Use pgoff_t instead of unsigned long
[safe/jmp/linux-2.6] / mm / filemap_xip.c
index 3b6e384..5e598c4 100644 (file)
 #include <linux/module.h>
 #include <linux/uio.h>
 #include <linux/rmap.h>
+#include <linux/sched.h>
 #include <asm/tlbflush.h>
-#include "filemap.h"
+
+/*
+ * We do use our own empty page to avoid interference with other users
+ * of ZERO_PAGE(), such as /dev/zero
+ */
+static struct page *__xip_sparse_page;
+
+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);
+               }
+       }
+       return __xip_sparse_page;
+}
 
 /*
  * This is a file read routine for execute in place files, and uses
@@ -32,7 +56,8 @@ do_xip_mapping_read(struct address_space *mapping,
                    read_actor_t actor)
 {
        struct inode *inode = mapping->host;
-       unsigned long index, end_index, offset;
+       pgoff_t index, end_index;
+       unsigned long offset;
        loff_t isize;
 
        BUG_ON(!mapping->a_ops->get_xip_page);
@@ -68,13 +93,12 @@ do_xip_mapping_read(struct address_space *mapping,
                if (unlikely(IS_ERR(page))) {
                        if (PTR_ERR(page) == -ENODATA) {
                                /* sparse */
-                               page = virt_to_page(empty_zero_page);
+                               page = ZERO_PAGE(0);
                        } else {
                                desc->error = PTR_ERR(page);
                                goto out;
                        }
-               } else
-                       BUG_ON(!PageUptodate(page));
+               }
 
                /* If users can be writing to this page using arbitrary
                 * virtual addresses, take care about potential aliasing
@@ -84,8 +108,7 @@ do_xip_mapping_read(struct address_space *mapping,
                        flush_dcache_page(page);
 
                /*
-                * Ok, we have the page, and it's up-to-date, so
-                * now we can copy it to user space...
+                * Ok, we have the page, so now we can copy it to user space...
                 *
                 * The actor routine returns how many bytes were actually used..
                 * NOTE! This may not be the same as how much of a user buffer
@@ -137,34 +160,12 @@ xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 }
 EXPORT_SYMBOL_GPL(xip_file_read);
 
-ssize_t
-xip_file_sendfile(struct file *in_file, loff_t *ppos,
-            size_t count, read_actor_t actor, void *target)
-{
-       read_descriptor_t desc;
-
-       if (!count)
-               return 0;
-
-       desc.written = 0;
-       desc.count = count;
-       desc.arg.data = target;
-       desc.error = 0;
-
-       do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
-                           ppos, &desc, actor);
-       if (desc.written)
-               return desc.written;
-       return desc.error;
-}
-EXPORT_SYMBOL_GPL(xip_file_sendfile);
-
 /*
  * __xip_unmap is invoked from xip_unmap and
  * xip_write
  *
  * This function walks all vmas of the address_space and unmaps the
- * empty_zero_page when found at pgoff. Should it go in rmap.c?
+ * __xip_sparse_page when found at pgoff.
  */
 static void
 __xip_unmap (struct address_space * mapping,
@@ -176,6 +177,12 @@ __xip_unmap (struct address_space * mapping,
        unsigned long address;
        pte_t *pte;
        pte_t pteval;
+       spinlock_t *ptl;
+       struct page *page;
+
+       page = __xip_sparse_page;
+       if (!page)
+               return;
 
        spin_lock(&mapping->i_mmap_lock);
        vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
@@ -183,81 +190,74 @@ __xip_unmap (struct address_space * mapping,
                address = vma->vm_start +
                        ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
                BUG_ON(address < vma->vm_start || address >= vma->vm_end);
-               /*
-                * We need the page_table_lock to protect us from page faults,
-                * munmap, fork, etc...
-                */
-               pte = page_check_address(virt_to_page(empty_zero_page), mm,
-                                        address);
-               if (!IS_ERR(pte)) {
+               pte = page_check_address(page, mm, address, &ptl);
+               if (pte) {
                        /* Nuke the page table entry. */
-                       flush_cache_page(vma, address, pte_pfn(pte));
+                       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);
                        BUG_ON(pte_dirty(pteval));
-                       pte_unmap(pte);
-                       spin_unlock(&mm->page_table_lock);
+                       pte_unmap_unlock(pte, ptl);
+                       page_cache_release(page);
                }
        }
        spin_unlock(&mapping->i_mmap_lock);
 }
 
 /*
- * xip_nopage() is invoked via the vma operations vector for a
+ * xip_fault() is invoked via the vma operations vector for a
  * mapped memory region to read in file data during a page fault.
  *
- * This function is derived from filemap_nopage, but used for execute in place
+ * This function is derived from filemap_fault, but used for execute in place
  */
-static struct page *
-xip_file_nopage(struct vm_area_struct * area,
-                  unsigned long address,
-                  int *type)
+static int xip_file_fault(struct vm_area_struct *area, struct vm_fault *vmf)
 {
        struct file *file = area->vm_file;
        struct address_space *mapping = file->f_mapping;
        struct inode *inode = mapping->host;
        struct page *page;
-       unsigned long size, pgoff, endoff;
+       pgoff_t size;
 
-       pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
-               + area->vm_pgoff;
-       endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
-               + area->vm_pgoff;
+       /* XXX: are VM_FAULT_ codes OK? */
 
        size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-       if (pgoff >= size) {
-               return NULL;
-       }
+       if (vmf->pgoff >= size)
+               return VM_FAULT_SIGBUS;
 
-       page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
-       if (!IS_ERR(page)) {
-               BUG_ON(!PageUptodate(page));
-               return page;
-       }
+       page = mapping->a_ops->get_xip_page(mapping,
+                                       vmf->pgoff*(PAGE_SIZE/512), 0);
+       if (!IS_ERR(page))
+               goto out;
        if (PTR_ERR(page) != -ENODATA)
-               return NULL;
+               return VM_FAULT_OOM;
 
        /* sparse block */
        if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
            (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
            (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
                /* maybe shared writable, allocate new block */
-               page = mapping->a_ops->get_xip_page (mapping,
-                       pgoff*(PAGE_SIZE/512), 1);
+               page = mapping->a_ops->get_xip_page(mapping,
+                                       vmf->pgoff*(PAGE_SIZE/512), 1);
                if (IS_ERR(page))
-                       return NULL;
-               BUG_ON(!PageUptodate(page));
+                       return VM_FAULT_SIGBUS;
                /* unmap page at pgoff from all other vmas */
-               __xip_unmap(mapping, pgoff);
+               __xip_unmap(mapping, vmf->pgoff);
        } else {
-               /* not shared and writable, use empty_zero_page */
-               page = virt_to_page(empty_zero_page);
+               /* not shared and writable, use xip_sparse_page() */
+               page = xip_sparse_page();
+               if (!page)
+                       return VM_FAULT_OOM;
        }
 
-       return page;
+out:
+       page_cache_get(page);
+       vmf->page = page;
+       return 0;
 }
 
 static struct vm_operations_struct xip_file_vm_ops = {
-       .nopage         = xip_file_nopage,
+       .fault  = xip_file_fault,
 };
 
 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
@@ -266,6 +266,7 @@ int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 
        file_accessed(file);
        vma->vm_ops = &xip_file_vm_ops;
+       vma->vm_flags |= VM_CAN_NONLINEAR;
        return 0;
 }
 EXPORT_SYMBOL_GPL(xip_file_mmap);
@@ -275,7 +276,7 @@ __xip_file_write(struct file *filp, const char __user *buf,
                  size_t count, loff_t pos, loff_t *ppos)
 {
        struct address_space * mapping = filp->f_mapping;
-       struct address_space_operations *a_ops = mapping->a_ops;
+       const struct address_space_operations *a_ops = mapping->a_ops;
        struct inode    *inode = mapping->host;
        long            status = 0;
        struct page     *page;
@@ -288,6 +289,7 @@ __xip_file_write(struct file *filp, const char __user *buf,
                unsigned long index;
                unsigned long offset;
                size_t copied;
+               char *kaddr;
 
                offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
                index = pos >> PAGE_CACHE_SHIFT;
@@ -295,14 +297,6 @@ __xip_file_write(struct file *filp, const char __user *buf,
                if (bytes > count)
                        bytes = count;
 
-               /*
-                * Bring in the user page that we will copy from _first_.
-                * Otherwise there's a nasty deadlock on copying from the
-                * same page as we're writing to, without it being marked
-                * up-to-date.
-                */
-               fault_in_pages_readable(buf, bytes);
-
                page = a_ops->get_xip_page(mapping,
                                           index*(PAGE_SIZE/512), 0);
                if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
@@ -319,10 +313,13 @@ __xip_file_write(struct file *filp, const char __user *buf,
                        break;
                }
 
-               BUG_ON(!PageUptodate(page));
-
-               copied = filemap_copy_from_user(page, offset, buf, bytes);
+               fault_in_pages_readable(buf, bytes);
+               kaddr = kmap_atomic(page, KM_USER0);
+               copied = bytes -
+                       __copy_from_user_inatomic_nocache(kaddr + offset, buf, bytes);
+               kunmap_atomic(kaddr, KM_USER0);
                flush_dcache_page(page);
+
                if (likely(copied > 0)) {
                        status = copied;
 
@@ -342,7 +339,7 @@ __xip_file_write(struct file *filp, const char __user *buf,
        *ppos = pos;
        /*
         * No need to use i_size_read() here, the i_size
-        * cannot change under us because we hold i_sem.
+        * cannot change under us because we hold i_mutex.
         */
        if (pos > inode->i_size) {
                i_size_write(inode, pos);
@@ -362,7 +359,7 @@ xip_file_write(struct file *filp, const char __user *buf, size_t len,
        loff_t pos;
        ssize_t ret;
 
-       down(&inode->i_sem);
+       mutex_lock(&inode->i_mutex);
 
        if (!access_ok(VERIFY_READ, buf, len)) {
                ret=-EFAULT;
@@ -383,18 +380,18 @@ xip_file_write(struct file *filp, const char __user *buf, size_t len,
        if (count == 0)
                goto out_backing;
 
-       ret = remove_suid(filp->f_dentry);
+       ret = remove_suid(filp->f_path.dentry);
        if (ret)
                goto out_backing;
 
-       inode_update_time(inode, 1);
+       file_update_time(filp);
 
        ret = __xip_file_write (filp, buf, count, pos, ppos);
 
  out_backing:
        current->backing_dev_info = NULL;
  out_up:
-       up(&inode->i_sem);
+       mutex_unlock(&inode->i_mutex);
        return ret;
 }
 EXPORT_SYMBOL_GPL(xip_file_write);
@@ -412,7 +409,6 @@ xip_truncate_page(struct address_space *mapping, loff_t from)
        unsigned blocksize;
        unsigned length;
        struct page *page;
-       void *kaddr;
 
        BUG_ON(!mapping->a_ops->get_xip_page);
 
@@ -435,13 +431,8 @@ xip_truncate_page(struct address_space *mapping, loff_t from)
                        return 0;
                else
                        return PTR_ERR(page);
-       } else
-               BUG_ON(!PageUptodate(page));
-       kaddr = kmap_atomic(page, KM_USER0);
-       memset(kaddr + offset, 0, length);
-       kunmap_atomic(kaddr, KM_USER0);
-
-       flush_dcache_page(page);
+       }
+       zero_user(page, offset, length);
        return 0;
 }
 EXPORT_SYMBOL_GPL(xip_truncate_page);