diff mbox series

[v3,45/55] ceph: Use sendmsg(MSG_SPLICE_PAGES) rather than sendpage()

Message ID 20230331160914.1608208-46-dhowells@redhat.com
State Superseded
Headers show
Series None | expand

Commit Message

David Howells March 31, 2023, 4:09 p.m. UTC
Use sendmsg() and MSG_SPLICE_PAGES rather than sendpage in ceph when
transmitting data.  For the moment, this can only transmit one page at a
time because of the architecture of net/ceph/, but if
write_partial_message_data() can be given a bvec[] at a time by the
iteration code, this would allow pages to be sent in a batch.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Ilya Dryomov <idryomov@gmail.com>
cc: Xiubo Li <xiubli@redhat.com>
cc: Jeff Layton <jlayton@kernel.org>
cc: "David S. Miller" <davem@davemloft.net>
cc: Eric Dumazet <edumazet@google.com>
cc: Jakub Kicinski <kuba@kernel.org>
cc: Paolo Abeni <pabeni@redhat.com>
cc: Jens Axboe <axboe@kernel.dk>
cc: Matthew Wilcox <willy@infradead.org>
cc: ceph-devel@vger.kernel.org
cc: netdev@vger.kernel.org
---
 net/ceph/messenger_v2.c | 89 +++++++++--------------------------------
 1 file changed, 18 insertions(+), 71 deletions(-)

Comments

Xiubo Li April 10, 2023, 12:20 p.m. UTC | #1
David,

With these patches I can reproduce the hang every time, and I haven't 
gotten a chance to debug it yet, the logs:


<3>[  727.042312] INFO: task kworker/u20:4:78 blocked for more than 245 
seconds.
<3>[  727.042381]       Tainted: G        W 6.3.0-rc3+ #9
<3>[  727.042417] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" 
disables this message.
<6>[  727.042449] task:kworker/u20:4   state:D stack:0 pid:78    
ppid:2      flags:0x00004000
<6>[  727.042496] Workqueue: writeback wb_workfn (flush-ceph-1)
<6>[  727.042558] Call Trace:
<6>[  727.042576]  <TASK>
<6>[  727.042630]  __schedule+0x4a8/0xa80
<6>[  727.042942]  ? __pfx___schedule+0x10/0x10
<6>[  727.042997]  ? __pfx___lock_release+0x10/0x10
<6>[  727.043118]  ? check_chain_key+0x205/0x2b0
<6>[  727.043299]  schedule+0x8e/0x120
<6>[  727.043372]  schedule_preempt_disabled+0x11/0x20
<6>[  727.043420]  __mutex_lock+0x97b/0x1270
<6>[  727.043532]  ? ceph_con_send+0xa4/0x310 [libceph]
<6>[  727.044348]  ? __pfx___mutex_lock+0x10/0x10
<6>[  727.044410]  ? encode_oloc+0x16e/0x1b0 [libceph]
<6>[  727.045473]  ? ceph_con_send+0xa4/0x310 [libceph]
<6>[  727.046298]  ceph_con_send+0xa4/0x310 [libceph]
<6>[  727.047200]  send_request+0x2b1/0x760 [libceph]
<6>[  727.048122]  ? __pfx___lock_release+0x10/0x10
<6>[  727.048186]  ? __lock_acquired+0x1ef/0x3d0
<6>[  727.048264]  ? __pfx_send_request+0x10/0x10 [libceph]
<6>[  727.049104]  ? check_chain_key+0x205/0x2b0
<6>[  727.049215]  ? link_request+0xcd/0x1a0 [libceph]
<6>[  727.050047]  ? do_raw_spin_unlock+0x99/0x100
<6>[  727.050159]  ? _raw_spin_unlock+0x1f/0x40
<6>[  727.050279]  __submit_request+0x2b7/0x4e0 [libceph]
<6>[  727.051259]  ceph_osdc_start_request+0x31/0x40 [libceph]
<6>[  727.052141]  ceph_writepages_start+0x1d13/0x2490 [ceph]
<6>[  727.053258]  ? __pfx_ceph_writepages_start+0x10/0x10 [ceph]
<6>[  727.053872]  ? validate_chain+0xa1/0x760
<6>[  727.054023]  ? __pfx_validate_chain+0x10/0x10
<6>[  727.054171]  ? check_chain_key+0x205/0x2b0
<6>[  727.054308]  ? __lock_acquire+0x7b3/0xfc0
<6>[  727.054726]  ? reacquire_held_locks+0x18b/0x290
<6>[  727.054790]  ? writeback_sb_inodes+0x263/0x7c0
<6>[  727.054990]  do_writepages+0x106/0x320
<6>[  727.055068]  ? __pfx_do_writepages+0x10/0x10
<6>[  727.055152]  ? __pfx___lock_release+0x10/0x10
<6>[  727.055205]  ? __lock_acquired+0x1ef/0x3d0
<6>[  727.055314]  ? check_chain_key+0x205/0x2b0
<6>[  727.055594]  __writeback_single_inode+0x95/0x450
<6>[  727.055725]  writeback_sb_inodes+0x392/0x7c0
<6>[  727.055912]  ? __pfx_writeback_sb_inodes+0x10/0x10
<6>[  727.055994]  ? __pfx_lock_acquire+0x10/0x10
<6>[  727.056045]  ? do_raw_spin_unlock+0x99/0x100
<6>[  727.056221]  ? __pfx_move_expired_inodes+0x10/0x10
<6>[  727.056541]  __writeback_inodes_wb+0x6a/0x130
<6>[  727.056674]  wb_writeback+0x45b/0x530
<6>[  727.056780]  ? __pfx_wb_writeback+0x10/0x10
<6>[  727.056910]  ? _find_next_bit+0x37/0xc0
<6>[  727.057102]  wb_do_writeback+0x434/0x4f0
<6>[  727.057216]  ? __pfx_wb_do_writeback+0x10/0x10
<6>[  727.057445]  ? __lock_acquire+0x7b3/0xfc0
<6>[  727.057614]  wb_workfn+0xe0/0x400
<6>[  727.057692]  ? __pfx_wb_workfn+0x10/0x10
<6>[  727.057752]  ? lock_acquire+0x15c/0x3e0
<6>[  727.057802]  ? process_one_work+0x436/0x990
<6>[  727.057903]  ? __pfx_lock_acquire+0x10/0x10
<6>[  727.058039]  ? check_chain_key+0x205/0x2b0
<6>[  727.058086]  ? __pfx_try_to_wake_up+0x10/0x10
<6>[  727.058138]  ? mark_held_locks+0x23/0x90
<6>[  727.058306]  process_one_work+0x505/0x990
<6>[  727.058610]  ? __pfx_process_one_work+0x10/0x10
<6>[  727.058731]  ? mark_held_locks+0x23/0x90
<6>[  727.058827]  ? worker_thread+0xce/0x670
<6>[  727.058958]  worker_thread+0x2dd/0x670
<6>[  727.059145]  ? __pfx_worker_thread+0x10/0x10
<6>[  727.059201]  kthread+0x16f/0x1a0
<6>[  727.059252]  ? __pfx_kthread+0x10/0x10
<6>[  727.059341]  ret_from_fork+0x2c/0x50
<6>[  727.059683]  </TASK>
<3>[  727.059964] INFO: task kworker/9:6:8586 blocked for more than 245 
seconds.
<3>[  727.060031]       Tainted: G        W 6.3.0-rc3+ #9
<3>[  727.060086] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" 
disables this message.
<6>[  727.060140] task:kworker/9:6     state:D stack:0 pid:8586  
ppid:2      flags:0x00004000
<6>[  727.060211] Workqueue: events handle_osds_timeout [libceph]
<6>[  727.060836] Call Trace:
<6>[  727.060892]  <TASK>
<6>[  727.061089]  __schedule+0x4a8/0xa80
<6>[  727.061243]  ? __pfx___schedule+0x10/0x10
<6>[  727.061488]  ? mark_held_locks+0x6b/0x90
<6>[  727.061581]  ? lockdep_hardirqs_on_prepare.part.0+0xea/0x1b0
<6>[  727.061696]  schedule+0x8e/0x120
<6>[  727.061767]  schedule_preempt_disabled+0x11/0x20
<6>[  727.061812]  rwsem_down_write_slowpath+0x2d6/0x840
<6>[  727.061944]  ? __pfx_rwsem_down_write_slowpath+0x10/0x10
<6>[  727.062240]  down_write+0x1bf/0x1d0
<6>[  727.062307]  ? __pfx_down_write+0x10/0x10
<6>[  727.062517]  ? check_chain_key+0x205/0x2b0
<6>[  727.062635]  handle_osds_timeout+0x6f/0x1b0 [libceph]
<6>[  727.063487]  process_one_work+0x505/0x990
<6>[  727.063657]  ? __pfx_process_one_work+0x10/0x10
<6>[  727.063789]  ? mark_held_locks+0x23/0x90
<6>[  727.063913]  ? worker_thread+0xce/0x670
<6>[  727.064024]  worker_thread+0x2dd/0x670
<6>[  727.064161]  ? __kthread_parkme+0xc9/0xe0
<6>[  727.064353]  ? __pfx_worker_thread+0x10/0x10
<6>[  727.064411]  kthread+0x16f/0x1a0
<6>[  727.064462]  ? __pfx_kthread+0x10/0x10
<6>[  727.064556]  ret_from_fork+0x2c/0x50
<6>[  727.064816]  </TASK>
<3>[  727.064853] INFO: task kworker/6:6:8587 blocked for more than 245 
seconds.
<3>[  727.064917]       Tainted: G        W 6.3.0-rc3+ #9
<3>[  727.064970] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" 
disables this message.
<6>[  727.065022] task:kworker/6:6     state:D stack:0 pid:8587  
ppid:2      flags:0x00004000
<6>[  727.065081] Workqueue: events delayed_work [ceph]
<6>[  727.065693] Call Trace:
<6>[  727.065724]  <TASK>
<6>[  727.065833]  __schedule+0x4a8/0xa80
<6>[  727.065932]  ? __pfx___schedule+0x10/0x10
<6>[  727.065983]  ? rwsem_down_read_slowpath+0x294/0x940
<6>[  727.066112]  ? mark_lock.part.0+0xd7/0x6c0
<6>[  727.066292]  ? _raw_spin_unlock_irq+0x24/0x40
<6>[  727.066404]  ? rwsem_down_read_slowpath+0x294/0x940
<6>[  727.066460]  schedule+0x8e/0x120
<6>[  727.066529]  schedule_preempt_disabled+0x11/0x20
<6>[  727.066574]  rwsem_down_read_slowpath+0x486/0x940
<6>[  727.066686]  ? __pfx_rwsem_down_read_slowpath+0x10/0x10
<6>[  727.066885]  ? lock_acquire+0x15c/0x3e0
<6>[  727.066949]  ? find_held_lock+0x8c/0xa0
<6>[  727.066994]  ? kvm_clock_read+0x14/0x30
<6>[  727.067036]  ? kvm_sched_clock_read+0x5/0x20
<6>[  727.067269]  __down_read_common+0xad/0x310
<6>[  727.067344]  ? __pfx___down_read_common+0x10/0x10
<6>[  727.067450]  ? ceph_send_cap_releases+0xbe/0x6a0 [ceph]
<6>[  727.068071]  down_read+0x7a/0x90
<6>[  727.068140]  ceph_send_cap_releases+0xbe/0x6a0 [ceph]
<6>[  727.068869]  ? mark_held_locks+0x6b/0x90
<6>[  727.068990]  ? __pfx_ceph_send_cap_releases+0x10/0x10 [ceph]
<6>[  727.069733]  delayed_work+0x30b/0x310 [ceph]
<6>[  727.070445]  process_one_work+0x505/0x990
<6>[  727.070625]  ? __pfx_process_one_work+0x10/0x10
<6>[  727.070769]  ? mark_held_locks+0x23/0x90
<6>[  727.070838]  ? worker_thread+0xce/0x670
<6>[  727.070916]  worker_thread+0x2dd/0x670
<6>[  727.071016]  ? __kthread_parkme+0xc9/0xe0
<6>[  727.071076]  ? __pfx_worker_thread+0x10/0x10
<6>[  727.071115]  kthread+0x16f/0x1a0
<6>[  727.071150]  ? __pfx_kthread+0x10/0x10
<6>[  727.071368]  ret_from_fork+0x2c/0x50
<6>[  727.071532]  </TASK>
<3>[  727.071593] INFO: task kworker/9:18:9987 blocked for more than 245 
seconds.
<3>[  727.071642]       Tainted: G        W 6.3.0-rc3+ #9
<3>[  727.071711] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" 
disables this message.
<6>[  727.071750] task:kworker/9:18    state:D stack:0 pid:9987  
ppid:2      flags:0x00004000
<6>[  727.071797] Workqueue: events handle_timeout [libceph]
<6>[  727.072375] Call Trace:
<6>[  727.072397]  <TASK>
<6>[  727.072457]  __schedule+0x4a8/0xa80
<6>[  727.072536]  ? __pfx___schedule+0x10/0x10
<6>[  727.072659]  ? mark_lock.part.0+0xd7/0x6c0
<6>[  727.072711]  ? _raw_spin_unlock_irq+0x24/0x40
<6>[  727.072790]  schedule+0x8e/0x120
<6>[  727.072839]  schedule_preempt_disabled+0x11/0x20
<6>[  727.072871]  rwsem_down_write_slowpath+0x2d6/0x840
<6>[  727.072945]  ? __pfx_rwsem_down_write_slowpath+0x10/0x10
<6>[  727.073161]  down_write+0x1bf/0x1d0
<6>[  727.073424]  ? __pfx_down_write+0x10/0x10
<6>[  727.073551]  handle_timeout+0x12a/0x6f0 [libceph]
<6>[  727.074072]  ? __pfx_lock_acquire+0x10/0x10
<6>[  727.074188]  ? __pfx_handle_timeout+0x10/0x10 [libceph]
<6>[  727.075342]  process_one_work+0x505/0x990
<6>[  727.075473]  ? __pfx_process_one_work+0x10/0x10
<6>[  727.075558]  ? mark_held_locks+0x23/0x90
<6>[  727.075655]  ? worker_thread+0xce/0x670
<6>[  727.075716]  worker_thread+0x2dd/0x670
<6>[  727.075820]  ? __kthread_parkme+0xc9/0xe0
<6>[  727.075865]  ? __pfx_worker_thread+0x10/0x10
<6>[  727.075895]  kthread+0x16f/0x1a0
<6>[  727.075923]  ? __pfx_kthread+0x10/0x10
<6>[  727.076036]  ret_from_fork+0x2c/0x50
<6>[  727.076169]  </TASK>
<3>[  727.076201] INFO: task ffsb:13945 blocked for more than 245 seconds.
<3>[  727.076239]       Tainted: G        W 6.3.0-rc3+ #9
<3>[  727.076272] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" 
disables this message.
<6>[  727.076302] task:ffsb            state:D stack:0 pid:13945 
ppid:10657  flags:0x00000002
<6>[  727.076340] Call Trace:
<6>[  727.076357]  <TASK>
<6>[  727.076400]  __schedule+0x4a8/0xa80
<6>[  727.076459]  ? __pfx___schedule+0x10/0x10
<6>[  727.076489]  ? asm_sysvec_apic_timer_interrupt+0x16/0x20
<6>[  727.076629]  ? rwsem_down_read_slowpath+0x294/0x940
<6>[  727.076663]  schedule+0x8e/0x120
<6>[  727.076701]  schedule_preempt_disabled+0x11/0x20
<6>[  727.076726]  rwsem_down_read_slowpath+0x486/0x940
<6>[  727.076789]  ? __pfx_rwsem_down_read_slowpath+0x10/0x10
<6>[  727.076887]  ? lock_acquire+0x15c/0x3e0
<6>[  727.076921]  ? find_held_lock+0x8c/0xa0
<6>[  727.077014]  ? kvm_clock_read+0x14/0x30
<6>[  727.077047]  ? kvm_sched_clock_read+0x5/0x20
<6>[  727.077110]  __down_read_common+0xad/0x310
<6>[  727.077147]  ? __pfx___down_read_common+0x10/0x10
<6>[  727.077182]  ? __pfx_generic_write_checks+0x10/0x10
<6>[  727.077231]  ? ceph_write_iter+0x33c/0xad0 [ceph]
<6>[  727.077557]  down_read+0x7a/0x90
<6>[  727.077596]  ceph_write_iter+0x33c/0xad0 [ceph]
<6>[  727.078041]  ? __pfx_ceph_write_iter+0x10/0x10 [ceph]
<6>[  727.078324]  ? __pfx_lock_acquire+0x10/0x10
<6>[  727.078358]  ? __might_resched+0x213/0x300
<6>[  727.078413]  ? inode_security+0x6d/0x90
<6>[  727.078454]  ? selinux_file_permission+0x1d5/0x210
<6>[  727.078558]  vfs_write+0x567/0x750
<6>[  727.078608]  ? __pfx_vfs_write+0x10/0x10
<6>[  727.078786]  ksys_write+0xc9/0x170
<6>[  727.078824]  ? __pfx_ksys_write+0x10/0x10
<6>[  727.078857]  ? ktime_get_coarse_real_ts64+0x100/0x110
<6>[  727.078883]  ? ktime_get_coarse_real_ts64+0xa4/0x110
<6>[  727.079029]  do_syscall_64+0x5c/0x90
<6>[  727.079071]  ? do_syscall_64+0x69/0x90
<6>[  727.079105]  ? lockdep_hardirqs_on_prepare.part.0+0xea/0x1b0
<6>[  727.079151]  ? do_syscall_64+0x69/0x90
<6>[  727.079184]  ? lockdep_hardirqs_on_prepare.part.0+0xea/0x1b0
<6>[  727.079234]  ? do_syscall_64+0x69/0x90
<6>[  727.079257]  ? lockdep_hardirqs_on_prepare.part.0+0xea/0x1b0
<6>[  727.079303]  entry_SYSCALL_64_after_hwframe+0x72/0xdc
<6>[  727.079336] RIP: 0033:0x7fa52913ebcf
<6>[  727.079361] RSP: 002b:00007fa522dfba30 EFLAGS: 00000293 ORIG_RAX: 
0000000000000001
<6>[  727.079396] RAX: ffffffffffffffda RBX: 0000000000001000 RCX: 
00007fa52913ebcf
<6>[  727.079418] RDX: 0000000000001000 RSI: 00007fa51c001450 RDI: 
0000000000000004
<6>[  727.079438] RBP: 0000000000000004 R08: 0000000000000000 R09: 
0000000001cc3580
<6>[  727.079456] R10: 00000000000001c0 R11: 0000000000000293 R12: 
0000000000000000
<6>[  727.079474] R13: 0000000001cc3580 R14: 0000000000001000 R15: 
00007fa51c001450
<6>[  727.079614]  </TASK>
<6>[  727.079633] Future hung task reports are suppressed, see sysctl 
kernel.hung_task_warnings
<4>[  727.079652]
<4>[  727.079652] Showing all locks held in the system:
<4>[  727.079671] 1 lock held by rcu_tasks_kthre/12:
<4>[  727.079691]  #0: ffffffff8bf1fc80 
(rcu_tasks.tasks_gp_mutex){+.+.}-{4:4}, at: rcu_tasks_one_gp+0x32/0x280
<4>[  727.079788] 1 lock held by rcu_tasks_rude_/13:
<4>[  727.079805]  #0: ffffffff8bf1f9a0 
(rcu_tasks_rude.tasks_gp_mutex){+.+.}-{4:4}, at: rcu_tasks_one_gp+0x32/0x280
<4>[  727.079896] 1 lock held by rcu_tasks_trace/14:
<4>[  727.079976]  #0: ffffffff8bf1f660 
(rcu_tasks_trace.tasks_gp_mutex){+.+.}-{4:4}, at: 
rcu_tasks_one_gp+0x32/0x280
<4>[  727.080085] 6 locks held by kworker/u20:4/78:
<4>[  727.080105]  #0: ffff8884644d3148 
((wq_completion)writeback){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.080192]  #1: ffff888101027de8 
((work_completion)(&(&wb->dwork)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.080282]  #2: ffff8881022740e8 
(&type->s_umount_key#62){.+.+}-{4:4}, at: trylock_super+0x16/0x70
<4>[  727.080385]  #3: ffff88811e378d70 (&osdc->lock){++++}-{4:4}, at: 
ceph_osdc_start_request+0x17/0x40 [libceph]
<4>[  727.080760]  #4: ffff88810c116980 (&osd->lock){+.+.}-{4:4}, at: 
__submit_request+0xfa/0x4e0 [libceph]
<4>[  727.081139]  #5: ffff88810c116178 (&con->mutex){+.+.}-{4:4}, at: 
ceph_con_send+0xa4/0x310 [libceph]
<4>[  727.081458] 3 locks held by kworker/3:1/90:
<4>[  727.081517]  #0: ffff888130cced48 
((wq_completion)ceph-msgr){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.081593]  #1: ffff888101117de8 
((work_completion)(&(&con->work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.081668]  #2: ffff88810c116178 (&con->mutex){+.+.}-{4:4}, at: 
ceph_con_workfn+0x3c/0x5c0 [libceph]
<4>[  727.082041] 1 lock held by khungtaskd/96:
<4>[  727.082057]  #0: ffffffff8bf20840 (rcu_read_lock){....}-{1:3}, at: 
debug_show_all_locks+0x29/0x230
<4>[  727.082149] 1 lock held by systemd-journal/687:
<4>[  727.082179] 5 locks held by in:imjournal/882:
<4>[  727.082270] 1 lock held by sshd/3493:
<4>[  727.082320] 3 locks held by kworker/3:6/8369:
<4>[  727.082337]  #0: ffff888130cced48 
((wq_completion)ceph-msgr){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.082413]  #1: ffff8881105cfde8 
((work_completion)(&(&con->work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.082502]  #2: ffff8881803ce098 (&s->s_mutex){+.+.}-{4:4}, at: 
send_mds_reconnect+0x13e/0x7c0 [ceph]
<4>[  727.082791] 3 locks held by kworker/9:6/8586:
<4>[  727.082807]  #0: ffff888100063548 
((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.082927]  #1: ffff88812a767de8 
((work_completion)(&(&osdc->osds_timeout_work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.083002]  #2: ffff88811e378d70 (&osdc->lock){++++}-{4:4}, at: 
handle_osds_timeout+0x6f/0x1b0 [libceph]
<4>[  727.083336] 4 locks held by kworker/6:6/8587:
<4>[  727.083352]  #0: ffff888100063548 
((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.083424]  #1: ffff8881248a7de8 
((work_completion)(&(&mdsc->delayed_work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.083509]  #2: ffff8881803ce098 (&s->s_mutex){+.+.}-{4:4}, at: 
delayed_work+0x1d8/0x310 [ceph]
<4>[  727.083794]  #3: ffff88811e378d70 (&osdc->lock){++++}-{4:4}, at: 
ceph_send_cap_releases+0xbe/0x6a0 [ceph]
<4>[  727.084151] 3 locks held by kworker/8:16/8663:
<4>[  727.084170]  #0: ffff888130cced48 
((wq_completion)ceph-msgr){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.084245]  #1: ffff88810aac7de8 
((work_completion)(&(&con->work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.084316]  #2: ffff888464998178 (&con->mutex){+.+.}-{4:4}, at: 
ceph_con_workfn+0x3c/0x5c0 [libceph]
<4>[  727.084684] 3 locks held by kworker/9:18/9987:
<4>[  727.084701]  #0: ffff888100063548 
((wq_completion)events){+.+.}-{0:0}, at: process_one_work+0x436/0x990
<4>[  727.084776]  #1: ffff888112b17de8 
((work_completion)(&(&osdc->timeout_work)->work)){+.+.}-{0:0}, at: 
process_one_work+0x436/0x990
<4>[  727.084938]  #2: ffff88811e378d70 (&osdc->lock){++++}-{4:4}, at: 
handle_timeout+0x12a/0x6f0 [libceph]
<4>[  727.085274] 2 locks held by less/11897:
<4>[  727.085291]  #0: ffff8881138800a0 (&tty->ldisc_sem){++++}-{0:0}, 
at: tty_ldisc_ref_wait+0x24/0x70
<4>[  727.085370]  #1: ffffc900019122f8 
(&ldata->atomic_read_lock){+.+.}-{4:4}, at: n_tty_read+0x849/0xa70
<4>[  727.085469] 3 locks held by cat/13283:
<4>[  727.085484] 4 locks held by ffsb/13945:
<4>[  727.085497]  #0: ffff88812b95ba78 (&f->f_pos_lock){+.+.}-{4:4}, 
at: __fdget_pos+0x75/0x80
<4>[  727.085558]  #1: ffff888102274480 (sb_writers#12){.+.+}-{0:0}, at: 
ksys_write+0xc9/0x170
<4>[  727.085654]  #2: ffff88812aeca1b8 
(&sb->s_type->i_mutex_key#15){++++}-{4:4}, at: 
ceph_start_io_write+0x15/0x30 [ceph]
<4>[  727.085963]  #3: ffff88811e378d70 (&osdc->lock){++++}-{4:4}, at: 
ceph_write_iter+0x33c/0xad0 [ceph]
<4>[  727.086198] 5 locks held by kworker/3:0/13962:
<4>[  727.086213]

Thanks

- Xiubo


On 4/1/23 00:09, David Howells wrote:
> Use sendmsg() and MSG_SPLICE_PAGES rather than sendpage in ceph when
> transmitting data.  For the moment, this can only transmit one page at a
> time because of the architecture of net/ceph/, but if
> write_partial_message_data() can be given a bvec[] at a time by the
> iteration code, this would allow pages to be sent in a batch.
>
> Signed-off-by: David Howells <dhowells@redhat.com>
> cc: Ilya Dryomov <idryomov@gmail.com>
> cc: Xiubo Li <xiubli@redhat.com>
> cc: Jeff Layton <jlayton@kernel.org>
> cc: "David S. Miller" <davem@davemloft.net>
> cc: Eric Dumazet <edumazet@google.com>
> cc: Jakub Kicinski <kuba@kernel.org>
> cc: Paolo Abeni <pabeni@redhat.com>
> cc: Jens Axboe <axboe@kernel.dk>
> cc: Matthew Wilcox <willy@infradead.org>
> cc: ceph-devel@vger.kernel.org
> cc: netdev@vger.kernel.org
> ---
>   net/ceph/messenger_v2.c | 89 +++++++++--------------------------------
>   1 file changed, 18 insertions(+), 71 deletions(-)
>
> diff --git a/net/ceph/messenger_v2.c b/net/ceph/messenger_v2.c
> index 301a991dc6a6..1637a0c21126 100644
> --- a/net/ceph/messenger_v2.c
> +++ b/net/ceph/messenger_v2.c
> @@ -117,91 +117,38 @@ static int ceph_tcp_recv(struct ceph_connection *con)
>   	return ret;
>   }
>   
> -static int do_sendmsg(struct socket *sock, struct iov_iter *it)
> -{
> -	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
> -	int ret;
> -
> -	msg.msg_iter = *it;
> -	while (iov_iter_count(it)) {
> -		ret = sock_sendmsg(sock, &msg);
> -		if (ret <= 0) {
> -			if (ret == -EAGAIN)
> -				ret = 0;
> -			return ret;
> -		}
> -
> -		iov_iter_advance(it, ret);
> -	}
> -
> -	WARN_ON(msg_data_left(&msg));
> -	return 1;
> -}
> -
> -static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
> -{
> -	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
> -	struct bio_vec bv;
> -	int ret;
> -
> -	if (WARN_ON(!iov_iter_is_bvec(it)))
> -		return -EINVAL;
> -
> -	while (iov_iter_count(it)) {
> -		/* iov_iter_iovec() for ITER_BVEC */
> -		bvec_set_page(&bv, it->bvec->bv_page,
> -			      min(iov_iter_count(it),
> -				  it->bvec->bv_len - it->iov_offset),
> -			      it->bvec->bv_offset + it->iov_offset);
> -
> -		/*
> -		 * sendpage cannot properly handle pages with
> -		 * page_count == 0, we need to fall back to sendmsg if
> -		 * that's the case.
> -		 *
> -		 * Same goes for slab pages: skb_can_coalesce() allows
> -		 * coalescing neighboring slab objects into a single frag
> -		 * which triggers one of hardened usercopy checks.
> -		 */
> -		if (sendpage_ok(bv.bv_page)) {
> -			ret = sock->ops->sendpage(sock, bv.bv_page,
> -						  bv.bv_offset, bv.bv_len,
> -						  CEPH_MSG_FLAGS);
> -		} else {
> -			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
> -			ret = sock_sendmsg(sock, &msg);
> -		}
> -		if (ret <= 0) {
> -			if (ret == -EAGAIN)
> -				ret = 0;
> -			return ret;
> -		}
> -
> -		iov_iter_advance(it, ret);
> -	}
> -
> -	return 1;
> -}
> -
>   /*
>    * Write as much as possible.  The socket is expected to be corked,
>    * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here.
>    *
>    * Return:
> - *   1 - done, nothing (else) to write
> + *  >0 - done, nothing (else) to write
>    *   0 - socket is full, need to wait
>    *  <0 - error
>    */
>   static int ceph_tcp_send(struct ceph_connection *con)
>   {
> +	struct msghdr msg = {
> +		.msg_iter	= con->v2.out_iter,
> +		.msg_flags	= CEPH_MSG_FLAGS,
> +	};
>   	int ret;
>   
> +	if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter)))
> +		return -EINVAL;
> +
> +	if (con->v2.out_iter_sendpage)
> +		msg.msg_flags |= MSG_SPLICE_PAGES;
> +
>   	dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
>   	     iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
> -	if (con->v2.out_iter_sendpage)
> -		ret = do_try_sendpage(con->sock, &con->v2.out_iter);
> -	else
> -		ret = do_sendmsg(con->sock, &con->v2.out_iter);
> +
> +	ret = sock_sendmsg(con->sock, &msg);
> +	if (ret > 0)
> +		iov_iter_advance(&con->v2.out_iter, ret);
> +	else if (ret == -EAGAIN)
> +		ret = 0;
> +
>   	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
>   	     iov_iter_count(&con->v2.out_iter));
>   	return ret;
>
diff mbox series

Patch

diff --git a/net/ceph/messenger_v2.c b/net/ceph/messenger_v2.c
index 301a991dc6a6..1637a0c21126 100644
--- a/net/ceph/messenger_v2.c
+++ b/net/ceph/messenger_v2.c
@@ -117,91 +117,38 @@  static int ceph_tcp_recv(struct ceph_connection *con)
 	return ret;
 }
 
-static int do_sendmsg(struct socket *sock, struct iov_iter *it)
-{
-	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
-	int ret;
-
-	msg.msg_iter = *it;
-	while (iov_iter_count(it)) {
-		ret = sock_sendmsg(sock, &msg);
-		if (ret <= 0) {
-			if (ret == -EAGAIN)
-				ret = 0;
-			return ret;
-		}
-
-		iov_iter_advance(it, ret);
-	}
-
-	WARN_ON(msg_data_left(&msg));
-	return 1;
-}
-
-static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
-{
-	struct msghdr msg = { .msg_flags = CEPH_MSG_FLAGS };
-	struct bio_vec bv;
-	int ret;
-
-	if (WARN_ON(!iov_iter_is_bvec(it)))
-		return -EINVAL;
-
-	while (iov_iter_count(it)) {
-		/* iov_iter_iovec() for ITER_BVEC */
-		bvec_set_page(&bv, it->bvec->bv_page,
-			      min(iov_iter_count(it),
-				  it->bvec->bv_len - it->iov_offset),
-			      it->bvec->bv_offset + it->iov_offset);
-
-		/*
-		 * sendpage cannot properly handle pages with
-		 * page_count == 0, we need to fall back to sendmsg if
-		 * that's the case.
-		 *
-		 * Same goes for slab pages: skb_can_coalesce() allows
-		 * coalescing neighboring slab objects into a single frag
-		 * which triggers one of hardened usercopy checks.
-		 */
-		if (sendpage_ok(bv.bv_page)) {
-			ret = sock->ops->sendpage(sock, bv.bv_page,
-						  bv.bv_offset, bv.bv_len,
-						  CEPH_MSG_FLAGS);
-		} else {
-			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
-			ret = sock_sendmsg(sock, &msg);
-		}
-		if (ret <= 0) {
-			if (ret == -EAGAIN)
-				ret = 0;
-			return ret;
-		}
-
-		iov_iter_advance(it, ret);
-	}
-
-	return 1;
-}
-
 /*
  * Write as much as possible.  The socket is expected to be corked,
  * so we don't bother with MSG_MORE/MSG_SENDPAGE_NOTLAST here.
  *
  * Return:
- *   1 - done, nothing (else) to write
+ *  >0 - done, nothing (else) to write
  *   0 - socket is full, need to wait
  *  <0 - error
  */
 static int ceph_tcp_send(struct ceph_connection *con)
 {
+	struct msghdr msg = {
+		.msg_iter	= con->v2.out_iter,
+		.msg_flags	= CEPH_MSG_FLAGS,
+	};
 	int ret;
 
+	if (WARN_ON(!iov_iter_is_bvec(&con->v2.out_iter)))
+		return -EINVAL;
+
+	if (con->v2.out_iter_sendpage)
+		msg.msg_flags |= MSG_SPLICE_PAGES;
+
 	dout("%s con %p have %zu try_sendpage %d\n", __func__, con,
 	     iov_iter_count(&con->v2.out_iter), con->v2.out_iter_sendpage);
-	if (con->v2.out_iter_sendpage)
-		ret = do_try_sendpage(con->sock, &con->v2.out_iter);
-	else
-		ret = do_sendmsg(con->sock, &con->v2.out_iter);
+
+	ret = sock_sendmsg(con->sock, &msg);
+	if (ret > 0)
+		iov_iter_advance(&con->v2.out_iter, ret);
+	else if (ret == -EAGAIN)
+		ret = 0;
+
 	dout("%s con %p ret %d left %zu\n", __func__, con, ret,
 	     iov_iter_count(&con->v2.out_iter));
 	return ret;