On Wed, 2022-08-24 at 15:27 -0400, Jeff Layton wrote: > On Wed, 2022-08-24 at 16:43 +0100, Luís Henriques wrote: > > On Fri, May 27, 2022 at 12:34:59PM -0400, Jeff Layton wrote: > > > Once the Ceph PR for this merge window has gone through, I'd like to > > > start merging in some of the preliminary fscrypt patches. In particular, > > > I'd like to merge these two patches into ceph-client/master so that they > > > go to linux-next: > > > > > > be2bc0698248 fscrypt: export fscrypt_fname_encrypt and fscrypt_fname_encrypted_size > > > 7feda88977b8 fscrypt: add fscrypt_context_for_new_inode > > > > > > I'd like to see these in ceph-client/testing, so that they start getting > > > some exposure in teuthology: > > > > > > 477944c2ed29 libceph: add spinlock around osd->o_requests > > > 355d9572686c libceph: define struct ceph_sparse_extent and add some helpers > > > 229a3e2cf1c7 libceph: add sparse read support to msgr2 crc state machine > > > a0a9795c2a2c libceph: add sparse read support to OSD client > > > 6a16e0951aaf libceph: support sparse reads on msgr2 secure codepath > > > 538b618f8726 libceph: add sparse read support to msgr1 > > > 7ef4c2c39f05 ceph: add new mount option to enable sparse reads > > > b609087729f4 ceph: preallocate inode for ops that may create one > > > e66323d65639 ceph: make ceph_msdc_build_path use ref-walk > > > > > > ...they don't add any new functionality (other than the sparse read > > > stuff), but they do change "normal" operation in some ways that we'll > > > need later, so I'd like to see them start being regularly tested. > > > > > > If that goes OK, then I'll plan to start merging another tranche a > > > couple of weeks after that. > > > > > > Does that sound OK? > > > -- > > > Jeff Layton <jlayton@xxxxxxxxxx> > > > > > > > Sorry for hijacking this thread but, after not looking at this branch for > > a few weeks, I did run a few fstests and just saw the splat bellow. > > Before start looking at it I wanted to ask if it looks familiar. It seems > > to be reliably trigger by running generic/013, and since I never saw this > > before, something probably broke in a recent rebase. > > > > [ 237.090462] kernel BUG at net/ceph/messenger.c:1116! > > [ 237.092299] invalid opcode: 0000 [#1] PREEMPT SMP PTI > > [ 237.093536] CPU: 1 PID: 21 Comm: kworker/1:0 Not tainted 5.19.0-rc6+ #99 > > [ 237.095169] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552-rebuilt.opensuse.org 04/01/2014 > > [ 237.097916] Workqueue: ceph-msgr ceph_con_workfn [libceph] > > [ 237.099350] RIP: 0010:ceph_msg_data_next+0x251/0x280 [libceph] > > [ 237.100778] Code: 00 10 00 00 48 89 0e 48 29 c8 48 8b 4f 10 48 39 c8 48 0f 47 c1 49 89 04 24 0f b7 4f 24 48 8b 42 08 48 8b 04 c8 e9 d8 fe ff ff <0f> 0b 0f 0b 0f 0b 0f 0bb > > [ 237.105190] RSP: 0018:ffffc900000bbc08 EFLAGS: 00010246 > > [ 237.106565] RAX: 0000000000000000 RBX: ffff888009354378 RCX: 0000000000000000 > > [ 237.108264] RDX: ffff8880064e0900 RSI: ffffc900000bbc48 RDI: ffff888009354378 > > [ 237.109956] RBP: ffffc900000bbc48 R08: 0000000073d74d75 R09: 0000000000000004 > > [ 237.111683] R10: ffff888009354378 R11: 0000000000000001 R12: ffffc900000bbc50 > > [ 237.113660] R13: 0000160000000000 R14: 0000000000001000 R15: ffff888009354378 > > [ 237.115380] FS: 0000000000000000(0000) GS:ffff88806f700000(0000) knlGS:0000000000000000 > > [ 237.117299] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > [ 237.118689] CR2: 000000000043d000 CR3: 000000007b456000 CR4: 00000000000006a0 > > [ 237.120397] Call Trace: > > [ 237.121005] <TASK> > > [ 237.121614] advance_cursor+0x4f/0x140 [libceph] > > [ 237.122942] osd_sparse_read+0x439/0x670 [libceph] > > [ 237.124310] prepare_sparse_read_cont+0xa6/0x510 [libceph] > > [ 237.125833] ? inet_recvmsg+0x56/0x130 > > [ 237.126959] ceph_con_v2_try_read+0x51d/0x1b60 [libceph] > > [ 237.128523] ? _raw_spin_unlock+0x12/0x30 > > [ 237.129862] ? finish_task_switch.isra.0+0xa3/0x270 > > [ 237.131266] ceph_con_workfn+0x2f9/0x760 [libceph] > > [ 237.132481] process_one_work+0x1c3/0x3c0 > > [ 237.133454] worker_thread+0x4d/0x3c0 > > [ 237.134369] ? rescuer_thread+0x380/0x380 > > [ 237.135298] kthread+0xe2/0x110 > > [ 237.136018] ? kthread_complete_and_exit+0x20/0x20 > > [ 237.137088] ret_from_fork+0x22/0x30 > > [ 237.137901] </TASK> > > [ 237.138441] Modules linked in: ceph libceph > > [ 237.139798] ---[ end trace 0000000000000000 ]--- > > [ 237.140970] RIP: 0010:ceph_msg_data_next+0x251/0x280 [libceph] > > [ 237.142216] Code: 00 10 00 00 48 89 0e 48 29 c8 48 8b 4f 10 48 39 c8 48 0f 47 c1 49 89 04 24 0f b7 4f 24 48 8b 42 08 48 8b 04 c8 e9 d8 fe ff ff <0f> 0b 0f 0b 0f 0b 0f 0bb > > [ 237.146797] RSP: 0018:ffffc900000bbc08 EFLAGS: 00010246 > > [ 237.148291] RAX: 0000000000000000 RBX: ffff888009354378 RCX: 0000000000000000 > > [ 237.149816] RDX: ffff8880064e0900 RSI: ffffc900000bbc48 RDI: ffff888009354378 > > [ 237.151332] RBP: ffffc900000bbc48 R08: 0000000073d74d75 R09: 0000000000000004 > > [ 237.152816] R10: ffff888009354378 R11: 0000000000000001 R12: ffffc900000bbc50 > > [ 237.154395] R13: 0000160000000000 R14: 0000000000001000 R15: ffff888009354378 > > [ 237.155890] FS: 0000000000000000(0000) GS:ffff88806f700000(0000) knlGS:0000000000000000 > > [ 237.157558] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > [ 237.158772] CR2: 000000000043d000 CR3: 000000007b456000 CR4: 00000000000006a0 > > [ 237.160258] note: kworker/1:0[21] exited with preempt_count 1 > > > > I think this is due to a missing wait in the ceph_sync_write rmw code > that probably crept in when it was rebased last time. It just needs: > > diff --git a/fs/ceph/file.c b/fs/ceph/file.c > index 8e809299f90b..9d0fe8e95ba8 100644 > --- a/fs/ceph/file.c > +++ b/fs/ceph/file.c > @@ -1756,6 +1756,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, > } > > ceph_osdc_start_request(osdc, req); > + ret = ceph_osdc_wait_request(osdc, req); > > /* FIXME: length field is wrong if there are 2 extents */ > ceph_update_read_metrics(&fsc->mdsc->metric, > > > > I went ahead and rebased the wip-fscrypt pile onto the ceph/testing > branch, and pushed the result to the ceph-fscrypt branch in my tree. > > Unfortunately, I'm still seeing a crash in the splice codepath when > running generic/013. We'll need to run that down as well: > > [ 119.170902] ================================================================== > [ 119.173416] BUG: KASAN: null-ptr-deref in iter_file_splice_write+0x454/0x5e0 > [ 119.175485] Read of size 8 at addr 0000000000000008 by task fsstress/1755 > [ 119.177524] > [ 119.178010] CPU: 3 PID: 1755 Comm: fsstress Tainted: G OE 6.0.0-rc1+ #383 > [ 119.180333] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.0-1.fc36 04/01/2014 > [ 119.182764] Call Trace: > [ 119.183527] <TASK> > [ 119.184198] dump_stack_lvl+0x5b/0x77 > [ 119.185308] kasan_report+0xb1/0xf0 > [ 119.186586] ? iter_file_splice_write+0x454/0x5e0 > [ 119.188072] iter_file_splice_write+0x454/0x5e0 > [ 119.189462] ? splice_from_pipe_next+0x290/0x290 > [ 119.190855] ? add_to_pipe+0x180/0x180 > [ 119.192073] ? static_obj+0x26/0x80 > [ 119.193175] direct_splice_actor+0x80/0xb0 > [ 119.194406] splice_direct_to_actor+0x1aa/0x400 > [ 119.195759] ? is_bpf_text_address+0x67/0xe0 > [ 119.197064] ? do_splice_direct+0x160/0x160 > [ 119.198351] ? do_splice_to+0x130/0x130 > [ 119.199526] do_splice_direct+0xf5/0x160 > [ 119.200710] ? splice_direct_to_actor+0x400/0x400 > [ 119.208599] ? kmem_cache_free+0x144/0x560 > [ 119.216004] generic_copy_file_range+0x74/0x90 > [ 119.223735] ? rw_verify_area+0xb0/0xb0 > [ 119.231082] ? vfs_fstatat+0x5b/0x70 > [ 119.238399] ? __lock_acquire+0x92e/0x2cf0 > [ 119.245382] ceph_copy_file_range+0x583/0x980 [ceph] > [ 119.252078] ? ceph_do_objects_copy.constprop.0+0xa30/0xa30 [ceph] > [ 119.255963] ? lockdep_hardirqs_on_prepare+0x220/0x220 > [ 119.259361] ? lock_acquire+0x167/0x3e0 > [ 119.262571] ? lock_downgrade+0x390/0x390 > [ 119.265763] ? _copy_to_user+0x4c/0x60 > [ 119.268933] ? inode_security+0x5c/0x80 > [ 119.272186] ? avc_policy_seqno+0x28/0x40 > [ 119.275361] ? lock_is_held_type+0xe3/0x140 > [ 119.278516] vfs_copy_file_range+0x2d3/0x900 > [ 119.281647] ? generic_file_rw_checks+0xd0/0xd0 > [ 119.284789] ? __do_sys_newfstatat+0x70/0xa0 > [ 119.287984] __do_sys_copy_file_range+0x12a/0x260 > [ 119.291130] ? vfs_copy_file_range+0x900/0x900 > [ 119.294017] ? lockdep_hardirqs_on_prepare+0x128/0x220 > [ 119.296956] ? syscall_enter_from_user_mode+0x22/0xc0 > [ 119.299837] ? __x64_sys_copy_file_range+0x5c/0x80 > [ 119.302742] do_syscall_64+0x3a/0x90 > [ 119.305550] entry_SYSCALL_64_after_hwframe+0x63/0xcd > [ 119.308474] RIP: 0033:0x7fa75610af3d > [ 119.311122] Code: 5d c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d b3 ce 0e 00 f7 d8 64 89 01 48 > [ 119.317912] RSP: 002b:00007ffd487125d8 EFLAGS: 00000206 ORIG_RAX: 0000000000000146 > [ 119.321306] RAX: ffffffffffffffda RBX: 00000000000326fa RCX: 00007fa75610af3d > [ 119.324518] RDX: 0000000000000004 RSI: 00007ffd48712620 RDI: 0000000000000003 > [ 119.327686] RBP: 000000000000008e R08: 000000000000ab2a R09: 0000000000000000 > [ 119.330790] R10: 00007ffd48712628 R11: 0000000000000206 R12: 0000000000000003 > [ 119.333883] R13: 0000000000400000 R14: 000000000000ab2a R15: 000000000001818d > [ 119.337074] </TASK> > [ 119.339644] ================================================================== > Ok, I ran this down too, and it's a bit more disturbing. This seems to be the result of either a long-standing bug in the kclient (since it was originally merged) or a change in OSD behavior. I'm not sure which. I just sent a patch to the ML that should fix this in ceph/testing regardless of the OSD's behavior. I've also rebased my ceph-fscrypt branch on top of that fix. I'm running tests on it now, but it seems to be behaving so far. -- Jeff Layton <jlayton@xxxxxxxxxx>
