const: mark struct vm_struct_operations
[linux-2.6.git] / mm / shmem.c
index d8ef7ba..356dd99 100644 (file)
  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  *
+ * tiny-shmem:
+ * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
+ *
  * This file is released under the GPL.
  */
 
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <linux/vfs.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/file.h>
+#include <linux/mm.h>
+#include <linux/module.h>
+#include <linux/swap.h>
+#include <linux/ima.h>
+
+static struct vfsmount *shm_mnt;
+
+#ifdef CONFIG_SHMEM
 /*
  * This virtual memory filesystem is heavily based on the ramfs. It
  * extends ramfs by the ability to use swap and honor resource limits
  * which makes it a completely usable filesystem.
  */
 
-#include <linux/module.h>
-#include <linux/init.h>
-#include <linux/fs.h>
 #include <linux/xattr.h>
 #include <linux/exportfs.h>
 #include <linux/generic_acl.h>
-#include <linux/mm.h>
 #include <linux/mman.h>
-#include <linux/file.h>
-#include <linux/swap.h>
-#include <linux/pagemap.h>
 #include <linux/string.h>
 #include <linux/slab.h>
 #include <linux/backing-dev.h>
 #include <linux/shmem_fs.h>
-#include <linux/mount.h>
 #include <linux/writeback.h>
-#include <linux/vfs.h>
 #include <linux/blkdev.h>
 #include <linux/security.h>
 #include <linux/swapops.h>
 #include <linux/migrate.h>
 #include <linux/highmem.h>
 #include <linux/seq_file.h>
+#include <linux/magic.h>
 
 #include <asm/uaccess.h>
 #include <asm/div64.h>
 #include <asm/pgtable.h>
 
-/* This magic number is used in glibc for posix shared memory */
-#define TMPFS_MAGIC    0x01021994
-
+/*
+ * The maximum size of a shmem/tmpfs file is limited by the maximum size of
+ * its triple-indirect swap vector - see illustration at shmem_swp_entry().
+ *
+ * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
+ * but one eighth of that on a 64-bit kernel.  With 8kB page size, maximum
+ * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
+ * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
+ *
+ * We use / and * instead of shifts in the definitions below, so that the swap
+ * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
+ */
 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
-#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
-#define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
+#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
+
+#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
+#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
 
-#define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
-#define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
+#define SHMEM_MAX_BYTES  min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
+#define SHMEM_MAX_INDEX  ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
 
+#define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
 
 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
@@ -163,13 +184,13 @@ static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
  */
 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 {
-       return (flags & VM_ACCOUNT)?
-               security_vm_enough_memory(VM_ACCT(size)): 0;
+       return (flags & VM_NORESERVE) ?
+               0 : security_vm_enough_memory_kern(VM_ACCT(size));
 }
 
 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 {
-       if (flags & VM_ACCOUNT)
+       if (!(flags & VM_NORESERVE))
                vm_unacct_memory(VM_ACCT(size));
 }
 
@@ -181,13 +202,13 @@ static inline void shmem_unacct_size(unsigned long flags, loff_t size)
  */
 static inline int shmem_acct_block(unsigned long flags)
 {
-       return (flags & VM_ACCOUNT)?
-               0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
+       return (flags & VM_NORESERVE) ?
+               security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 }
 
 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 {
-       if (!(flags & VM_ACCOUNT))
+       if (flags & VM_NORESERVE)
                vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 }
 
@@ -197,11 +218,11 @@ static const struct file_operations shmem_file_operations;
 static const struct inode_operations shmem_inode_operations;
 static const struct inode_operations shmem_dir_inode_operations;
 static const struct inode_operations shmem_special_inode_operations;
-static struct vm_operations_struct shmem_vm_ops;
+static const struct vm_operations_struct shmem_vm_ops;
 
 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,
+       .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
        .unplug_io_fn   = default_unplug_io_fn,
 };
 
@@ -922,20 +943,30 @@ found:
        error = 1;
        if (!inode)
                goto out;
-       /* Precharge page while we can wait, compensate afterwards */
+       /*
+        * Charge page using GFP_KERNEL while we can wait.
+        * Charged back to the user(not to caller) when swap account is used.
+        * add_to_page_cache() will be called with GFP_NOWAIT.
+        */
        error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
        if (error)
                goto out;
        error = radix_tree_preload(GFP_KERNEL);
-       if (error)
-               goto uncharge;
+       if (error) {
+               mem_cgroup_uncharge_cache_page(page);
+               goto out;
+       }
        error = 1;
 
        spin_lock(&info->lock);
        ptr = shmem_swp_entry(info, idx, NULL);
-       if (ptr && ptr->val == entry.val)
-               error = add_to_page_cache(page, inode->i_mapping,
+       if (ptr && ptr->val == entry.val) {
+               error = add_to_page_cache_locked(page, inode->i_mapping,
                                                idx, GFP_NOWAIT);
+               /* does mem_cgroup_uncharge_cache_page on error */
+       } else  /* we must compensate for our precharge above */
+               mem_cgroup_uncharge_cache_page(page);
+
        if (error == -EEXIST) {
                struct page *filepage = find_get_page(inode->i_mapping, idx);
                error = 1;
@@ -961,8 +992,6 @@ found:
                shmem_swp_unmap(ptr);
        spin_unlock(&info->lock);
        radix_tree_preload_end();
-uncharge:
-       mem_cgroup_uncharge_page(page);
 out:
        unlock_page(page);
        page_cache_release(page);
@@ -1017,8 +1046,9 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
         * sync from ever calling shmem_writepage; but a stacking filesystem
         * may use the ->writepage of its underlying filesystem, in which case
         * tmpfs should write out to swap only in response to memory pressure,
-        * and not for pdflush or sync.  However, in those cases, we do still
-        * want to check if there's a redundant swappage to be discarded.
+        * and not for the writeback threads or sync.  However, in those cases,
+        * we do still want to check if there's a redundant swappage to be
+        * discarded.
         */
        if (wbc->for_reclaim)
                swap = get_swap_page();
@@ -1053,8 +1083,7 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
                swap_duplicate(swap);
                BUG_ON(page_mapped(page));
                page_cache_release(page);       /* pagecache ref */
-               set_page_dirty(page);
-               unlock_page(page);
+               swap_writepage(page, wbc);
                if (inode) {
                        mutex_lock(&shmem_swaplist_mutex);
                        /* move instead of add in case we're racing */
@@ -1068,7 +1097,11 @@ static int shmem_writepage(struct page *page, struct writeback_control *wbc)
        shmem_swp_unmap(entry);
 unlock:
        spin_unlock(&info->lock);
-       swap_free(swap);
+       /*
+        * add_to_swap_cache() doesn't return -EEXIST, so we can safely
+        * clear SWAP_HAS_CACHE flag.
+        */
+       swapcache_free(swap, NULL);
 redirty:
        set_page_dirty(page);
        if (wbc->for_reclaim)
@@ -1079,105 +1112,47 @@ redirty:
 
 #ifdef CONFIG_NUMA
 #ifdef CONFIG_TMPFS
-static int shmem_parse_mpol(char *value, unsigned short *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 (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
-                       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 online nodes with memory if no nodelist
-                */
-               if (!nodelist)
-                       *policy_nodes = node_states[N_HIGH_MEMORY];
-               err = 0;
-       }
-out:
-       /* Restore string for error message */
-       if (nodelist)
-               *--nodelist = ':';
-       return err;
-}
-
-static void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
-                           const nodemask_t policy_nodes)
+static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 {
-       char *policy_string;
+       char buffer[64];
 
-       switch (policy) {
-       case MPOL_PREFERRED:
-               policy_string = "prefer";
-               break;
-       case MPOL_BIND:
-               policy_string = "bind";
-               break;
-       case MPOL_INTERLEAVE:
-               policy_string = "interleave";
-               break;
-       default:
-               /* MPOL_DEFAULT */
-               return;
-       }
+       if (!mpol || mpol->mode == MPOL_DEFAULT)
+               return;         /* show nothing */
 
-       seq_printf(seq, ",mpol=%s", policy_string);
+       mpol_to_str(buffer, sizeof(buffer), mpol, 1);
 
-       if (policy != MPOL_INTERLEAVE ||
-           !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) {
-               char buffer[64];
-               int len;
+       seq_printf(seq, ",mpol=%s", buffer);
+}
 
-               len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes);
-               if (len < sizeof(buffer))
-                       seq_printf(seq, ":%s", buffer);
-               else
-                       seq_printf(seq, ":?");
+static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
+{
+       struct mempolicy *mpol = NULL;
+       if (sbinfo->mpol) {
+               spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
+               mpol = sbinfo->mpol;
+               mpol_get(mpol);
+               spin_unlock(&sbinfo->stat_lock);
        }
+       return mpol;
 }
 #endif /* CONFIG_TMPFS */
 
 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
                        struct shmem_inode_info *info, unsigned long idx)
 {
+       struct mempolicy mpol, *spol;
        struct vm_area_struct pvma;
        struct page *page;
 
+       spol = mpol_cond_copy(&mpol,
+                               mpol_shared_policy_lookup(&info->policy, idx));
+
        /* Create a pseudo vma that just contains the policy */
        pvma.vm_start = 0;
        pvma.vm_pgoff = idx;
        pvma.vm_ops = NULL;
-       pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
+       pvma.vm_policy = spol;
        page = swapin_readahead(entry, gfp, &pvma, 0);
-       mpol_free(pvma.vm_policy);
        return page;
 }
 
@@ -1185,27 +1160,21 @@ static struct page *shmem_alloc_page(gfp_t gfp,
                        struct shmem_inode_info *info, unsigned long idx)
 {
        struct vm_area_struct pvma;
-       struct page *page;
 
        /* Create a pseudo vma that just contains the policy */
        pvma.vm_start = 0;
        pvma.vm_pgoff = idx;
        pvma.vm_ops = NULL;
        pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
-       page = alloc_page_vma(gfp, &pvma, 0);
-       mpol_free(pvma.vm_policy);
-       return page;
+
+       /*
+        * alloc_page_vma() will drop the shared policy reference
+        */
+       return alloc_page_vma(gfp, &pvma, 0);
 }
 #else /* !CONFIG_NUMA */
 #ifdef CONFIG_TMPFS
-static inline int shmem_parse_mpol(char *value, unsigned short *policy,
-                                               nodemask_t *policy_nodes)
-{
-       return 1;
-}
-
-static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
-                           const nodemask_t policy_nodes)
+static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
 {
 }
 #endif /* CONFIG_TMPFS */
@@ -1223,6 +1192,13 @@ static inline struct page *shmem_alloc_page(gfp_t gfp,
 }
 #endif /* CONFIG_NUMA */
 
+#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
+static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
+{
+       return NULL;
+}
+#endif
+
 /*
  * shmem_getpage - either get the page from swap or allocate a new one
  *
@@ -1318,7 +1294,7 @@ repeat:
                }
 
                /* We have to do this with page locked to prevent races */
-               if (TestSetPageLocked(swappage)) {
+               if (!trylock_page(swappage)) {
                        shmem_swp_unmap(entry);
                        spin_unlock(&info->lock);
                        wait_on_page_locked(swappage);
@@ -1354,8 +1330,8 @@ repeat:
                        SetPageUptodate(filepage);
                        set_page_dirty(filepage);
                        swap_free(swap);
-               } else if (!(error = add_to_page_cache(
-                               swappage, mapping, idx, GFP_NOWAIT))) {
+               } else if (!(error = add_to_page_cache_locked(swappage, mapping,
+                                       idx, GFP_NOWAIT))) {
                        info->flags |= SHMEM_PAGEIN;
                        shmem_swp_set(info, entry, 0);
                        shmem_swp_unmap(entry);
@@ -1367,17 +1343,22 @@ repeat:
                } else {
                        shmem_swp_unmap(entry);
                        spin_unlock(&info->lock);
-                       unlock_page(swappage);
                        if (error == -ENOMEM) {
-                               /* allow reclaim from this memory cgroup */
-                               error = mem_cgroup_cache_charge(swappage,
-                                       current->mm, gfp & ~__GFP_HIGHMEM);
+                               /*
+                                * reclaim from proper memory cgroup and
+                                * call memcg's OOM if needed.
+                                */
+                               error = mem_cgroup_shmem_charge_fallback(
+                                                               swappage,
+                                                               current->mm,
+                                                               gfp);
                                if (error) {
+                                       unlock_page(swappage);
                                        page_cache_release(swappage);
                                        goto failed;
                                }
-                               mem_cgroup_uncharge_page(swappage);
                        }
+                       unlock_page(swappage);
                        page_cache_release(swappage);
                        goto repeat;
                }
@@ -1385,7 +1366,7 @@ repeat:
                shmem_swp_unmap(entry);
                filepage = find_get_page(mapping, idx);
                if (filepage &&
-                   (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
+                   (!PageUptodate(filepage) || !trylock_page(filepage))) {
                        spin_unlock(&info->lock);
                        wait_on_page_locked(filepage);
                        page_cache_release(filepage);
@@ -1415,6 +1396,8 @@ repeat:
                }
 
                if (!filepage) {
+                       int ret;
+
                        spin_unlock(&info->lock);
                        filepage = shmem_alloc_page(gfp, info, idx);
                        if (!filepage) {
@@ -1423,10 +1406,11 @@ repeat:
                                error = -ENOMEM;
                                goto failed;
                        }
+                       SetPageSwapBacked(filepage);
 
                        /* Precharge page while we can wait, compensate after */
                        error = mem_cgroup_cache_charge(filepage, current->mm,
-                                                       gfp & ~__GFP_HIGHMEM);
+                                       GFP_KERNEL);
                        if (error) {
                                page_cache_release(filepage);
                                shmem_unacct_blocks(info->flags, 1);
@@ -1443,10 +1427,18 @@ repeat:
                                swap = *entry;
                                shmem_swp_unmap(entry);
                        }
-                       if (error || swap.val || 0 != add_to_page_cache_lru(
-                                       filepage, mapping, idx, GFP_NOWAIT)) {
+                       ret = error || swap.val;
+                       if (ret)
+                               mem_cgroup_uncharge_cache_page(filepage);
+                       else
+                               ret = add_to_page_cache_lru(filepage, mapping,
+                                               idx, GFP_NOWAIT);
+                       /*
+                        * At add_to_page_cache_lru() failure, uncharge will
+                        * be done automatically.
+                        */
+                       if (ret) {
                                spin_unlock(&info->lock);
-                               mem_cgroup_uncharge_page(filepage);
                                page_cache_release(filepage);
                                shmem_unacct_blocks(info->flags, 1);
                                shmem_free_blocks(inode, 1);
@@ -1455,7 +1447,6 @@ repeat:
                                        goto failed;
                                goto repeat;
                        }
-                       mem_cgroup_uncharge_page(filepage);
                        info->flags |= SHMEM_PAGEIN;
                }
 
@@ -1492,7 +1483,6 @@ static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
        if (error)
                return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
 
-       mark_page_accessed(vmf->page);
        return ret | VM_FAULT_LOCKED;
 }
 
@@ -1525,12 +1515,16 @@ int shmem_lock(struct file *file, int lock, struct user_struct *user)
                if (!user_shm_lock(inode->i_size, user))
                        goto out_nomem;
                info->flags |= VM_LOCKED;
+               mapping_set_unevictable(file->f_mapping);
        }
        if (!lock && (info->flags & VM_LOCKED) && user) {
                user_shm_unlock(inode->i_size, user);
                info->flags &= ~VM_LOCKED;
+               mapping_clear_unevictable(file->f_mapping);
+               scan_mapping_unevictable_pages(file->f_mapping);
        }
        retval = 0;
+
 out_nomem:
        spin_unlock(&info->lock);
        return retval;
@@ -1544,8 +1538,8 @@ static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
        return 0;
 }
 
-static struct inode *
-shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
+static struct inode *shmem_get_inode(struct super_block *sb, int mode,
+                                       dev_t dev, unsigned long flags)
 {
        struct inode *inode;
        struct shmem_inode_info *info;
@@ -1557,17 +1551,18 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
        inode = new_inode(sb);
        if (inode) {
                inode->i_mode = mode;
-               inode->i_uid = current->fsuid;
-               inode->i_gid = current->fsgid;
+               inode->i_uid = current_fsuid();
+               inode->i_gid = current_fsgid();
                inode->i_blocks = 0;
-               inode->i_mapping->a_ops = &shmem_aops;
                inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
                inode->i_generation = get_seconds();
                info = SHMEM_I(inode);
                memset(info, 0, (char *)inode - (char *)info);
                spin_lock_init(&info->lock);
+               info->flags = flags & VM_NORESERVE;
                INIT_LIST_HEAD(&info->swaplist);
+               cache_no_acl(inode);
 
                switch (mode & S_IFMT) {
                default:
@@ -1575,10 +1570,11 @@ shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
                        init_special_inode(inode, mode, dev);
                        break;
                case S_IFREG:
+                       inode->i_mapping->a_ops = &shmem_aops;
                        inode->i_op = &shmem_inode_operations;
                        inode->i_fop = &shmem_file_operations;
-                       mpol_shared_policy_init(&info->policy, sbinfo->policy,
-                                                       &sbinfo->policy_nodes);
+                       mpol_shared_policy_init(&info->policy,
+                                                shmem_get_sbmpol(sbinfo));
                        break;
                case S_IFDIR:
                        inc_nlink(inode);
@@ -1592,8 +1588,7 @@ 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_DEFAULT,
-                                               NULL);
+                       mpol_shared_policy_init(&info->policy, NULL);
                        break;
                }
        } else
@@ -1639,8 +1634,8 @@ shmem_write_end(struct file *file, struct address_space *mapping,
        if (pos + copied > inode->i_size)
                i_size_write(inode, pos + copied);
 
-       unlock_page(page);
        set_page_dirty(page);
+       unlock_page(page);
        page_cache_release(page);
 
        return copied;
@@ -1748,26 +1743,38 @@ static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_
        file_accessed(filp);
 }
 
-static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
+static ssize_t shmem_file_aio_read(struct kiocb *iocb,
+               const struct iovec *iov, unsigned long nr_segs, loff_t pos)
 {
-       read_descriptor_t desc;
+       struct file *filp = iocb->ki_filp;
+       ssize_t retval;
+       unsigned long seg;
+       size_t count;
+       loff_t *ppos = &iocb->ki_pos;
 
-       if ((ssize_t) count < 0)
-               return -EINVAL;
-       if (!access_ok(VERIFY_WRITE, buf, count))
-               return -EFAULT;
-       if (!count)
-               return 0;
+       retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
+       if (retval)
+               return retval;
 
-       desc.written = 0;
-       desc.count = count;
-       desc.arg.buf = buf;
-       desc.error = 0;
+       for (seg = 0; seg < nr_segs; seg++) {
+               read_descriptor_t desc;
 
-       do_shmem_file_read(filp, ppos, &desc, file_read_actor);
-       if (desc.written)
-               return desc.written;
-       return desc.error;
+               desc.written = 0;
+               desc.arg.buf = iov[seg].iov_base;
+               desc.count = iov[seg].iov_len;
+               if (desc.count == 0)
+                       continue;
+               desc.error = 0;
+               do_shmem_file_read(filp, ppos, &desc, file_read_actor);
+               retval += desc.written;
+               if (desc.error) {
+                       retval = retval ?: desc.error;
+                       break;
+               }
+               if (desc.count > 0)
+                       break;
+       }
+       return retval;
 }
 
 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
@@ -1797,9 +1804,10 @@ static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
 static int
 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
 {
-       struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
+       struct inode *inode;
        int error = -ENOSPC;
 
+       inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE);
        if (inode) {
                error = security_inode_init_security(inode, dir, NULL, NULL,
                                                     NULL);
@@ -1938,7 +1946,7 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
        if (len > PAGE_CACHE_SIZE)
                return -ENAMETOOLONG;
 
-       inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
+       inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
        if (!inode)
                return -ENOSPC;
 
@@ -1964,12 +1972,13 @@ static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *s
                        iput(inode);
                        return error;
                }
-               unlock_page(page);
+               inode->i_mapping->a_ops = &shmem_aops;
                inode->i_op = &shmem_symlink_inode_operations;
                kaddr = kmap_atomic(page, KM_USER0);
                memcpy(kaddr, symname, len);
                kunmap_atomic(kaddr, KM_USER0);
                set_page_dirty(page);
+               unlock_page(page);
                page_cache_release(page);
        }
        if (dir->i_mode & S_ISGID)
@@ -2208,8 +2217,7 @@ static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
                        if (*rest)
                                goto bad_val;
                } else if (!strcmp(this_char,"mpol")) {
-                       if (shmem_parse_mpol(value, &sbinfo->policy,
-                                            &sbinfo->policy_nodes))
+                       if (mpol_parse_str(value, &sbinfo->mpol, 1))
                                goto bad_val;
                } else {
                        printk(KERN_ERR "tmpfs: Bad mount option %s\n",
@@ -2260,8 +2268,9 @@ static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
        sbinfo->free_blocks = config.max_blocks - blocks;
        sbinfo->max_inodes  = config.max_inodes;
        sbinfo->free_inodes = config.max_inodes - inodes;
-       sbinfo->policy      = config.policy;
-       sbinfo->policy_nodes = config.policy_nodes;
+
+       mpol_put(sbinfo->mpol);
+       sbinfo->mpol        = config.mpol;      /* transfers initial ref */
 out:
        spin_unlock(&sbinfo->stat_lock);
        return error;
@@ -2282,7 +2291,7 @@ static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
                seq_printf(seq, ",uid=%u", sbinfo->uid);
        if (sbinfo->gid != 0)
                seq_printf(seq, ",gid=%u", sbinfo->gid);
-       shmem_show_mpol(seq, sbinfo->policy, sbinfo->policy_nodes);
+       shmem_show_mpol(seq, sbinfo->mpol);
        return 0;
 }
 #endif /* CONFIG_TMPFS */
@@ -2293,8 +2302,7 @@ static void shmem_put_super(struct super_block *sb)
        sb->s_fs_info = NULL;
 }
 
-static int shmem_fill_super(struct super_block *sb,
-                           void *data, int silent)
+int shmem_fill_super(struct super_block *sb, void *data, int silent)
 {
        struct inode *inode;
        struct dentry *root;
@@ -2302,18 +2310,14 @@ static int shmem_fill_super(struct super_block *sb,
        int err = -ENOMEM;
 
        /* Round up to L1_CACHE_BYTES to resist false sharing */
-       sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
+       sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
                                L1_CACHE_BYTES), GFP_KERNEL);
        if (!sbinfo)
                return -ENOMEM;
 
-       sbinfo->max_blocks = 0;
-       sbinfo->max_inodes = 0;
        sbinfo->mode = S_IRWXUGO | S_ISVTX;
-       sbinfo->uid = current->fsuid;
-       sbinfo->gid = current->fsgid;
-       sbinfo->policy = MPOL_DEFAULT;
-       sbinfo->policy_nodes = node_states[N_HIGH_MEMORY];
+       sbinfo->uid = current_fsuid();
+       sbinfo->gid = current_fsgid();
        sb->s_fs_info = sbinfo;
 
 #ifdef CONFIG_TMPFS
@@ -2350,7 +2354,7 @@ static int shmem_fill_super(struct super_block *sb,
        sb->s_flags |= MS_POSIXACL;
 #endif
 
-       inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0);
+       inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
        if (!inode)
                goto failed;
        inode->i_uid = sbinfo->uid;
@@ -2385,19 +2389,14 @@ static void shmem_destroy_inode(struct inode *inode)
                /* only struct inode is valid if it's an inline symlink */
                mpol_free_shared_policy(&SHMEM_I(inode)->policy);
        }
-       shmem_acl_destroy_inode(inode);
        kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
 }
 
-static void init_once(struct kmem_cache *cachep, void *foo)
+static void init_once(void *foo)
 {
        struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
 
        inode_init_once(&p->vfs_inode);
-#ifdef CONFIG_TMPFS_POSIX_ACL
-       p->i_acl = NULL;
-       p->i_default_acl = NULL;
-#endif
 }
 
 static int init_inodecache(void)
@@ -2422,14 +2421,16 @@ static const struct address_space_operations shmem_aops = {
        .write_end      = shmem_write_end,
 #endif
        .migratepage    = migrate_page,
+       .error_remove_page = generic_error_remove_page,
 };
 
 static const struct file_operations shmem_file_operations = {
        .mmap           = shmem_mmap,
 #ifdef CONFIG_TMPFS
        .llseek         = generic_file_llseek,
-       .read           = shmem_file_read,
+       .read           = do_sync_read,
        .write          = do_sync_write,
+       .aio_read       = shmem_file_aio_read,
        .aio_write      = generic_file_aio_write,
        .fsync          = simple_sync_file,
        .splice_read    = generic_file_splice_read,
@@ -2446,7 +2447,7 @@ static const struct inode_operations shmem_inode_operations = {
        .getxattr       = generic_getxattr,
        .listxattr      = generic_listxattr,
        .removexattr    = generic_removexattr,
-       .permission     = shmem_permission,
+       .check_acl      = shmem_check_acl,
 #endif
 
 };
@@ -2469,7 +2470,7 @@ static const struct inode_operations shmem_dir_inode_operations = {
        .getxattr       = generic_getxattr,
        .listxattr      = generic_listxattr,
        .removexattr    = generic_removexattr,
-       .permission     = shmem_permission,
+       .check_acl      = shmem_check_acl,
 #endif
 };
 
@@ -2480,7 +2481,7 @@ static const struct inode_operations shmem_special_inode_operations = {
        .getxattr       = generic_getxattr,
        .listxattr      = generic_listxattr,
        .removexattr    = generic_removexattr,
-       .permission     = shmem_permission,
+       .check_acl      = shmem_check_acl,
 #endif
 };
 
@@ -2497,7 +2498,7 @@ static const struct super_operations shmem_ops = {
        .put_super      = shmem_put_super,
 };
 
-static struct vm_operations_struct shmem_vm_ops = {
+static const struct vm_operations_struct shmem_vm_ops = {
        .fault          = shmem_fault,
 #ifdef CONFIG_NUMA
        .set_policy     = shmem_set_policy,
@@ -2518,9 +2519,8 @@ static struct file_system_type tmpfs_fs_type = {
        .get_sb         = shmem_get_sb,
        .kill_sb        = kill_litter_super,
 };
-static struct vfsmount *shm_mnt;
 
-static int __init init_tmpfs(void)
+int __init init_tmpfs(void)
 {
        int error;
 
@@ -2557,15 +2557,64 @@ out4:
        shm_mnt = ERR_PTR(error);
        return error;
 }
-module_init(init_tmpfs)
+
+#else /* !CONFIG_SHMEM */
+
+/*
+ * tiny-shmem: simple shmemfs and tmpfs using ramfs code
+ *
+ * This is intended for small system where the benefits of the full
+ * shmem code (swap-backed and resource-limited) are outweighed by
+ * their complexity. On systems without swap this code should be
+ * effectively equivalent, but much lighter weight.
+ */
+
+#include <linux/ramfs.h>
+
+static struct file_system_type tmpfs_fs_type = {
+       .name           = "tmpfs",
+       .get_sb         = ramfs_get_sb,
+       .kill_sb        = kill_litter_super,
+};
+
+int __init init_tmpfs(void)
+{
+       BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
+
+       shm_mnt = kern_mount(&tmpfs_fs_type);
+       BUG_ON(IS_ERR(shm_mnt));
+
+       return 0;
+}
+
+int shmem_unuse(swp_entry_t entry, struct page *page)
+{
+       return 0;
+}
+
+int shmem_lock(struct file *file, int lock, struct user_struct *user)
+{
+       return 0;
+}
+
+#define shmem_vm_ops                           generic_file_vm_ops
+#define shmem_file_operations                  ramfs_file_operations
+#define shmem_get_inode(sb, mode, dev, flags)  ramfs_get_inode(sb, mode, dev)
+#define shmem_acct_size(flags, size)           0
+#define shmem_unacct_size(flags, size)         do {} while (0)
+#define SHMEM_MAX_BYTES                                MAX_LFS_FILESIZE
+
+#endif /* CONFIG_SHMEM */
+
+/* common code */
 
 /**
  * shmem_file_setup - get an unlinked file living in tmpfs
  * @name: name for dentry (to be seen in /proc/<pid>/maps
  * @size: size to be set for the file
- * @flags: vm_flags
+ * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
  */
-struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
+struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
 {
        int error;
        struct file *file;
@@ -2597,16 +2646,22 @@ struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
                goto put_dentry;
 
        error = -ENOSPC;
-       inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
+       inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
        if (!inode)
                goto close_file;
 
-       SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
        d_instantiate(dentry, inode);
        inode->i_size = size;
        inode->i_nlink = 0;     /* It is unlinked */
        init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
-                       &shmem_file_operations);
+                 &shmem_file_operations);
+
+#ifndef CONFIG_MMU
+       error = ramfs_nommu_expand_for_mapping(inode, size);
+       if (error)
+               goto close_file;
+#endif
+       ima_counts_get(file);
        return file;
 
 close_file:
@@ -2617,6 +2672,7 @@ put_memory:
        shmem_unacct_size(flags, size);
        return ERR_PTR(error);
 }
+EXPORT_SYMBOL_GPL(shmem_file_setup);
 
 /**
  * shmem_zero_setup - setup a shared anonymous mapping