9 LUSTRE=${LUSTRE:-`dirname $0`/..}
11 . $LUSTRE/tests/test-framework.sh
15 . ${CONFIG:=$LUSTRE/tests/cfg/lmv.sh}
21 # Allow us to override the setup if we already have a mounted system by
22 # setting SETUP=" " and CLEANUP=" "
23 SETUP=${SETUP:-"setup"}
24 CLEANUP=${CLEANUP:-"cleanup"}
29 if [ "$MDSCOUNT" -gt 1 ]; then
31 for mds in `mds_list`; do
32 MDSDEV=$TMP/${mds}-`hostname`
33 add_mds $mds --dev $MDSDEV --size $MDSSIZE --lmv lmv1_svc
35 add_lov_to_lmv lov1 lmv1_svc --stripe_sz $STRIPE_BYTES \
36 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
39 add_mds mds1 --dev $MDSDEV --size $MDSSIZE
40 add_lov lov1 mds1 --stripe_sz $STRIPE_BYTES \
41 --stripe_cnt $STRIPES_PER_OBJ --stripe_pattern 0
46 add_ost ost --lov lov1 --dev $OSTDEV --size $OSTSIZE
47 add_ost ost2 --lov lov1 --dev ${OSTDEV}-2 --size $OSTSIZE
48 add_client client ${MDS} --lov lov1 --path $MOUNT
53 start_krb5_kdc || exit 1
54 start ost --reformat $OSTLCONFARGS
55 start ost2 --reformat $OSTLCONFARGS
56 start_lsvcgssd || exit 2
58 [ "$DAEMONFILE" ] && $LCTL debug_daemon start $DAEMONFILE $DAEMONSIZE
59 for mds in `mds_list`; do
60 start $mds --reformat $MDSLCONFARGS
62 grep " $MOUNT " /proc/mounts || zconf_mount `hostname` $MOUNT
66 zconf_umount `hostname` $MOUNT
67 for mds in `mds_list`; do
68 stop $mds ${FORCE} $MDSLCONFARGS
72 stop ost2 ${FORCE} --dump cleanup.log
73 stop ost ${FORCE} --dump cleanup.log
76 if [ ! -z "$EVAL" ]; then
81 if [ "$ONLY" == "setup" ]; then
86 if [ "$ONLY" == "cleanup" ]; then
87 sysctl -w portals.debug=0 || true
92 REFORMAT=--reformat $SETUP
96 drop_request "mcreate $MOUNT/1" || return 1
97 drop_reint_reply "mcreate $MOUNT/2" || return 2
99 run_test 1 "mcreate: drop req, drop rep"
102 drop_request "tchmod 111 $MOUNT/2" || return 1
103 drop_reint_reply "tchmod 666 $MOUNT/2" || return 2
105 run_test 2 "chmod: drop req, drop rep"
108 drop_request "statone $MOUNT/2" || return 1
109 drop_reply "statone $MOUNT/2" || return 2
111 run_test 3 "stat: drop req, drop rep"
114 do_facet client "cp /etc/resolv.conf $MOUNT/resolv.conf" || return 1
115 drop_request "cat $MOUNT/resolv.conf > /dev/null" || return 2
116 drop_reply "cat $MOUNT/resolv.conf > /dev/null" || return 3
118 run_test 4 "open: drop req, drop rep"
121 drop_request "mv $MOUNT/resolv.conf $MOUNT/renamed" || return 1
122 drop_reint_reply "mv $MOUNT/renamed $MOUNT/renamed-again" || return 2
123 do_facet client "checkstat -v $MOUNT/renamed-again" || return 3
125 run_test 5 "rename: drop req, drop rep"
128 drop_request "mlink $MOUNT/renamed-again $MOUNT/link1" || return 1
129 drop_reint_reply "mlink $MOUNT/renamed-again $MOUNT/link2" || return 2
131 run_test 6 "link: drop req, drop rep"
134 drop_request "munlink $MOUNT/link1" || return 1
135 drop_reint_reply "munlink $MOUNT/link2" || return 2
137 run_test 7 "unlink: drop req, drop rep"
141 drop_reint_reply "touch $MOUNT/renamed" || return 1
143 run_test 8 "touch: drop rep (bug 1423)"
147 pause_bulk "cp /etc/profile $MOUNT" || return 1
148 do_facet client "cp /etc/termcap $MOUNT" || return 2
149 do_facet client "sync"
150 do_facet client "rm $MOUNT/termcap $MOUNT/profile" || return 3
152 run_test 9 "pause bulk on OST (bug 1420)"
156 do_facet client mcreate $MOUNT/f10 || return 1
157 drop_bl_callback "chmod 0777 $MOUNT/f10" || return 2
158 # wait for the mds to evict the client
159 #echo "sleep $(($TIMEOUT*2))"
160 #sleep $(($TIMEOUT*2))
161 do_facet client touch $MOUNT/f10 || echo "touch failed, evicted"
162 do_facet client checkstat -v -p 0777 $MOUNT/f10 || return 3
163 do_facet client "munlink $MOUNT/f10"
165 run_test 10 "finish request on server after client eviction (bug 1521)"
168 # wake up a thead waiting for completion after eviction
170 do_facet client multiop $MOUNT/$tfile Ow || return 1
171 do_facet client multiop $MOUNT/$tfile or || return 2
175 do_facet client multiop $MOUNT/$tfile or || return 3
176 drop_bl_callback multiop $MOUNT/$tfile Ow ||
177 echo "client evicted, as expected"
179 do_facet client munlink $MOUNT/$tfile || return 4
181 run_test 11 "wake up a thead waiting for completion after eviction (b=2460)"
185 $LCTL mark multiop $MOUNT/$tfile OS_c
186 do_facet mds "sysctl -w lustre.fail_loc=0x115"
187 clear_failloc mds $((TIMEOUT * 2)) &
188 multiop $MOUNT/$tfile OS_c &
190 #define OBD_FAIL_MDS_CLOSE_NET 0x115
193 echo "waiting for multiop $PID"
194 wait $PID || return 2
195 do_facet client munlink $MOUNT/$tfile || return 3
197 run_test 12 "recover from timed out resend in ptlrpcd (b=2494)"
199 # Bug 113, check that readdir lost recv timeout works.
201 mkdir /mnt/lustre/readdir
202 touch /mnt/lustre/readdir/newentry
203 # OBD_FAIL_MDS_READPAGE_NET|OBD_FAIL_ONCE
204 do_facet mds "sysctl -w lustre.fail_loc=0x80000104"
205 ls /mnt/lustre/readdir || return 1
206 do_facet mds "sysctl -w lustre.fail_loc=0"
207 rm -rf /mnt/lustre/readdir
209 run_test 13 "mdc_readpage restart test (bug 1138)"
211 # Bug 113, check that readdir lost send timeout works.
213 mkdir /mnt/lustre/readdir
214 touch /mnt/lustre/readdir/newentry
215 # OBD_FAIL_MDS_SENDPAGE|OBD_FAIL_ONCE
216 do_facet mds "sysctl -w lustre.fail_loc=0x80000106"
217 ls /mnt/lustre/readdir || return 1
218 do_facet mds "sysctl -w lustre.fail_loc=0"
220 run_test 14 "mdc_readpage resend test (bug 1138)"
223 do_facet mds "sysctl -w lustre.fail_loc=0x80000128"
224 touch $DIR/$tfile && return 1
227 run_test 15 "failed open (-ENOMEM)"
230 for f in /proc/fs/lustre/llite/*/read_ahead; do
236 for f in /proc/fs/lustre/llite/*/read_ahead; do
242 do_facet client cp /etc/termcap $MOUNT
246 #define OBD_FAIL_PTLRPC_BULK_PUT_NET 0x504 | OBD_FAIL_ONCE
247 sysctl -w lustre.fail_loc=0x80000504
249 # will get evicted here
250 do_facet client "cmp /etc/termcap $MOUNT/termcap" && return 1
251 sysctl -w lustre.fail_loc=0
252 # give recovery a chance to finish (shouldn't take long)
254 do_facet client "cmp /etc/termcap $MOUNT/termcap" || return 2
257 run_test 16 "timeout bulk put, evict client (2732)"
260 # OBD_FAIL_PTLRPC_BULK_GET_NET 0x0503 | OBD_FAIL_ONCE
261 # client will get evicted here
262 sysctl -w lustre.fail_loc=0x80000503
263 do_facet client cp /etc/termcap $DIR/$tfile
266 sysctl -w lustre.fail_loc=0
267 do_facet client "df $DIR"
269 do_facet client "cmp /etc/termcap $DIR/$tfile" && return 1
270 do_facet client "rm $DIR/$tfile" || return 2
273 run_test 17 "timeout bulk get, evict client (2732)"
276 do_facet client mkdir -p $MOUNT/$tdir
277 f=$MOUNT/$tdir/$tfile
280 pgcache_empty || return 1
283 lfs setstripe $f $((128 * 1024)) 1 1
285 do_facet client cp /etc/termcap $f
287 local osc2_dev=`$LCTL device_list | \
288 awk '(/ost2.*client_facet/){print $4}' `
289 $LCTL --device %$osc2_dev deactivate
290 # my understanding is that there should be nothing in the page
291 # cache after the client reconnects?
293 pgcache_empty || rc=2
294 $LCTL --device %$osc2_dev activate
298 run_test 18a "manual ost invalidate clears page cache immediately"
301 # OBD_FAIL_PTLRPC_BULK_PUT_NET|OBD_FAIL_ONCE
302 do_facet client mkdir -p $MOUNT/$tdir
303 f=$MOUNT/$tdir/$tfile
304 f2=$MOUNT/$tdir/${tfile}-2
307 pgcache_empty || return 1
309 # shouldn't have to set stripe size of count==1
310 lfs setstripe $f $((128 * 1024)) 0 1
311 lfs setstripe $f2 $((128 * 1024)) 0 1
313 do_facet client cp /etc/termcap $f
315 # just use this write to trigger the client's eviction from the ost
316 sysctl -w lustre.fail_loc=0x80000503
317 do_facet client dd if=/dev/zero of=$f2 bs=4k count=1
319 sysctl -w lustre.fail_loc=0
320 # allow recovery to complete
321 sleep $((TIMEOUT + 2))
322 # my understanding is that there should be nothing in the page
323 # cache after the client reconnects?
325 pgcache_empty || rc=2
329 run_test 18b "eviction and reconnect clears page cache (2766)"
333 do_facet client mcreate $f || return 1
334 drop_ldlm_cancel "chmod 0777 $f" || echo evicted
336 do_facet client checkstat -v -p 0777 $f || echo evicted
337 # let the client reconnect
339 do_facet client "munlink $f"
341 run_test 19a "test expired_lock_main on mds (2867)"
345 do_facet client multiop $f Ow || return 1
346 do_facet client multiop $f or || return 2
350 do_facet client multiop $f or || return 3
351 drop_ldlm_cancel multiop $f Ow || echo "client evicted, as expected"
353 do_facet client munlink $f || return 4
355 run_test 19b "test expired_lock_main on ost (2867)"
357 test_20a() { # bug 2983 - ldlm_handle_enqueue cleanup
359 multiop $DIR/$tdir/${tfile} O_wc &
363 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
364 do_facet ost sysctl -w lustre.fail_loc=0x80000308
365 kill -USR1 $MULTI_PID
368 [ $rc -eq 0 ] && error "multiop didn't fail enqueue: rc $rc" || true
370 run_test 20a "ldlm_handle_enqueue error (should return error)"
372 test_20b() { # bug 2986 - ldlm_handle_enqueue error during open
374 touch $DIR/$tdir/${tfile}
376 #define OBD_FAIL_LDLM_ENQUEUE_EXTENT_ERR 0x308
377 do_facet ost sysctl -w lustre.fail_loc=0x80000308
378 dd if=/etc/hosts of=$DIR/$tdir/$tfile && \
379 error "didn't fail open enqueue" || true
381 run_test 20b "ldlm_handle_enqueue error (should return error)"