Skip to contentSkip to conversation

grub-fstest: Show error message if command causes grub_errno

PATCH In review Thomas Schmitt via Grub-devel 1 participant 4 messages Jun 20, 2024 – Jun 21, 2024

lore
Share & tools
lore
#0 lore raw

cover grub-fstest: Show error message if command causes grub_errno

Most return values of function execute_command() are ignored in util/grub-fstest.c. Only one of two "loopback" occasions and the command "cryptomount" care. They perform grub_util_error() if the return value is non-zero.
So there seems to be a general lack of handling of problems found in the drivers or command implementations.

On the other hand, all tests which use grub-fstest are accustomed to this silence about problems. Internal processing of grub-fstest does possibly important things after an error occurs and before the program ends. It seems inappropriate to immediately bail out when execute_command() encounters grub_errno != GRUB_ERROR_NONE.

So i propose a patch which just counts the occurences of grub_errno in execute_command(), reports grub_errmsg, and resets grub_errno. Processing goes on as is tradition. But at the end of the program, a non-zero error count causes a non-zero exit value.
Example:

  $ gunzip <tests/iso9660_ce_loop2.iso.gz >/tmp/iso9660_ce_loop2.iso
  $ ./grub-fstest /tmp/iso9660_ce_loop2.iso ls /
  ls : GRUB error 9 with message 'suspecting endless CE loop'
  ./grub-fstest: error: encountered 1 error during command execution.
  $ echo $?
  1

(I will submit an addon patch which changes the grub_errno display to
ls : GRUB_ERR_BAD_FS with message 'suspecting endless CE loop'
by adding a new function grub_errno_to_name() to err.[ch].)

A new program option -E re-activates the current behavior. I.e. it disables counting, reporting, and non-zero exit caused by grub_errno.

After installing help2man and running ./configure i get from "make" a file grub-fstest.1 which mentions the new option. (I fail to find file grub-fstest.info which is promised by the text in grub-fstest.1. Package texinfo is installed.)

I duely checked the impact of this change on "make check:

Installed are the Debian 12 packages which INSTALL prescribes. "make check" was run as superuser with the old and the new version of grub-fstest. Then i compared the outputs about PASS and FAIL.

The only test which newly failed is iso9660_ce_loop in tests/iso9660_test.in because it takes non-zero exit as indication of timeout, but will now regularly receive non-zero because iso9660_ce_loop.iso is intentionally bad.
So patch 2/2 adds -E to the run of grub-fstest with iso9660_ce_loop. This makes tests/iso9660_test PASS again.

4 tests yielded SKIP due to my platform "efi" and target "x86_64":
tests/fddboot_test.in
tests/netboot_test.in
tests/pseries_test.in
tests/core_compress_test.in
They are small and i can see no reference to grub-fstest in them.

6 tests yielded FAIL with both versions of grub-fstest. They seem unrelated to grub-fstest:

- erofs_test says:

    mkfs.erofs -Eforce-inode-compact -L g;/éт 莭😁 /tmp/grub-fs-tester.2024062018

0702880665192.erofs_compact.XeJ/erofs_compact_512_4096_1_0.img /tmp/grub-fs-tester.20240620180702880665192.erofs_compact.XeJ/master
mkfs.erofs: invalid option -- 'L'
The test seems to expect something other than erofs-utils_1.5-1_amd64.deb
of Debian 12.

- exfat_test shows the help text of mkfs.exfat. In grub-fs-tester i see

    x"exfat")
        "mkfs.$fs" -s $((BLKSIZE/512)) -n "$FSLABEL" "${MOUNTDEVICE}"

But mkfs.exfat has neither option -s nor -n in help text or man page. mkexfatfs offers: -n volume-name -s sectors-per-cluster

- hfs_test says:
LABEL FAIL
Device loop0: Filesystem type hfs - Label `untitled' ...
grub-fs-tester does:

    FSLABEL="grub_t;/estéàèèéie fiucnree";;
    "mkfs.hfs" -b $BLKSIZE -v "`echo $FSLABEL |recode utf8..macroman`" -h "${MOUNTDEVICE}"

man mkfs.hfs shows man newfs_hfs, where -v is mentioned as:
-v volume-name
Volume name (file system name) in ascii or UTF-8 format.
(Was already so in Debian 11.)
So the recode filter seems unneeded and unhelpful.

- luks1_test says:

    ./grub-probe: error: disk `lvm/grub-fs-tester.20230401111443076506440.luks1.PDo' not found.
    UUID FAIL
    96efa19ce89c41279912b6a3b26a99d3

- luks2_test says:

    ./grub-probe: error: disk `lvm/grub-fs-tester.20230401111450278223161.luks2.NWP' not found.
    UUID FAIL
    8a1f1e83695d4dd1b6b73fcb542b73da

- grub_func_test reports checksum problems

    tests/video_checksum.c:checksum:615: assert failed: 0 Checksum cmdline_cat_2560x1440xrgba8888:44 failed: 0x62031fea vs 0x8071678a

-----------------------------------------------------------------------

Thomas Schmitt (2):
grub-fstest: Show error message if command causes grub_errno
tests: Use new grub-fstest option -E with iso9660_ce_loop test

 tests/iso9660_test.in |  2 +-
 util/grub-fstest.c    | 30 +++++++++++++++++++++++++++++-
 2 files changed, 30 insertions(+), 2 deletions(-)

--
2.39.2

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#1 lore raw

PATCH 1/2 grub-fstest: Show error message if command causes grub_errno

Check grub_errno after a command was performed. If not GRUB_ERR_NONE then print grub_errno and grub_errmsg. Count the errors and exit 1 with a message if the count is larger than zero when function fs_test() ends. The number of printed messages is restricted to 3.

Introduce a new option -E which re-enables the old behavior which did not show grub_errmsg and did not exit with non-zero value if grub_errno was found not being GRUB_ERR_NONE.

---
 util/grub-fstest.c | 30 +++++++++++++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)
util/grub-fstest.c +29 -1
index 7ff9037b8..145b40a0e 100644
@@ -47,16 +47,33 @@
 #pragma GCC diagnostic error "-Wmissing-prototypes"
 #pragma GCC diagnostic error "-Wmissing-declarations"
 
+static int execute_error_count = 0;
+static int no_cmd_error = 0;
+
 static grub_err_t
 execute_command (const char *name, int n, char **args)
 {
   grub_command_t cmd;
+  grub_err_t ret;
 
   cmd = grub_command_find (name);
   if (! cmd)
     grub_util_error (_("can't find command `%s'"), name);
 
-  return (cmd->func) (cmd, n, args);
+  ret = (cmd->func) (cmd, n, args);
+  if (grub_errno != GRUB_ERR_NONE && !no_cmd_error)
+    {
+      execute_error_count++;
+      if (execute_error_count <= 3)
+	grub_err_printf (_("%s : GRUB error %d with message '%s'\n"),
+			 name, (int) grub_errno, grub_errmsg);
+      else if (execute_error_count == 4)
+	grub_err_printf (_("%s : more GRUB errors encountered but not shown\n"),
+			 name);
+      grub_errno = GRUB_ERR_NONE;
+      grub_errmsg[0] = 0;
+    }
+  return ret;
 }
 
 enum {
@@ -510,6 +527,12 @@ fstest (int n)
       grub_free (loop_name);
       grub_free (argv[0]);
     }
+
+  if (execute_error_count == 1)
+    grub_util_error (_("encountered 1 error during command execution"));
+  else if (execute_error_count > 0)
+    grub_util_error (_("encountered %d errors during command execution"),
+		     execute_error_count);
 }
 
 static struct argp_option options[] = {
@@ -535,6 +558,7 @@ static struct argp_option options[] = {
    N_("FILE|prompt"), 0, N_("Load zfs crypto key."),                 2},
   {"verbose",   'v', NULL, 0, N_("print verbose messages."), 2},
   {"uncompress", 'u', NULL, 0, N_("Uncompress data."), 2},
+  {"no-cmd-error", 'E', NULL, 0, N_("Ignore errors of command execution."), 2},
   {0, 0, 0, 0, 0, 0}
 };
 
@@ -638,6 +662,10 @@ argp_parser (int key, char *arg, struct argp_state *state)
       uncompress = 1;
       return 0;
 
+    case 'E':
+      no_cmd_error = 1;
+      return 0;
+
     case ARGP_KEY_END:
       if (args_count < num_disks)
 	{

--
2.39.2

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#2 lore raw

PATCH 2/2 tests: Use new grub-fstest option -E with iso9660_ce_loop test

In reply to Thomas Schmitt via Grub-devel

This part of tests/iso9660_test works on an intentionally bad ISO which with older versions of GRUB will cause an endless cycle. Therefore the test runs grub-fstest under the program timeout. A non-zero exit value is taken as indication that the timeout was triggered. Non-empty output from grub-fstest is taken as indication that a CE entry was skipped by an old bug, thus breaking the endless cycle by mistake.

A change in grub-fstest causes output of an error message and a non-zero exit value when the bad ISO is handled properly.

So add the new grub-fstest option -E to the run of grub-fstest on iso9660_ce_loop.iso and iso9660_ce_loop2.iso. This suppresses the error message and the non-zero exit value.

---
 tests/iso9660_test.in | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tests/iso9660_test.in +1 -1
index a1f752adf..677c777d9 100644
@@ -20,7 +20,7 @@ for fs in iso9660_ce_loop iso9660_ce_loop2; do
     { echo "Failed to make temporary directory"; exit 99; }
   gunzip <"$srcdir"/tests/${fs}.iso.gz >"${tempdir}/${fs}.iso" || exit 99
   output=$(LC_ALL=C timeout -s KILL "60" \
-    "@builddir@/grub-fstest" "${tempdir}/${fs}.iso" ls / ) || ret=$?
+    "@builddir@/grub-fstest" -E "${tempdir}/${fs}.iso" ls / ) || ret=$?
   rm -rf "$tempdir"
   if [ "${ret:-0}" -ne 0 -o -n "$output" ]; then
     echo "FAIL ($fs)"

--
2.39.2

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel

#3 lore raw
In reply to Thomas Schmitt via Grub-devel

Hi,

somehow i managed to omit the first paragraph of the cover letter:
---------------------------------------------------------------------

grub-fstest describes itself as "debug tool for filesystem driver" but fails to forward to the user the grub_errno status and grub_errmsg of the drivers under command "ls". Further it does not indicate such errors by its exit status.
Reason is that function grub_cmd_ls() in grub-core/commands/ls.c unconditionally returns 0 and that grub_errno is not used anywhere in util/grub-fstest.c.

---------------------------------------------------------------------

Have a nice day :)

Thomas

_______________________________________________
Grub-devel mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/grub-devel