[RESEND PATCH v3 1/2] cli: Correct several bugs in cli_getch()
This function does not behave as expected when unknown escape sequences are sent to it:
- it fails to store (and thus echo) the last character of the invalid sequence - it fails to set esc_len to 0 when it finishes emitting the invalid sequence, meaning that the following character will appear to be part of a new escape sequence - it processes the first character of the rejected sequence as a valid character, just starting the sequence all over again
The last two bugs conspire to produce an "impossible condition #876" message which is the main symptom of this behaviour.
Fix these bugs and add a test to verify the behaviour.
Signed-off-by: Simon Glass sjg@chromium.org Reported-by: Heinrich Schuchardt xypron.glpk@gmx.de ---
(no changes since v1)
common/cli_getch.c | 5 +++-- test/common/Makefile | 1 + test/common/cread.c | 48 ++++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 52 insertions(+), 2 deletions(-) create mode 100644 test/common/cread.c
diff --git a/common/cli_getch.c b/common/cli_getch.c index 87c23edcf4b4..61d4cb261b81 100644 --- a/common/cli_getch.c +++ b/common/cli_getch.c @@ -129,7 +129,7 @@ static int cli_ch_esc(struct cli_ch_state *cch, int ichar,
*actp = act;
- return act == ESC_CONVERTED ? ichar : 0; + return ichar; }
int cli_ch_process(struct cli_ch_state *cch, int ichar) @@ -145,6 +145,7 @@ int cli_ch_process(struct cli_ch_state *cch, int ichar) return cch->esc_save[cch->emit_upto++]; cch->emit_upto = 0; cch->emitting = false; + cch->esc_len = 0; } return 0; } else if (ichar == -ETIMEDOUT) { @@ -185,7 +186,7 @@ int cli_ch_process(struct cli_ch_state *cch, int ichar) cch->esc_save[cch->esc_len++] = ichar; ichar = cch->esc_save[cch->emit_upto++]; cch->emitting = true; - break; + return ichar; case ESC_CONVERTED: /* valid escape sequence, return the resulting char */ cch->esc_len = 0; diff --git a/test/common/Makefile b/test/common/Makefile index cc918f64e544..a5ab10f6f69b 100644 --- a/test/common/Makefile +++ b/test/common/Makefile @@ -3,3 +3,4 @@ obj-y += cmd_ut_common.o obj-$(CONFIG_AUTOBOOT) += test_autoboot.o obj-$(CONFIG_CYCLIC) += cyclic.o obj-$(CONFIG_EVENT) += event.o +obj-y += cread.o diff --git a/test/common/cread.c b/test/common/cread.c new file mode 100644 index 000000000000..3dce4bdb0ef3 --- /dev/null +++ b/test/common/cread.c @@ -0,0 +1,48 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Copyright 2023 Google LLC + */ + +#include <common.h> +#include <cli.h> +#include <test/common.h> +#include <test/test.h> +#include <test/ut.h> + +static int cli_ch_test(struct unit_test_state *uts) +{ + struct cli_ch_state s_cch, *cch = &s_cch; + + cli_ch_init(cch); + + /* should be nothing to return at first */ + ut_asserteq(0, cli_ch_process(cch, 0)); + + /* check normal entry */ + ut_asserteq('a', cli_ch_process(cch, 'a')); + ut_asserteq('b', cli_ch_process(cch, 'b')); + ut_asserteq('c', cli_ch_process(cch, 'c')); + ut_asserteq(0, cli_ch_process(cch, 0)); + + /* send an invalid escape sequence */ + ut_asserteq(0, cli_ch_process(cch, '\e')); + ut_asserteq(0, cli_ch_process(cch, '[')); + + /* + * with the next char it sees that the sequence is invalid, so starts + * emitting it + */ + ut_asserteq('\e', cli_ch_process(cch, 'X')); + + /* now we set 0 bytes to empty the buffer */ + ut_asserteq('[', cli_ch_process(cch, 0)); + ut_asserteq('X', cli_ch_process(cch, 0)); + ut_asserteq(0, cli_ch_process(cch, 0)); + + /* things are normal again */ + ut_asserteq('a', cli_ch_process(cch, 'a')); + ut_asserteq(0, cli_ch_process(cch, 0)); + + return 0; +} +COMMON_TEST(cli_ch_test, 0);
The second call to cli_ch_process() is in the wrong place, meaning that the one of the characters of an invalid escape sequence is swallowed instead of being returned.
Fix the bug and add a test to cover this.
This behaviour matches that of the code before cli_getch() was introduced. This was verified on the commit before b08e9d4b66 i.e.:
7d850f85aad ("sandbox: Enable mmc command and legacy images")
Signed-off-by: Simon Glass sjg@chromium.org Reported-by: Heinrich Schuchardt xypron.glpk@gmx.de ---
Changes in v3: - Update commit message
Changes in v2: - Fix a failling test
common/cli_readline.c | 3 +-- include/cli.h | 4 ++-- test/common/cread.c | 45 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 48 insertions(+), 4 deletions(-)
diff --git a/common/cli_readline.c b/common/cli_readline.c index 709e9c3d38b1..e83743e90c92 100644 --- a/common/cli_readline.c +++ b/common/cli_readline.c @@ -284,10 +284,9 @@ static int cread_line(const char *const prompt, char *buf, unsigned int *len, }
ichar = getcmd_getch(); + ichar = cli_ch_process(cch, ichar); }
- ichar = cli_ch_process(cch, ichar); - /* ichar=0x0 when error occurs in U-Boot getc */ if (!ichar) continue; diff --git a/include/cli.h b/include/cli.h index c777c90313f2..094a6602d70e 100644 --- a/include/cli.h +++ b/include/cli.h @@ -98,8 +98,8 @@ int cli_readline(const char *const prompt); * * @prompt: Prompt to display * @buffer: Place to put the line that is entered - * @timeout: Timeout in milliseconds, 0 if none - * Return: command line length excluding terminator, or -ve on error: of the + * @timeout: Timeout in seconds, 0 if none + * Return: command line length excluding terminator, or -ve on error: if the * timeout is exceeded (either CONFIG_BOOT_RETRY_TIME or the timeout * parameter), then -2 is returned. If a break is detected (Ctrl-C) then * -1 is returned. diff --git a/test/common/cread.c b/test/common/cread.c index 3dce4bdb0ef3..2fdd29a265f7 100644 --- a/test/common/cread.c +++ b/test/common/cread.c @@ -46,3 +46,48 @@ static int cli_ch_test(struct unit_test_state *uts) return 0; } COMMON_TEST(cli_ch_test, 0); + +static int cread_test(struct unit_test_state *uts) +{ + int duration; + ulong start; + char buf[10]; + + /* + * useful for debugging + * + * gd->flags &= ~GD_FLG_RECORD; + * print_buffer(0, buf, 1, 7, 0); + */ + + console_record_reset_enable(); + + /* simple input */ + *buf = '\0'; + ut_asserteq(4, console_in_puts("abc\n")); + ut_asserteq(3, cli_readline_into_buffer("-> ", buf, 1)); + ut_asserteq_str("abc", buf); + + /* try an escape sequence (cursor left after the 'c') */ + *buf = '\0'; + ut_asserteq(8, console_in_puts("abc\e[Dx\n")); + ut_asserteq(4, cli_readline_into_buffer("-> ", buf, 1)); + ut_asserteq_str("abxc", buf); + + /* invalid escape sequence */ + *buf = '\0'; + ut_asserteq(8, console_in_puts("abc\e[Xx\n")); + ut_asserteq(7, cli_readline_into_buffer("-> ", buf, 1)); + ut_asserteq_str("abc\e[Xx", buf); + + /* check timeout, should be between 1000 and 1050ms */ + start = get_timer(0); + *buf = '\0'; + ut_asserteq(-2, cli_readline_into_buffer("-> ", buf, 1)); + duration = get_timer(start) - 1000; + ut_assert(duration >= 0); + ut_assert(duration < 50); + + return 0; +} +COMMON_TEST(cread_test, 0);
On 3/27/23 21:34, Simon Glass wrote:
The second call to cli_ch_process() is in the wrong place, meaning that the one of the characters of an invalid escape sequence is swallowed instead of being returned.
Fix the bug and add a test to cover this.
This behaviour matches that of the code before cli_getch() was introduced. This was verified on the commit before b08e9d4b66 i.e.:
7d850f85aad ("sandbox: Enable mmc command and legacy images")Signed-off-by: Simon Glass sjg@chromium.org Reported-by: Heinrich Schuchardt xypron.glpk@gmx.de
Thanks for the fix. Some strange behavior still remains like <SHIFT><CTRL><F4> giving '6S' on the command line. But at least no error code.
Best regards
Heinrich
Hi Heinrich,
On Wed, 29 Mar 2023 at 05:43, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 3/27/23 21:34, Simon Glass wrote:
The second call to cli_ch_process() is in the wrong place, meaning that the one of the characters of an invalid escape sequence is swallowed instead of being returned.
Fix the bug and add a test to cover this.
This behaviour matches that of the code before cli_getch() was introduced. This was verified on the commit before b08e9d4b66 i.e.:
7d850f85aad ("sandbox: Enable mmc command and legacy images")Signed-off-by: Simon Glass sjg@chromium.org Reported-by: Heinrich Schuchardt xypron.glpk@gmx.de
Thanks for the fix. Some strange behavior still remains like <SHIFT><CTRL><F4> giving '6S' on the command line. But at least no error code.
Yes, this matches the previous behaviour, which has existed a long time. We could easily suppress echoing of unknown escape sequence. Should we suppress them being entered also, so that U-Boot ignores them? I'm really not sure. There was a reverted patch[1] which indicates that this is not trivial to figure out.
Regards, Simon
[1] d2e64d29c44d cli_readline: Only insert printable chars
Am 29. März 2023 22:02:11 MESZ schrieb Simon Glass sjg@chromium.org:
Hi Heinrich,
On Wed, 29 Mar 2023 at 05:43, Heinrich Schuchardt xypron.glpk@gmx.de wrote:
On 3/27/23 21:34, Simon Glass wrote:
The second call to cli_ch_process() is in the wrong place, meaning that the one of the characters of an invalid escape sequence is swallowed instead of being returned.
Fix the bug and add a test to cover this.
This behaviour matches that of the code before cli_getch() was introduced. This was verified on the commit before b08e9d4b66 i.e.:
7d850f85aad ("sandbox: Enable mmc command and legacy images")Signed-off-by: Simon Glass sjg@chromium.org Reported-by: Heinrich Schuchardt xypron.glpk@gmx.de
Thanks for the fix. Some strange behavior still remains like <SHIFT><CTRL><F4> giving '6S' on the command line. But at least no error code.
Yes, this matches the previous behaviour, which has existed a long time. We could easily suppress echoing of unknown escape sequence. Should we suppress them being entered also, so that U-Boot ignores them? I'm really not sure. There was a reverted patch[1] which indicates that this is not trivial to figure out.
We already have code in lib/efi_loader/efi_condole.c that supports many more escape sequences. We should deduplicate the code.
Best regards
Heinrich
Regards, Simon
[1] d2e64d29c44d cli_readline: Only insert printable chars
On Tue, 28 Mar 2023 08:34:13 +1300, Simon Glass wrote:
This function does not behave as expected when unknown escape sequences are sent to it:
- it fails to store (and thus echo) the last character of the invalid sequence
- it fails to set esc_len to 0 when it finishes emitting the invalid sequence, meaning that the following character will appear to be part of a new escape sequence
- it processes the first character of the rejected sequence as a valid character, just starting the sequence all over again
[...]
Applied to u-boot/master, thanks!
participants (3)
-
Heinrich Schuchardt -
Simon Glass -
Tom Rini