When a module creates a file in procfs and a program uses the file with mmap, the .owner field of the file_operations structure is ignored. This allows removing the module while the file is still being used. Therefore this sequence of events leads to a kernel oops: * load a module which creates a file in procfs with an mmap operation * open the file * use mmap on it * rmmod the module * call unmap Here are parts of the oops caused by unmap: [ 1234.337725] BUG: unable to handle kernel paging request at ffffffffa030a268 [ 1234.338007] IP: [<ffffffff811bf054>] remove_vma+0x24/0x70 [ 1234.338007] PGD 1c17067 PUD 1c18063 PMD 3d713067 PTE 0 [ 1234.338007] Oops: 0000 [#1] SMP [ 1234.338007] Modules linked in: fuse rpcsec_gss_krb5 nfsv4 dns_resolver nfs fscache cfg80211 rfkill ppdev bochs_drm virtio_net serio_raw parport_pc ttm pvpanic parport drm_kms_helper drm i2c_piix4 nfsd auth_rpcgss nfs_acl lockd sunrpc virtio_blk virtio_pci virtio_ring virtio ata_generic pata_acpi [last unloaded: procfs_mmap] [ 1234.338007] Call Trace: [ 1234.338007] [<ffffffff811c155f>] do_munmap+0x27f/0x3b0 [ 1234.338007] [<ffffffff811c16d1>] vm_munmap+0x41/0x60 [ 1234.338007] [<ffffffff811c2652>] SyS_munmap+0x22/0x30 [ 1234.338007] [<ffffffff817301a9>] system_call_fastpath+0x16/0x1b Fix this by making proc_reg_open grab a reference to the module owning pde->proc_fops. More information and example code to reproduce this oops can be found on https://bugzilla.kernel.org/show_bug.cgi?id=92511 Signed-off-by: Nicolas Iooss <nicolas.iooss_linux@xxxxxxx> --- fs/proc/inode.c | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/fs/proc/inode.c b/fs/proc/inode.c index 8420a2f80811..5df17cb730fa 100644 --- a/fs/proc/inode.c +++ b/fs/proc/inode.c @@ -329,12 +329,16 @@ static int proc_reg_open(struct inode *inode, struct file *file) kfree(pdeo); return -ENOENT; } + fops_get(pde->proc_fops); open = pde->proc_fops->open; release = pde->proc_fops->release; if (open) rv = open(inode, file); + if (rv != 0) + fops_put(pde->proc_fops); + if (rv == 0 && release) { /* To know what to release. */ pdeo->file = file; @@ -361,6 +365,7 @@ static int proc_reg_release(struct inode *inode, struct file *file) } } spin_unlock(&pde->pde_unload_lock); + fops_put(pde->proc_fops); return 0; } -- 2.3.0 -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
