9 LUSTRE=${LUSTRE:-`dirname $0`/..}
10 UPCALL=${UPCALL:-$PWD/recovery-small-upcall.sh}
11 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/local.sh}
20 # Allow us to override the setup if we already have a mounted system by
21 # setting SETUP=" " and CLEANUP=" "
22 SETUP=${SETUP:-"setup"}
23 CLEANUP=${CLEANUP:-"cleanup"}
28 add_mds mds --dev $MDSDEV --size $MDSSIZE
29 add_lov lov1 mds --stripe_sz $STRIPE_BYTES\
30 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
31 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
32 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
33 add_client client mds --lov lov1 --path $MOUNT
38 start ost --reformat $OSTLCONFARGS
39 start ost2 --reformat $OSTLCONFARGS
40 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
41 start mds $MDSLCONFARGS --reformat
42 zconf_mount `hostname` $MOUNT
46 zconf_umount `hostname` $MOUNT
47 stop mds ${FORCE} $MDSLCONFARGS
48 stop ost2 ${FORCE} --dump cleanup.log
49 stop ost ${FORCE} --dump cleanup.log
54 do_mds 'echo -e "device \$mds1\\nprobe\\nnotransno\\nreadonly" | lctl'
59 do_client "df -h $MOUNT" # trigger failover, if we haven't already
62 if [ ! -z "$EVAL" ]; then
67 if [ "$ONLY" == "cleanup" ]; then
68 sysctl -w portals.debug=0 || true
73 REFORMAT=--reformat $SETUP
77 drop_request "mcreate $MOUNT/1" || return 1
78 drop_reply "mcreate $MOUNT/2" || return 2
80 run_test 1 "mcreate: drop req, drop rep"
83 drop_request "tchmod 111 $MOUNT/2" || return 1
84 drop_reply "tchmod 666 $MOUNT/2" || return 2
86 run_test 2 "chmod: drop req, drop rep"
89 drop_request "statone $MOUNT/2" || return 1
90 drop_reply "statone $MOUNT/2" || return 2
92 run_test 3 "stat: drop req, drop rep"
95 do_facet client "cp /etc/resolv.conf $MOUNT/resolv.conf" || return 1
96 drop_request "cat $MOUNT/resolv.conf > /dev/null" || return 2
97 drop_reply "cat $MOUNT/resolv.conf > /dev/null" || return 3
99 run_test 4 "open: drop req, drop rep"
102 drop_request "mv $MOUNT/resolv.conf $MOUNT/renamed" || return 1
103 drop_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
104 do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
106 run_test 5 "rename: drop req, drop rep"
109 drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
110 drop_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
112 run_test 6 "link: drop req, drop rep"
115 drop_request "munlink $MOUNT/link1" || return 1
116 drop_reply "munlink $MOUNT/link2" || return 2
118 run_test 7 "unlink: drop req, drop rep"
123 drop_reply "touch $MOUNT/renamed" || return 1
125 run_test 8 "touch: drop rep (bug 1423)"
130 pause_bulk "cp /etc/profile $MOUNT" || return 1
131 do_facet client "cp /etc/termcap $MOUNT" || return 2
132 do_facet client "sync"
133 do_facet client "rm $MOUNT/termcap $MOUNT/profile" || return 3
135 run_test 9 "pause bulk on OST (bug 1420)"
139 do_facet client mcreate $MOUNT/f10 || return 1
140 drop_bl_callback "chmod 0777 $MOUNT/f10" || return 2
141 # wait for the mds to evict the client
142 #echo "sleep $(($TIMEOUT*2))"
143 #sleep $(($TIMEOUT*2))
144 do_facet client touch $MOUNT/f10 || echo "touch failed, evicted"
145 do_facet client checkstat -v -p 0777 $MOUNT/f10 || return 3
146 do_facet client "munlink $MOUNT/f10"
148 run_test 10 "finish request on server after client eviction (bug 1521)"
151 # wake up a thead waiting for completion after eviction
153 do_facet client multiop $MOUNT/$tfile Ow || return 1
154 do_facet client multiop $MOUNT/$tfile or || return 2
158 do_facet client multiop $MOUNT/$tfile or || return 3
159 drop_bl_callback multiop $MOUNT/$tfile Ow ||
160 echo "client evicted, as expected"
162 do_facet client munlink $MOUNT/$tfile || return 4
164 run_test 11 "wake up a thead waiting for completion after eviction (b=2460)"
168 $LCTL mark multiop $MOUNT/$tfile OS_c
169 do_facet mds "sysctl -w lustre.fail_loc=0x115"
170 clear_failloc mds $((TIMEOUT * 2)) &
171 multiop $MOUNT/$tfile OS_c &
173 #define OBD_FAIL_MDS_CLOSE_NET 0x115
176 echo "waiting for multiop $PID"
177 wait $PID || return 2
178 do_facet client munlink $MOUNT/$tfile || return 3
180 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
182 # Bug 113, check that readdir lost recv timeout works.
184 mkdir /mnt/lustre/readdir
185 touch /mnt/lustre/readdir/newentry
186 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
187 do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
188 ls /mnt/lustre/readdir || return 1
189 do_facet mds "sysctl -w lustre.fail_loc=0"
190 rm -rf /mnt/lustre/readdir
192 run_test 13 "mdc_readpage restart test (bug 1138)"
194 # Bug 113, check that readdir lost send timeout works.
196 mkdir /mnt/lustre/readdir
197 touch /mnt/lustre/readdir/newentry
198 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
199 do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
200 ls /mnt/lustre/readdir || return 1
201 do_facet mds "sysctl -w lustre.fail_loc=0"
203 run_test 14 "mdc_readpage resend test (bug 1138)"
206 do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
207 touch $DIR/$tfile && return 1
210 run_test 15 "failed open (-ENOMEM)"
213 # OBD_FAIL_PTLRPC_BULK_PUT_NET | OBD_FAIL_ONCE
214 do_facet client cp /etc/termcap $MOUNT
217 sysctl -w lustre.fail_loc=0x80000504
219 # wil get evicted here
220 do_facet client "diff /etc/termcap $MOUNT/termcap" && return 1
221 sysctl -w lustre.fail_loc=0
222 do_facet client "diff /etc/termcap $MOUNT/termcap" || return 2
225 run_test 16 "timeout bulk put, evict client (2732)"
228 # OBD_FAIL_PTLRPC_BULK_GET_NET | OBD_FAIL_ONCE
229 # wil get evicted here
230 sysctl -w lustre.fail_loc=0x80000503
231 do_facet client cp /etc/termcap $MOUNT && return 1
233 do_facet client "diff /etc/termcap $MOUNT/termcap" && return 1
234 sysctl -w lustre.fail_loc=0
235 do_facet client "diff /etc/termcap $MOUNT/termcap" || return 2
238 run_test 17 "timeout bulk get, evict client (2732)"
241 # OBD_FAIL_PTLRPC_BULK_PUT_NET|OBD_FAIL_ONCE
242 do_facet client mkdir -p $MOUNT/$tdir
243 f=$MOUNT/$tdir/$tfile
244 f2=$MOUNT/$tdir/${tfile}-2
247 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
248 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
249 echo there is still data in page cache $a ?
255 # shouldn't have to set stripe size of count==1
256 lfs setstripe $f $((128 * 1024)) 0 1
257 lfs setstripe $f2 $((128 * 1024)) 0 1
259 do_facet client cp /etc/termcap $f
261 # just use this write to trigger the client's eviction from the ost
262 sysctl -w lustre.fail_loc=0x80000503
263 do_facet client dd if=/dev/zero of=$f2 bs=4k count=1
265 sysctl -w lustre.fail_loc=0
266 # allow recovery to complete
268 # my understanding is that there should be nothing in the page
269 # cache after the client reconnects?
270 for a in /proc/fs/lustre/llite/*/dump_page_cache; do
271 if [ `wc -l $a | awk '{print $1}'` -gt 1 ]; then
272 echo there is still data in page cache $a ?
278 run_test 18 "eviction and reconnect clears page cache (2766)"