Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / mm / shmem.c
index e8dc5af..c066617 100644 (file)
@@ -6,8 +6,8 @@
  *              2000-2001 Christoph Rohland
  *              2000-2001 SAP AG
  *              2002 Red Hat Inc.
- * Copyright (C) 2002-2004 Hugh Dickins.
- * Copyright (C) 2002-2004 VERITAS Software Corporation.
+ * Copyright (C) 2002-2005 Hugh Dickins.
+ * Copyright (C) 2002-2005 VERITAS Software Corporation.
  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  *
  * Extended attribute support for tmpfs:
@@ -45,7 +45,9 @@
 #include <linux/swapops.h>
 #include <linux/mempolicy.h>
 #include <linux/namei.h>
-#include <linux/xattr.h>
+#include <linux/ctype.h>
+#include <linux/migrate.h>
+
 #include <asm/uaccess.h>
 #include <asm/div64.h>
 #include <asm/pgtable.h>
@@ -71,9 +73,6 @@
 /* Pretend that each entry is of this size in directory's i_size */
 #define BOGO_DIRENT_SIZE 20
 
-/* Keep swapped page count in private field of indirect struct page */
-#define nr_swapped             private
-
 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
 enum sgp_type {
        SGP_QUICK,      /* don't try more than file page cache lookup */
@@ -85,7 +84,7 @@ enum sgp_type {
 static int shmem_getpage(struct inode *inode, unsigned long idx,
                         struct page **pagep, enum sgp_type sgp, int *type);
 
-static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
+static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
 {
        /*
         * The above definition of ENTRIES_PER_PAGE, and the use of
@@ -178,10 +177,9 @@ static struct address_space_operations shmem_aops;
 static struct file_operations shmem_file_operations;
 static struct inode_operations shmem_inode_operations;
 static struct inode_operations shmem_dir_inode_operations;
-static struct inode_operations shmem_special_inode_operations;
 static struct vm_operations_struct shmem_vm_ops;
 
-static struct backing_dev_info shmem_backing_dev_info = {
+static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
        .ra_pages       = 0,    /* No readahead */
        .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
        .unplug_io_fn   = default_unplug_io_fn,
@@ -193,7 +191,7 @@ static DEFINE_SPINLOCK(shmem_swaplist_lock);
 static void shmem_free_blocks(struct inode *inode, long pages)
 {
        struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
-       if (sbinfo) {
+       if (sbinfo->max_blocks) {
                spin_lock(&sbinfo->stat_lock);
                sbinfo->free_blocks += pages;
                inode->i_blocks -= pages*BLOCKS_PER_PAGE;
@@ -325,8 +323,10 @@ static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, uns
 
        entry->val = value;
        info->swapped += incdec;
-       if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
-               kmap_atomic_to_page(entry)->nr_swapped += incdec;
+       if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
+               struct page *page = kmap_atomic_to_page(entry);
+               set_page_private(page, page_private(page) + incdec);
+       }
 }
 
 /*
@@ -356,7 +356,7 @@ static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long
                 * page (and perhaps indirect index pages) yet to allocate:
                 * a waste to allocate index if we cannot allocate data.
                 */
-               if (sbinfo) {
+               if (sbinfo->max_blocks) {
                        spin_lock(&sbinfo->stat_lock);
                        if (sbinfo->free_blocks <= 1) {
                                spin_unlock(&sbinfo->stat_lock);
@@ -369,9 +369,8 @@ static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long
 
                spin_unlock(&info->lock);
                page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
-               if (page) {
-                       page->nr_swapped = 0;
-               }
+               if (page)
+                       set_page_private(page, 0);
                spin_lock(&info->lock);
 
                if (!page) {
@@ -460,7 +459,7 @@ static void shmem_free_pages(struct list_head *next)
        } while (next);
 }
 
-static void shmem_truncate(struct inode *inode)
+static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 {
        struct shmem_inode_info *info = SHMEM_I(inode);
        unsigned long idx;
@@ -478,18 +477,27 @@ static void shmem_truncate(struct inode *inode)
        long nr_swaps_freed = 0;
        int offset;
        int freed;
+       int punch_hole = 0;
 
        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
-       idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+       idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
        if (idx >= info->next_index)
                return;
 
        spin_lock(&info->lock);
        info->flags |= SHMEM_TRUNCATE;
-       limit = info->next_index;
-       info->next_index = idx;
+       if (likely(end == (loff_t) -1)) {
+               limit = info->next_index;
+               info->next_index = idx;
+       } else {
+               limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+               if (limit > info->next_index)
+                       limit = info->next_index;
+               punch_hole = 1;
+       }
+
        topdir = info->i_indirect;
-       if (topdir && idx <= SHMEM_NR_DIRECT) {
+       if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
                info->i_indirect = NULL;
                nr_pages_to_free++;
                list_add(&topdir->lru, &pages_to_free);
@@ -562,7 +570,7 @@ static void shmem_truncate(struct inode *inode)
                        diroff = 0;
                }
                subdir = dir[diroff];
-               if (subdir && subdir->nr_swapped) {
+               if (subdir && page_private(subdir)) {
                        size = limit - idx;
                        if (size > ENTRIES_PER_PAGE)
                                size = ENTRIES_PER_PAGE;
@@ -573,14 +581,15 @@ static void shmem_truncate(struct inode *inode)
                        nr_swaps_freed += freed;
                        if (offset)
                                spin_lock(&info->lock);
-                       subdir->nr_swapped -= freed;
+                       set_page_private(subdir, page_private(subdir) - freed);
                        if (offset)
                                spin_unlock(&info->lock);
-                       BUG_ON(subdir->nr_swapped > offset);
+                       if (!punch_hole)
+                               BUG_ON(page_private(subdir) > offset);
                }
                if (offset)
                        offset = 0;
-               else if (subdir) {
+               else if (subdir && !page_private(subdir)) {
                        dir[diroff] = NULL;
                        nr_pages_to_free++;
                        list_add(&subdir->lru, &pages_to_free);
@@ -597,7 +606,7 @@ done2:
                 * Also, though shmem_getpage checks i_size before adding to
                 * cache, no recheck after: so fix the narrow window there too.
                 */
-               truncate_inode_pages(inode->i_mapping, inode->i_size);
+               truncate_inode_pages_range(inode->i_mapping, start, end);
        }
 
        spin_lock(&info->lock);
@@ -617,6 +626,11 @@ done2:
        }
 }
 
+static void shmem_truncate(struct inode *inode)
+{
+       shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
+}
+
 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 {
        struct inode *inode = dentry->d_inode;
@@ -667,6 +681,7 @@ static void shmem_delete_inode(struct inode *inode)
        struct shmem_inode_info *info = SHMEM_I(inode);
 
        if (inode->i_op->truncate == shmem_truncate) {
+               truncate_inode_pages(inode->i_mapping, 0);
                shmem_unacct_size(info->flags, inode->i_size);
                inode->i_size = 0;
                shmem_truncate(inode);
@@ -676,8 +691,8 @@ static void shmem_delete_inode(struct inode *inode)
                        spin_unlock(&shmem_swaplist_lock);
                }
        }
-       if (sbinfo) {
-               BUG_ON(inode->i_blocks);
+       BUG_ON(inode->i_blocks);
+       if (sbinfo->max_inodes) {
                spin_lock(&sbinfo->stat_lock);
                sbinfo->free_inodes++;
                spin_unlock(&sbinfo->stat_lock);
@@ -743,7 +758,7 @@ static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, s
                        dir = shmem_dir_map(subdir);
                }
                subdir = *dir;
-               if (subdir && subdir->nr_swapped) {
+               if (subdir && page_private(subdir)) {
                        ptr = shmem_swp_map(subdir);
                        size = limit - idx;
                        if (size > ENTRIES_PER_PAGE)
@@ -857,10 +872,55 @@ unlock:
        swap_free(swap);
 redirty:
        set_page_dirty(page);
-       return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
+       return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
 }
 
 #ifdef CONFIG_NUMA
+static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
+{
+       char *nodelist = strchr(value, ':');
+       int err = 1;
+
+       if (nodelist) {
+               /* NUL-terminate policy string */
+               *nodelist++ = '\0';
+               if (nodelist_parse(nodelist, *policy_nodes))
+                       goto out;
+       }
+       if (!strcmp(value, "default")) {
+               *policy = MPOL_DEFAULT;
+               /* Don't allow a nodelist */
+               if (!nodelist)
+                       err = 0;
+       } else if (!strcmp(value, "prefer")) {
+               *policy = MPOL_PREFERRED;
+               /* Insist on a nodelist of one node only */
+               if (nodelist) {
+                       char *rest = nodelist;
+                       while (isdigit(*rest))
+                               rest++;
+                       if (!*rest)
+                               err = 0;
+               }
+       } else if (!strcmp(value, "bind")) {
+               *policy = MPOL_BIND;
+               /* Insist on a nodelist */
+               if (nodelist)
+                       err = 0;
+       } else if (!strcmp(value, "interleave")) {
+               *policy = MPOL_INTERLEAVE;
+               /* Default to nodes online if no nodelist */
+               if (!nodelist)
+                       *policy_nodes = node_online_map;
+               err = 0;
+       }
+out:
+       /* Restore string for error message */
+       if (nodelist)
+               *--nodelist = ':';
+       return err;
+}
+
 static struct page *shmem_swapin_async(struct shared_policy *p,
                                       swp_entry_t entry, unsigned long idx)
 {
@@ -898,7 +958,7 @@ struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 }
 
 static struct page *
-shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
+shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
                 unsigned long idx)
 {
        struct vm_area_struct pvma;
@@ -913,6 +973,11 @@ shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
        return page;
 }
 #else
+static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
+{
+       return 1;
+}
+
 static inline struct page *
 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 {
@@ -921,8 +986,7 @@ shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 }
 
 static inline struct page *
-shmem_alloc_page(unsigned int __nocast gfp,struct shmem_inode_info *info,
-                                unsigned long idx)
+shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
 {
        return alloc_page(gfp | __GFP_ZERO);
 }
@@ -1016,6 +1080,14 @@ repeat:
                        page_cache_release(swappage);
                        goto repeat;
                }
+               if (!PageSwapCache(swappage)) {
+                       /* Page migration has occured */
+                       shmem_swp_unmap(entry);
+                       spin_unlock(&info->lock);
+                       unlock_page(swappage);
+                       page_cache_release(swappage);
+                       goto repeat;
+               }
                if (PageWriteback(swappage)) {
                        shmem_swp_unmap(entry);
                        spin_unlock(&info->lock);
@@ -1079,7 +1151,7 @@ repeat:
        } else {
                shmem_swp_unmap(entry);
                sbinfo = SHMEM_SB(inode->i_sb);
-               if (sbinfo) {
+               if (sbinfo->max_blocks) {
                        spin_lock(&sbinfo->stat_lock);
                        if (sbinfo->free_blocks == 0 ||
                            shmem_acct_block(info->flags)) {
@@ -1194,6 +1266,7 @@ static int shmem_populate(struct vm_area_struct *vma,
                err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
                if (err)
                        return err;
+               /* Page may still be null, but only if nonblock was set. */
                if (page) {
                        mark_page_accessed(page);
                        err = install_page(mm, vma, addr, page, prot);
@@ -1201,7 +1274,10 @@ static int shmem_populate(struct vm_area_struct *vma,
                                page_cache_release(page);
                                return err;
                        }
-               } else if (nonblock) {
+               } else if (vma->vm_flags & VM_NONLINEAR) {
+                       /* No page was found just because we can't read it in
+                        * now (being here implies nonblock != 0), but the page
+                        * may exist, so set the PTE to fault it in later. */
                        err = install_file_pte(mm, vma, addr, pgoff, prot);
                        if (err)
                                return err;
@@ -1254,7 +1330,7 @@ out_nomem:
        return retval;
 }
 
-static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
+int shmem_mmap(struct file *file, struct vm_area_struct *vma)
 {
        file_accessed(file);
        vma->vm_ops = &shmem_vm_ops;
@@ -1268,7 +1344,7 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
        struct shmem_inode_info *info;
        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 
-       if (sbinfo) {
+       if (sbinfo->max_inodes) {
                spin_lock(&sbinfo->stat_lock);
                if (!sbinfo->free_inodes) {
                        spin_unlock(&sbinfo->stat_lock);
@@ -1295,13 +1371,13 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
 
                switch (mode & S_IFMT) {
                default:
-                       inode->i_op = &shmem_special_inode_operations;
                        init_special_inode(inode, mode, dev);
                        break;
                case S_IFREG:
                        inode->i_op = &shmem_inode_operations;
                        inode->i_fop = &shmem_file_operations;
-                       mpol_shared_policy_init(&info->policy);
+                       mpol_shared_policy_init(&info->policy, sbinfo->policy,
+                                                       &sbinfo->policy_nodes);
                        break;
                case S_IFDIR:
                        inode->i_nlink++;
@@ -1315,10 +1391,11 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
                         * Must not load anything in the rbtree,
                         * mpol_free_shared_policy will not be called.
                         */
-                       mpol_shared_policy_init(&info->policy);
+                       mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
+                                               NULL);
                        break;
                }
-       } else if (sbinfo) {
+       } else if (sbinfo->max_inodes) {
                spin_lock(&sbinfo->stat_lock);
                sbinfo->free_inodes++;
                spin_unlock(&sbinfo->stat_lock);
@@ -1327,31 +1404,6 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
 }
 
 #ifdef CONFIG_TMPFS
-
-static int shmem_set_size(struct shmem_sb_info *sbinfo,
-                         unsigned long max_blocks, unsigned long max_inodes)
-{
-       int error;
-       unsigned long blocks, inodes;
-
-       spin_lock(&sbinfo->stat_lock);
-       blocks = sbinfo->max_blocks - sbinfo->free_blocks;
-       inodes = sbinfo->max_inodes - sbinfo->free_inodes;
-       error = -EINVAL;
-       if (max_blocks < blocks)
-               goto out;
-       if (max_inodes < inodes)
-               goto out;
-       error = 0;
-       sbinfo->max_blocks  = max_blocks;
-       sbinfo->free_blocks = max_blocks - blocks;
-       sbinfo->max_inodes  = max_inodes;
-       sbinfo->free_inodes = max_inodes - inodes;
-out:
-       spin_unlock(&sbinfo->stat_lock);
-       return error;
-}
-
 static struct inode_operations shmem_symlink_inode_operations;
 static struct inode_operations shmem_symlink_inline_operations;
 
@@ -1380,7 +1432,7 @@ shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t
        if (!access_ok(VERIFY_READ, buf, count))
                return -EFAULT;
 
-       down(&inode->i_sem);
+       mutex_lock(&inode->i_mutex);
 
        pos = *ppos;
        written = 0;
@@ -1465,7 +1517,7 @@ shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t
        if (written)
                err = written;
 out:
-       up(&inode->i_sem);
+       mutex_unlock(&inode->i_mutex);
        return err;
 }
 
@@ -1501,7 +1553,7 @@ static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_
 
                /*
                 * We must evaluate after, since reads (unlike writes)
-                * are called without i_sem protection against truncate
+                * are called without i_mutex protection against truncate
                 */
                nr = PAGE_CACHE_SIZE;
                i_size = i_size_read(inode);
@@ -1529,8 +1581,10 @@ static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_
                         */
                        if (!offset)
                                mark_page_accessed(page);
-               } else
+               } else {
                        page = ZERO_PAGE(0);
+                       page_cache_get(page);
+               }
 
                /*
                 * Ok, we have the page, and it's up-to-date, so
@@ -1606,15 +1660,17 @@ static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
        buf->f_type = TMPFS_SUPER_MAGIC;
        buf->f_bsize = PAGE_CACHE_SIZE;
        buf->f_namelen = NAME_MAX;
-       if (sbinfo) {
-               spin_lock(&sbinfo->stat_lock);
+       spin_lock(&sbinfo->stat_lock);
+       if (sbinfo->max_blocks) {
                buf->f_blocks = sbinfo->max_blocks;
                buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
+       }
+       if (sbinfo->max_inodes) {
                buf->f_files = sbinfo->max_inodes;
                buf->f_ffree = sbinfo->free_inodes;
-               spin_unlock(&sbinfo->stat_lock);
        }
        /* else leave those fields 0 like simple_statfs */
+       spin_unlock(&sbinfo->stat_lock);
        return 0;
 }
 
@@ -1628,6 +1684,15 @@ shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
        int error = -ENOSPC;
 
        if (inode) {
+               error = security_inode_init_security(inode, dir, NULL, NULL,
+                                                    NULL);
+               if (error) {
+                       if (error != -EOPNOTSUPP) {
+                               iput(inode);
+                               return error;
+                       }
+                       error = 0;
+               }
                if (dir->i_mode & S_ISGID) {
                        inode->i_gid = dir->i_gid;
                        if (S_ISDIR(mode))
@@ -1637,7 +1702,6 @@ shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
                dir->i_ctime = dir->i_mtime = CURRENT_TIME;
                d_instantiate(dentry, inode);
                dget(dentry); /* Extra count - pin the dentry in core */
-               error = 0;
        }
        return error;
 }
@@ -1671,7 +1735,7 @@ static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentr
         * but each new link needs a new dentry, pinning lowmem, and
         * tmpfs dentries cannot be pruned until they are unlinked.
         */
-       if (sbinfo) {
+       if (sbinfo->max_inodes) {
                spin_lock(&sbinfo->stat_lock);
                if (!sbinfo->free_inodes) {
                        spin_unlock(&sbinfo->stat_lock);
@@ -1696,7 +1760,7 @@ static int shmem_unlink(struct inode *dir, struct dentry *dentry)
 
        if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
                struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
-               if (sbinfo) {
+               if (sbinfo->max_inodes) {
                        spin_lock(&sbinfo->stat_lock);
                        sbinfo->free_inodes++;
                        spin_unlock(&sbinfo->stat_lock);
@@ -1715,6 +1779,7 @@ static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
        if (!simple_empty(dentry))
                return -ENOTEMPTY;
 
+       dentry->d_inode->i_nlink--;
        dir->i_nlink--;
        return shmem_unlink(dir, dentry);
 }
@@ -1767,6 +1832,16 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
        if (!inode)
                return -ENOSPC;
 
+       error = security_inode_init_security(inode, dir, NULL, NULL,
+                                            NULL);
+       if (error) {
+               if (error != -EOPNOTSUPP) {
+                       iput(inode);
+                       return error;
+               }
+               error = 0;
+       }
+
        info = SHMEM_I(inode);
        inode->i_size = len-1;
        if (len <= (char *)inode - (char *)info) {
@@ -1795,44 +1870,33 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
        return 0;
 }
 
-static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
+static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
 {
        nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
-       return 0;
+       return NULL;
 }
 
-static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
+static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
        struct page *page = NULL;
        int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
        nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
-       return 0;
+       return page;
 }
 
-static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
+static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
 {
        if (!IS_ERR(nd_get_link(nd))) {
-               struct page *page;
-
-               page = find_get_page(dentry->d_inode->i_mapping, 0);
-               if (!page)
-                       BUG();
+               struct page *page = cookie;
                kunmap(page);
                mark_page_accessed(page);
                page_cache_release(page);
-               page_cache_release(page);
        }
 }
 
 static struct inode_operations shmem_symlink_inline_operations = {
        .readlink       = generic_readlink,
        .follow_link    = shmem_follow_link_inline,
-#ifdef CONFIG_TMPFS_XATTR
-       .setxattr       = generic_setxattr,
-       .getxattr       = generic_getxattr,
-       .listxattr      = generic_listxattr,
-       .removexattr    = generic_removexattr,
-#endif
 };
 
 static struct inode_operations shmem_symlink_inode_operations = {
@@ -1840,19 +1904,31 @@ static struct inode_operations shmem_symlink_inode_operations = {
        .readlink       = generic_readlink,
        .follow_link    = shmem_follow_link,
        .put_link       = shmem_put_link,
-#ifdef CONFIG_TMPFS_XATTR
-       .setxattr       = generic_setxattr,
-       .getxattr       = generic_getxattr,
-       .listxattr      = generic_listxattr,
-       .removexattr    = generic_removexattr,
-#endif
 };
 
-static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
+static int shmem_parse_options(char *options, int *mode, uid_t *uid,
+       gid_t *gid, unsigned long *blocks, unsigned long *inodes,
+       int *policy, nodemask_t *policy_nodes)
 {
        char *this_char, *value, *rest;
 
-       while ((this_char = strsep(&options, ",")) != NULL) {
+       while (options != NULL) {
+               this_char = options;
+               for (;;) {
+                       /*
+                        * NUL-terminate this option: unfortunately,
+                        * mount options form a comma-separated list,
+                        * but mpol's nodelist may also contain commas.
+                        */
+                       options = strchr(options, ',');
+                       if (options == NULL)
+                               break;
+                       options++;
+                       if (!isdigit(*options)) {
+                               options[-1] = '\0';
+                               break;
+                       }
+               }
                if (!*this_char)
                        continue;
                if ((value = strchr(this_char,'=')) != NULL) {
@@ -1902,6 +1978,9 @@ static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid,
                        *gid = simple_strtoul(value,&rest,0);
                        if (*rest)
                                goto bad_val;
+               } else if (!strcmp(this_char,"mpol")) {
+                       if (shmem_parse_mpol(value,policy,policy_nodes))
+                               goto bad_val;
                } else {
                        printk(KERN_ERR "tmpfs: Bad mount option %s\n",
                               this_char);
@@ -1920,22 +1999,46 @@ bad_val:
 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
 {
        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
-       unsigned long max_blocks = 0;
-       unsigned long max_inodes = 0;
+       unsigned long max_blocks = sbinfo->max_blocks;
+       unsigned long max_inodes = sbinfo->max_inodes;
+       int policy = sbinfo->policy;
+       nodemask_t policy_nodes = sbinfo->policy_nodes;
+       unsigned long blocks;
+       unsigned long inodes;
+       int error = -EINVAL;
+
+       if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
+                               &max_inodes, &policy, &policy_nodes))
+               return error;
 
-       if (sbinfo) {
-               max_blocks = sbinfo->max_blocks;
-               max_inodes = sbinfo->max_inodes;
-       }
-       if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
-               return -EINVAL;
-       /* Keep it simple: disallow limited <-> unlimited remount */
-       if ((max_blocks || max_inodes) == !sbinfo)
-               return -EINVAL;
-       /* But allow the pointless unlimited -> unlimited remount */
-       if (!sbinfo)
-               return 0;
-       return shmem_set_size(sbinfo, max_blocks, max_inodes);
+       spin_lock(&sbinfo->stat_lock);
+       blocks = sbinfo->max_blocks - sbinfo->free_blocks;
+       inodes = sbinfo->max_inodes - sbinfo->free_inodes;
+       if (max_blocks < blocks)
+               goto out;
+       if (max_inodes < inodes)
+               goto out;
+       /*
+        * Those tests also disallow limited->unlimited while any are in
+        * use, so i_blocks will always be zero when max_blocks is zero;
+        * but we must separately disallow unlimited->limited, because
+        * in that case we have no record of how much is already in use.
+        */
+       if (max_blocks && !sbinfo->max_blocks)
+               goto out;
+       if (max_inodes && !sbinfo->max_inodes)
+               goto out;
+
+       error = 0;
+       sbinfo->max_blocks  = max_blocks;
+       sbinfo->free_blocks = max_blocks - blocks;
+       sbinfo->max_inodes  = max_inodes;
+       sbinfo->free_inodes = max_inodes - inodes;
+       sbinfo->policy = policy;
+       sbinfo->policy_nodes = policy_nodes;
+out:
+       spin_unlock(&sbinfo->stat_lock);
+       return error;
 }
 #endif
 
@@ -1945,12 +2048,6 @@ static void shmem_put_super(struct super_block *sb)
        sb->s_fs_info = NULL;
 }
 
-#ifdef CONFIG_TMPFS_XATTR
-static struct xattr_handler *shmem_xattr_handlers[];
-#else
-#define shmem_xattr_handlers NULL
-#endif
-
 static int shmem_fill_super(struct super_block *sb,
                            void *data, int silent)
 {
@@ -1960,11 +2057,13 @@ static int shmem_fill_super(struct super_block *sb,
        uid_t uid = current->fsuid;
        gid_t gid = current->fsgid;
        int err = -ENOMEM;
-
-#ifdef CONFIG_TMPFS
+       struct shmem_sb_info *sbinfo;
        unsigned long blocks = 0;
        unsigned long inodes = 0;
+       int policy = MPOL_DEFAULT;
+       nodemask_t policy_nodes = node_online_map;
 
+#ifdef CONFIG_TMPFS
        /*
         * Per default we only allow half of the physical ram per
         * tmpfs instance, limiting inodes to one per page of lowmem;
@@ -1975,34 +2074,36 @@ static int shmem_fill_super(struct super_block *sb,
                inodes = totalram_pages - totalhigh_pages;
                if (inodes > blocks)
                        inodes = blocks;
-
-               if (shmem_parse_options(data, &mode,
-                                       &uid, &gid, &blocks, &inodes))
+               if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
+                                       &inodes, &policy, &policy_nodes))
                        return -EINVAL;
        }
-
-       if (blocks || inodes) {
-               struct shmem_sb_info *sbinfo;
-               sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
-               if (!sbinfo)
-                       return -ENOMEM;
-               sb->s_fs_info = sbinfo;
-               spin_lock_init(&sbinfo->stat_lock);
-               sbinfo->max_blocks = blocks;
-               sbinfo->free_blocks = blocks;
-               sbinfo->max_inodes = inodes;
-               sbinfo->free_inodes = inodes;
-       }
-       sb->s_xattr = shmem_xattr_handlers;
 #else
        sb->s_flags |= MS_NOUSER;
 #endif
 
+       /* Round up to L1_CACHE_BYTES to resist false sharing */
+       sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
+                               L1_CACHE_BYTES), GFP_KERNEL);
+       if (!sbinfo)
+               return -ENOMEM;
+
+       spin_lock_init(&sbinfo->stat_lock);
+       sbinfo->max_blocks = blocks;
+       sbinfo->free_blocks = blocks;
+       sbinfo->max_inodes = inodes;
+       sbinfo->free_inodes = inodes;
+       sbinfo->policy = policy;
+       sbinfo->policy_nodes = policy_nodes;
+
+       sb->s_fs_info = sbinfo;
        sb->s_maxbytes = SHMEM_MAX_BYTES;
        sb->s_blocksize = PAGE_CACHE_SIZE;
        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
        sb->s_magic = TMPFS_SUPER_MAGIC;
        sb->s_op = &shmem_ops;
+       sb->s_time_gran = 1;
+
        inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
        if (!inode)
                goto failed;
@@ -2021,7 +2122,7 @@ failed:
        return err;
 }
 
-static kmem_cache_t *shmem_inode_cachep;
+static struct kmem_cache *shmem_inode_cachep;
 
 static struct inode *shmem_alloc_inode(struct super_block *sb)
 {
@@ -2041,7 +2142,8 @@ static void shmem_destroy_inode(struct inode *inode)
        kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
 }
 
-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 shmem_inode_info *p = (struct shmem_inode_info *) foo;
 
@@ -2074,6 +2176,7 @@ static struct address_space_operations shmem_aops = {
        .prepare_write  = shmem_prepare_write,
        .commit_write   = simple_commit_write,
 #endif
+       .migratepage    = migrate_page,
 };
 
 static struct file_operations shmem_file_operations = {
@@ -2090,12 +2193,7 @@ static struct file_operations shmem_file_operations = {
 static struct inode_operations shmem_inode_operations = {
        .truncate       = shmem_truncate,
        .setattr        = shmem_notify_change,
-#ifdef CONFIG_TMPFS_XATTR
-       .setxattr       = generic_setxattr,
-       .getxattr       = generic_getxattr,
-       .listxattr      = generic_listxattr,
-       .removexattr    = generic_removexattr,
-#endif
+       .truncate_range = shmem_truncate_range,
 };
 
 static struct inode_operations shmem_dir_inode_operations = {
@@ -2109,21 +2207,6 @@ static struct inode_operations shmem_dir_inode_operations = {
        .rmdir          = shmem_rmdir,
        .mknod          = shmem_mknod,
        .rename         = shmem_rename,
-#ifdef CONFIG_TMPFS_XATTR
-       .setxattr       = generic_setxattr,
-       .getxattr       = generic_getxattr,
-       .listxattr      = generic_listxattr,
-       .removexattr    = generic_removexattr,
-#endif
-#endif
-};
-
-static struct inode_operations shmem_special_inode_operations = {
-#ifdef CONFIG_TMPFS_XATTR
-       .setxattr       = generic_setxattr,
-       .getxattr       = generic_getxattr,
-       .listxattr      = generic_listxattr,
-       .removexattr    = generic_removexattr,
 #endif
 };
 
@@ -2149,48 +2232,6 @@ static struct vm_operations_struct shmem_vm_ops = {
 };
 
 
-#ifdef CONFIG_TMPFS_SECURITY
-
-static size_t shmem_xattr_security_list(struct inode *inode, char *list, size_t list_len,
-                                       const char *name, size_t name_len)
-{
-       return security_inode_listsecurity(inode, list, list_len);
-}
-
-static int shmem_xattr_security_get(struct inode *inode, const char *name, void *buffer, size_t size)
-{
-       if (strcmp(name, "") == 0)
-               return -EINVAL;
-       return security_inode_getsecurity(inode, name, buffer, size);
-}
-
-static int shmem_xattr_security_set(struct inode *inode, const char *name, const void *value, size_t size, int flags)
-{
-       if (strcmp(name, "") == 0)
-               return -EINVAL;
-       return security_inode_setsecurity(inode, name, value, size, flags);
-}
-
-static struct xattr_handler shmem_xattr_security_handler = {
-       .prefix = XATTR_SECURITY_PREFIX,
-       .list   = shmem_xattr_security_list,
-       .get    = shmem_xattr_security_get,
-       .set    = shmem_xattr_security_set,
-};
-
-#endif /* CONFIG_TMPFS_SECURITY */
-
-#ifdef CONFIG_TMPFS_XATTR
-
-static struct xattr_handler *shmem_xattr_handlers[] = {
-#ifdef CONFIG_TMPFS_SECURITY
-       &shmem_xattr_security_handler,
-#endif
-       NULL
-};
-
-#endif /* CONFIG_TMPFS_XATTR */
-
 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
        int flags, const char *dev_name, void *data)
 {