Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
Merge tag 'ceph-for-4.11-rc2' of git://github.com/ceph/ceph-client
Pull ceph fixes from Ilya Dryomov:
- a fix for the recently discovered misdirected requests bug present in
jewel and later on the server side and all stable kernels
- a fixup for -rc1 CRUSH changes
- two usability enhancements: osd_request_timeout option and
supported_features bus attribute.
* tag 'ceph-for-4.11-rc2' of git://github.com/ceph/ceph-client:
libceph: osd_request_timeout option
rbd: supported_features bus attribute
libceph: don't set weight to IN when OSD is destroyed
libceph: fix crush_decode() for older maps
+66
-8
+12
-4
drivers/block/rbd.c
+12
-4
drivers/block/rbd.c
···
120
120
121
121
/* Feature bits */
122
122
123
-
#define RBD_FEATURE_LAYERING (1<<0)
124
-
#define RBD_FEATURE_STRIPINGV2 (1<<1)
125
-
#define RBD_FEATURE_EXCLUSIVE_LOCK (1<<2)
126
-
#define RBD_FEATURE_DATA_POOL (1<<7)
123
+
#define RBD_FEATURE_LAYERING (1ULL<<0)
124
+
#define RBD_FEATURE_STRIPINGV2 (1ULL<<1)
125
+
#define RBD_FEATURE_EXCLUSIVE_LOCK (1ULL<<2)
126
+
#define RBD_FEATURE_DATA_POOL (1ULL<<7)
127
+
127
128
#define RBD_FEATURES_ALL (RBD_FEATURE_LAYERING | \
128
129
RBD_FEATURE_STRIPINGV2 | \
129
130
RBD_FEATURE_EXCLUSIVE_LOCK | \
···
500
499
return is_lock_owner;
501
500
}
502
501
502
+
static ssize_t rbd_supported_features_show(struct bus_type *bus, char *buf)
503
+
{
504
+
return sprintf(buf, "0x%llx\n", RBD_FEATURES_SUPPORTED);
505
+
}
506
+
503
507
static BUS_ATTR(add, S_IWUSR, NULL, rbd_add);
504
508
static BUS_ATTR(remove, S_IWUSR, NULL, rbd_remove);
505
509
static BUS_ATTR(add_single_major, S_IWUSR, NULL, rbd_add_single_major);
506
510
static BUS_ATTR(remove_single_major, S_IWUSR, NULL, rbd_remove_single_major);
511
+
static BUS_ATTR(supported_features, S_IRUGO, rbd_supported_features_show, NULL);
507
512
508
513
static struct attribute *rbd_bus_attrs[] = {
509
514
&bus_attr_add.attr,
510
515
&bus_attr_remove.attr,
511
516
&bus_attr_add_single_major.attr,
512
517
&bus_attr_remove_single_major.attr,
518
+
&bus_attr_supported_features.attr,
513
519
NULL,
514
520
};
515
521
+2
include/linux/ceph/libceph.h
+2
include/linux/ceph/libceph.h
···
48
48
unsigned long mount_timeout; /* jiffies */
49
49
unsigned long osd_idle_ttl; /* jiffies */
50
50
unsigned long osd_keepalive_timeout; /* jiffies */
51
+
unsigned long osd_request_timeout; /* jiffies */
51
52
52
53
/*
53
54
* any type that can't be simply compared or doesn't need need
···
69
68
#define CEPH_MOUNT_TIMEOUT_DEFAULT msecs_to_jiffies(60 * 1000)
70
69
#define CEPH_OSD_KEEPALIVE_DEFAULT msecs_to_jiffies(5 * 1000)
71
70
#define CEPH_OSD_IDLE_TTL_DEFAULT msecs_to_jiffies(60 * 1000)
71
+
#define CEPH_OSD_REQUEST_TIMEOUT_DEFAULT 0 /* no timeout */
72
72
73
73
#define CEPH_MONC_HUNT_INTERVAL msecs_to_jiffies(3 * 1000)
74
74
#define CEPH_MONC_PING_INTERVAL msecs_to_jiffies(10 * 1000)
+1
include/linux/ceph/osd_client.h
+1
include/linux/ceph/osd_client.h
+15
net/ceph/ceph_common.c
+15
net/ceph/ceph_common.c
···
230
230
Opt_osdkeepalivetimeout,
231
231
Opt_mount_timeout,
232
232
Opt_osd_idle_ttl,
233
+
Opt_osd_request_timeout,
233
234
Opt_last_int,
234
235
/* int args above */
235
236
Opt_fsid,
···
257
256
{Opt_osdkeepalivetimeout, "osdkeepalive=%d"},
258
257
{Opt_mount_timeout, "mount_timeout=%d"},
259
258
{Opt_osd_idle_ttl, "osd_idle_ttl=%d"},
259
+
{Opt_osd_request_timeout, "osd_request_timeout=%d"},
260
260
/* int args above */
261
261
{Opt_fsid, "fsid=%s"},
262
262
{Opt_name, "name=%s"},
···
363
361
opt->osd_keepalive_timeout = CEPH_OSD_KEEPALIVE_DEFAULT;
364
362
opt->mount_timeout = CEPH_MOUNT_TIMEOUT_DEFAULT;
365
363
opt->osd_idle_ttl = CEPH_OSD_IDLE_TTL_DEFAULT;
364
+
opt->osd_request_timeout = CEPH_OSD_REQUEST_TIMEOUT_DEFAULT;
366
365
367
366
/* get mon ip(s) */
368
367
/* ip1[:port1][,ip2[:port2]...] */
···
476
473
}
477
474
opt->mount_timeout = msecs_to_jiffies(intval * 1000);
478
475
break;
476
+
case Opt_osd_request_timeout:
477
+
/* 0 is "wait forever" (i.e. infinite timeout) */
478
+
if (intval < 0 || intval > INT_MAX / 1000) {
479
+
pr_err("osd_request_timeout out of range\n");
480
+
err = -EINVAL;
481
+
goto out;
482
+
}
483
+
opt->osd_request_timeout = msecs_to_jiffies(intval * 1000);
484
+
break;
479
485
480
486
case Opt_share:
481
487
opt->flags &= ~CEPH_OPT_NOSHARE;
···
569
557
if (opt->osd_keepalive_timeout != CEPH_OSD_KEEPALIVE_DEFAULT)
570
558
seq_printf(m, "osdkeepalivetimeout=%d,",
571
559
jiffies_to_msecs(opt->osd_keepalive_timeout) / 1000);
560
+
if (opt->osd_request_timeout != CEPH_OSD_REQUEST_TIMEOUT_DEFAULT)
561
+
seq_printf(m, "osd_request_timeout=%d,",
562
+
jiffies_to_msecs(opt->osd_request_timeout) / 1000);
572
563
573
564
/* drop redundant comma */
574
565
if (m->count != pos)
+35
-1
net/ceph/osd_client.c
+35
-1
net/ceph/osd_client.c
···
1709
1709
1710
1710
req->r_flags |= CEPH_OSD_FLAG_ONDISK;
1711
1711
atomic_inc(&req->r_osdc->num_requests);
1712
+
1713
+
req->r_start_stamp = jiffies;
1712
1714
}
1713
1715
1714
1716
static void submit_request(struct ceph_osd_request *req, bool wrlocked)
···
1789
1787
finish_request(req);
1790
1788
complete_all(&req->r_completion);
1791
1789
ceph_osdc_put_request(req);
1790
+
}
1791
+
1792
+
static void abort_request(struct ceph_osd_request *req, int err)
1793
+
{
1794
+
dout("%s req %p tid %llu err %d\n", __func__, req, req->r_tid, err);
1795
+
1796
+
cancel_map_check(req);
1797
+
complete_request(req, err);
1792
1798
}
1793
1799
1794
1800
static void check_pool_dne(struct ceph_osd_request *req)
···
2497
2487
container_of(work, struct ceph_osd_client, timeout_work.work);
2498
2488
struct ceph_options *opts = osdc->client->options;
2499
2489
unsigned long cutoff = jiffies - opts->osd_keepalive_timeout;
2490
+
unsigned long expiry_cutoff = jiffies - opts->osd_request_timeout;
2500
2491
LIST_HEAD(slow_osds);
2501
2492
struct rb_node *n, *p;
2502
2493
···
2513
2502
struct ceph_osd *osd = rb_entry(n, struct ceph_osd, o_node);
2514
2503
bool found = false;
2515
2504
2516
-
for (p = rb_first(&osd->o_requests); p; p = rb_next(p)) {
2505
+
for (p = rb_first(&osd->o_requests); p; ) {
2517
2506
struct ceph_osd_request *req =
2518
2507
rb_entry(p, struct ceph_osd_request, r_node);
2508
+
2509
+
p = rb_next(p); /* abort_request() */
2519
2510
2520
2511
if (time_before(req->r_stamp, cutoff)) {
2521
2512
dout(" req %p tid %llu on osd%d is laggy\n",
2522
2513
req, req->r_tid, osd->o_osd);
2523
2514
found = true;
2515
+
}
2516
+
if (opts->osd_request_timeout &&
2517
+
time_before(req->r_start_stamp, expiry_cutoff)) {
2518
+
pr_err_ratelimited("tid %llu on osd%d timeout\n",
2519
+
req->r_tid, osd->o_osd);
2520
+
abort_request(req, -ETIMEDOUT);
2524
2521
}
2525
2522
}
2526
2523
for (p = rb_first(&osd->o_linger_requests); p; p = rb_next(p)) {
···
2547
2528
2548
2529
if (found)
2549
2530
list_move_tail(&osd->o_keepalive_item, &slow_osds);
2531
+
}
2532
+
2533
+
if (opts->osd_request_timeout) {
2534
+
for (p = rb_first(&osdc->homeless_osd.o_requests); p; ) {
2535
+
struct ceph_osd_request *req =
2536
+
rb_entry(p, struct ceph_osd_request, r_node);
2537
+
2538
+
p = rb_next(p); /* abort_request() */
2539
+
2540
+
if (time_before(req->r_start_stamp, expiry_cutoff)) {
2541
+
pr_err_ratelimited("tid %llu on osd%d timeout\n",
2542
+
req->r_tid, osdc->homeless_osd.o_osd);
2543
+
abort_request(req, -ETIMEDOUT);
2544
+
}
2545
+
}
2550
2546
}
2551
2547
2552
2548
if (atomic_read(&osdc->num_homeless) || !list_empty(&slow_osds))
+1
-3
net/ceph/osdmap.c
+1
-3
net/ceph/osdmap.c
···
390
390
dout("crush decode tunable chooseleaf_stable = %d\n",
391
391
c->chooseleaf_stable);
392
392
393
-
crush_finalize(c);
394
-
395
393
done:
394
+
crush_finalize(c);
396
395
dout("crush_decode success\n");
397
396
return c;
398
397
···
1379
1380
if ((map->osd_state[osd] & CEPH_OSD_EXISTS) &&
1380
1381
(xorstate & CEPH_OSD_EXISTS)) {
1381
1382
pr_info("osd%d does not exist\n", osd);
1382
-
map->osd_weight[osd] = CEPH_OSD_IN;
1383
1383
ret = set_primary_affinity(map, osd,
1384
1384
CEPH_OSD_DEFAULT_PRIMARY_AFFINITY);
1385
1385
if (ret)