Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / cramfs / inode.c
diff --git a/kernel/fs/cramfs/inode.c b/kernel/fs/cramfs/inode.c
new file mode 100644 (file)
index 0000000..355c522
--- /dev/null
@@ -0,0 +1,611 @@
+/*
+ * Compressed rom filesystem for Linux.
+ *
+ * Copyright (C) 1999 Linus Torvalds.
+ *
+ * This file is released under the GPL.
+ */
+
+/*
+ * These are the VFS interfaces to the compressed rom filesystem.
+ * The actual compression is based on zlib, see the other files.
+ */
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/blkdev.h>
+#include <linux/slab.h>
+#include <linux/vfs.h>
+#include <linux/mutex.h>
+#include <uapi/linux/cramfs_fs.h>
+#include <linux/uaccess.h>
+
+#include "internal.h"
+
+/*
+ * cramfs super-block data in memory
+ */
+struct cramfs_sb_info {
+       unsigned long magic;
+       unsigned long size;
+       unsigned long blocks;
+       unsigned long files;
+       unsigned long flags;
+};
+
+static inline struct cramfs_sb_info *CRAMFS_SB(struct super_block *sb)
+{
+       return sb->s_fs_info;
+}
+
+static const struct super_operations cramfs_ops;
+static const struct inode_operations cramfs_dir_inode_operations;
+static const struct file_operations cramfs_directory_operations;
+static const struct address_space_operations cramfs_aops;
+
+static DEFINE_MUTEX(read_mutex);
+
+
+/* These macros may change in future, to provide better st_ino semantics. */
+#define OFFSET(x)      ((x)->i_ino)
+
+static unsigned long cramino(const struct cramfs_inode *cino, unsigned int offset)
+{
+       if (!cino->offset)
+               return offset + 1;
+       if (!cino->size)
+               return offset + 1;
+
+       /*
+        * The file mode test fixes buggy mkcramfs implementations where
+        * cramfs_inode->offset is set to a non zero value for entries
+        * which did not contain data, like devices node and fifos.
+        */
+       switch (cino->mode & S_IFMT) {
+       case S_IFREG:
+       case S_IFDIR:
+       case S_IFLNK:
+               return cino->offset << 2;
+       default:
+               break;
+       }
+       return offset + 1;
+}
+
+static struct inode *get_cramfs_inode(struct super_block *sb,
+       const struct cramfs_inode *cramfs_inode, unsigned int offset)
+{
+       struct inode *inode;
+       static struct timespec zerotime;
+
+       inode = iget_locked(sb, cramino(cramfs_inode, offset));
+       if (!inode)
+               return ERR_PTR(-ENOMEM);
+       if (!(inode->i_state & I_NEW))
+               return inode;
+
+       switch (cramfs_inode->mode & S_IFMT) {
+       case S_IFREG:
+               inode->i_fop = &generic_ro_fops;
+               inode->i_data.a_ops = &cramfs_aops;
+               break;
+       case S_IFDIR:
+               inode->i_op = &cramfs_dir_inode_operations;
+               inode->i_fop = &cramfs_directory_operations;
+               break;
+       case S_IFLNK:
+               inode->i_op = &page_symlink_inode_operations;
+               inode->i_data.a_ops = &cramfs_aops;
+               break;
+       default:
+               init_special_inode(inode, cramfs_inode->mode,
+                               old_decode_dev(cramfs_inode->size));
+       }
+
+       inode->i_mode = cramfs_inode->mode;
+       i_uid_write(inode, cramfs_inode->uid);
+       i_gid_write(inode, cramfs_inode->gid);
+
+       /* if the lower 2 bits are zero, the inode contains data */
+       if (!(inode->i_ino & 3)) {
+               inode->i_size = cramfs_inode->size;
+               inode->i_blocks = (cramfs_inode->size - 1) / 512 + 1;
+       }
+
+       /* Struct copy intentional */
+       inode->i_mtime = inode->i_atime = inode->i_ctime = zerotime;
+       /* inode->i_nlink is left 1 - arguably wrong for directories,
+          but it's the best we can do without reading the directory
+          contents.  1 yields the right result in GNU find, even
+          without -noleaf option. */
+
+       unlock_new_inode(inode);
+
+       return inode;
+}
+
+/*
+ * We have our own block cache: don't fill up the buffer cache
+ * with the rom-image, because the way the filesystem is set
+ * up the accesses should be fairly regular and cached in the
+ * page cache and dentry tree anyway..
+ *
+ * This also acts as a way to guarantee contiguous areas of up to
+ * BLKS_PER_BUF*PAGE_CACHE_SIZE, so that the caller doesn't need to
+ * worry about end-of-buffer issues even when decompressing a full
+ * page cache.
+ */
+#define READ_BUFFERS (2)
+/* NEXT_BUFFER(): Loop over [0..(READ_BUFFERS-1)]. */
+#define NEXT_BUFFER(_ix) ((_ix) ^ 1)
+
+/*
+ * BLKS_PER_BUF_SHIFT should be at least 2 to allow for "compressed"
+ * data that takes up more space than the original and with unlucky
+ * alignment.
+ */
+#define BLKS_PER_BUF_SHIFT     (2)
+#define BLKS_PER_BUF           (1 << BLKS_PER_BUF_SHIFT)
+#define BUFFER_SIZE            (BLKS_PER_BUF*PAGE_CACHE_SIZE)
+
+static unsigned char read_buffers[READ_BUFFERS][BUFFER_SIZE];
+static unsigned buffer_blocknr[READ_BUFFERS];
+static struct super_block *buffer_dev[READ_BUFFERS];
+static int next_buffer;
+
+/*
+ * Returns a pointer to a buffer containing at least LEN bytes of
+ * filesystem starting at byte offset OFFSET into the filesystem.
+ */
+static void *cramfs_read(struct super_block *sb, unsigned int offset, unsigned int len)
+{
+       struct address_space *mapping = sb->s_bdev->bd_inode->i_mapping;
+       struct page *pages[BLKS_PER_BUF];
+       unsigned i, blocknr, buffer;
+       unsigned long devsize;
+       char *data;
+
+       if (!len)
+               return NULL;
+       blocknr = offset >> PAGE_CACHE_SHIFT;
+       offset &= PAGE_CACHE_SIZE - 1;
+
+       /* Check if an existing buffer already has the data.. */
+       for (i = 0; i < READ_BUFFERS; i++) {
+               unsigned int blk_offset;
+
+               if (buffer_dev[i] != sb)
+                       continue;
+               if (blocknr < buffer_blocknr[i])
+                       continue;
+               blk_offset = (blocknr - buffer_blocknr[i]) << PAGE_CACHE_SHIFT;
+               blk_offset += offset;
+               if (blk_offset + len > BUFFER_SIZE)
+                       continue;
+               return read_buffers[i] + blk_offset;
+       }
+
+       devsize = mapping->host->i_size >> PAGE_CACHE_SHIFT;
+
+       /* Ok, read in BLKS_PER_BUF pages completely first. */
+       for (i = 0; i < BLKS_PER_BUF; i++) {
+               struct page *page = NULL;
+
+               if (blocknr + i < devsize) {
+                       page = read_mapping_page(mapping, blocknr + i, NULL);
+                       /* synchronous error? */
+                       if (IS_ERR(page))
+                               page = NULL;
+               }
+               pages[i] = page;
+       }
+
+       for (i = 0; i < BLKS_PER_BUF; i++) {
+               struct page *page = pages[i];
+
+               if (page) {
+                       wait_on_page_locked(page);
+                       if (!PageUptodate(page)) {
+                               /* asynchronous error */
+                               page_cache_release(page);
+                               pages[i] = NULL;
+                       }
+               }
+       }
+
+       buffer = next_buffer;
+       next_buffer = NEXT_BUFFER(buffer);
+       buffer_blocknr[buffer] = blocknr;
+       buffer_dev[buffer] = sb;
+
+       data = read_buffers[buffer];
+       for (i = 0; i < BLKS_PER_BUF; i++) {
+               struct page *page = pages[i];
+
+               if (page) {
+                       memcpy(data, kmap(page), PAGE_CACHE_SIZE);
+                       kunmap(page);
+                       page_cache_release(page);
+               } else
+                       memset(data, 0, PAGE_CACHE_SIZE);
+               data += PAGE_CACHE_SIZE;
+       }
+       return read_buffers[buffer] + offset;
+}
+
+static void cramfs_kill_sb(struct super_block *sb)
+{
+       struct cramfs_sb_info *sbi = CRAMFS_SB(sb);
+
+       kill_block_super(sb);
+       kfree(sbi);
+}
+
+static int cramfs_remount(struct super_block *sb, int *flags, char *data)
+{
+       sync_filesystem(sb);
+       *flags |= MS_RDONLY;
+       return 0;
+}
+
+static int cramfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+       int i;
+       struct cramfs_super super;
+       unsigned long root_offset;
+       struct cramfs_sb_info *sbi;
+       struct inode *root;
+
+       sb->s_flags |= MS_RDONLY;
+
+       sbi = kzalloc(sizeof(struct cramfs_sb_info), GFP_KERNEL);
+       if (!sbi)
+               return -ENOMEM;
+       sb->s_fs_info = sbi;
+
+       /* Invalidate the read buffers on mount: think disk change.. */
+       mutex_lock(&read_mutex);
+       for (i = 0; i < READ_BUFFERS; i++)
+               buffer_blocknr[i] = -1;
+
+       /* Read the first block and get the superblock from it */
+       memcpy(&super, cramfs_read(sb, 0, sizeof(super)), sizeof(super));
+       mutex_unlock(&read_mutex);
+
+       /* Do sanity checks on the superblock */
+       if (super.magic != CRAMFS_MAGIC) {
+               /* check for wrong endianness */
+               if (super.magic == CRAMFS_MAGIC_WEND) {
+                       if (!silent)
+                               pr_err("wrong endianness\n");
+                       return -EINVAL;
+               }
+
+               /* check at 512 byte offset */
+               mutex_lock(&read_mutex);
+               memcpy(&super, cramfs_read(sb, 512, sizeof(super)), sizeof(super));
+               mutex_unlock(&read_mutex);
+               if (super.magic != CRAMFS_MAGIC) {
+                       if (super.magic == CRAMFS_MAGIC_WEND && !silent)
+                               pr_err("wrong endianness\n");
+                       else if (!silent)
+                               pr_err("wrong magic\n");
+                       return -EINVAL;
+               }
+       }
+
+       /* get feature flags first */
+       if (super.flags & ~CRAMFS_SUPPORTED_FLAGS) {
+               pr_err("unsupported filesystem features\n");
+               return -EINVAL;
+       }
+
+       /* Check that the root inode is in a sane state */
+       if (!S_ISDIR(super.root.mode)) {
+               pr_err("root is not a directory\n");
+               return -EINVAL;
+       }
+       /* correct strange, hard-coded permissions of mkcramfs */
+       super.root.mode |= (S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP | S_IROTH | S_IXOTH);
+
+       root_offset = super.root.offset << 2;
+       if (super.flags & CRAMFS_FLAG_FSID_VERSION_2) {
+               sbi->size = super.size;
+               sbi->blocks = super.fsid.blocks;
+               sbi->files = super.fsid.files;
+       } else {
+               sbi->size = 1<<28;
+               sbi->blocks = 0;
+               sbi->files = 0;
+       }
+       sbi->magic = super.magic;
+       sbi->flags = super.flags;
+       if (root_offset == 0)
+               pr_info("empty filesystem");
+       else if (!(super.flags & CRAMFS_FLAG_SHIFTED_ROOT_OFFSET) &&
+                ((root_offset != sizeof(struct cramfs_super)) &&
+                 (root_offset != 512 + sizeof(struct cramfs_super))))
+       {
+               pr_err("bad root offset %lu\n", root_offset);
+               return -EINVAL;
+       }
+
+       /* Set it all up.. */
+       sb->s_op = &cramfs_ops;
+       root = get_cramfs_inode(sb, &super.root, 0);
+       if (IS_ERR(root))
+               return PTR_ERR(root);
+       sb->s_root = d_make_root(root);
+       if (!sb->s_root)
+               return -ENOMEM;
+       return 0;
+}
+
+static int cramfs_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+       struct super_block *sb = dentry->d_sb;
+       u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
+
+       buf->f_type = CRAMFS_MAGIC;
+       buf->f_bsize = PAGE_CACHE_SIZE;
+       buf->f_blocks = CRAMFS_SB(sb)->blocks;
+       buf->f_bfree = 0;
+       buf->f_bavail = 0;
+       buf->f_files = CRAMFS_SB(sb)->files;
+       buf->f_ffree = 0;
+       buf->f_fsid.val[0] = (u32)id;
+       buf->f_fsid.val[1] = (u32)(id >> 32);
+       buf->f_namelen = CRAMFS_MAXPATHLEN;
+       return 0;
+}
+
+/*
+ * Read a cramfs directory entry.
+ */
+static int cramfs_readdir(struct file *file, struct dir_context *ctx)
+{
+       struct inode *inode = file_inode(file);
+       struct super_block *sb = inode->i_sb;
+       char *buf;
+       unsigned int offset;
+
+       /* Offset within the thing. */
+       if (ctx->pos >= inode->i_size)
+               return 0;
+       offset = ctx->pos;
+       /* Directory entries are always 4-byte aligned */
+       if (offset & 3)
+               return -EINVAL;
+
+       buf = kmalloc(CRAMFS_MAXPATHLEN, GFP_KERNEL);
+       if (!buf)
+               return -ENOMEM;
+
+       while (offset < inode->i_size) {
+               struct cramfs_inode *de;
+               unsigned long nextoffset;
+               char *name;
+               ino_t ino;
+               umode_t mode;
+               int namelen;
+
+               mutex_lock(&read_mutex);
+               de = cramfs_read(sb, OFFSET(inode) + offset, sizeof(*de)+CRAMFS_MAXPATHLEN);
+               name = (char *)(de+1);
+
+               /*
+                * Namelengths on disk are shifted by two
+                * and the name padded out to 4-byte boundaries
+                * with zeroes.
+                */
+               namelen = de->namelen << 2;
+               memcpy(buf, name, namelen);
+               ino = cramino(de, OFFSET(inode) + offset);
+               mode = de->mode;
+               mutex_unlock(&read_mutex);
+               nextoffset = offset + sizeof(*de) + namelen;
+               for (;;) {
+                       if (!namelen) {
+                               kfree(buf);
+                               return -EIO;
+                       }
+                       if (buf[namelen-1])
+                               break;
+                       namelen--;
+               }
+               if (!dir_emit(ctx, buf, namelen, ino, mode >> 12))
+                       break;
+
+               ctx->pos = offset = nextoffset;
+       }
+       kfree(buf);
+       return 0;
+}
+
+/*
+ * Lookup and fill in the inode data..
+ */
+static struct dentry *cramfs_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
+{
+       unsigned int offset = 0;
+       struct inode *inode = NULL;
+       int sorted;
+
+       mutex_lock(&read_mutex);
+       sorted = CRAMFS_SB(dir->i_sb)->flags & CRAMFS_FLAG_SORTED_DIRS;
+       while (offset < dir->i_size) {
+               struct cramfs_inode *de;
+               char *name;
+               int namelen, retval;
+               int dir_off = OFFSET(dir) + offset;
+
+               de = cramfs_read(dir->i_sb, dir_off, sizeof(*de)+CRAMFS_MAXPATHLEN);
+               name = (char *)(de+1);
+
+               /* Try to take advantage of sorted directories */
+               if (sorted && (dentry->d_name.name[0] < name[0]))
+                       break;
+
+               namelen = de->namelen << 2;
+               offset += sizeof(*de) + namelen;
+
+               /* Quick check that the name is roughly the right length */
+               if (((dentry->d_name.len + 3) & ~3) != namelen)
+                       continue;
+
+               for (;;) {
+                       if (!namelen) {
+                               inode = ERR_PTR(-EIO);
+                               goto out;
+                       }
+                       if (name[namelen-1])
+                               break;
+                       namelen--;
+               }
+               if (namelen != dentry->d_name.len)
+                       continue;
+               retval = memcmp(dentry->d_name.name, name, namelen);
+               if (retval > 0)
+                       continue;
+               if (!retval) {
+                       inode = get_cramfs_inode(dir->i_sb, de, dir_off);
+                       break;
+               }
+               /* else (retval < 0) */
+               if (sorted)
+                       break;
+       }
+out:
+       mutex_unlock(&read_mutex);
+       if (IS_ERR(inode))
+               return ERR_CAST(inode);
+       d_add(dentry, inode);
+       return NULL;
+}
+
+static int cramfs_readpage(struct file *file, struct page *page)
+{
+       struct inode *inode = page->mapping->host;
+       u32 maxblock;
+       int bytes_filled;
+       void *pgdata;
+
+       maxblock = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
+       bytes_filled = 0;
+       pgdata = kmap(page);
+
+       if (page->index < maxblock) {
+               struct super_block *sb = inode->i_sb;
+               u32 blkptr_offset = OFFSET(inode) + page->index*4;
+               u32 start_offset, compr_len;
+
+               start_offset = OFFSET(inode) + maxblock*4;
+               mutex_lock(&read_mutex);
+               if (page->index)
+                       start_offset = *(u32 *) cramfs_read(sb, blkptr_offset-4,
+                               4);
+               compr_len = (*(u32 *) cramfs_read(sb, blkptr_offset, 4) -
+                       start_offset);
+               mutex_unlock(&read_mutex);
+
+               if (compr_len == 0)
+                       ; /* hole */
+               else if (unlikely(compr_len > (PAGE_CACHE_SIZE << 1))) {
+                       pr_err("bad compressed blocksize %u\n",
+                               compr_len);
+                       goto err;
+               } else {
+                       mutex_lock(&read_mutex);
+                       bytes_filled = cramfs_uncompress_block(pgdata,
+                                PAGE_CACHE_SIZE,
+                                cramfs_read(sb, start_offset, compr_len),
+                                compr_len);
+                       mutex_unlock(&read_mutex);
+                       if (unlikely(bytes_filled < 0))
+                               goto err;
+               }
+       }
+
+       memset(pgdata + bytes_filled, 0, PAGE_CACHE_SIZE - bytes_filled);
+       flush_dcache_page(page);
+       kunmap(page);
+       SetPageUptodate(page);
+       unlock_page(page);
+       return 0;
+
+err:
+       kunmap(page);
+       ClearPageUptodate(page);
+       SetPageError(page);
+       unlock_page(page);
+       return 0;
+}
+
+static const struct address_space_operations cramfs_aops = {
+       .readpage = cramfs_readpage
+};
+
+/*
+ * Our operations:
+ */
+
+/*
+ * A directory can only readdir
+ */
+static const struct file_operations cramfs_directory_operations = {
+       .llseek         = generic_file_llseek,
+       .read           = generic_read_dir,
+       .iterate        = cramfs_readdir,
+};
+
+static const struct inode_operations cramfs_dir_inode_operations = {
+       .lookup         = cramfs_lookup,
+};
+
+static const struct super_operations cramfs_ops = {
+       .remount_fs     = cramfs_remount,
+       .statfs         = cramfs_statfs,
+};
+
+static struct dentry *cramfs_mount(struct file_system_type *fs_type,
+       int flags, const char *dev_name, void *data)
+{
+       return mount_bdev(fs_type, flags, dev_name, data, cramfs_fill_super);
+}
+
+static struct file_system_type cramfs_fs_type = {
+       .owner          = THIS_MODULE,
+       .name           = "cramfs",
+       .mount          = cramfs_mount,
+       .kill_sb        = cramfs_kill_sb,
+       .fs_flags       = FS_REQUIRES_DEV,
+};
+MODULE_ALIAS_FS("cramfs");
+
+static int __init init_cramfs_fs(void)
+{
+       int rv;
+
+       rv = cramfs_uncompress_init();
+       if (rv < 0)
+               return rv;
+       rv = register_filesystem(&cramfs_fs_type);
+       if (rv < 0)
+               cramfs_uncompress_exit();
+       return rv;
+}
+
+static void __exit exit_cramfs_fs(void)
+{
+       cramfs_uncompress_exit();
+       unregister_filesystem(&cramfs_fs_type);
+}
+
+module_init(init_cramfs_fs)
+module_exit(exit_cramfs_fs)
+MODULE_LICENSE("GPL");