the commit message. As long as the cutoff is at least a day later than the minimum value of our future signed 'timestamp_t', the condition does the right thing. And considering that oldest time a signed 64 bit timestamp can represent far exceeds the age of the universe, and the oldest value of even a signed 32 bit timestamp is almost half the age of the Earth, I wasn't too worried.

It's still going to trip up static analysers though isn't it? Also it will confuse readers who have to reason that it does not overflow in practice as we (probably?) never use very negative values

Best Wishes

Phillip

Quoted 31 lines
Quoted 31 lines

You need this on top:

builtin/name-rev.c +2 -2
index a4d8d312ab..2d83c2b172 100644
@@ -487,10 +487,10 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 
  	if (cutoff) {
  		/* check for undeflow */
-		if (cutoff - CUTOFF_DATE_SLOP < cutoff)
+		if (cutoff > TIME_MIN + CUTOFF_DATE_SLOP)
  			cutoff = cutoff - CUTOFF_DATE_SLOP;
  		else
-			cutoff = 0;
+			cutoff = TIME_MIN;
  	}
  	for_each_ref(name_ref, &data);
 
git-compat-util.h +1 -0
index c68c61d07c..1bdc21a069 100644
@@ -344,6 +344,7 @@ typedef uintmax_t timestamp_t;
  #define PRItime PRIuMAX
  #define parse_timestamp strtoumax
  #define TIME_MAX UINTMAX_MAX
+#define TIME_MIN 0
 
  #ifndef PATH_SEP
  #define PATH_SEP ':'
#6 lore raw
In reply to SZEDER Gábor

Am 23.09.19 um 10:37 schrieb SZEDER Gábor:

Quoted 9 lines · Johannes Sixt
Quoted 7 lines

Huh? For signed cutoff and positive CUTOFF_DATE_SLOP, cutoff - CUTOFF_DATE_SLOP < cutoff is ALWAYS true. Signed interger underflow is undefined behavior and signed integer arithmetic does not wrap around!

IOW, the new condition makes only sense today, because cutoff is an unsigned type, but breaks down should we switch to a signed type.

Quoted 7 lines

Yeah, that's what I meant with worrying about signed underflow in the commit message. As long as the cutoff is at least a day later than the minimum value of our future signed 'timestamp_t', the condition does the right thing. And considering that oldest time a signed 64 bit timestamp can represent far exceeds the age of the universe, and the oldest value of even a signed 32 bit timestamp is almost half the age of the Earth, I wasn't too worried.

Note that commits and timestamps can be forged easily. I'm not worried that Git does not work "correctly" with forged timestamps (as long as it is not exploitable); but when it is simple to make foolproof, we should do it.

BTW, 32-bit timestamps cover just ~135 years (not half the age of Earth). That's too little for people who want to store historic documents in a Git repository.

-- Hannes

#7 lore raw
Quoted 18 lines · Johannes Sixt

Am 23.09.19 um 10:37 schrieb SZEDER Gábor:

Quoted 9 lines · Johannes Sixt
Quoted 7 lines

Huh? For signed cutoff and positive CUTOFF_DATE_SLOP, cutoff - CUTOFF_DATE_SLOP < cutoff is ALWAYS true. Signed interger underflow is undefined behavior and signed integer arithmetic does not wrap around!

IOW, the new condition makes only sense today, because cutoff is an unsigned type, but breaks down should we switch to a signed type.

Quoted 7 lines

Yeah, that's what I meant with worrying about signed underflow in the commit message. As long as the cutoff is at least a day later than the minimum value of our future signed 'timestamp_t', the condition does the right thing. And considering that oldest time a signed 64 bit timestamp can represent far exceeds the age of the universe, and the oldest value of even a signed 32 bit timestamp is almost half the age of the Earth, I wasn't too worried.

Quoted 7 lines

Note that commits and timestamps can be forged easily. I'm not worried that Git does not work "correctly" with forged timestamps (as long as it is not exploitable); but when it is simple to make foolproof, we should do it.

BTW, 32-bit timestamps cover just ~135 years (not half the age of Earth).

Gah, forgot the division with seconds/year when calculating the range of the 32bit timestamp.

Quoted 2 lines

That's too little for people who want to store historic documents in a Git repository.

Indeed, but 'timestamp_t' is defined as 'uintmax_t', and we have the Makefile knob 'NO_UINTMAX_T', in which case 'uintmax_t' is defined as 'uint32_t'...

#10 lore raw

PATCH 1/1 name-rev: avoid cutoff timestamp underflow

In reply to SZEDER Gábor

When 'git name-rev' is invoked with commit-ish parameters, it tries to save some work, and doesn't visit commits older than the committer date of the oldest given commit minus a one day worth of slop. Since our 'timestamp_t' is an unsigned type, this leads to a timestamp underflow when the committer date of the oldest given commit is within a day of the UNIX epoch. As a result the cutoff timestamp ends up far-far in the future, and 'git name-rev' doesn't visit any commits, and names each given commit as 'undefined'.

Check whether subtracting the slop from the oldest committer date would lead to an underflow, and use no cutoff in that case. We don't have a TIME_MIN constant, dddbad728c (timestamp_t: a new data type for timestamps, 2017-04-26) didn't add one, so do it now.

Note that the type of the cutoff timestamp variable used to be signed before 5589e87fd8 (name-rev: change a "long" variable to timestamp_t, 2017-05-20). The behavior was still the same even back then, but the underflow didn't happen when substracting the slop from the oldest committer date, but when comparing the signed cutoff timestamp with unsigned committer dates in name_rev(). IOW, this underflow bug is as old as 'git name-rev' itself.

---
Range-diff:
1: 3bef6bdf7b ! 1: b053181af9 name-rev: avoid cutoff timestamp underflow
@@ Commit message
far-far in the future, and 'git name-rev' doesn't visit any commits, and names each given commit as 'undefined'.

- Check whether substacting the slop from the oldest committer date
- would lead to an underflow, and use a 0 as cutoff in that case. This
- way it will handle commits shortly after the epoch even if we were to
- switch to a signed 'timestamp_t' (but then we'll have to worry about
- signed underflow for very old commits).
+ Check whether subtracting the slop from the oldest committer date
+ would lead to an underflow, and use no cutoff in that case. We don't
+ have a TIME_MIN constant, dddbad728c (timestamp_t: a new data type for
+ timestamps, 2017-04-26) didn't add one, so do it now.

Note that the type of the cutoff timestamp variable used to be signed before 5589e87fd8 (name-rev: change a "long" variable to timestamp_t,
@@ Commit message
unsigned committer dates in name_rev(). IOW, this underflow bug is as old as 'git name-rev' itself.

+ Helped-by: Johannes Sixt <[email protected]>

## builtin/name-rev.c ##
@@ builtin/name-rev.c

-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
-+ * One day. See the 'name a rev close to epoch' test in t6120 when ++ * One day. See the 'name a rev shortly after epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
@@ builtin/name-rev.c: int cmd_name_rev(int argc, const char **argv, const char *pr
- cutoff = cutoff - CUTOFF_DATE_SLOP;
+ if (cutoff) {
+ /* check for undeflow */
-+ if (cutoff - CUTOFF_DATE_SLOP < cutoff)
++ if (cutoff > TIME_MIN + CUTOFF_DATE_SLOP)
+ cutoff = cutoff - CUTOFF_DATE_SLOP;
+ else
-+ cutoff = 0;
++ cutoff = TIME_MIN;
+ }
for_each_ref(name_ref, &data);

if (transform_stdin) {

+ ## git-compat-util.h ##
+@@ git-compat-util.h: typedef uintmax_t timestamp_t;
+ #define PRItime PRIuMAX
+ #define parse_timestamp strtoumax
+ #define TIME_MAX UINTMAX_MAX
++#define TIME_MIN 0
+
+ #ifndef PATH_SEP
+ #define PATH_SEP ':'
+
## t/t6120-describe.sh ##
@@ t/t6120-describe.sh: test_expect_success 'describe complains about missing object' '
test_must_fail git describe $ZERO_OID
@@ t/t6120-describe.sh: test_expect_success 'describe complains about missing objec
+ # in builtin/name-rev.c.
+ GIT_COMMITTER_DATE="@1234 +0000" \
+ git commit -m "committer date shortly after epoch" &&
-+ near_commit_oid=$(git rev-parse HEAD) &&
++ old_commit_oid=$(git rev-parse HEAD) &&
+
-+ echo "$near_commit_oid no-timestamp-underflow" >expect &&
-+ git name-rev $near_commit_oid >actual &&
++ echo "$old_commit_oid no-timestamp-underflow" >expect &&
++ git name-rev $old_commit_oid >actual &&
+ test_cmp expect actual
+'
+

 builtin/name-rev.c  | 15 ++++++++++++---
 git-compat-util.h   |  1 +
 t/t6120-describe.sh | 15 +++++++++++++++
 3 files changed, 28 insertions(+), 3 deletions(-)
builtin/name-rev.c +12 -3
index c785fe16ba..b0f0776947 100644
@@ -9,7 +9,11 @@
 #include "sha1-lookup.h"
 #include "commit-slab.h"
 
-#define CUTOFF_DATE_SLOP 86400 /* one day */
+/*
+ * One day.  See the 'name a rev shortly after epoch' test in t6120 when
+ * changing this value
+ */
+#define CUTOFF_DATE_SLOP 86400
 
 typedef struct rev_name {
 	const char *tip_name;
@@ -481,8 +485,13 @@ int cmd_name_rev(int argc, const char **argv, const char *prefix)
 		add_object_array(object, *argv, &revs);
 	}
 
-	if (cutoff)
-		cutoff = cutoff - CUTOFF_DATE_SLOP;
+	if (cutoff) {
+		/* check for undeflow */
+		if (cutoff > TIME_MIN + CUTOFF_DATE_SLOP)
+			cutoff = cutoff - CUTOFF_DATE_SLOP;
+		else
+			cutoff = TIME_MIN;
+	}
 	for_each_ref(name_ref, &data);
 
 	if (transform_stdin) {
git-compat-util.h +1 -0
index f0d13e4e28..d2e884e46a 100644
@@ -344,6 +344,7 @@ typedef uintmax_t timestamp_t;
 #define PRItime PRIuMAX
 #define parse_timestamp strtoumax
 #define TIME_MAX UINTMAX_MAX
+#define TIME_MIN 0
 
 #ifndef PATH_SEP
 #define PATH_SEP ':'
t/t6120-describe.sh +15 -0
index 2b883d8174..45047d0a72 100755
@@ -424,4 +424,19 @@ test_expect_success 'describe complains about missing object' '
 	test_must_fail git describe $ZERO_OID
 '
 
+test_expect_success 'name-rev a rev shortly after epoch' '
+	test_when_finished "git checkout master" &&
+
+	git checkout --orphan no-timestamp-underflow &&
+	# Any date closer to epoch than the CUTOFF_DATE_SLOP constant
+	# in builtin/name-rev.c.
+	GIT_COMMITTER_DATE="@1234 +0000" \
+	git commit -m "committer date shortly after epoch" &&
+	old_commit_oid=$(git rev-parse HEAD) &&
+
+	echo "$old_commit_oid no-timestamp-underflow" >expect &&
+	git name-rev $old_commit_oid >actual &&
+	test_cmp expect actual
+'
+
 test_done
Signature · 1 line
2.23.0.597.gb58f4577a1