Re: [PATCH] ipc/shm.c add ->pagesize function to shm_vm_ops

From: Davidlohr Bueso
Date: Mon Jul 30 2018 - 12:45:14 EST


On Fri, 27 Jul 2018, Jane Chu wrote:

Commit 05ea88608d4e13 (mm, hugetlbfs: introduce ->pagesize() to
vm_operations_struct) adds a new ->pagesize() function to
hugetlb_vm_ops, intended to cover all hugetlbfs backed files.

With System V shared memory model, if "huge page" is specified,
the "shared memory" is backed by hugetlbfs files, but the mappings
initiated via shmget/shmat have their original vm_ops overwritten
with shm_vm_ops, so we need to add a ->pagesize function to shm_vm_ops.
Otherwise, vma_kernel_pagesize() returns PAGE_SIZE given a hugetlbfs
backed vma, result in below BUG:

fs/hugetlbfs/inode.c
443 if (unlikely(page_mapped(page))) {
444 BUG_ON(truncate_op);

[ 242.268342] hugetlbfs: oracle (4592): Using mlock ulimits for SHM_HUGETLB is deprecated
[ 282.653208] ------------[ cut here ]------------
[ 282.708447] kernel BUG at fs/hugetlbfs/inode.c:444!
[ 282.818957] Modules linked in: nfsv3 rpcsec_gss_krb5 nfsv4 ...
[ 284.025873] CPU: 35 PID: 5583 Comm: oracle_5583_sbt Not tainted 4.14.35-1829.el7uek.x86_64 #2
[ 284.246609] task: ffff9bf0507aaf80 task.stack: ffffa9e625628000
[ 284.317455] RIP: 0010:remove_inode_hugepages+0x3db/0x3e2
....
[ 285.292389] Call Trace:
[ 285.321630] hugetlbfs_evict_inode+0x1e/0x3e
[ 285.372707] evict+0xdb/0x1af
[ 285.408185] iput+0x1a2/0x1f7
[ 285.443661] dentry_unlink_inode+0xc6/0xf0
[ 285.492661] __dentry_kill+0xd8/0x18d
[ 285.536459] dput+0x1b5/0x1ed
[ 285.571939] __fput+0x18b/0x216
[ 285.609495] ____fput+0xe/0x10
[ 285.646030] task_work_run+0x90/0xa7
[ 285.688788] exit_to_usermode_loop+0xdd/0x116
[ 285.740905] do_syscall_64+0x187/0x1ae
[ 285.785740] entry_SYSCALL_64_after_hwframe+0x150/0x0

Suggested-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
Signed-off-by: Jane Chu <jane.chu@xxxxxxxxxx>

Acked-by: Davidlohr Bueso <dbueso@xxxxxxx>