]> nv-tegra.nvidia Code Review - linux-2.6.git/blobdiff - mm/filemap_xip.c
config: tegra3: enable /dev mount with ACL
[linux-2.6.git] / mm / filemap_xip.c
index 3e744abcce9daa8aada1487bd42e31b769c414bf..93356cd12828a40eb4d635a9e0e0bea3d6ba0790 100644 (file)
 #include <linux/module.h>
 #include <linux/uio.h>
 #include <linux/rmap.h>
+#include <linux/mmu_notifier.h>
 #include <linux/sched.h>
+#include <linux/seqlock.h>
+#include <linux/mutex.h>
+#include <linux/gfp.h>
 #include <asm/tlbflush.h>
 #include <asm/io.h>
 
  * We do use our own empty page to avoid interference with other users
  * of ZERO_PAGE(), such as /dev/zero
  */
+static DEFINE_MUTEX(xip_sparse_mutex);
+static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
 static struct page *__xip_sparse_page;
 
+/* called under xip_sparse_mutex */
 static struct page *xip_sparse_page(void)
 {
        if (!__xip_sparse_page) {
                struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
 
-               if (page) {
-                       static DEFINE_SPINLOCK(xip_alloc_lock);
-                       spin_lock(&xip_alloc_lock);
-                       if (!__xip_sparse_page)
-                               __xip_sparse_page = page;
-                       else
-                               __free_page(page);
-                       spin_unlock(&xip_alloc_lock);
-               }
+               if (page)
+                       __xip_sparse_page = page;
        }
        return __xip_sparse_page;
 }
@@ -90,8 +90,8 @@ do_xip_mapping_read(struct address_space *mapping,
                        }
                }
                nr = nr - offset;
-               if (nr > len)
-                       nr = len;
+               if (nr > len - copied)
+                       nr = len - copied;
 
                error = mapping->a_ops->get_xip_mem(mapping, index, 0,
                                                        &xip_mem, &xip_pfn);
@@ -173,30 +173,43 @@ __xip_unmap (struct address_space * mapping,
        pte_t pteval;
        spinlock_t *ptl;
        struct page *page;
+       unsigned count;
+       int locked = 0;
+
+       count = read_seqcount_begin(&xip_sparse_seq);
 
        page = __xip_sparse_page;
        if (!page)
                return;
 
-       spin_lock(&mapping->i_mmap_lock);
+retry:
+       mutex_lock(&mapping->i_mmap_mutex);
        vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
                mm = vma->vm_mm;
                address = vma->vm_start +
                        ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
                BUG_ON(address < vma->vm_start || address >= vma->vm_end);
-               pte = page_check_address(page, mm, address, &ptl);
+               pte = page_check_address(page, mm, address, &ptl, 1);
                if (pte) {
                        /* Nuke the page table entry. */
                        flush_cache_page(vma, address, pte_pfn(*pte));
-                       pteval = ptep_clear_flush(vma, address, pte);
-                       page_remove_rmap(page, vma);
-                       dec_mm_counter(mm, file_rss);
+                       pteval = ptep_clear_flush_notify(vma, address, pte);
+                       page_remove_rmap(page);
+                       dec_mm_counter(mm, MM_FILEPAGES);
                        BUG_ON(pte_dirty(pteval));
                        pte_unmap_unlock(pte, ptl);
                        page_cache_release(page);
                }
        }
-       spin_unlock(&mapping->i_mmap_lock);
+       mutex_unlock(&mapping->i_mmap_mutex);
+
+       if (locked) {
+               mutex_unlock(&xip_sparse_mutex);
+       } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
+               mutex_lock(&xip_sparse_mutex);
+               locked = 1;
+               goto retry;
+       }
 }
 
 /*
@@ -217,7 +230,7 @@ static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
        int error;
 
        /* XXX: are VM_FAULT_ codes OK? */
-
+again:
        size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
        if (vmf->pgoff >= size)
                return VM_FAULT_SIGBUS;
@@ -236,8 +249,10 @@ static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
                int err;
 
                /* maybe shared writable, allocate new block */
+               mutex_lock(&xip_sparse_mutex);
                error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
                                                        &xip_mem, &xip_pfn);
+               mutex_unlock(&xip_sparse_mutex);
                if (error)
                        return VM_FAULT_SIGBUS;
                /* unmap sparse mappings at pgoff from all other vmas */
@@ -251,18 +266,38 @@ found:
                BUG_ON(err);
                return VM_FAULT_NOPAGE;
        } else {
+               int err, ret = VM_FAULT_OOM;
+
+               mutex_lock(&xip_sparse_mutex);
+               write_seqcount_begin(&xip_sparse_seq);
+               error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
+                                                       &xip_mem, &xip_pfn);
+               if (unlikely(!error)) {
+                       write_seqcount_end(&xip_sparse_seq);
+                       mutex_unlock(&xip_sparse_mutex);
+                       goto again;
+               }
+               if (error != -ENODATA)
+                       goto out;
                /* not shared and writable, use xip_sparse_page() */
                page = xip_sparse_page();
                if (!page)
-                       return VM_FAULT_OOM;
+                       goto out;
+               err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
+                                                       page);
+               if (err == -ENOMEM)
+                       goto out;
 
-               page_cache_get(page);
-               vmf->page = page;
-               return 0;
+               ret = VM_FAULT_NOPAGE;
+out:
+               write_seqcount_end(&xip_sparse_seq);
+               mutex_unlock(&xip_sparse_mutex);
+
+               return ret;
        }
 }
 
-static struct vm_operations_struct xip_file_vm_ops = {
+static const struct vm_operations_struct xip_file_vm_ops = {
        .fault  = xip_file_fault,
 };
 
@@ -307,8 +342,10 @@ __xip_file_write(struct file *filp, const char __user *buf,
                                                &xip_mem, &xip_pfn);
                if (status == -ENODATA) {
                        /* we allocate a new page unmap it */
+                       mutex_lock(&xip_sparse_mutex);
                        status = a_ops->get_xip_mem(mapping, index, 1,
                                                        &xip_mem, &xip_pfn);
+                       mutex_unlock(&xip_sparse_mutex);
                        if (!status)
                                /* unmap page at pgoff from all other vmas */
                                __xip_unmap(mapping, index);
@@ -380,7 +417,7 @@ xip_file_write(struct file *filp, const char __user *buf, size_t len,
        if (count == 0)
                goto out_backing;
 
-       ret = remove_suid(filp->f_path.dentry);
+       ret = file_remove_suid(filp);
        if (ret)
                goto out_backing;