Bug 1979814

+++ This bug was initially created as a clone of Bug #1962515 +++

Description of problem:

Without -a, duplicates listed for remote filesystems under same device (same exported fsid), test case for NFS :


## RHEL7 :

[root@rhel77 ~]# rpm -q coreutils
coreutils-8.22-24.el7.x86_64
[root@rhel77 ~]# mkdir -p /testdir ; mkdir -p /mnt/testdir
[root@rhel77 ~]# mkdir -p /mnt/testdir/share1 ; mkdir -p /mnt/testdir/share2
[root@rhel77 ~]# mount -o bind /mnt/testdir /testdir
[root@rhel77 ~]# mount -o vers=4.1 rhel73:/share /testdir/share1
[root@rhel77 ~]# mount -o vers=4.1 rhel73:/workvol /testdir/share2
[root@rhel77 ~]# df -h -t nfs4
Filesystem       Size  Used Avail Use% Mounted on
rhel73:/share    8.0G  5.4G  2.7G  67% /testdir/share1
rhel73:/workvol  8.0G  5.4G  2.7G  67% /testdir/share2

[root@rhel77 ~]# cat /proc/fs/nfsfs/volumes 
NV SERVER   PORT DEV          FSID                              FSC
v4 ac1701e1  801 0:41         628fdfac389143b8:b4ef233ffd082203 no 
[root@rhel77 ~]# grep "0:41" /proc/self/mountinfo 
88 84 0:41 / /testdir/share1 rw,relatime shared:67 - nfs4 ...
91 39 0:41 / /mnt/testdir/share1 rw,relatime shared:67 - nfs4 ...
89 84 0:41 / /testdir/share2 rw,relatime shared:72 - nfs4 ...
95 39 0:41 / /mnt/testdir/share2 rw,relatime shared:72 - ...


## RHEL8 :

[root@rhel8 ~]# rpm -q coreutils
coreutils-8.30-8.el8.x86_64
[root@rhel8 ~]# mkdir -p /testdir ; mkdir -p /mnt/testdir
[root@rhel8 ~]# mkdir -p /mnt/testdir/share1 ; mkdir -p /mnt/testdir/share2
[root@rhel8 ~]# mount -o bind /mnt/testdir /testdir
[root@rhel8 ~]# mount -o vers=4.1 rhel73:/share /testdir/share1
[root@rhel8 ~]# mount -o vers=4.1 rhel73:/workvol /testdir/share2
[root@rhel8 ~]# df -h -t nfs4
Filesystem                        Size  Used Avail Use% Mounted on
rhel73:/share                     8.0G  5.4G  2.7G  67% /testdir/share1
rhel73:/workvol                   8.0G  5.4G  2.7G  67% /testdir/share2
rhel73:/workvol                   8.0G  5.4G  2.7G  67% /mnt/testdir/share2 <<<--- dupe

[root@rhel8 coreutils]# cat /proc/fs/nfsfs/volumes 
NV SERVER   PORT DEV          FSID                              FSC
v4 ac1701e1  801 0:48         628fdfac389143b8:b4ef233ffd082203 no 
[root@rhel8 coreutils]# grep "0:48" /proc/self/mountinfo 
335 326 0:48 / /testdir/share1 rw,relatime shared:177 - nfs4 ...
338 92 0:48 / /mnt/testdir/share1 rw,relatime shared:177 - nfs4 ...
336 326 0:48 / /testdir/share2 rw,relatime shared:186 - nfs4 ...
356 92 0:48 / /mnt/testdir/share2 rw,relatime shared:186 - nfs4 ...


This is based on a concrete field setup including the bind mount which sort of replicate mount tree.

Difference in behaviour between rhel7 and rhel8 is due to inclusion of this upstream commit :

1c17f61ef df: improve performance with many mount points

Which essentially uses a hash table on filter_mount_list() instead of a linked list.
As a result, when only introduce one entry on the hash table for first filesystem with
same device and we compare "devname" to that, that would be "seen_dev", when using linked 
list we simply compare with the latest element added to the list "device_list", that's why, 
for rhel7, and in this example no dupes are listed, since entries are consecutive.

I've tried this patch to make the behaviour rhel7-like for rhel8, resinserting 
on the table latest element found if already present :

diff --git a/src/df.c b/src/df.c
index 42cfeed2c..d011c6088 100644
--- a/src/df.c
+++ b/src/df.c
@@ -795,12 +795,28 @@ filter_mount_list (bool devices_only)
         {
           /* Add the device number to the device_table.  */
           struct devlist *devlist = xmalloc (sizeof *devlist);
+         union {
+                 const void *ptr;
+                 struct devlist *my_dev;
+          } entry;
+
           devlist->me = me;
           devlist->dev_num = buf.st_dev;
           devlist->next = device_list;
           device_list = devlist;
-          if (hash_insert (devlist_table, devlist) == NULL)
-            xalloc_die ();
+          switch (hash_insert_if_absent (devlist_table, devlist, &entry.ptr))
+            {
+                   case -1:
+                         xalloc_die ();
+                         break;
+                   case 0:
+                         if ((hash_remove(devlist_table, entry.my_dev) == NULL) ||
+                             (hash_insert (devlist_table, devlist) == NULL))
+                                 xalloc_die ();
+                         break;
+                   default:
+                         break;
+           }
 
           me = me->me_next;
         }


Seems to work :

[root@rhel8 coreutils]# /tmp/df -h -t nfs4   <<------- df with above patch
Filesystem       Size  Used Avail Use% Mounted on
rhel73:/share    8.0G  5.4G  2.7G  67% /testdir/share1
rhel73:/workvol  8.0G  5.4G  2.7G  67% /testdir/share2
[root@rhel8 coreutils]# df -h -t nfs4
Filesystem       Size  Used Avail Use% Mounted on
rhel73:/share    8.0G  5.4G  2.7G  67% /testdir/share1
rhel73:/workvol  8.0G  5.4G  2.7G  67% /testdir/share2
rhel73:/workvol  8.0G  5.4G  2.7G  67% /mnt/testdir/share2

But maybe it should be considered a more general case and modify devlist_compare and/or devlist_hash
to consider devname for hashing comparing purposes


Version-Release number of selected component (if applicable): 8.3, coreutils-8.30-8.el8.x86_64


How reproducible: 100%


Steps to Reproduce: see above


--- Additional comment from Kamil Dudka on 2021-07-07 09:22:27 CEST ---

upstream commit: https://git.savannah.gnu.org/gitweb/?p=coreutils.git;a=commitdiff;h=v8.32-169-gd6125af095c