NFSD: don't report compiled-out versions as present
[safe/jmp/linux-2.6] / fs / hugetlbfs / inode.c
index dbd01d2..a0bbd3d 100644 (file)
 #include <linux/pagevec.h>
 #include <linux/parser.h>
 #include <linux/mman.h>
-#include <linux/quotaops.h>
 #include <linux/slab.h>
 #include <linux/dnotify.h>
 #include <linux/statfs.h>
 #include <linux/security.h>
+#include <linux/magic.h>
 
 #include <asm/uaccess.h>
 
-/* some random number */
-#define HUGETLBFS_MAGIC        0x958458f6
-
 static const struct super_operations hugetlbfs_ops;
 static const struct address_space_operations hugetlbfs_aops;
 const struct file_operations hugetlbfs_file_operations;
@@ -44,6 +41,7 @@ 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 = {
+       .name           = "hugetlbfs",
        .ra_pages       = 0,    /* No readahead */
        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK,
 };
@@ -57,7 +55,7 @@ enum {
        Opt_err,
 };
 
-static match_table_t tokens = {
+static const match_table_t tokens = {
        {Opt_size,      "size=%s"},
        {Opt_nr_inodes, "nr_inodes=%s"},
        {Opt_mode,      "mode=%o"},
@@ -108,7 +106,8 @@ static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
 
        if (hugetlb_reserve_pages(inode,
                                vma->vm_pgoff >> huge_page_order(h),
-                               len >> huge_page_shift(h), vma))
+                               len >> huge_page_shift(h), vma,
+                               vma->vm_flags))
                goto out;
 
        ret = 0;
@@ -252,6 +251,7 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
        for (;;) {
                struct page *page;
                unsigned long nr, ret;
+               int ra;
 
                /* nr is the maximum number of bytes to copy from this page */
                nr = huge_page_size(h);
@@ -274,16 +274,19 @@ static ssize_t hugetlbfs_read(struct file *filp, char __user *buf,
                         */
                        ret = len < nr ? len : nr;
                        if (clear_user(buf, ret))
-                               ret = -EFAULT;
+                               ra = -EFAULT;
+                       else
+                               ra = 0;
                } else {
                        /*
                         * We have the page, copy it to user space buffer.
                         */
-                       ret = hugetlbfs_read_actor(page, offset, buf, len, nr);
+                       ra = hugetlbfs_read_actor(page, offset, buf, len, nr);
+                       ret = ra;
                }
-               if (ret < 0) {
+               if (ra < 0) {
                        if (retval == 0)
-                               retval = ret;
+                               retval = ra;
                        if (page)
                                page_cache_release(page);
                        goto out;
@@ -308,16 +311,6 @@ out:
        return retval;
 }
 
-/*
- * Read a page. Again trivial. If it didn't already exist
- * in the page cache, it is zero-filled.
- */
-static int hugetlbfs_readpage(struct file *file, struct page * page)
-{
-       unlock_page(page);
-       return -EINVAL;
-}
-
 static int hugetlbfs_write_begin(struct file *file,
                        struct address_space *mapping,
                        loff_t pos, unsigned len, unsigned flags,
@@ -386,36 +379,11 @@ static void hugetlbfs_delete_inode(struct inode *inode)
 
 static void hugetlbfs_forget_inode(struct inode *inode) __releases(inode_lock)
 {
-       struct super_block *sb = inode->i_sb;
-
-       if (!hlist_unhashed(&inode->i_hash)) {
-               if (!(inode->i_state & (I_DIRTY|I_SYNC)))
-                       list_move(&inode->i_list, &inode_unused);
-               inodes_stat.nr_unused++;
-               if (!sb || (sb->s_flags & MS_ACTIVE)) {
-                       spin_unlock(&inode_lock);
-                       return;
-               }
-               inode->i_state |= I_WILL_FREE;
-               spin_unlock(&inode_lock);
-               /*
-                * write_inode_now is a noop as we set BDI_CAP_NO_WRITEBACK
-                * in our backing_dev_info.
-                */
-               write_inode_now(inode, 1);
-               spin_lock(&inode_lock);
-               inode->i_state &= ~I_WILL_FREE;
-               inodes_stat.nr_unused--;
-               hlist_del_init(&inode->i_hash);
+       if (generic_detach_inode(inode)) {
+               truncate_hugepages(inode, 0);
+               clear_inode(inode);
+               destroy_inode(inode);
        }
-       list_del_init(&inode->i_list);
-       list_del_init(&inode->i_sb_list);
-       inode->i_state |= I_FREEING;
-       inodes_stat.nr_inodes--;
-       spin_unlock(&inode_lock);
-       truncate_hugepages(inode, 0);
-       clear_inode(inode);
-       destroy_inode(inode);
 }
 
 static void hugetlbfs_drop_inode(struct inode *inode)
@@ -506,12 +474,18 @@ 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_blocks = 0;
                inode->i_mapping->a_ops = &hugetlbfs_aops;
                inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
                INIT_LIST_HEAD(&inode->i_mapping->private_list);
                info = HUGETLBFS_I(inode);
+               /*
+                * The policy is initialized here even if we are creating a
+                * private inode because initialization simply creates an
+                * an empty rb tree and calls spin_lock_init(), later when we
+                * call mpol_free_shared_policy() it will just return because
+                * the rb tree will still be empty.
+                */
                mpol_shared_policy_init(&info->policy, NULL);
                switch (mode & S_IFMT) {
                default:
@@ -551,9 +525,9 @@ static int hugetlbfs_mknod(struct inode *dir,
                if (S_ISDIR(mode))
                        mode |= S_ISGID;
        } else {
-               gid = current->fsgid;
+               gid = current_fsgid();
        }
-       inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid, gid, mode, dev);
+       inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(), gid, mode, dev);
        if (inode) {
                dir->i_ctime = dir->i_mtime = CURRENT_TIME;
                d_instantiate(dentry, inode);
@@ -586,9 +560,9 @@ static int hugetlbfs_symlink(struct inode *dir,
        if (dir->i_mode & S_ISGID)
                gid = dir->i_gid;
        else
-               gid = current->fsgid;
+               gid = current_fsgid();
 
-       inode = hugetlbfs_get_inode(dir->i_sb, current->fsuid,
+       inode = hugetlbfs_get_inode(dir->i_sb, current_fsuid(),
                                        gid, S_IFLNK|S_IRWXUGO, 0);
        if (inode) {
                int l = strlen(symname)+1;
@@ -698,14 +672,13 @@ static void hugetlbfs_destroy_inode(struct inode *inode)
 }
 
 static const struct address_space_operations hugetlbfs_aops = {
-       .readpage       = hugetlbfs_readpage,
        .write_begin    = hugetlbfs_write_begin,
        .write_end      = hugetlbfs_write_end,
        .set_page_dirty = hugetlbfs_set_page_dirty,
 };
 
 
-static void init_once(struct kmem_cache *cachep, void *foo)
+static void init_once(void *foo)
 {
        struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
 
@@ -838,7 +811,7 @@ hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
 bad_val:
        printk(KERN_ERR "hugetlbfs: Bad value '%s' for mount option '%s'\n",
               args[0].from, p);
-       return 1;
+       return -EINVAL;
 }
 
 static int
@@ -854,8 +827,8 @@ hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
 
        config.nr_blocks = -1; /* No limit on size by default */
        config.nr_inodes = -1; /* No limit on number of inodes by default */
-       config.uid = current->fsuid;
-       config.gid = current->fsgid;
+       config.uid = current_fsuid();
+       config.gid = current_fsgid();
        config.mode = 0755;
        config.hstate = &default_hstate;
        ret = hugetlbfs_parse_options(data, &config);
@@ -939,54 +912,61 @@ static struct vfsmount *hugetlbfs_vfsmount;
 
 static int can_do_hugetlb_shm(void)
 {
-       return likely(capable(CAP_IPC_LOCK) ||
-                       in_group_p(sysctl_hugetlb_shm_group) ||
-                       can_do_mlock());
+       return capable(CAP_IPC_LOCK) || in_group_p(sysctl_hugetlb_shm_group);
 }
 
-struct file *hugetlb_file_setup(const char *name, size_t size)
+struct file *hugetlb_file_setup(const char *name, size_t size, int acctflag,
+                               struct user_struct **user, int creat_flags)
 {
        int error = -ENOMEM;
        struct file *file;
        struct inode *inode;
-       struct dentry *dentry, *root;
+       struct path path;
+       struct dentry *root;
        struct qstr quick_string;
 
+       *user = NULL;
        if (!hugetlbfs_vfsmount)
                return ERR_PTR(-ENOENT);
 
-       if (!can_do_hugetlb_shm())
-               return ERR_PTR(-EPERM);
-
-       if (!user_shm_lock(size, current->user))
-               return ERR_PTR(-ENOMEM);
+       if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
+               *user = current_user();
+               if (user_shm_lock(size, *user)) {
+                       WARN_ONCE(1,
+                         "Using mlock ulimits for SHM_HUGETLB deprecated\n");
+               } else {
+                       *user = NULL;
+                       return ERR_PTR(-EPERM);
+               }
+       }
 
        root = hugetlbfs_vfsmount->mnt_root;
        quick_string.name = name;
        quick_string.len = strlen(quick_string.name);
        quick_string.hash = 0;
-       dentry = d_alloc(root, &quick_string);
-       if (!dentry)
+       path.dentry = d_alloc(root, &quick_string);
+       if (!path.dentry)
                goto out_shm_unlock;
 
+       path.mnt = mntget(hugetlbfs_vfsmount);
        error = -ENOSPC;
-       inode = hugetlbfs_get_inode(root->d_sb, current->fsuid,
-                               current->fsgid, S_IFREG | S_IRWXUGO, 0);
+       inode = hugetlbfs_get_inode(root->d_sb, current_fsuid(),
+                               current_fsgid(), S_IFREG | S_IRWXUGO, 0);
        if (!inode)
                goto out_dentry;
 
        error = -ENOMEM;
        if (hugetlb_reserve_pages(inode, 0,
-                       size >> huge_page_shift(hstate_inode(inode)), NULL))
+                       size >> huge_page_shift(hstate_inode(inode)), NULL,
+                       acctflag))
                goto out_inode;
 
-       d_instantiate(dentry, inode);
+       d_instantiate(path.dentry, inode);
        inode->i_size = size;
        inode->i_nlink = 0;
 
        error = -ENFILE;
-       file = alloc_file(hugetlbfs_vfsmount, dentry,
-                       FMODE_WRITE | FMODE_READ,
+       file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
                        &hugetlbfs_file_operations);
        if (!file)
                goto out_dentry; /* inode is already attached */
@@ -996,9 +976,12 @@ struct file *hugetlb_file_setup(const char *name, size_t size)
 out_inode:
        iput(inode);
 out_dentry:
-       dput(dentry);
+       path_put(&path);
 out_shm_unlock:
-       user_shm_unlock(size, current->user);
+       if (*user) {
+               user_shm_unlock(size, *user);
+               *user = NULL;
+       }
        return ERR_PTR(error);
 }