Skip to content

Commit 0dfb36b

Browse files
torvaldstorvalds
committed
Merge tag 'ceph-for-6.19-rc1' of https://github.com/ceph/ceph-client
Pull ceph updates from Ilya Dryomov: "We have a patch that adds an initial set of tracepoints to the MDS client from Max, a fix that hardens osdmap parsing code from myself (marked for stable) and a few assorted fixups" * tag 'ceph-for-6.19-rc1' of https://github.com/ceph/ceph-client: rbd: stop selecting CRC32, CRYPTO, and CRYPTO_AES ceph: stop selecting CRC32, CRYPTO, and CRYPTO_AES libceph: make decode_pool() more resilient against corrupted osdmaps libceph: Amend checking to fix `make W=1` build breakage ceph: Amend checking to fix `make W=1` build breakage ceph: add trace points to the MDS client libceph: fix log output race condition in OSD client
2 parents 4cfc214 + 21c1466 commit 0dfb36b

9 files changed

Lines changed: 316 additions & 79 deletions

File tree

‎drivers/block/Kconfig‎

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -316,9 +316,6 @@ config BLK_DEV_RBD
316316
tristate "Rados block device (RBD)"
317317
depends on INET && BLOCK
318318
select CEPH_LIB
319-
select CRC32
320-
select CRYPTO_AES
321-
select CRYPTO
322319
help
323320
Say Y here if you want include the Rados block device, which stripes
324321
a block device over objects stored in the Ceph distributed object

‎fs/ceph/Kconfig‎

Lines changed: 0 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -3,9 +3,6 @@ config CEPH_FS
33
tristate "Ceph distributed file system"
44
depends on INET
55
select CEPH_LIB
6-
select CRC32
7-
select CRYPTO_AES
8-
select CRYPTO
96
select NETFS_SUPPORT
107
select FS_ENCRYPTION_ALGS if FS_ENCRYPTION
118
default n

‎fs/ceph/caps.c‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -18,6 +18,7 @@
1818
#include "crypto.h"
1919
#include <linux/ceph/decode.h>
2020
#include <linux/ceph/messenger.h>
21+
#include <trace/events/ceph.h>
2122

2223
/*
2324
* Capability management
@@ -4452,6 +4453,9 @@ void ceph_handle_caps(struct ceph_mds_session *session,
44524453
session->s_mds, ceph_cap_op_name(op), vino.ino, vino.snap, inode,
44534454
seq, issue_seq, mseq);
44544455

4456+
trace_ceph_handle_caps(mdsc, session, op, &vino, ceph_inode(inode),
4457+
seq, issue_seq, mseq);
4458+
44554459
mutex_lock(&session->s_mutex);
44564460

44574461
if (!inode) {

‎fs/ceph/mds_client.c‎

Lines changed: 18 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -24,6 +24,7 @@
2424
#include <linux/ceph/pagelist.h>
2525
#include <linux/ceph/auth.h>
2626
#include <linux/ceph/debugfs.h>
27+
#include <trace/events/ceph.h>
2728

2829
#define RECONNECT_MAX_SIZE (INT_MAX - PAGE_SIZE)
2930

@@ -3288,6 +3289,8 @@ static void complete_request(struct ceph_mds_client *mdsc,
32883289
{
32893290
req->r_end_latency = ktime_get();
32903291

3292+
trace_ceph_mdsc_complete_request(mdsc, req);
3293+
32913294
if (req->r_callback)
32923295
req->r_callback(mdsc, req);
32933296
complete_all(&req->r_completion);
@@ -3419,6 +3422,8 @@ static int __send_request(struct ceph_mds_session *session,
34193422
{
34203423
int err;
34213424

3425+
trace_ceph_mdsc_send_request(session, req);
3426+
34223427
err = __prepare_send_request(session, req, drop_cap_releases);
34233428
if (!err) {
34243429
ceph_msg_get(req->r_request);
@@ -3470,6 +3475,8 @@ static void __do_request(struct ceph_mds_client *mdsc,
34703475
}
34713476
if (mdsc->mdsmap->m_epoch == 0) {
34723477
doutc(cl, "no mdsmap, waiting for map\n");
3478+
trace_ceph_mdsc_suspend_request(mdsc, session, req,
3479+
ceph_mdsc_suspend_reason_no_mdsmap);
34733480
list_add(&req->r_wait, &mdsc->waiting_for_map);
34743481
return;
34753482
}
@@ -3491,6 +3498,8 @@ static void __do_request(struct ceph_mds_client *mdsc,
34913498
goto finish;
34923499
}
34933500
doutc(cl, "no mds or not active, waiting for map\n");
3501+
trace_ceph_mdsc_suspend_request(mdsc, session, req,
3502+
ceph_mdsc_suspend_reason_no_active_mds);
34943503
list_add(&req->r_wait, &mdsc->waiting_for_map);
34953504
return;
34963505
}
@@ -3536,9 +3545,11 @@ static void __do_request(struct ceph_mds_client *mdsc,
35363545
* it to the mdsc queue.
35373546
*/
35383547
if (session->s_state == CEPH_MDS_SESSION_REJECTED) {
3539-
if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER))
3548+
if (ceph_test_mount_opt(mdsc->fsc, CLEANRECOVER)) {
3549+
trace_ceph_mdsc_suspend_request(mdsc, session, req,
3550+
ceph_mdsc_suspend_reason_rejected);
35403551
list_add(&req->r_wait, &mdsc->waiting_for_map);
3541-
else
3552+
} else
35423553
err = -EACCES;
35433554
goto out_session;
35443555
}
@@ -3552,6 +3563,8 @@ static void __do_request(struct ceph_mds_client *mdsc,
35523563
if (random)
35533564
req->r_resend_mds = mds;
35543565
}
3566+
trace_ceph_mdsc_suspend_request(mdsc, session, req,
3567+
ceph_mdsc_suspend_reason_session);
35553568
list_add(&req->r_wait, &session->s_waiting);
35563569
goto out_session;
35573570
}
@@ -3652,6 +3665,7 @@ static void __wake_requests(struct ceph_mds_client *mdsc,
36523665
list_del_init(&req->r_wait);
36533666
doutc(cl, " wake request %p tid %llu\n", req,
36543667
req->r_tid);
3668+
trace_ceph_mdsc_resume_request(mdsc, req);
36553669
__do_request(mdsc, req);
36563670
}
36573671
}
@@ -3678,6 +3692,7 @@ static void kick_requests(struct ceph_mds_client *mdsc, int mds)
36783692
req->r_session->s_mds == mds) {
36793693
doutc(cl, " kicking tid %llu\n", req->r_tid);
36803694
list_del_init(&req->r_wait);
3695+
trace_ceph_mdsc_resume_request(mdsc, req);
36813696
__do_request(mdsc, req);
36823697
}
36833698
}
@@ -3724,6 +3739,7 @@ int ceph_mdsc_submit_request(struct ceph_mds_client *mdsc, struct inode *dir,
37243739
doutc(cl, "submit_request on %p for inode %p\n", req, dir);
37253740
mutex_lock(&mdsc->mutex);
37263741
__register_request(mdsc, req, dir);
3742+
trace_ceph_mdsc_submit_request(mdsc, req);
37273743
__do_request(mdsc, req);
37283744
err = req->r_err;
37293745
mutex_unlock(&mdsc->mutex);

‎fs/ceph/snap.c‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -374,7 +374,7 @@ static int build_snap_context(struct ceph_mds_client *mdsc,
374374

375375
/* alloc new snap context */
376376
err = -ENOMEM;
377-
if (num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64))
377+
if ((size_t)num > (SIZE_MAX - sizeof(*snapc)) / sizeof(u64))
378378
goto fail;
379379
snapc = ceph_create_snap_context(num, GFP_NOFS);
380380
if (!snapc)

‎fs/ceph/super.c‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -30,6 +30,9 @@
3030

3131
#include <uapi/linux/magic.h>
3232

33+
#define CREATE_TRACE_POINTS
34+
#include <trace/events/ceph.h>
35+
3336
static DEFINE_SPINLOCK(ceph_fsc_lock);
3437
static LIST_HEAD(ceph_fsc_list);
3538

‎include/trace/events/ceph.h‎

Lines changed: 234 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,234 @@
1+
/* SPDX-License-Identifier: GPL-2.0-or-later */
2+
/* Ceph filesystem support module tracepoints
3+
*
4+
* Copyright (C) 2025 IONOS SE. All Rights Reserved.
5+
* Written by Max Kellermann (max.kellermann@ionos.com)
6+
*/
7+
#undef TRACE_SYSTEM
8+
#define TRACE_SYSTEM ceph
9+
10+
#if !defined(_TRACE_CEPH_H) || defined(TRACE_HEADER_MULTI_READ)
11+
#define _TRACE_CEPH_H
12+
13+
#include <linux/tracepoint.h>
14+
15+
#define ceph_mdsc_suspend_reasons \
16+
EM(ceph_mdsc_suspend_reason_no_mdsmap, "no-mdsmap") \
17+
EM(ceph_mdsc_suspend_reason_no_active_mds, "no-active-mds") \
18+
EM(ceph_mdsc_suspend_reason_rejected, "rejected") \
19+
E_(ceph_mdsc_suspend_reason_session, "session")
20+
21+
#ifndef __NETFS_DECLARE_TRACE_ENUMS_ONCE_ONLY
22+
#define __NETFS_DECLARE_TRACE_ENUMS_ONCE_ONLY
23+
24+
#undef EM
25+
#undef E_
26+
#define EM(a, b) a,
27+
#define E_(a, b) a
28+
29+
enum ceph_mdsc_suspend_reason { ceph_mdsc_suspend_reasons } __mode(byte);
30+
31+
#endif
32+
33+
/*
34+
* Export enum symbols via userspace.
35+
*/
36+
#undef EM
37+
#undef E_
38+
#define EM(a, b) TRACE_DEFINE_ENUM(a);
39+
#define E_(a, b) TRACE_DEFINE_ENUM(a);
40+
41+
ceph_mdsc_suspend_reasons;
42+
43+
/*
44+
* Now redefine the EM() and E_() macros to map the enums to the strings that
45+
* will be printed in the output.
46+
*/
47+
#undef EM
48+
#undef E_
49+
#define EM(a, b) { a, b },
50+
#define E_(a, b) { a, b }
51+
52+
TRACE_EVENT(ceph_mdsc_submit_request,
53+
TP_PROTO(struct ceph_mds_client *mdsc,
54+
struct ceph_mds_request *req),
55+
56+
TP_ARGS(mdsc, req),
57+
58+
TP_STRUCT__entry(
59+
__field(u64, tid)
60+
__field(int, op)
61+
__field(u64, ino)
62+
__field(u64, snap)
63+
),
64+
65+
TP_fast_assign(
66+
struct inode *inode;
67+
68+
__entry->tid = req->r_tid;
69+
__entry->op = req->r_op;
70+
71+
inode = req->r_inode;
72+
if (inode == NULL && req->r_dentry)
73+
inode = d_inode(req->r_dentry);
74+
75+
if (inode) {
76+
__entry->ino = ceph_ino(inode);
77+
__entry->snap = ceph_snap(inode);
78+
} else {
79+
__entry->ino = __entry->snap = 0;
80+
}
81+
),
82+
83+
TP_printk("R=%llu op=%s ino=%llx,%llx",
84+
__entry->tid,
85+
ceph_mds_op_name(__entry->op),
86+
__entry->ino, __entry->snap)
87+
);
88+
89+
TRACE_EVENT(ceph_mdsc_suspend_request,
90+
TP_PROTO(struct ceph_mds_client *mdsc,
91+
struct ceph_mds_session *session,
92+
struct ceph_mds_request *req,
93+
enum ceph_mdsc_suspend_reason reason),
94+
95+
TP_ARGS(mdsc, session, req, reason),
96+
97+
TP_STRUCT__entry(
98+
__field(u64, tid)
99+
__field(int, op)
100+
__field(int, mds)
101+
__field(enum ceph_mdsc_suspend_reason, reason)
102+
),
103+
104+
TP_fast_assign(
105+
__entry->tid = req->r_tid;
106+
__entry->op = req->r_op;
107+
__entry->mds = session ? session->s_mds : -1;
108+
__entry->reason = reason;
109+
),
110+
111+
TP_printk("R=%llu op=%s reason=%s",
112+
__entry->tid,
113+
ceph_mds_op_name(__entry->op),
114+
__print_symbolic(__entry->reason, ceph_mdsc_suspend_reasons))
115+
);
116+
117+
TRACE_EVENT(ceph_mdsc_resume_request,
118+
TP_PROTO(struct ceph_mds_client *mdsc,
119+
struct ceph_mds_request *req),
120+
121+
TP_ARGS(mdsc, req),
122+
123+
TP_STRUCT__entry(
124+
__field(u64, tid)
125+
__field(int, op)
126+
),
127+
128+
TP_fast_assign(
129+
__entry->tid = req->r_tid;
130+
__entry->op = req->r_op;
131+
),
132+
133+
TP_printk("R=%llu op=%s",
134+
__entry->tid,
135+
ceph_mds_op_name(__entry->op))
136+
);
137+
138+
TRACE_EVENT(ceph_mdsc_send_request,
139+
TP_PROTO(struct ceph_mds_session *session,
140+
struct ceph_mds_request *req),
141+
142+
TP_ARGS(session, req),
143+
144+
TP_STRUCT__entry(
145+
__field(u64, tid)
146+
__field(int, op)
147+
__field(int, mds)
148+
),
149+
150+
TP_fast_assign(
151+
__entry->tid = req->r_tid;
152+
__entry->op = req->r_op;
153+
__entry->mds = session->s_mds;
154+
),
155+
156+
TP_printk("R=%llu op=%s mds=%d",
157+
__entry->tid,
158+
ceph_mds_op_name(__entry->op),
159+
__entry->mds)
160+
);
161+
162+
TRACE_EVENT(ceph_mdsc_complete_request,
163+
TP_PROTO(struct ceph_mds_client *mdsc,
164+
struct ceph_mds_request *req),
165+
166+
TP_ARGS(mdsc, req),
167+
168+
TP_STRUCT__entry(
169+
__field(u64, tid)
170+
__field(int, op)
171+
__field(int, err)
172+
__field(unsigned long, latency_ns)
173+
),
174+
175+
TP_fast_assign(
176+
__entry->tid = req->r_tid;
177+
__entry->op = req->r_op;
178+
__entry->err = req->r_err;
179+
__entry->latency_ns = req->r_end_latency - req->r_start_latency;
180+
),
181+
182+
TP_printk("R=%llu op=%s err=%d latency_ns=%lu",
183+
__entry->tid,
184+
ceph_mds_op_name(__entry->op),
185+
__entry->err,
186+
__entry->latency_ns)
187+
);
188+
189+
TRACE_EVENT(ceph_handle_caps,
190+
TP_PROTO(struct ceph_mds_client *mdsc,
191+
struct ceph_mds_session *session,
192+
int op,
193+
const struct ceph_vino *vino,
194+
struct ceph_inode_info *inode,
195+
u32 seq, u32 mseq, u32 issue_seq),
196+
197+
TP_ARGS(mdsc, session, op, vino, inode, seq, mseq, issue_seq),
198+
199+
TP_STRUCT__entry(
200+
__field(int, mds)
201+
__field(int, op)
202+
__field(u64, ino)
203+
__field(u64, snap)
204+
__field(u32, seq)
205+
__field(u32, mseq)
206+
__field(u32, issue_seq)
207+
),
208+
209+
TP_fast_assign(
210+
__entry->mds = session->s_mds;
211+
__entry->op = op;
212+
__entry->ino = vino->ino;
213+
__entry->snap = vino->snap;
214+
__entry->seq = seq;
215+
__entry->mseq = mseq;
216+
__entry->issue_seq = issue_seq;
217+
),
218+
219+
TP_printk("mds=%d op=%s vino=%llx.%llx seq=%u iseq=%u mseq=%u",
220+
__entry->mds,
221+
ceph_cap_op_name(__entry->op),
222+
__entry->ino,
223+
__entry->snap,
224+
__entry->seq,
225+
__entry->issue_seq,
226+
__entry->mseq)
227+
);
228+
229+
#undef EM
230+
#undef E_
231+
#endif /* _TRACE_CEPH_H */
232+
233+
/* This part must be outside protection */
234+
#include <trace/define_trace.h>

0 commit comments

Comments
 (0)