get_unmapped_area handles MAP_FIXED in hugetlbfs
[safe/jmp/linux-2.6] / fs / hugetlbfs / inode.c
index e6410d8..fe625cd 100644 (file)
@@ -22,6 +22,7 @@
 #include <linux/backing-dev.h>
 #include <linux/hugetlb.h>
 #include <linux/pagevec.h>
+#include <linux/mman.h>
 #include <linux/quotaops.h>
 #include <linux/slab.h>
 #include <linux/dnotify.h>
 /* some random number */
 #define HUGETLBFS_MAGIC        0x958458f6
 
-static struct super_operations hugetlbfs_ops;
-static struct address_space_operations hugetlbfs_aops;
+static const struct super_operations hugetlbfs_ops;
+static const struct address_space_operations hugetlbfs_aops;
 const struct file_operations hugetlbfs_file_operations;
-static struct inode_operations hugetlbfs_dir_inode_operations;
-static struct inode_operations hugetlbfs_inode_operations;
+static const struct inode_operations hugetlbfs_dir_inode_operations;
+static const struct inode_operations hugetlbfs_inode_operations;
 
 static struct backing_dev_info hugetlbfs_backing_dev_info = {
        .ra_pages       = 0,    /* No readahead */
@@ -58,33 +59,26 @@ static void huge_pagevec_release(struct pagevec *pvec)
 
 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 {
-       struct inode *inode = file->f_dentry->d_inode;
+       struct inode *inode = file->f_path.dentry->d_inode;
        loff_t len, vma_len;
        int ret;
 
-       if (vma->vm_pgoff & (HPAGE_SIZE / PAGE_SIZE - 1))
-               return -EINVAL;
-
-       if (vma->vm_start & ~HPAGE_MASK)
-               return -EINVAL;
-
-       if (vma->vm_end & ~HPAGE_MASK)
-               return -EINVAL;
-
-       if (vma->vm_end - vma->vm_start < HPAGE_SIZE)
-               return -EINVAL;
+       /*
+        * vma alignment has already been checked by prepare_hugepage_range.
+        * If you add any error returns here, do so after setting VM_HUGETLB,
+        * so is_vm_hugetlb_page tests below unmap_region go the right way
+        * when do_mmap_pgoff unwinds (may be important on powerpc and ia64).
+        */
+       vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
+       vma->vm_ops = &hugetlb_vm_ops;
 
        vma_len = (loff_t)(vma->vm_end - vma->vm_start);
 
        mutex_lock(&inode->i_mutex);
        file_accessed(file);
-       vma->vm_flags |= VM_HUGETLB | VM_RESERVED;
-       vma->vm_ops = &hugetlb_vm_ops;
 
        ret = -ENOMEM;
        len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
-       if (!(vma->vm_flags & VM_WRITE) && len > inode->i_size)
-               goto out;
 
        if (vma->vm_flags & VM_MAYSHARE &&
            hugetlb_reserve_pages(inode, vma->vm_pgoff >> (HPAGE_SHIFT-PAGE_SHIFT),
@@ -93,7 +87,7 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
        ret = 0;
        hugetlb_prefault_arch_hook(vma->vm_mm);
-       if (inode->i_size < len)
+       if (vma->vm_flags & VM_WRITE && inode->i_size < len)
                inode->i_size = len;
 out:
        mutex_unlock(&inode->i_mutex);
@@ -105,10 +99,7 @@ out:
  * Called under down_write(mmap_sem).
  */
 
-#ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
-unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
-               unsigned long len, unsigned long pgoff, unsigned long flags);
-#else
+#ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
 static unsigned long
 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
                unsigned long len, unsigned long pgoff, unsigned long flags)
@@ -122,6 +113,12 @@ hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
        if (len > TASK_SIZE)
                return -ENOMEM;
 
+       if (flags & MAP_FIXED) {
+               if (prepare_hugepage_range(addr, len, pgoff))
+                       return -EINVAL;
+               return addr;
+       }
+
        if (addr) {
                addr = ALIGN(addr, HPAGE_SIZE);
                vma = find_vma(mm, addr);
@@ -183,7 +180,7 @@ static int hugetlbfs_commit_write(struct file *file,
 
 static void truncate_huge_page(struct page *page)
 {
-       clear_page_dirty(page);
+       cancel_dirty_page(page, /* No IO accounting for huge pages? */0);
        ClearPageUptodate(page);
        remove_from_page_cache(page);
        put_page(page);
@@ -231,7 +228,7 @@ static void hugetlbfs_delete_inode(struct inode *inode)
        clear_inode(inode);
 }
 
-static void hugetlbfs_forget_inode(struct inode *inode)
+static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 {
        struct super_block *sb = inode->i_sb;
 
@@ -273,29 +270,27 @@ static void hugetlbfs_drop_inode(struct inode *inode)
                hugetlbfs_forget_inode(inode);
 }
 
-/*
- * h_pgoff is in HPAGE_SIZE units.
- * vma->vm_pgoff is in PAGE_SIZE units.
- */
 static inline void
-hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
+hugetlb_vmtruncate_list(struct prio_tree_root *root, pgoff_t pgoff)
 {
        struct vm_area_struct *vma;
        struct prio_tree_iter iter;
 
-       vma_prio_tree_foreach(vma, &iter, root, h_pgoff, ULONG_MAX) {
-               unsigned long h_vm_pgoff;
+       vma_prio_tree_foreach(vma, &iter, root, pgoff, ULONG_MAX) {
                unsigned long v_offset;
 
-               h_vm_pgoff = vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT);
-               v_offset = (h_pgoff - h_vm_pgoff) << HPAGE_SHIFT;
                /*
-                * Is this VMA fully outside the truncation point?
+                * Can the expression below overflow on 32-bit arches?
+                * No, because the prio_tree returns us only those vmas
+                * which overlap the truncated area starting at pgoff,
+                * and no vma on a 32-bit arch can span beyond the 4GB.
                 */
-               if (h_vm_pgoff >= h_pgoff)
+               if (vma->vm_pgoff < pgoff)
+                       v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT;
+               else
                        v_offset = 0;
 
-               unmap_hugepage_range(vma,
+               __unmap_hugepage_range(vma,
                                vma->vm_start + v_offset, vma->vm_end);
        }
 }
@@ -305,14 +300,14 @@ hugetlb_vmtruncate_list(struct prio_tree_root *root, unsigned long h_pgoff)
  */
 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
 {
-       unsigned long pgoff;
+       pgoff_t pgoff;
        struct address_space *mapping = inode->i_mapping;
 
        if (offset > inode->i_size)
                return -EINVAL;
 
        BUG_ON(offset & ~HPAGE_MASK);
-       pgoff = offset >> HPAGE_SHIFT;
+       pgoff = offset >> PAGE_SHIFT;
 
        inode->i_size = offset;
        spin_lock(&mapping->i_mmap_lock);
@@ -359,7 +354,6 @@ static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
                inode->i_mode = mode;
                inode->i_uid = uid;
                inode->i_gid = gid;
-               inode->i_blksize = HPAGE_SIZE;
                inode->i_blocks = 0;
                inode->i_mapping->a_ops = &hugetlbfs_aops;
                inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
@@ -380,7 +374,7 @@ static struct inode *hugetlbfs_get_inode(struct super_block *sb, uid_t uid,
                        inode->i_fop = &simple_dir_operations;
 
                        /* directory inodes start off with i_nlink == 2 (for "." entry) */
-                       inode->i_nlink++;
+                       inc_nlink(inode);
                        break;
                case S_IFLNK:
                        inode->i_op = &page_symlink_inode_operations;
@@ -421,7 +415,7 @@ static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 {
        int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
        if (!retval)
-               dir->i_nlink++;
+               inc_nlink(dir);
        return retval;
 }
 
@@ -459,10 +453,13 @@ static int hugetlbfs_symlink(struct inode *dir,
 }
 
 /*
- * For direct-IO reads into hugetlb pages
+ * mark the head page dirty
  */
 static int hugetlbfs_set_page_dirty(struct page *page)
 {
+       struct page *head = compound_head(page);
+
+       SetPageDirty(head);
        return 0;
 }
 
@@ -523,7 +520,7 @@ static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
 }
 
 
-static kmem_cache_t *hugetlbfs_inode_cachep;
+static struct kmem_cache *hugetlbfs_inode_cachep;
 
 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 {
@@ -532,7 +529,7 @@ static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 
        if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
                return NULL;
-       p = kmem_cache_alloc(hugetlbfs_inode_cachep, SLAB_KERNEL);
+       p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
        if (unlikely(!p)) {
                hugetlbfs_inc_free_inodes(sbinfo);
                return NULL;
@@ -547,7 +544,7 @@ static void hugetlbfs_destroy_inode(struct inode *inode)
        kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 }
 
-static struct address_space_operations hugetlbfs_aops = {
+static const struct address_space_operations hugetlbfs_aops = {
        .readpage       = hugetlbfs_readpage,
        .prepare_write  = hugetlbfs_prepare_write,
        .commit_write   = hugetlbfs_commit_write,
@@ -555,7 +552,7 @@ static struct address_space_operations hugetlbfs_aops = {
 };
 
 
-static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
+static void init_once(void *foo, struct kmem_cache *cachep, unsigned long flags)
 {
        struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 
@@ -570,7 +567,7 @@ const struct file_operations hugetlbfs_file_operations = {
        .get_unmapped_area      = hugetlb_get_unmapped_area,
 };
 
-static struct inode_operations hugetlbfs_dir_inode_operations = {
+static const struct inode_operations hugetlbfs_dir_inode_operations = {
        .create         = hugetlbfs_create,
        .lookup         = simple_lookup,
        .link           = simple_link,
@@ -583,11 +580,11 @@ static struct inode_operations hugetlbfs_dir_inode_operations = {
        .setattr        = hugetlbfs_setattr,
 };
 
-static struct inode_operations hugetlbfs_inode_operations = {
+static const struct inode_operations hugetlbfs_inode_operations = {
        .setattr        = hugetlbfs_setattr,
 };
 
-static struct super_operations hugetlbfs_ops = {
+static const struct super_operations hugetlbfs_ops = {
        .alloc_inode    = hugetlbfs_alloc_inode,
        .destroy_inode  = hugetlbfs_destroy_inode,
        .statfs         = hugetlbfs_statfs,
@@ -627,7 +624,6 @@ hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
                                do_div(size, 100);
                                rest++;
                        }
-                       size &= HPAGE_MASK;
                        pconfig->nr_blocks = (size >> HPAGE_SHIFT);
                        value = rest;
                } else if (!strcmp(opt,"nr_inodes")) {
@@ -785,8 +781,8 @@ struct file *hugetlb_zero_setup(size_t size)
        d_instantiate(dentry, inode);
        inode->i_size = size;
        inode->i_nlink = 0;
-       file->f_vfsmnt = mntget(hugetlbfs_vfsmount);
-       file->f_dentry = dentry;
+       file->f_path.mnt = mntget(hugetlbfs_vfsmount);
+       file->f_path.dentry = dentry;
        file->f_mapping = inode->i_mapping;
        file->f_op = &hugetlbfs_file_operations;
        file->f_mode = FMODE_WRITE | FMODE_READ;