From 339b84ab6b1d66900c27bd999271cb2ae40ce812 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Thu, 20 Jun 2024 09:45:09 -0400 Subject: closures: Change BUG_ON() to WARN_ON() If a BUG_ON() can be hit in the wild, it shouldn't be a BUG_ON() For reference, this has popped up once in the CI, and we'll need more info to debug it: 03240 ------------[ cut here ]------------ 03240 kernel BUG at lib/closure.c:21! 03240 kernel BUG at lib/closure.c:21! 03240 Internal error: Oops - BUG: 00000000f2000800 [#1] SMP 03240 Modules linked in: 03240 CPU: 15 PID: 40534 Comm: kworker/u80:1 Not tainted 6.10.0-rc4-ktest-ga56da69799bd #25570 03240 Hardware name: linux,dummy-virt (DT) 03240 Workqueue: btree_update btree_interior_update_work 03240 pstate: 00001005 (nzcv daif -PAN -UAO -TCO -DIT +SSBS BTYPE=--) 03240 pc : closure_put+0x224/0x2a0 03240 lr : closure_put+0x24/0x2a0 03240 sp : ffff0000d12071c0 03240 x29: ffff0000d12071c0 x28: dfff800000000000 x27: ffff0000d1207360 03240 x26: 0000000000000040 x25: 0000000000000040 x24: 0000000000000040 03240 x23: ffff0000c1f20180 x22: 0000000000000000 x21: ffff0000c1f20168 03240 x20: 0000000040000000 x19: ffff0000c1f20140 x18: 0000000000000001 03240 x17: 0000000000003aa0 x16: 0000000000003ad0 x15: 1fffe0001c326974 03240 x14: 0000000000000a1e x13: 0000000000000000 x12: 1fffe000183e402d 03240 x11: ffff6000183e402d x10: dfff800000000000 x9 : ffff6000183e402e 03240 x8 : 0000000000000001 x7 : 00009fffe7c1bfd3 x6 : ffff0000c1f2016b 03240 x5 : ffff0000c1f20168 x4 : ffff6000183e402e x3 : ffff800081391954 03240 x2 : 0000000000000001 x1 : 0000000000000000 x0 : 00000000a8000000 03240 Call trace: 03240 closure_put+0x224/0x2a0 03240 bch2_check_for_deadlock+0x910/0x1028 03240 bch2_six_check_for_deadlock+0x1c/0x30 03240 six_lock_slowpath.isra.0+0x29c/0xed0 03240 six_lock_ip_waiter+0xa8/0xf8 03240 __bch2_btree_node_lock_write+0x14c/0x298 03240 bch2_trans_lock_write+0x6d4/0xb10 03240 __bch2_trans_commit+0x135c/0x5520 03240 btree_interior_update_work+0x1248/0x1c10 03240 process_scheduled_works+0x53c/0xd90 03240 worker_thread+0x370/0x8c8 03240 kthread+0x258/0x2e8 03240 ret_from_fork+0x10/0x20 03240 Code: aa1303e0 d63f0020 a94363f7 17ffff8c (d4210000) 03240 ---[ end trace 0000000000000000 ]--- 03240 Kernel panic - not syncing: Oops - BUG: Fatal exception 03240 SMP: stopping secondary CPUs 03241 SMP: failed to stop secondary CPUs 13,15 03241 Kernel Offset: disabled 03241 CPU features: 0x00,00000003,80000008,4240500b 03241 Memory Limit: none 03241 ---[ end Kernel panic - not syncing: Oops - BUG: Fatal exception ]--- 03246 ========= FAILED TIMEOUT copygc_torture_no_checksum in 7200s Signed-off-by: Kent Overstreet --- lib/closure.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'lib/closure.c') diff --git a/lib/closure.c b/lib/closure.c index 07409e9e35a5..2e1ee9fdec08 100644 --- a/lib/closure.c +++ b/lib/closure.c @@ -17,12 +17,18 @@ static inline void closure_put_after_sub(struct closure *cl, int flags) { int r = flags & CLOSURE_REMAINING_MASK; - BUG_ON(flags & CLOSURE_GUARD_MASK); - BUG_ON(!r && (flags & ~CLOSURE_DESTRUCTOR)); + if (WARN(flags & CLOSURE_GUARD_MASK, + "closure has guard bits set: %x (%u)", + flags & CLOSURE_GUARD_MASK, (unsigned) __fls(r))) + r &= ~CLOSURE_GUARD_MASK; if (!r) { smp_acquire__after_ctrl_dep(); + WARN(flags & ~CLOSURE_DESTRUCTOR, + "closure ref hit 0 with incorrect flags set: %x (%u)", + flags & ~CLOSURE_DESTRUCTOR, (unsigned) __fls(flags)); + cl->closure_get_happened = false; if (cl->fn && !(flags & CLOSURE_DESTRUCTOR)) { -- cgit From 06efa5f30c28eaf237247ca8c4cb46eb62cb6bd9 Mon Sep 17 00:00:00 2001 From: Kent Overstreet Date: Sat, 22 Jun 2024 21:38:58 -0400 Subject: closures: closure_get_not_zero(), closure_return_sync() Provide new primitives for solving a lifetime issue with bcachefs btree_trans objects. closure_sync_return(): like closure_sync(), wait synchronously for any outstanding gets. like closure_return, the closure is considered "finished" and the ref left at 0. closure_get_not_zero(): get a ref on a closure if it's alive, i.e. the ref is not zero. Signed-off-by: Kent Overstreet --- lib/closure.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 46 insertions(+), 6 deletions(-) (limited to 'lib/closure.c') diff --git a/lib/closure.c b/lib/closure.c index 2e1ee9fdec08..c971216d9d77 100644 --- a/lib/closure.c +++ b/lib/closure.c @@ -13,7 +13,7 @@ #include #include -static inline void closure_put_after_sub(struct closure *cl, int flags) +static inline void closure_put_after_sub_checks(int flags) { int r = flags & CLOSURE_REMAINING_MASK; @@ -22,12 +22,17 @@ static inline void closure_put_after_sub(struct closure *cl, int flags) flags & CLOSURE_GUARD_MASK, (unsigned) __fls(r))) r &= ~CLOSURE_GUARD_MASK; - if (!r) { - smp_acquire__after_ctrl_dep(); + WARN(!r && (flags & ~CLOSURE_DESTRUCTOR), + "closure ref hit 0 with incorrect flags set: %x (%u)", + flags & ~CLOSURE_DESTRUCTOR, (unsigned) __fls(flags)); +} + +static inline void closure_put_after_sub(struct closure *cl, int flags) +{ + closure_put_after_sub_checks(flags); - WARN(flags & ~CLOSURE_DESTRUCTOR, - "closure ref hit 0 with incorrect flags set: %x (%u)", - flags & ~CLOSURE_DESTRUCTOR, (unsigned) __fls(flags)); + if (!(flags & CLOSURE_REMAINING_MASK)) { + smp_acquire__after_ctrl_dep(); cl->closure_get_happened = false; @@ -145,6 +150,41 @@ void __sched __closure_sync(struct closure *cl) } EXPORT_SYMBOL(__closure_sync); +/* + * closure_return_sync - finish running a closure, synchronously (i.e. waiting + * for outstanding get()s to finish) and returning once closure refcount is 0. + * + * Unlike closure_sync() this doesn't reinit the ref to 1; subsequent + * closure_get_not_zero() calls waill fail. + */ +void __sched closure_return_sync(struct closure *cl) +{ + struct closure_syncer s = { .task = current }; + + cl->s = &s; + set_closure_fn(cl, closure_sync_fn, NULL); + + unsigned flags = atomic_sub_return_release(1 + CLOSURE_RUNNING - CLOSURE_DESTRUCTOR, + &cl->remaining); + + closure_put_after_sub_checks(flags); + + if (unlikely(flags & CLOSURE_REMAINING_MASK)) { + while (1) { + set_current_state(TASK_UNINTERRUPTIBLE); + if (s.done) + break; + schedule(); + } + + __set_current_state(TASK_RUNNING); + } + + if (cl->parent) + closure_put(cl->parent); +} +EXPORT_SYMBOL(closure_return_sync); + int __sched __closure_sync_timeout(struct closure *cl, unsigned long timeout) { struct closure_syncer s = { .task = current }; -- cgit