[LU-5690] Unable to parse the mdt.nosquash_nids parameter when using commas in expr_list

• Type: Bug
• Resolution: Fixed
• Priority: Minor
• Fix Version/s: Lustre 2.8.0
• Affects Version/s: Lustre 2.4.3, Lustre 2.5.3
• Environment: Lustre 2.4.3 w/ patch LU-4460
• Severity: 2
• Assignee: Jian Yu
• Reporter: Bruno Travouillon (Inactive)

Description

One of our support engineer is reporting that Lustre is unable to start the MDS when using a comma in the setting of mdt.nosquash_nids.

We can find the LNET NID range syntax in the documentation (23.2.3. Tips on Using Root Squash). Setting mdt.nosquash_nids=192.168.0.[13,14]@tcp0 should be valid.

<nidlist>     :== <nidrange> [ ' ' <nidrange> ]
<nidrange>   :== <addrrange> '@' <net>
<addrrange>  :== '*' |
           <ipaddr_range> |
           <numaddr_range>
<ipaddr_range>       :==
<numaddr_range>.<numaddr_range>.<numaddr_range>.<numaddr_range>
<numaddr_range>      :== <number> |
                   <expr_list>
<expr_list>  :== '[' <range_expr> [ ',' <range_expr>] ']'
<range_expr> :== <number> |
           <number> '-' <number> |
           <number> '-' <number> '/' <number>
<net>        :== <netname> | <netname><number>
<netname>    :== "lo" | "tcp" | "o2ib" | "cib" | "openib" | "iib" | 
           "vib" | "ra" | "elan" | "gm" | "mx" | "ptl"
<number>     :== <nonnegative decimal> | <hexadecimal>

I was able to reproduce this issue on my test lab. Here are the traces.

[root@mds1 ~]# tunefs.lustre --erase-params --param mgsnode=192.168.122.41@tcp --param lov.stripecount=1 --param lov.stripesize=1048576 --param  "mdt.nosquash_nids=192.168.0.[13,14]@tcp0" /dev/vdb checking for existing Lustre data: found
Reading CONFIGS/mountdata

   Read previous values:
Target:     scratch-MDT0000
Index:      0
Lustre FS:  scratch
Mount type: ldiskfs
Flags:      0x1
              (MDT )
Persistent mount opts: user_xattr,errors=remount-ro
Parameters: mgsnode=192.168.122.41@tcp lov.stripecount=1 lov.stripesize=1048576 mdt.nosquash_nids=192.168.0.13@tcp0


   Permanent disk data:
Target:     scratch-MDT0000
Index:      0
Lustre FS:  scratch
Mount type: ldiskfs
Flags:      0x41
              (MDT update )
Persistent mount opts: user_xattr,errors=remount-ro
Parameters: mgsnode=192.168.122.41@tcp lov.stripecount=1 lov.stripesize=1048576 mdt.nosquash_nids=192.168.0.[13,14]@tcp0

Writing CONFIGS/mountdata
[root@mds1 ~]# mount -v -t lustre /dev/vdb /mnt/fs/scratch/mdt/
arg[0] = /sbin/mount.lustre
arg[1] = -v
arg[2] = -o
arg[3] = rw
arg[4] = /dev/vdb
arg[5] = /mnt/fs/scratch/mdt/
source = /dev/vdb (/dev/vdb), target = /mnt/fs/scratch/mdt
options = rw
checking for existing Lustre data: found
Reading CONFIGS/mountdata
mounting device /dev/vdb at /mnt/fs/scratch/mdt, flags=0x1000000 options=osd=osd-ldiskfs,user_xattr,errors=remount-ro,mgsnode=192.168.122.41@tcp,param=mgsnode=192.168.122.41@tcp,param=lov.stripecount=1,param=lov.stripesize=1048576,param=mdt.nosquash_nids=192.168.0.[13,14]@tcp0,svname=scratch-MDT0000,device=/dev/vdb
mount.lustre: mount /dev/vdb at /mnt/fs/scratch/mdt failed: Invalid argument retries left: 0
mount.lustre: mount /dev/vdb at /mnt/fs/scratch/mdt failed: Invalid argument
This may have multiple causes.
Are the mount options correct?
Check the syslog for more info.
[root@mds1 ~]# dmesg |tail
LDISKFS-fs (vdb): barriers disabled
LDISKFS-fs (vdb): mounted filesystem with ordered data mode. quota=on. Opts:
LDISKFS-fs (vdb): barriers disabled
LDISKFS-fs (vdb): mounted filesystem with ordered data mode. quota=on. Opts:
LDISKFS-fs (vdb): Unrecognized mount option "14]@tcp0" or missing value   <=== NOT GOOD :-(
LustreError: 8846:0:(osd_handler.c:5481:osd_mount()) scratch-MDT0000-osd: can't mount /dev/vdb: -22
LustreError: 8846:0:(obd_config.c:572:class_setup()) setup scratch-MDT0000-osd failed (-22)
LustreError: 8846:0:(obd_mount.c:201:lustre_start_simple()) scratch-MDT0000-osd setup error -22
LustreError: 8846:0:(obd_mount_server.c:1686:server_fill_super()) Unable to start osd on /dev/vdb: -22
LustreError: 8846:0:(obd_mount.c:1324:lustre_fill_super()) Unable to mount  (-22)

In the debug trace:

00000020:01000004:0.0:1411043156.536717:0:8846:0:(obd_mount.c:839:lmd_print()) options: user_xattr,errors=remount-ro,14]@tcp0

Obviously, something is going wrong in the way the parameters are parsed.

[root@mds1 ~]# mount -t ldiskfs /dev/vdb /mnt/fs/scratch/mdt/
[root@mds1 ~]# hexdump -C /mnt/fs/scratch/mdt/CONFIGS/mountdata
00000000  01 00 d0 1d 00 00 00 00  00 00 00 00 00 00 00 00  |................|
00000010  01 00 00 00 41 00 00 00  00 00 00 00 01 00 00 00  |....A...........|
00000020  73 63 72 61 74 63 68 00  00 00 00 00 00 00 00 00  |scratch.........|
00000030  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00000060  73 63 72 61 74 63 68 2d  4d 44 54 30 30 30 30 00  |scratch-MDT0000.|
00000070  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00001000  75 73 65 72 5f 78 61 74  74 72 2c 65 72 72 6f 72  |user_xattr,error|
00001010  73 3d 72 65 6d 6f 75 6e  74 2d 72 6f 00 00 00 00  |s=remount-ro....|
00001020  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00002000  20 6d 67 73 6e 6f 64 65  3d 31 39 32 2e 31 36 38  | mgsnode=192.168|
00002010  2e 31 32 32 2e 34 31 40  74 63 70 20 6c 6f 76 2e  |.122.41@tcp lov.|
00002020  73 74 72 69 70 65 63 6f  75 6e 74 3d 31 20 6c 6f  |stripecount=1 lo|
00002030  76 2e 73 74 72 69 70 65  73 69 7a 65 3d 31 30 34  |v.stripesize=104|
00002040  38 35 37 36 20 6d 64 74  2e 6e 6f 73 71 75 61 73  |8576 mdt.nosquas|
00002050  68 5f 6e 69 64 73 3d 31  39 32 2e 31 36 38 2e 30  |h_nids=192.168.0|
00002060  2e 5b 31 33 2c 31 34 5d  40 74 63 70 30 00 5d 40  |.[13,14]@tcp0.]@|
00002070  74 63 70 00 00 00 00 00  00 00 00 00 00 00 00 00  |tcp.............|
00002080  00 00 00 00 00 00 00 00  00 00 00 00 00 00 00 00  |................|
*
00003000

The flaw is in the way lmd_parse() parses the param flags in lustre/obdclass/obd_mount.c (source from Lustre 2.5.3):

1143         } else if (strncmp(s1, "param=", 6) == 0) {
1144             int length;
1145             char *tail = strchr(s1 + 6, ',');
1146             if (tail == NULL) {
1147                 length = strlen(s1);
1148             } else {
1149                 lnet_nid_t nid;
1150                 char      *param_str = tail + 1;
1151                 int        supplementary = 1;
1152 
1153                 while (class_parse_nid_quiet(param_str, &nid,
1154                                  &param_str) == 0) {
1155                     supplementary = 0;
1156                 }
1157                 length = param_str - s1 - supplementary;
1158             }
1159             length -= 6;
1160             strncat(lmd->lmd_params, s1 + 6, length);
1161             strcat(lmd->lmd_params, " ");
1162             s3 = s1 + 6 + length;
1163             clear++;

This part of code has been introduced by LU-4460. However, reverting the patch is not enough as the parsing will still fail because of line 1145 (introduced in commit 6a65619724).

A workaround is to use mdt.nosquash_nids='192.168.0.13@tcp0 192.168.0.14@tcp0' (or of course 192.168.0.[13-14]@tcp0 here, but this is an example )

Issue Links

• is related to: LU-8311 Target does not mount with the new mgsnode parameter format in case of multirail configuration
• is related to: LU-7566 mounting lustre emits bogus parse_nidrange() errors to the console