Fix pagemap_read() use of struct mm_walk
Andrew Morton [Sat, 5 Jul 2008 08:02:01 +0000 (01:02 -0700)]
Fix some issues in pagemap_read noted by Alexey:

- initialize pagemap_walk.mm to "mm" , so the code starts working as
  advertised

- initialize ->private to "&pm" so it wouldn't immediately oops in
  pagemap_pte_hole()

- unstatic struct pagemap_walk, so two threads won't fsckup each other
  (including those started by root, including flipping ->mm when you don't
  have permissions)

- pagemap_read() contains two calls to ptrace_may_attach(), second one
  looks unneeded.

- avoid possible kmalloc(0) and integer wraparound.

Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
[ Personally, I'd just remove the functionality entirely  - Linus ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/proc/task_mmu.c

index 05053d7..c492449 100644 (file)
@@ -602,11 +602,6 @@ static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
        return err;
 }
 
-static struct mm_walk pagemap_walk = {
-       .pmd_entry = pagemap_pte_range,
-       .pte_hole = pagemap_pte_hole
-};
-
 /*
  * /proc/pid/pagemap - an array mapping virtual pages to pfns
  *
@@ -641,6 +636,11 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
        struct pagemapread pm;
        int pagecount;
        int ret = -ESRCH;
+       struct mm_walk pagemap_walk;
+       unsigned long src;
+       unsigned long svpfn;
+       unsigned long start_vaddr;
+       unsigned long end_vaddr;
 
        if (!task)
                goto out;
@@ -659,11 +659,15 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
        if (!mm)
                goto out_task;
 
-       ret = -ENOMEM;
+
        uaddr = (unsigned long)buf & PAGE_MASK;
        uend = (unsigned long)(buf + count);
        pagecount = (PAGE_ALIGN(uend) - uaddr) / PAGE_SIZE;
-       pages = kmalloc(pagecount * sizeof(struct page *), GFP_KERNEL);
+       ret = 0;
+       if (pagecount == 0)
+               goto out_mm;
+       pages = kcalloc(pagecount, sizeof(struct page *), GFP_KERNEL);
+       ret = -ENOMEM;
        if (!pages)
                goto out_mm;
 
@@ -684,33 +688,33 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
        pm.out = (u64 *)buf;
        pm.end = (u64 *)(buf + count);
 
-       if (!ptrace_may_attach(task)) {
-               ret = -EIO;
-       } else {
-               unsigned long src = *ppos;
-               unsigned long svpfn = src / PM_ENTRY_BYTES;
-               unsigned long start_vaddr = svpfn << PAGE_SHIFT;
-               unsigned long end_vaddr = TASK_SIZE_OF(task);
-
-               /* watch out for wraparound */
-               if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
-                       start_vaddr = end_vaddr;
-
-               /*
-                * The odds are that this will stop walking way
-                * before end_vaddr, because the length of the
-                * user buffer is tracked in "pm", and the walk
-                * will stop when we hit the end of the buffer.
-                */
-               ret = walk_page_range(start_vaddr, end_vaddr,
-                                       &pagemap_walk);
-               if (ret == PM_END_OF_BUFFER)
-                       ret = 0;
-               /* don't need mmap_sem for these, but this looks cleaner */
-               *ppos += (char *)pm.out - buf;
-               if (!ret)
-                       ret = (char *)pm.out - buf;
-       }
+       pagemap_walk.pmd_entry = pagemap_pte_range;
+       pagemap_walk.pte_hole = pagemap_pte_hole;
+       pagemap_walk.mm = mm;
+       pagemap_walk.private = &pm;
+
+       src = *ppos;
+       svpfn = src / PM_ENTRY_BYTES;
+       start_vaddr = svpfn << PAGE_SHIFT;
+       end_vaddr = TASK_SIZE_OF(task);
+
+       /* watch out for wraparound */
+       if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT)
+               start_vaddr = end_vaddr;
+
+       /*
+        * The odds are that this will stop walking way
+        * before end_vaddr, because the length of the
+        * user buffer is tracked in "pm", and the walk
+        * will stop when we hit the end of the buffer.
+        */
+       ret = walk_page_range(start_vaddr, end_vaddr, &pagemap_walk);
+       if (ret == PM_END_OF_BUFFER)
+               ret = 0;
+       /* don't need mmap_sem for these, but this looks cleaner */
+       *ppos += (char *)pm.out - buf;
+       if (!ret)
+               ret = (char *)pm.out - buf;
 
 out_pages:
        for (; pagecount; pagecount--) {