linux系统报xfs_vm_releasepage警告问题的处理方法

问题说明

最近的几台机器在同一天的不同时段都出现以下警告信息:

mar 26 20:55:03 host1 kernel: warning: at fs/xfs/xfs_aops.c:1045 xfs_vm_releasepage+0xcb/0x100 [xfs]()
mar 26 20:55:03 host1 kernel: modules linked in: nf_conntrack_ipv4 nf_defrag_ipv4 xt_conntrack nf_conntrack iptable_filter ip_tables ebtable_filter ebtables ip6table_
filter ip6_tables devlink bridge stp llc xt_multiport sunrpc dm_mirror dm_region_hash dm_log dm_mod intel_powerclamp coretemp intel_rapl iosf_mbi kvm_intel kvm irqbypa
ss crc32_pclmul ghash_clmulni_intel aesni_intel lrw gf128mul glue_helper ablk_helper cryptd itco_wdt itco_vendor_support dcdbas ipmi_devintf ipmi_si sg pcspkr ipmi_msg
handler shpchp i2c_i801 lpc_ich nfit libnvdimm acpi_power_meter kgwttm(oe) xfs libcrc32c sd_mod crc_t10dif crct10dif_generic crct10dif_pclmul crct10dif_common crc32c_i
ntel mgag200 drm_kms_helper igb syscopyarea sysfillrect sysimgblt ptp fb_sys_fops ttm pps_core dca ahci drm i2c_algo_bit libahci megaraid_sas i2c_core libata
mar 26 20:55:03 host1 kernel: fjes [last unloaded: nf_defrag_ipv4]
mar 26 20:55:03 host1 kernel: cpu: 10 pid: 224 comm: kswapd0 tainted: g      oe ------------  3.10.0-514.21.2.el7.x86_64 #1
mar 26 20:55:03 host1 kernel: hardware name: dell inc. poweredge r640/0w23h8, bios 1.3.7 02/08/2018
mar 26 20:55:03 host1 kernel: 0000000000000000 00000000e02a0d05 ffff88103c7ebaa0 ffffffff81687073
mar 26 20:55:03 host1 kernel: ffff88103c7ebad8 ffffffff81085cb0 ffffea0000687620 ffffea0000687600
mar 26 20:55:03 host1 kernel: ffff88004a71daf8 ffff88103c7ebda0 ffffea0000687600 ffff88103c7ebae8
mar 26 20:55:03 host1 kernel: call trace:
mar 26 20:55:03 host1 kernel: [<ffffffff81687073>] dump_stack+0x19/0x1b
mar 26 20:55:03 host1 kernel: [<ffffffff81085cb0>] warn_slowpath_common+0x70/0xb0
mar 26 20:55:03 host1 kernel: [<ffffffff81085dfa>] warn_slowpath_null+0x1a/0x20
mar 26 20:55:03 host1 kernel: [<ffffffffa038bfdb>] xfs_vm_releasepage+0xcb/0x100 [xfs]
mar 26 20:55:03 host1 kernel: [<ffffffff81180b22>] try_to_release_page+0x32/0x50
mar 26 20:55:03 host1 kernel: [<ffffffff81196ad6>] shrink_active_list+0x3d6/0x3e0
mar 26 20:55:03 host1 kernel: [<ffffffff81196ed1>] shrink_lruvec+0x3f1/0x770
mar 26 20:55:03 host1 kernel: [<ffffffff811972c6>] shrink_zone+0x76/0x1a0
mar 26 20:55:03 host1 kernel: [<ffffffff8119857c>] balance_pgdat+0x48c/0x5e0
mar 26 20:55:03 host1 kernel: [<ffffffff81198843>] kswapd+0x173/0x450
mar 26 20:55:03 host1 kernel: [<ffffffff810b1b20>] ? wake_up_atomic_t+0x30/0x30
mar 26 20:55:03 host1 kernel: [<ffffffff811986d0>] ? balance_pgdat+0x5e0/0x5e0
mar 26 20:55:03 host1 kernel: [<ffffffff810b0a4f>] kthread+0xcf/0xe0
mar 26 20:55:03 host1 kernel: [<ffffffff810b0980>] ? kthread_create_on_node+0x140/0x140
mar 26 20:55:03 host1 kernel: [<ffffffff81697698>] ret_from_fork+0x58/0x90
mar 26 20:55:03 host1 kernel: [<ffffffff810b0980>] ? kthread_create_on_node+0x140/0x140
mar 26 20:55:03 host1 kernel: ---[ end trace 24823c5c7a1ea2be ]---

这几台机器的 kernel 及应用程序等崩溃信息由 abrtd 服务接管, 可以通过 abrt-cli 查看概要信息:

# abrt-cli list --since 1547518209
id 2181dce8f72761585cb6a904dbff1806c1315c27
reason:     warning: at fs/xfs/xfs_aops.c:1045 xfs_vm_releasepage+0xcb/0x100 [xfs]()
time:      sat 23 mar 2019 08:30:45 pm cst
cmdline:    boot_image=/boot/vmlinuz-3.10.0-514.16.1.el7.x86_64 root=/dev/sda1 ro crashkernel=auto net.ifnames=0 biosdevname=0
package:    kernel
uid:      0 (root)
count:     1
directory:   /var/spool/abrt/oops-2019-03-23-20:30:45-163925-0

内核版本如下:

centos7
linux host1 3.10.0-514.21.2.el7.x86_64

分析处理

红帽知识库

参考红帽知识库文档, xfs 的这类警告信息在 xfs 模块遍历代码路径的时候会打印该信息, 不影响主机使用. 可升级内核到 kernel-3.10.0-693.el7 版本避免该警告信息, 详细参见:

root cause:

the messages were informational and they do not affect the system in a negative manner. they are seen because the xfs module is traversing through xfs code path.

代码分析

红帽知识库中并未提到内存回收的相关信息, 不过从堆栈信息来看, 像是因为内核回收内存而引起的, 查看对应时间点的内存使用情况如下所示:

04:30:01 pm kbmemfree kbmemused %memused kbbuffers kbcached kbcommit  %commit kbactive  kbinact  kbdirty
......
08:40:01 pm  513940 130976220   99.61    876 104616380 28610584   21.76 92439660 34840920    524
08:50:01 pm  479896 131010264   99.64    876 104666496 28557292   21.72 92513872 34804240    400
09:00:01 pm  455948 131034212   99.65    876 104675712 28588852   21.74 92418724 34926132    572
09:10:01 pm  556980 130933180   99.58    876 104610352 28552656   21.71 94287212 32983892    900

# sysctl vm.min_free_kbytes
vm.min_free_kbytes = 90112

20:50 到 21:00 之间的可用内存并没有增加, 这意味着系统可能没有做内存回收操作, 我们按照 kernel 日志的堆栈信息来看函数的调用关系:

shrink_active_list -> try_to_release_page -> xfs_vm_releasepage

//source/mm/filemap.c
3225 int try_to_release_page(struct page *page, gfp_t gfp_mask)
3226 {
3227   struct address_space * const mapping = page->mapping;
......
3233   if (mapping && mapping->a_ops->releasepage)
3234     return mapping->a_ops->releasepage(page, gfp_mask);  xfs_vm_releasepage
3235   return try_to_free_buffers(page);
3236 }

//source/fs/xfs/xfs_aops.c
1034 static int
1035 xfs_vm_releasepage(
1036   struct page   *page,
1037   gfp_t      gfp_mask)
1038 {
1039   int     delalloc, unwritten;
1040 
1041   trace_xfs_releasepage(page->mapping->host, page, 0, 0);
1042 
1043   xfs_count_page_state(page, &delalloc, &unwritten);
1044 
1045   if (warn_on_once(delalloc))
1046     return 0;
1047   if (warn_on_once(unwritten))
1048     return 0;
1049 
1050   return try_to_free_buffers(page);
1051 }
......
1827 const struct address_space_operations xfs_address_space_operations = {
1833   .releasepage    = xfs_vm_releasepage,

对应 kernel 日志 kernel: warning: at fs/xfs/xfs_aops.c:1045 即可看出源文件 source/fs/xfs/xfs_aops.c 的 1045 行打印出了该堆栈信息, 实际上并没有执行 try_to_free_buffers 就已经返回:

1045   if (warn_on_once(delalloc))
1046     return 0;

warn_on_once 则相对简单, 在源文件 source/include/asm-generic/bug.h 即可找到:

73 #define __warn()    warn_slowpath_null(__file__, __line__)

85 #define warn_on(condition) ({            \
...
88     __warn();            \

136 #define warn_on_once(condition) ({       \
....
140   if (unlikely(__ret_warn_once))       \
141     if (warn_on(!__warned))       \

__warn 函数则调用了堆栈信息里的 warn_slowpath_null 函数, 进而调用 warn_slowpath_common 函数打印了堆栈信息:

//source/kernel/panic.c
517 void warn_slowpath_null(const char *file, int line)
518 {
519   warn_slowpath_common(file, line, __builtin_return_address(0),
520         taint_warn, null);
521 }

463 static void warn_slowpath_common(const char *file, int line, void *caller,
464         unsigned taint, struct slowpath_args *args)
465 {
466   disable_trace_on_warning();
467 
468   printk(kern_warning "------------[ cut here ]------------\n");
469   printk(kern_warning "warning: at %s:%d %ps()\n", file, line, caller);
470 
471   if (args)
472     vprintk(args->fmt, args->args);
......
485   print_modules();
486   dump_stack();
487   print_oops_end_marker();

我们大致可以看出这个堆栈信息只是警告, 和红帽知识库中描述的一致, 并不影响主机的使用.

总结说明

从上面源文件的函数来看, 只要 kswapd 内存回收的时候调用了 xfs_vm_releasepage 就有可能打印堆栈信息, 如果打印堆栈则不会执行 try_to_free_buffers 操作, 所以查看内存使用的时候可用内存并没有增加. 如果不希望出现堆栈信息可以开启 disable_trace_on_warning 函数对应的 kernel.traceoff_on_warning 内核参数关闭堆栈提示, 不过关闭后其他的内核信息也就不会再打印, 所以从这方面来看只有升级内核版本才会避免出现这个信息.

好了，以上就是这篇文章的全部内容了，希望本文的内容对大家的学习或者工作具有一定的参考学习价值，谢谢大家对的支持。