Skip to contentSkip to conversation

generic/386: check the correct field from df output

PATCH In review Theodore Ts'o 4 participants 5 messages Dec 22, 2019 – Dec 29, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 generic/386: check the correct field from df output

The generic/386 test was checking the "Available" field when it should have been checking the "1k-blocks" field, which represents the project quota's hard limit. On xfs, an empty directory takes no space, so it doesn't matter. But for ext4, an empty directory still takes 4k (or whatever the file system's block size happens to be):

Filesystem 1K-blocks Used Available Use% Pathname
/dev/vdc 512000 4 511996 0% /vdc/test

This causes generic/386 to falsely fail. There was a confusing comment claiming that for a very long device name, the df output would have a line break, and for that reason, the test would extract the field using $(NF-2). However, looking at xfsprogs's quota command, I see no evidence that there is any line breaking logic. Since we now want to use the second field, even if there was some line breaking going on, using $2 should be a better choice.

This fix is needed to fix generic/386 from failing on ext4:

    hard limit 524283904 bytes, expected 524288000
---
 tests/generic/386 | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/386 +1 -3
index 0c44c80e..37e9b943 100755
@@ -62,8 +62,6 @@ _require_scratch
 # both the "df" and the "report" output.  For "report", the line we're
 # interested in contains our project name in the first field.  For "df"
 # it contains our project directory in the last field.
-# But if the device name is too long, the "df" output is broke into two
-# lines, the fourth field is not correct, so take $(NF-2) of "df"
 _filter_quota_rpt() {
 	awk '
 	BEGIN {
@@ -89,7 +87,7 @@ _filter_quota_rpt() {
 			bsize = byte_size($4);
 		} else if ($NF ~ proj_dir) {
 			# this is the "df" output
-			bsize = byte_size($(NF-2));
+			bsize = byte_size($2));
 		} else {
 			next;
 		}
Signature · 1 line
2.24.1
#1 lore raw
Quoted 2 lines · Theodore Ts'o

... However, looking at xfsprogs's quota command, I see no evidence that there is any line breaking logic.

Eric, since you're the xfsprogs maintainer --- am I missing something? Was there a line breaking feature in the past?

- Ted

Reviewed-by Dave Chinner
#2 lore raw
Quoted 9 lines · Theodore Y. Ts'o
Quoted 2 lines · Theodore Ts'o

... However, looking at xfsprogs's quota command, I see no evidence that there is any line breaking logic.

Eric, since you're the xfsprogs maintainer --- am I missing something? Was there a line breaking feature in the past?

- Ted

The logic is in the free_space() function.

        if (flags & HUMAN_FLAG) {
                count = fprintf(fp, "%-12s", path->fs_name);
                if (count > 13)
                        fprintf(fp, "\n%12s", " ");
        } else {
                count = fprintf(fp, "%-19s", path->fs_name);
                if (count > 20)
                        fprintf(fp, "\n%19s", " ");
        }

eguan made this change to xfstests to handle it:

commit 50f2346560487303b381d11bbe52b93cd0a887bd

Author: Eryu Guan <[email protected]>
Date:   Tue Oct 14 22:59:38 2014 +1100

xfs/262: update filter to deal with long device name correctly

If the device name is too long, the output of xfs_quota -c "df" will be broke into two lines as

Filesystem 1K-blocks Used Available Use% Pathname
/dev/mapper/rhel_hp--dl388eg8--01-testlv2
15718400 32932 15685468 0% /mnt/testarea/scratch
/dev/mapper/rhel_hp--dl388eg8--01-testlv2
512000 0 512000 0% /mnt/testarea/scratch/test

and _filter_quota_rpt() couldn't catch the correct available number and test will fail as

[root@hp-dl388g8-01 xfstests]# diff -u tests/xfs/262.out /root/xfstests/results//xfs/262.out.bad
--- tests/xfs/262.out 2014-10-08 20:16:19.000000000 +0800
+++ /root/xfstests/results//xfs/262.out.bad 2014-10-09 14:29:38.795813323 +0800 @@ -1,2 +1,4 @@
QA output created by 262
Silence is golden.
+hard limit 0 bytes, expected 524288000
+hard limit 0 bytes, expected 524288000

Update the filter so it could catch the correct value.

Does:

- bsize = byte_size($(NF-2));
+ bsize = byte_size($(NF-4));

fix it properly on ext4?

#3 lore raw
In reply to Theodore Ts'o
Quoted 17 lines · Theodore Ts'o

The generic/386 test was checking the "Available" field when it should have been checking the "1k-blocks" field, which represents the project quota's hard limit. On xfs, an empty directory takes no space, so it doesn't matter. But for ext4, an empty directory still takes 4k (or whatever the file system's block size happens to be):

Filesystem 1K-blocks Used Available Use% Pathname
/dev/vdc 512000 4 511996 0% /vdc/test

This causes generic/386 to falsely fail. There was a confusing comment claiming that for a very long device name, the df output would have a line break, and for that reason, the test would extract the field using $(NF-2). However, looking at xfsprogs's quota command, I see no evidence that there is any line breaking logic. Since we now want to use the second field, even if there was some line breaking going on, using $2 should be a better choice.

Quoted 1 line

This fix is needed to fix generic/386 from failing on ext4:

Hi Theodore

I have a same fix but I use $(NF-4) field of df command. Also I wrongly think project quota uses a block size. url:
https://patchwork.kernel.org/patch/11296657/

Best Regards
Yang Xu

Quoted 31 lines
     hard limit 524283904 bytes, expected 524288000
---
  tests/generic/386 | 4 +---
  1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/386 +1 -3
index 0c44c80e..37e9b943 100755
@@ -62,8 +62,6 @@ _require_scratch
  # both the "df" and the "report" output.  For "report", the line we're
  # interested in contains our project name in the first field.  For "df"
  # it contains our project directory in the last field.
-# But if the device name is too long, the "df" output is broke into two
-# lines, the fourth field is not correct, so take $(NF-2) of "df"
  _filter_quota_rpt() {
  	awk '
  	BEGIN {
@@ -89,7 +87,7 @@ _filter_quota_rpt() {
  			bsize = byte_size($4);
  		} else if ($NF ~ proj_dir) {
  			# this is the "df" output
-			bsize = byte_size($(NF-2));
+			bsize = byte_size($2));
  		} else {
  			next;
  		}
#4 lore raw
Quoted 20 lines · Yang Xu
Quoted 17 lines · Theodore Ts'o

The generic/386 test was checking the "Available" field when it should have been checking the "1k-blocks" field, which represents the project quota's hard limit. On xfs, an empty directory takes no space, so it doesn't matter. But for ext4, an empty directory still takes 4k (or whatever the file system's block size happens to be):

Filesystem 1K-blocks Used Available Use% Pathname
/dev/vdc 512000 4 511996 0% /vdc/test

This causes generic/386 to falsely fail. There was a confusing comment claiming that for a very long device name, the df output would have a line break, and for that reason, the test would extract the field using $(NF-2). However, looking at xfsprogs's quota command, I see no evidence that there is any line breaking logic. Since we now want to use the second field, even if there was some line breaking going on, using $2 should be a better choice.

Quoted 7 lines
Quoted 1 line

This fix is needed to fix generic/386 from failing on ext4:

Hi Theodore

I have a same fix but I use $(NF-4) field of df command. Also I wrongly think project quota uses a block size.
url:
https://patchwork.kernel.org/patch/11296657/

Yes, $(NF-4) works, as Eric suggested as well. I applied Ted's patch (with $2->$(NF-4) fix), as the commit log is explaining well what's the problem and why taking "1k blocks" is correct. And thank you for the fix all the same!

Thanks,
Eryu

Quoted 36 lines

Best Regards
Yang Xu

Quoted 31 lines
     hard limit 524283904 bytes, expected 524288000
---
  tests/generic/386 | 4 +---
  1 file changed, 1 insertion(+), 3 deletions(-)
tests/generic/386 +1 -3
index 0c44c80e..37e9b943 100755
@@ -62,8 +62,6 @@ _require_scratch
  # both the "df" and the "report" output.  For "report", the line we're
  # interested in contains our project name in the first field.  For "df"
  # it contains our project directory in the last field.
-# But if the device name is too long, the "df" output is broke into two
-# lines, the fourth field is not correct, so take $(NF-2) of "df"
  _filter_quota_rpt() {
  	awk '
  	BEGIN {
@@ -89,7 +87,7 @@ _filter_quota_rpt() {
  			bsize = byte_size($4);
  		} else if ($NF ~ proj_dir) {
  			# this is the "df" output
-			bsize = byte_size($(NF-2));
+			bsize = byte_size($2));
  		} else {
  			next;
  		}