Re: ppc elf_map breakage with MAP_FIXED_NOREPLACE
From: Anshuman Khandual
Date: Sun Jan 28 2018 - 21:48:08 EST
On 01/26/2018 07:34 PM, Michal Hocko wrote:
> On Fri 26-01-18 18:04:27, Anshuman Khandual wrote:
> [...]
>> I tried to instrument mmap_region() for a single instance of 'sed'
>> binary and traced all it's VMA creation. But there is no trace when
>> that 'anon' VMA got created which suddenly shows up during subsequent
>> elf_map() call eventually failing it. Please note that the following
>> VMA was never created through call into map_region() in the process
>> which is strange.
>
> Could you share your debugging patch?
Please find the debug patch at the end.
>
>> =================================================================
>> [ 9.076867] Details for VMA[3] c000001fce42b7c0
>> [ 9.076925] vma c000001fce42b7c0 start 0000000010030000 end 0000000010040000
>> next c000001fce42b580 prev c000001fce42b880 mm c000001fce40fa00
>> prot 8000000000000104 anon_vma (null) vm_ops (null)
>> pgoff 1003 file (null) private_data (null)
>> flags: 0x100073(read|write|mayread|maywrite|mayexec|account)
>> =================================================================
>
> Isn't this vdso or some other special mapping? It is not really an
> anonymous vma. Please hook into __install_special_mapping
Yeah, will do. Its not an anon mapping as it does not have a anon_vma
structure ?
>
>> VMA creation for 'sed' binary
>> =============================
>> [ 9.071902] XXX: mm c000001fce40fa00 registered
>>
>> [ 9.071971] Total VMAs 2 on MM c000001fce40fa00
>> ----
>> [ 9.072010] Details for VMA[1] c000001fce42bdc0
>> [ 9.072064] vma c000001fce42bdc0 start 0000000010000000 end 0000000010020000
>> next c000001fce42b580 prev (null) mm c000001fce40fa00
>> prot 8000000000000105 anon_vma (null) vm_ops c008000011ddca18
>> pgoff 0 file c000001fe2969a00 private_data (null)
>> flags: 0x875(read|exec|mayread|maywrite|mayexec|denywrite)
>
> This one doesn't have any stack trace either... Yet it is a file
> mapping obviously. Special mappings shouldn't have any file associated.
> Strange...
IIUC, the first VMA (which seems to be an anon VMA) did not have any
stack trace and not sure how it got created.
[ 9.077335] vma c000001fce42b580 start 00007fffcafe0000 end 00007fffcb010000
next (null) prev c000001fce42b7c0 mm c000001fce40fa00
prot 8000000000000104 anon_vma c000001fce4456f0 vm_ops (null)
pgoff 1fffffffd file (null) private_data (null)
flags: 0x100173(read|write|mayread|maywrite|mayexec|growsdown|account)
the subsequent ones, this
[ 9.072010] Details for VMA[1] c000001fce42bdc0
[ 9.072064] vma c000001fce42bdc0 start 0000000010000000 end 0000000010020000
next c000001fce42b580 prev (null) mm c000001fce40fa00
prot 8000000000000105 anon_vma (null) vm_ops c008000011ddca18
pgoff 0 file c000001fe2969a00 private_data (null)
flags: 0x875(read|exec|mayread|maywrite|mayexec|denywrite)
and this (both are file mapping for sure and getting loaded from elf)
[ 9.074170] Details for VMA[2] c000001fce42b880
[ 9.074236] vma c000001fce42b880 start 0000000010020000 end 0000000010030000
next c000001fce42b580 prev c000001fce42bdc0 mm c000001fce40fa00
prot 8000000000000104 anon_vma (null) vm_ops c008000011ddca18
pgoff 1 file c000001fe2969a00 private_data (null)
flags: 0x100873(read|write|mayread|maywrite|mayexec|denywrite|account)
have similar stack traces
[ 9.072839] CPU: 48 PID: 7544 Comm: sed Not tainted 4.14.0-dirty #154
[ 9.072928] Call Trace:
[ 9.072952] [c000001fbef37840] [c000000000b17a00] dump_stack+0xb0/0xf0 (unreliable)
[ 9.073021] [c000001fbef37880] [c0000000002dbc48] mmap_region+0x718/0x720
[ 9.073097] [c000001fbef37970] [c0000000002dc034] do_mmap+0x3e4/0x480
[ 9.073179] [c000001fbef379f0] [c0000000002a96c8] vm_mmap_pgoff+0xe8/0x120
[ 9.073268] [c000001fbef37ac0] [c0000000003cf378] elf_map+0x98/0x270
[ 9.073326] [c000001fbef37b60] [c0000000003d1258] load_elf_binary+0x6f8/0x158c
[ 9.073416] [c000001fbef37c80] [c00000000035d320] search_binary_handler+0xd0/0x270
[ 9.073510] [c000001fbef37d10] [c00000000035f278] do_execveat_common.isra.31+0x658/0x890
[ 9.073599] [c000001fbef37df0] [c00000000035f8c0] SyS_execve+0x40/0x50
[ 9.073673] [c000001fbef37e30] [c00000000000b220] system_call+0x58/0x6c
and then again this one (which causes the collision subsequently) is neither
a anon VMA nor a file VMA and does not have a stack trace either.
[ 9.076867] Details for VMA[3] c000001fce42b7c0
[ 9.076925] vma c000001fce42b7c0 start 0000000010030000 end 0000000010040000
next c000001fce42b580 prev c000001fce42b880 mm c000001fce40fa00
prot 8000000000000104 anon_vma (null) vm_ops (null)
pgoff 1003 file (null) private_data (null)
flags: 0x100073(read|write|mayread|maywrite|mayexec|account)
Will double check the debug patch.
----------------------------------------------
diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
index d8c5657..ccef8fd 100644
--- a/fs/binfmt_elf.c
+++ b/fs/binfmt_elf.c
@@ -41,6 +41,7 @@
#include <linux/cred.h>
#include <linux/dax.h>
#include <linux/uaccess.h>
+#include <linux/mmdebug.h>
#include <asm/param.h>
#include <asm/page.h>
@@ -341,6 +342,10 @@ static int padzero(unsigned long elf_bss)
#ifndef elf_map
+extern struct mm_struct *mm_ptr;
+extern bool just_init;
+extern void dump_mm_vmas(const struct mm_struct *mm);
+
static unsigned long elf_map(struct file *filep, unsigned long addr,
struct elf_phdr *eppnt, int prot, int type,
unsigned long total_size)
@@ -372,11 +377,21 @@ static unsigned long elf_map(struct file *filep, unsigned long addr,
} else
map_addr = vm_mmap(filep, addr, size, prot, type, off);
- if ((type & MAP_FIXED_NOREPLACE) && BAD_ADDR(map_addr))
- pr_info("%d (%s): Uhuuh, elf segment at %p requested but the memory is mapped already\n",
+ if ((type & MAP_FIXED_NOREPLACE) && BAD_ADDR(map_addr)) {
+ struct vm_area_struct *vma;
+
+ if (strcmp(current->comm, "sed"))
+ return(map_addr);
+
+ vma = find_vma(current->mm, addr);
+ if (just_init && (mm_ptr == vma->vm_mm)) {
+ pr_info("%d (%s): Uhuuh, elf segment at %p requested but the memory is mapped already\n",
task_pid_nr(current), current->comm,
(void *)addr);
+ dump_mm_vmas(vma->vm_mm);
+ }
+ }
return(map_addr);
}
diff --git a/mm/mmap.c b/mm/mmap.c
index ca7b1cf..b427a5b 100644
--- a/mm/mmap.c
+++ b/mm/mmap.c
@@ -45,6 +45,7 @@
#include <linux/moduleparam.h>
#include <linux/pkeys.h>
#include <linux/oom.h>
+#include <linux/mmdebug.h>
#include <linux/uaccess.h>
#include <asm/cacheflush.h>
@@ -1611,6 +1612,25 @@ static inline int accountable_mapping(struct file *file, vm_flags_t vm_flags)
return (vm_flags & (VM_NORESERVE | VM_SHARED | VM_WRITE)) == VM_WRITE;
}
+struct mm_struct *mm_ptr;
+bool just_init;
+EXPORT_SYMBOL(mm_ptr);
+EXPORT_SYMBOL(just_init);
+
+void dump_mm_vmas(const struct mm_struct *mm)
+{
+ struct vm_area_struct *vma = mm->mmap;
+ int count;
+
+ printk("Total VMAs %d on MM %lx\n", mm->map_count, (unsigned long) mm);
+
+ for (count = 0; vma && count < mm->map_count; count++, vma = vma->vm_next) {
+ printk("Details for VMA[%d] %lx\n", count + 1, (unsigned long) vma);
+ dump_vma(vma);
+ }
+}
+EXPORT_SYMBOL(dump_mm_vmas);
+
unsigned long mmap_region(struct file *file, unsigned long addr,
unsigned long len, vm_flags_t vm_flags, unsigned long pgoff,
struct list_head *uf)
@@ -1754,6 +1774,21 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
vma_set_page_prot(vma);
+ if (!strcmp(current->comm, "sed")) {
+ if (!just_init) {
+ just_init = 1;
+ mm_ptr = vma->vm_mm;
+ printk("XXX: mm %lx registered\n", (unsigned long) mm_ptr);
+ dump_mm_vmas(vma->vm_mm);
+ dump_stack();
+ } else {
+ if(mm_ptr == vma->vm_mm) {
+ dump_mm_vmas(vma->vm_mm);
+ dump_stack();
+ }
+ }
+ }
+
return addr;
unmap_and_free_vma: