| Message ID | 20240612133357.2596-1-linus.luessing@c0d3.blue (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | Revert "batman-adv: prefer kfree_rcu() over call_rcu() with free-only callbacks" | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On Wed, Jun 12, 2024 at 03:33:57PM +0200, Linus Lüssing wrote: > This reverts commit cf9845d4566698731c23cd7a02db956706d058a1. > > This change seems to result in a memory leak / RCU race and the following > kernel splat when the batman-adv kernel module is unloaded: > > ``` > [ 112.208633] ============================================================================= > [ 112.210359] BUG batadv_tl_cache (Tainted: G OE ): Objects remaining in batadv_tl_cache on __kmem_cache_shutdown() > [ 112.211943] ----------------------------------------------------------------------------- > > [ 112.212517] Slab 0xffffe8afc0216d00 objects=16 used=1 fp=0xffff93f4085b4340 flags=0xfffffc0000a00(workingset|slab|node=0|zone=1|lastcpupid=0x1fffff) > [ 112.212517] CPU: 1 PID: 776 Comm: rmmod Tainted: G OE 6.8.12-amd64 #1 Debian 6.8.12-1 > [ 112.212517] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 > [ 112.212517] Call Trace: > [ 112.212517] <TASK> > [ 112.212517] dump_stack_lvl+0x64/0x80 > [ 112.212517] slab_err+0xe6/0x120 > [ 112.212517] __kmem_cache_shutdown+0x160/0x2e0 > [ 112.212517] kmem_cache_destroy+0x55/0x160 > [ 112.220849] batadv_tt_cache_destroy+0x15/0x60 [batman_adv] > [ 112.220849] __do_sys_delete_module+0x1d5/0x320 > [ 112.220849] do_syscall_64+0x83/0x190 > [ 112.220849] ? do_syscall_64+0x8f/0x190 > [ 112.220849] ? exc_page_fault+0x7f/0x180 > [ 112.220849] entry_SYSCALL_64_after_hwframe+0x78/0x80 > [ 112.224478] RIP: 0033:0x7f2ac8434977 > [ 112.224478] Code: 73 01 c3 48 8b 0d a9 94 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 79 94 0c 00 f7 d8 64 89 01 48 > [ 112.224478] RSP: 002b:00007ffe0adf6138 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 > [ 112.224478] RAX: ffffffffffffffda RBX: 000055db9018e770 RCX: 00007f2ac8434977 > [ 112.224478] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 000055db9018e7d8 > [ 112.224478] RBP: 0000000000000000 R08: 1999999999999999 R09: 0000000000000000 > [ 112.224478] R10: 00007f2ac84a6ac0 R11: 0000000000000206 R12: 00007ffe0adf6390 > [ 112.224478] R13: 000055db9018e770 R14: 000055db9018d2a0 R15: 0000000000000000 > [ 112.233961] </TASK> > [ 112.233961] Disabling lock debugging due to kernel taint > [ 112.233961] Object 0xffff93f4085b4140 @offset=320 > [ 112.233961] Allocated in batadv_tt_local_add+0x297/0xa20 [batman_adv] age=15835 cpu=1 pid=755 > [ 112.233961] batadv_tt_local_add+0x297/0xa20 [batman_adv] > [ 112.233961] batadv_interface_set_mac_addr+0xf6/0x120 [batman_adv] > [ 112.233961] dev_set_mac_address+0xde/0x140 > [ 112.233961] dev_set_mac_address_user+0x30/0x50 > [ 112.233961] do_setlink+0x261/0x12d0 > [ 112.233961] rtnl_setlink+0x11f/0x1d0 > [ 112.233961] rtnetlink_rcv_msg+0x152/0x3c0 > [ 112.241772] netlink_rcv_skb+0x5b/0x110 > [ 112.241772] netlink_unicast+0x1a6/0x290 > [ 112.241772] netlink_sendmsg+0x223/0x490 > [ 112.241772] __sys_sendto+0x1df/0x1f0 > [ 112.241772] __x64_sys_sendto+0x24/0x30 > [ 112.241772] do_syscall_64+0x83/0x190 > [ 112.241772] entry_SYSCALL_64_after_hwframe+0x78/0x80 > [ 112.245994] ------------[ cut here ]------------ > [ 112.246650] kmem_cache_destroy batadv_tl_cache: Slab cache still has objects when called from batadv_tt_cache_destroy+0x15/0x60 [batman_adv] > [ 112.246668] WARNING: CPU: 1 PID: 776 at mm/slab_common.c:493 kmem_cache_destroy+0x14d/0x160 > [ 112.249584] Modules linked in: veth batman_adv(OE-) cfg80211 rfkill bridge stp llc libcrc32c crc32c_generic crc16 rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver binfmt_misc pcspkr button joydev evdev serio_raw loop dm_mod efi_pstore nfnetlink vsock_loopback vmw_vsock_virtio_transport_common vmw_vsock_vmci_transport vsock vmw_vmci qemu_fw_cfg ip_tables x_tables autofs4 nfsv3 nfs_acl nfs lockd grace sunrpc 9pnet_rdma rdma_cm iw_cm ib_cm ib_core configfs 9p netfs ata_generic ata_piix libata psmouse scsi_mod 9pnet_virtio i2c_piix4 9pnet e1000 scsi_common floppy crypto_simd cryptd > [ 112.256555] CPU: 1 PID: 776 Comm: rmmod Tainted: G B OE 6.8.12-amd64 #1 Debian 6.8.12-1 > [ 112.258457] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 > [ 112.260410] RIP: 0010:kmem_cache_destroy+0x14d/0x160 > [ 112.261687] Code: 00 eb be 5b 5d 41 5c 41 5d c3 cc cc cc cc 48 8b 53 60 48 8b 4c 24 20 48 c7 c6 60 d5 e3 98 48 c7 c7 b8 ec 2d 99 e8 43 0d d8 ff <0f> 0b e9 e2 fe ff ff c3 cc cc cc cc 0f 1f 80 00 00 00 00 90 90 90 > [ 112.265219] RSP: 0018:ffffb3b2806e7e48 EFLAGS: 00010282 > [ 112.266044] RAX: 0000000000000000 RBX: ffff93f4270a2640 RCX: 0000000000000027 > [ 112.267157] RDX: ffff93f43c521708 RSI: 0000000000000001 RDI: ffff93f43c521700 > [ 112.268268] RBP: 000055db9018e7d8 R08: 0000000000000000 R09: ffffb3b2806e7cd8 > [ 112.269418] R10: ffffb3b2806e7cd0 R11: 0000000000000003 R12: 0000000080012d00 > [ 112.270572] R13: ffffb3b2806e7f58 R14: 0000000000000000 R15: 0000000000000000 > [ 112.271699] FS: 00007f2ac8308440(0000) GS:ffff93f43c500000(0000) knlGS:0000000000000000 > [ 112.273001] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > [ 112.273923] CR2: 00005584ef830110 CR3: 000000000787c000 CR4: 00000000000006f0 > [ 112.275050] Call Trace: > [ 112.275464] <TASK> > [ 112.275810] ? kmem_cache_destroy+0x14d/0x160 > [ 112.276518] ? __warn+0x81/0x130 > [ 112.277043] ? kmem_cache_destroy+0x14d/0x160 > [ 112.277730] ? report_bug+0x171/0x1a0 > [ 112.278315] ? prb_read_valid+0x1b/0x30 > [ 112.278919] ? handle_bug+0x3c/0x80 > [ 112.279467] ? exc_invalid_op+0x17/0x70 > [ 112.280071] ? asm_exc_invalid_op+0x1a/0x20 > [ 112.280741] ? kmem_cache_destroy+0x14d/0x160 > [ 112.281603] ? kmem_cache_destroy+0x14d/0x160 > [ 112.282489] batadv_tt_cache_destroy+0x15/0x60 [batman_adv] > [ 112.283373] __do_sys_delete_module+0x1d5/0x320 > [ 112.284080] do_syscall_64+0x83/0x190 > [ 112.284696] ? do_syscall_64+0x8f/0x190 > [ 112.285315] ? exc_page_fault+0x7f/0x180 > [ 112.285970] entry_SYSCALL_64_after_hwframe+0x78/0x80 > [ 112.286768] RIP: 0033:0x7f2ac8434977 > [ 112.287355] Code: 73 01 c3 48 8b 0d a9 94 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 79 94 0c 00 f7 d8 64 89 01 48 > [ 112.290282] RSP: 002b:00007ffe0adf6138 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 > [ 112.291465] RAX: ffffffffffffffda RBX: 000055db9018e770 RCX: 00007f2ac8434977 > [ 112.292595] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 000055db9018e7d8 > [ 112.293724] RBP: 0000000000000000 R08: 1999999999999999 R09: 0000000000000000 > [ 112.294863] R10: 00007f2ac84a6ac0 R11: 0000000000000206 R12: 00007ffe0adf6390 > [ 112.295982] R13: 000055db9018e770 R14: 000055db9018d2a0 R15: 0000000000000000 > [ 112.297103] </TASK> > [ 112.297465] ---[ end trace 0000000000000000 ]--- > ``` > > So far, after some debugging, the actual cause for this could > not immediately be found within the batman-adv code. > Therefore reverting this for now until the underlying issue can be > found and better understood. > > Some additional debugging information and discussions can be found > on our Redmine bugtracker, linked below. > > Link: https://www.open-mesh.org/issues/428 > Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue> > --- > Some help / guidance for further debugging from people more > knowledgeable in the RCU core internals would be appreciated. > > Specifically why the rcu_barrier() call in batadv_exit() does not > seem to catch the kfree_rcu() which the reverted patch was > originally adding. Let me make sure that I understand... You need rcu_barrier() to wait for any memory passed to kfree_rcu() to actually be freed? If so, please explain why you need this, as in what bad thing happens if the actual kfree() happens later. (I could imagine something involving OOM avoidance, but I need to hear your code's needs rather than my imaginations.) Thanx, Paul > compat.h | 8 ----- > net/batman-adv/translation-table.c | 47 ++++++++++++++++++++++++++++-- > 2 files changed, 44 insertions(+), 11 deletions(-) > > diff --git a/compat.h b/compat.h > index d823eeb7bc70..638bfc54f3d0 100644 > --- a/compat.h > +++ b/compat.h > @@ -16,14 +16,6 @@ > #include "compat-autoconf.h" > > > -#if LINUX_VERSION_IS_LESS(6, 4, 0) > - > -#if IS_ENABLED(CONFIG_SLOB) > -#error kfree_rcu for kmem_cache not supported when SLOB is enabled > -#endif > - > -#endif /* LINUX_VERSION_IS_LESS(6, 4, 0) */ > - > #endif /* __KERNEL__ */ > > #endif /* _NET_BATMAN_ADV_COMPAT_H_ */ > diff --git a/net/batman-adv/translation-table.c b/net/batman-adv/translation-table.c > index b21ff3c36b07..2243cec18ecc 100644 > --- a/net/batman-adv/translation-table.c > +++ b/net/batman-adv/translation-table.c > @@ -208,6 +208,20 @@ batadv_tt_global_hash_find(struct batadv_priv *bat_priv, const u8 *addr, > return tt_global_entry; > } > > +/** > + * batadv_tt_local_entry_free_rcu() - free the tt_local_entry > + * @rcu: rcu pointer of the tt_local_entry > + */ > +static void batadv_tt_local_entry_free_rcu(struct rcu_head *rcu) > +{ > + struct batadv_tt_local_entry *tt_local_entry; > + > + tt_local_entry = container_of(rcu, struct batadv_tt_local_entry, > + common.rcu); > + > + kmem_cache_free(batadv_tl_cache, tt_local_entry); > +} > + > /** > * batadv_tt_local_entry_release() - release tt_local_entry from lists and queue > * for free after rcu grace period > @@ -222,7 +236,7 @@ static void batadv_tt_local_entry_release(struct kref *ref) > > batadv_softif_vlan_put(tt_local_entry->vlan); > > - kfree_rcu(tt_local_entry, common.rcu); > + call_rcu(&tt_local_entry->common.rcu, batadv_tt_local_entry_free_rcu); > } > > /** > @@ -240,6 +254,20 @@ batadv_tt_local_entry_put(struct batadv_tt_local_entry *tt_local_entry) > batadv_tt_local_entry_release); > } > > +/** > + * batadv_tt_global_entry_free_rcu() - free the tt_global_entry > + * @rcu: rcu pointer of the tt_global_entry > + */ > +static void batadv_tt_global_entry_free_rcu(struct rcu_head *rcu) > +{ > + struct batadv_tt_global_entry *tt_global_entry; > + > + tt_global_entry = container_of(rcu, struct batadv_tt_global_entry, > + common.rcu); > + > + kmem_cache_free(batadv_tg_cache, tt_global_entry); > +} > + > /** > * batadv_tt_global_entry_release() - release tt_global_entry from lists and > * queue for free after rcu grace period > @@ -254,7 +282,7 @@ void batadv_tt_global_entry_release(struct kref *ref) > > batadv_tt_global_del_orig_list(tt_global_entry); > > - kfree_rcu(tt_global_entry, common.rcu); > + call_rcu(&tt_global_entry->common.rcu, batadv_tt_global_entry_free_rcu); > } > > /** > @@ -379,6 +407,19 @@ static void batadv_tt_global_size_dec(struct batadv_orig_node *orig_node, > batadv_tt_global_size_mod(orig_node, vid, -1); > } > > +/** > + * batadv_tt_orig_list_entry_free_rcu() - free the orig_entry > + * @rcu: rcu pointer of the orig_entry > + */ > +static void batadv_tt_orig_list_entry_free_rcu(struct rcu_head *rcu) > +{ > + struct batadv_tt_orig_list_entry *orig_entry; > + > + orig_entry = container_of(rcu, struct batadv_tt_orig_list_entry, rcu); > + > + kmem_cache_free(batadv_tt_orig_cache, orig_entry); > +} > + > /** > * batadv_tt_orig_list_entry_release() - release tt orig entry from lists and > * queue for free after rcu grace period > @@ -392,7 +433,7 @@ static void batadv_tt_orig_list_entry_release(struct kref *ref) > refcount); > > batadv_orig_node_put(orig_entry->orig_node); > - kfree_rcu(orig_entry, rcu); > + call_rcu(&orig_entry->rcu, batadv_tt_orig_list_entry_free_rcu); > } > > /** > -- > 2.45.1 > >
On Wed, Jun 12, 2024 at 07:06:04AM -0700, Paul E. McKenney wrote: > Let me make sure that I understand... > > You need rcu_barrier() to wait for any memory passed to kfree_rcu() > to actually be freed? If so, please explain why you need this, as > in what bad thing happens if the actual kfree() happens later. > > (I could imagine something involving OOM avoidance, but I need to > hear your code's needs rather than my imaginations.) > > Thanx, Paul We have allocated a kmem-cache for some objects, which are like batman-adv's version of a bridge's FDB entry. The very last thing we do before unloading the module is free'ing/destroying this kmem-cache with a call to kmem_cache_destroy(). As far as I understand before calling kmem_cache_destroy() we need to ensure that all previously allocated objects on this kmem-cache were free'd. At least we get this kernel splat (from Slub?) otherwise. I'm not quite sure if any other bad things other than this noise in dmesg would occur though. Other than a stale, zero objects entry remaining in /proc/slabinfo maybe. Which gets duplicated everytime we repeat loading+unloading the module. At least these entries would be a memory leak I suppose? ``` # after insmod/rmmod'ing batman-adv 6 times: $ cat /proc/slabinfo | grep batadv_tl_cache batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 ``` That's why we added this rcu_barrier() call on module shutdown in the batman-adv module __exit function right before the kmem_cache_destroy() calls. Hoping that this would wait for all call_rcu() / kfree_rcu() callbacks and their final kfree() to finish. This worked when we were using call_rcu() with our own callback with a kfree(). However for kfree_rcu() this somehow does not seem to be the case anymore (- or more likely I'm missing something else, some other bug within the batman-adv code?). Regards, Linus
On Wed, Jun 12, 2024 at 04:39:15PM +0200, Linus Lüssing wrote: > On Wed, Jun 12, 2024 at 07:06:04AM -0700, Paul E. McKenney wrote: > > Let me make sure that I understand... > > > > You need rcu_barrier() to wait for any memory passed to kfree_rcu() > > to actually be freed? If so, please explain why you need this, as > > in what bad thing happens if the actual kfree() happens later. > > > > (I could imagine something involving OOM avoidance, but I need to > > hear your code's needs rather than my imaginations.) > > > > Thanx, Paul > [...] > As far as I understand before calling kmem_cache_destroy() > we need to ensure that all previously allocated objects on this > kmem-cache were free'd. At least we get this kernel splat > (from Slub?) otherwise. I'm not quite sure if any other bad things > other than this noise in dmesg would occur though. Other than a > [...] I guess, without knowing the details of RCU and Slub, that at least nothing super serious, like a segfault, can happen when the remaining execution is just a kfree(), which won't need access to batman-adv internal functions anymore.
On Wed, Jun 12, 2024 at 04:39:15PM +0200, Linus Lüssing wrote: > On Wed, Jun 12, 2024 at 07:06:04AM -0700, Paul E. McKenney wrote: > > Let me make sure that I understand... > > > > You need rcu_barrier() to wait for any memory passed to kfree_rcu() > > to actually be freed? If so, please explain why you need this, as > > in what bad thing happens if the actual kfree() happens later. > > > > (I could imagine something involving OOM avoidance, but I need to > > hear your code's needs rather than my imaginations.) > > > > Thanx, Paul > > We have allocated a kmem-cache for some objects, which are like > batman-adv's version of a bridge's FDB entry. > > The very last thing we do before unloading the module is > free'ing/destroying this kmem-cache with a call to > kmem_cache_destroy(). > > As far as I understand before calling kmem_cache_destroy() > we need to ensure that all previously allocated objects on this > kmem-cache were free'd. At least we get this kernel splat > (from Slub?) otherwise. I'm not quite sure if any other bad things > other than this noise in dmesg would occur though. Other than a > stale, zero objects entry remaining in /proc/slabinfo maybe. Which > gets duplicated everytime we repeat loading+unloading the module. > At least these entries would be a memory leak I suppose? > > ``` > # after insmod/rmmod'ing batman-adv 6 times: > $ cat /proc/slabinfo | grep batadv_tl_cache > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > ``` > > That's why we added this rcu_barrier() call on module > shutdown in the batman-adv module __exit function right before the > kmem_cache_destroy() calls. Hoping that this would wait for all > call_rcu() / kfree_rcu() callbacks and their final kfree() to finish. > This worked when we were using call_rcu() with our own callback > with a kfree(). However for kfree_rcu() this somehow does not seem > to be the case anymore (- or more likely I'm missing something else, > some other bug within the batman-adv code?). It is quite possible that some of the recent energy-saving changes have caused rcu_barrier() to not wait for all kfree_rcu() memory to be freed. Which is timely, given a bunch of recently proposed changes that seemed like a good idea to me at the time. ;-) Thanx, Paul
On Wed, Jun 12, 2024 at 04:54:49PM +0200, Linus Lüssing wrote: > On Wed, Jun 12, 2024 at 04:39:15PM +0200, Linus Lüssing wrote: > > On Wed, Jun 12, 2024 at 07:06:04AM -0700, Paul E. McKenney wrote: > > > Let me make sure that I understand... > > > > > > You need rcu_barrier() to wait for any memory passed to kfree_rcu() > > > to actually be freed? If so, please explain why you need this, as > > > in what bad thing happens if the actual kfree() happens later. > > > > > > (I could imagine something involving OOM avoidance, but I need to > > > hear your code's needs rather than my imaginations.) > > > > > > Thanx, Paul > > [...] > > As far as I understand before calling kmem_cache_destroy() > > we need to ensure that all previously allocated objects on this > > kmem-cache were free'd. At least we get this kernel splat > > (from Slub?) otherwise. I'm not quite sure if any other bad things > > other than this noise in dmesg would occur though. Other than a > > [...] > > I guess, without knowing the details of RCU and Slub, that at > least nothing super serious, like a segfault, can happen when > the remaining execution is just a kfree(), which won't need > access to batman-adv internal functions anymore. We are looking into nice ways of solving this, but in the meantime, yes, if you are RCU-freeing slab objects into a slab that is destroyed at module-unload time, you currently need to stick with call_rcu() and rcu_barrier(). We do have some potential solutions to allow use of kfree_rcu() with this sort of slab, but they are still strictly potential. Apologies for my having failed to foresee this particular trap! Thanx, Paul
+ Vlastimil Babka + Matthew Wilcox > On Wed, Jun 12, 2024 at 07:06:04AM -0700, Paul E. McKenney wrote: > > Let me make sure that I understand... > > > > You need rcu_barrier() to wait for any memory passed to kfree_rcu() > > to actually be freed? If so, please explain why you need this, as > > in what bad thing happens if the actual kfree() happens later. > > > > (I could imagine something involving OOM avoidance, but I need to > > hear your code's needs rather than my imaginations.) > > > > Thanx, Paul > > We have allocated a kmem-cache for some objects, which are like > batman-adv's version of a bridge's FDB entry. > > The very last thing we do before unloading the module is > free'ing/destroying this kmem-cache with a call to > kmem_cache_destroy(). > > As far as I understand before calling kmem_cache_destroy() > we need to ensure that all previously allocated objects on this > kmem-cache were free'd. At least we get this kernel splat > (from Slub?) otherwise. I'm not quite sure if any other bad things > other than this noise in dmesg would occur though. Other than a > stale, zero objects entry remaining in /proc/slabinfo maybe. Which > gets duplicated everytime we repeat loading+unloading the module. > At least these entries would be a memory leak I suppose? > > ``` > # after insmod/rmmod'ing batman-adv 6 times: > $ cat /proc/slabinfo | grep batadv_tl_cache > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > batadv_tl_cache 0 16 256 16 1 : tunables 0 0 0 : slabdata 1 1 0 > ``` > > That's why we added this rcu_barrier() call on module > shutdown in the batman-adv module __exit function right before the > kmem_cache_destroy() calls. Hoping that this would wait for all > call_rcu() / kfree_rcu() callbacks and their final kfree() to finish. > This worked when we were using call_rcu() with our own callback > with a kfree(). However for kfree_rcu() this somehow does not seem > to be the case anymore (- or more likely I'm missing something else, > some other bug within the batman-adv code?). > Some background: Before kfree_rcu() could deal only with "internal system slab caches" which are static and live forever. After removing SLAB allocator it become possible to free a memory over RCU using kfree_rcu() without specifying a kmem-cache an object belongs to because two remaining allocators are capable of convert an object to its cache internally. So, now, kfree_rcu() does not need to be aware of any cache and this is something new. In your scenario the cache is destroyed and after that kfree_rcu() started to free objects into non-existing cache which is a problem. I have a question to Vlastimil. Is kmem_cached_destroy() removes the cache right away even though there are allocated objects which have not yet been freed? If so, is that possible to destroy the cache only when usage counter becomes zero? i.e. when all objects were returned back to the cache. Thank you! -- Uladzislau Rezki
On Wed, Jun 12, 2024 at 09:06:25AM -0700, Paul E. McKenney wrote: > We are looking into nice ways of solving this, but in the meantime, > yes, if you are RCU-freeing slab objects into a slab that is destroyed > at module-unload time, you currently need to stick with call_rcu() > and rcu_barrier(). > > We do have some potential solutions to allow use of kfree_rcu() with > this sort of slab, but they are still strictly potential. > > Apologies for my having failed to foresee this particular trap! > > Thanx, Paul No worries, thanks for the help and clarification! This at least restored my sanity, was starting to doubt my understanding of RCU and the batman-adv code the longer I tried to find the issue in batman-adv :D.
On Wednesday, 12 June 2024 18:31:57 CEST Linus Lüssing wrote: > On Wed, Jun 12, 2024 at 09:06:25AM -0700, Paul E. McKenney wrote: > > We are looking into nice ways of solving this, but in the meantime, > > yes, if you are RCU-freeing slab objects into a slab that is destroyed > > at module-unload time, you currently need to stick with call_rcu() > > and rcu_barrier(). > > > > We do have some potential solutions to allow use of kfree_rcu() with > > this sort of slab, but they are still strictly potential. > > > > Apologies for my having failed to foresee this particular trap! > > No worries, thanks for the help and clarification! This at least > restored my sanity, was starting to doubt my understanding of RCU > and the batman-adv code the longer I tried to find the issue in > batman-adv :D. Thanks Linus and Paul. I've queued up the revert. But feel free to submit a version with updated text in case you want to incorporate information from this thread. Kind regards, Sven
diff --git a/compat.h b/compat.h index d823eeb7bc70..638bfc54f3d0 100644 --- a/compat.h +++ b/compat.h @@ -16,14 +16,6 @@ #include "compat-autoconf.h" -#if LINUX_VERSION_IS_LESS(6, 4, 0) - -#if IS_ENABLED(CONFIG_SLOB) -#error kfree_rcu for kmem_cache not supported when SLOB is enabled -#endif - -#endif /* LINUX_VERSION_IS_LESS(6, 4, 0) */ - #endif /* __KERNEL__ */ #endif /* _NET_BATMAN_ADV_COMPAT_H_ */ diff --git a/net/batman-adv/translation-table.c b/net/batman-adv/translation-table.c index b21ff3c36b07..2243cec18ecc 100644 --- a/net/batman-adv/translation-table.c +++ b/net/batman-adv/translation-table.c @@ -208,6 +208,20 @@ batadv_tt_global_hash_find(struct batadv_priv *bat_priv, const u8 *addr, return tt_global_entry; } +/** + * batadv_tt_local_entry_free_rcu() - free the tt_local_entry + * @rcu: rcu pointer of the tt_local_entry + */ +static void batadv_tt_local_entry_free_rcu(struct rcu_head *rcu) +{ + struct batadv_tt_local_entry *tt_local_entry; + + tt_local_entry = container_of(rcu, struct batadv_tt_local_entry, + common.rcu); + + kmem_cache_free(batadv_tl_cache, tt_local_entry); +} + /** * batadv_tt_local_entry_release() - release tt_local_entry from lists and queue * for free after rcu grace period @@ -222,7 +236,7 @@ static void batadv_tt_local_entry_release(struct kref *ref) batadv_softif_vlan_put(tt_local_entry->vlan); - kfree_rcu(tt_local_entry, common.rcu); + call_rcu(&tt_local_entry->common.rcu, batadv_tt_local_entry_free_rcu); } /** @@ -240,6 +254,20 @@ batadv_tt_local_entry_put(struct batadv_tt_local_entry *tt_local_entry) batadv_tt_local_entry_release); } +/** + * batadv_tt_global_entry_free_rcu() - free the tt_global_entry + * @rcu: rcu pointer of the tt_global_entry + */ +static void batadv_tt_global_entry_free_rcu(struct rcu_head *rcu) +{ + struct batadv_tt_global_entry *tt_global_entry; + + tt_global_entry = container_of(rcu, struct batadv_tt_global_entry, + common.rcu); + + kmem_cache_free(batadv_tg_cache, tt_global_entry); +} + /** * batadv_tt_global_entry_release() - release tt_global_entry from lists and * queue for free after rcu grace period @@ -254,7 +282,7 @@ void batadv_tt_global_entry_release(struct kref *ref) batadv_tt_global_del_orig_list(tt_global_entry); - kfree_rcu(tt_global_entry, common.rcu); + call_rcu(&tt_global_entry->common.rcu, batadv_tt_global_entry_free_rcu); } /** @@ -379,6 +407,19 @@ static void batadv_tt_global_size_dec(struct batadv_orig_node *orig_node, batadv_tt_global_size_mod(orig_node, vid, -1); } +/** + * batadv_tt_orig_list_entry_free_rcu() - free the orig_entry + * @rcu: rcu pointer of the orig_entry + */ +static void batadv_tt_orig_list_entry_free_rcu(struct rcu_head *rcu) +{ + struct batadv_tt_orig_list_entry *orig_entry; + + orig_entry = container_of(rcu, struct batadv_tt_orig_list_entry, rcu); + + kmem_cache_free(batadv_tt_orig_cache, orig_entry); +} + /** * batadv_tt_orig_list_entry_release() - release tt orig entry from lists and * queue for free after rcu grace period @@ -392,7 +433,7 @@ static void batadv_tt_orig_list_entry_release(struct kref *ref) refcount); batadv_orig_node_put(orig_entry->orig_node); - kfree_rcu(orig_entry, rcu); + call_rcu(&orig_entry->rcu, batadv_tt_orig_list_entry_free_rcu); } /**
This reverts commit cf9845d4566698731c23cd7a02db956706d058a1. This change seems to result in a memory leak / RCU race and the following kernel splat when the batman-adv kernel module is unloaded: ``` [ 112.208633] ============================================================================= [ 112.210359] BUG batadv_tl_cache (Tainted: G OE ): Objects remaining in batadv_tl_cache on __kmem_cache_shutdown() [ 112.211943] ----------------------------------------------------------------------------- [ 112.212517] Slab 0xffffe8afc0216d00 objects=16 used=1 fp=0xffff93f4085b4340 flags=0xfffffc0000a00(workingset|slab|node=0|zone=1|lastcpupid=0x1fffff) [ 112.212517] CPU: 1 PID: 776 Comm: rmmod Tainted: G OE 6.8.12-amd64 #1 Debian 6.8.12-1 [ 112.212517] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 [ 112.212517] Call Trace: [ 112.212517] <TASK> [ 112.212517] dump_stack_lvl+0x64/0x80 [ 112.212517] slab_err+0xe6/0x120 [ 112.212517] __kmem_cache_shutdown+0x160/0x2e0 [ 112.212517] kmem_cache_destroy+0x55/0x160 [ 112.220849] batadv_tt_cache_destroy+0x15/0x60 [batman_adv] [ 112.220849] __do_sys_delete_module+0x1d5/0x320 [ 112.220849] do_syscall_64+0x83/0x190 [ 112.220849] ? do_syscall_64+0x8f/0x190 [ 112.220849] ? exc_page_fault+0x7f/0x180 [ 112.220849] entry_SYSCALL_64_after_hwframe+0x78/0x80 [ 112.224478] RIP: 0033:0x7f2ac8434977 [ 112.224478] Code: 73 01 c3 48 8b 0d a9 94 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 79 94 0c 00 f7 d8 64 89 01 48 [ 112.224478] RSP: 002b:00007ffe0adf6138 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 [ 112.224478] RAX: ffffffffffffffda RBX: 000055db9018e770 RCX: 00007f2ac8434977 [ 112.224478] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 000055db9018e7d8 [ 112.224478] RBP: 0000000000000000 R08: 1999999999999999 R09: 0000000000000000 [ 112.224478] R10: 00007f2ac84a6ac0 R11: 0000000000000206 R12: 00007ffe0adf6390 [ 112.224478] R13: 000055db9018e770 R14: 000055db9018d2a0 R15: 0000000000000000 [ 112.233961] </TASK> [ 112.233961] Disabling lock debugging due to kernel taint [ 112.233961] Object 0xffff93f4085b4140 @offset=320 [ 112.233961] Allocated in batadv_tt_local_add+0x297/0xa20 [batman_adv] age=15835 cpu=1 pid=755 [ 112.233961] batadv_tt_local_add+0x297/0xa20 [batman_adv] [ 112.233961] batadv_interface_set_mac_addr+0xf6/0x120 [batman_adv] [ 112.233961] dev_set_mac_address+0xde/0x140 [ 112.233961] dev_set_mac_address_user+0x30/0x50 [ 112.233961] do_setlink+0x261/0x12d0 [ 112.233961] rtnl_setlink+0x11f/0x1d0 [ 112.233961] rtnetlink_rcv_msg+0x152/0x3c0 [ 112.241772] netlink_rcv_skb+0x5b/0x110 [ 112.241772] netlink_unicast+0x1a6/0x290 [ 112.241772] netlink_sendmsg+0x223/0x490 [ 112.241772] __sys_sendto+0x1df/0x1f0 [ 112.241772] __x64_sys_sendto+0x24/0x30 [ 112.241772] do_syscall_64+0x83/0x190 [ 112.241772] entry_SYSCALL_64_after_hwframe+0x78/0x80 [ 112.245994] ------------[ cut here ]------------ [ 112.246650] kmem_cache_destroy batadv_tl_cache: Slab cache still has objects when called from batadv_tt_cache_destroy+0x15/0x60 [batman_adv] [ 112.246668] WARNING: CPU: 1 PID: 776 at mm/slab_common.c:493 kmem_cache_destroy+0x14d/0x160 [ 112.249584] Modules linked in: veth batman_adv(OE-) cfg80211 rfkill bridge stp llc libcrc32c crc32c_generic crc16 rpcsec_gss_krb5 auth_rpcgss nfsv4 dns_resolver binfmt_misc pcspkr button joydev evdev serio_raw loop dm_mod efi_pstore nfnetlink vsock_loopback vmw_vsock_virtio_transport_common vmw_vsock_vmci_transport vsock vmw_vmci qemu_fw_cfg ip_tables x_tables autofs4 nfsv3 nfs_acl nfs lockd grace sunrpc 9pnet_rdma rdma_cm iw_cm ib_cm ib_core configfs 9p netfs ata_generic ata_piix libata psmouse scsi_mod 9pnet_virtio i2c_piix4 9pnet e1000 scsi_common floppy crypto_simd cryptd [ 112.256555] CPU: 1 PID: 776 Comm: rmmod Tainted: G B OE 6.8.12-amd64 #1 Debian 6.8.12-1 [ 112.258457] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.3-debian-1.16.3-2 04/01/2014 [ 112.260410] RIP: 0010:kmem_cache_destroy+0x14d/0x160 [ 112.261687] Code: 00 eb be 5b 5d 41 5c 41 5d c3 cc cc cc cc 48 8b 53 60 48 8b 4c 24 20 48 c7 c6 60 d5 e3 98 48 c7 c7 b8 ec 2d 99 e8 43 0d d8 ff <0f> 0b e9 e2 fe ff ff c3 cc cc cc cc 0f 1f 80 00 00 00 00 90 90 90 [ 112.265219] RSP: 0018:ffffb3b2806e7e48 EFLAGS: 00010282 [ 112.266044] RAX: 0000000000000000 RBX: ffff93f4270a2640 RCX: 0000000000000027 [ 112.267157] RDX: ffff93f43c521708 RSI: 0000000000000001 RDI: ffff93f43c521700 [ 112.268268] RBP: 000055db9018e7d8 R08: 0000000000000000 R09: ffffb3b2806e7cd8 [ 112.269418] R10: ffffb3b2806e7cd0 R11: 0000000000000003 R12: 0000000080012d00 [ 112.270572] R13: ffffb3b2806e7f58 R14: 0000000000000000 R15: 0000000000000000 [ 112.271699] FS: 00007f2ac8308440(0000) GS:ffff93f43c500000(0000) knlGS:0000000000000000 [ 112.273001] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 112.273923] CR2: 00005584ef830110 CR3: 000000000787c000 CR4: 00000000000006f0 [ 112.275050] Call Trace: [ 112.275464] <TASK> [ 112.275810] ? kmem_cache_destroy+0x14d/0x160 [ 112.276518] ? __warn+0x81/0x130 [ 112.277043] ? kmem_cache_destroy+0x14d/0x160 [ 112.277730] ? report_bug+0x171/0x1a0 [ 112.278315] ? prb_read_valid+0x1b/0x30 [ 112.278919] ? handle_bug+0x3c/0x80 [ 112.279467] ? exc_invalid_op+0x17/0x70 [ 112.280071] ? asm_exc_invalid_op+0x1a/0x20 [ 112.280741] ? kmem_cache_destroy+0x14d/0x160 [ 112.281603] ? kmem_cache_destroy+0x14d/0x160 [ 112.282489] batadv_tt_cache_destroy+0x15/0x60 [batman_adv] [ 112.283373] __do_sys_delete_module+0x1d5/0x320 [ 112.284080] do_syscall_64+0x83/0x190 [ 112.284696] ? do_syscall_64+0x8f/0x190 [ 112.285315] ? exc_page_fault+0x7f/0x180 [ 112.285970] entry_SYSCALL_64_after_hwframe+0x78/0x80 [ 112.286768] RIP: 0033:0x7f2ac8434977 [ 112.287355] Code: 73 01 c3 48 8b 0d a9 94 0c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 b8 b0 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 79 94 0c 00 f7 d8 64 89 01 48 [ 112.290282] RSP: 002b:00007ffe0adf6138 EFLAGS: 00000206 ORIG_RAX: 00000000000000b0 [ 112.291465] RAX: ffffffffffffffda RBX: 000055db9018e770 RCX: 00007f2ac8434977 [ 112.292595] RDX: 0000000000000000 RSI: 0000000000000800 RDI: 000055db9018e7d8 [ 112.293724] RBP: 0000000000000000 R08: 1999999999999999 R09: 0000000000000000 [ 112.294863] R10: 00007f2ac84a6ac0 R11: 0000000000000206 R12: 00007ffe0adf6390 [ 112.295982] R13: 000055db9018e770 R14: 000055db9018d2a0 R15: 0000000000000000 [ 112.297103] </TASK> [ 112.297465] ---[ end trace 0000000000000000 ]--- ``` So far, after some debugging, the actual cause for this could not immediately be found within the batman-adv code. Therefore reverting this for now until the underlying issue can be found and better understood. Some additional debugging information and discussions can be found on our Redmine bugtracker, linked below. Link: https://www.open-mesh.org/issues/428 Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue> --- Some help / guidance for further debugging from people more knowledgeable in the RCU core internals would be appreciated. Specifically why the rcu_barrier() call in batadv_exit() does not seem to catch the kfree_rcu() which the reverted patch was originally adding. compat.h | 8 ----- net/batman-adv/translation-table.c | 47 ++++++++++++++++++++++++++++-- 2 files changed, 44 insertions(+), 11 deletions(-)