xfs: use memdup_user()
[safe/jmp/linux-2.6] / fs / fuse / inode.c
index 3141690..459b73d 100644 (file)
@@ -1,6 +1,6 @@
 /*
   FUSE: Filesystem in Userspace
-  Copyright (C) 2001-2006  Miklos Szeredi <miklos@szeredi.hu>
+  Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
 
   This program can be distributed under the terms of the GNU GPL.
   See the file COPYING.
@@ -18,6 +18,7 @@
 #include <linux/statfs.h>
 #include <linux/random.h>
 #include <linux/sched.h>
+#include <linux/exportfs.h>
 
 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
 MODULE_DESCRIPTION("Filesystem in Userspace");
@@ -36,10 +37,10 @@ struct fuse_mount_data {
        unsigned rootmode;
        unsigned user_id;
        unsigned group_id;
-       unsigned fd_present : 1;
-       unsigned rootmode_present : 1;
-       unsigned user_id_present : 1;
-       unsigned group_id_present : 1;
+       unsigned fd_present:1;
+       unsigned rootmode_present:1;
+       unsigned user_id_present:1;
+       unsigned group_id_present:1;
        unsigned flags;
        unsigned max_read;
        unsigned blksize;
@@ -93,7 +94,7 @@ void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
        req->in.numargs = 1;
        req->in.args[0].size = sizeof(struct fuse_forget_in);
        req->in.args[0].value = inarg;
-       request_send_noreply(fc, req);
+       fuse_request_send_noreply(fc, req);
 }
 
 static void fuse_clear_inode(struct inode *inode)
@@ -249,7 +250,7 @@ struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 
        fi = get_fuse_inode(inode);
        spin_lock(&fc->lock);
-       fi->nlookup ++;
+       fi->nlookup++;
        spin_unlock(&fc->lock);
        fuse_change_attributes(inode, attr, attr_valid, attr_version);
 
@@ -268,7 +269,7 @@ static void fuse_send_destroy(struct fuse_conn *fc)
                fc->destroy_req = NULL;
                req->in.h.opcode = FUSE_DESTROY;
                req->force = 1;
-               request_send(fc, req);
+               fuse_request_send(fc, req);
                fuse_put_request(fc, req);
        }
 }
@@ -291,6 +292,7 @@ static void fuse_put_super(struct super_block *sb)
        list_del(&fc->entry);
        fuse_ctl_remove_conn(fc);
        mutex_unlock(&fuse_mutex);
+       bdi_destroy(&fc->bdi);
        fuse_conn_put(fc);
 }
 
@@ -333,7 +335,7 @@ static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
        req->out.args[0].size =
                fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
        req->out.args[0].value = &outarg;
-       request_send(fc, req);
+       fuse_request_send(fc, req);
        err = req->out.h.error;
        if (!err)
                convert_fuse_statfs(buf, &outarg.st);
@@ -353,7 +355,7 @@ enum {
        OPT_ERR
 };
 
-static match_table_t tokens = {
+static const match_table_t tokens = {
        {OPT_FD,                        "fd=%u"},
        {OPT_ROOTMODE,                  "rootmode=%o"},
        {OPT_USER_ID,                   "user_id=%u"},
@@ -461,68 +463,69 @@ static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
        return 0;
 }
 
-static struct fuse_conn *new_conn(struct super_block *sb)
+int fuse_conn_init(struct fuse_conn *fc, struct super_block *sb)
 {
-       struct fuse_conn *fc;
        int err;
 
-       fc = kzalloc(sizeof(*fc), GFP_KERNEL);
-       if (fc) {
-               spin_lock_init(&fc->lock);
-               mutex_init(&fc->inst_mutex);
-               atomic_set(&fc->count, 1);
-               init_waitqueue_head(&fc->waitq);
-               init_waitqueue_head(&fc->blocked_waitq);
-               init_waitqueue_head(&fc->reserved_req_waitq);
-               INIT_LIST_HEAD(&fc->pending);
-               INIT_LIST_HEAD(&fc->processing);
-               INIT_LIST_HEAD(&fc->io);
-               INIT_LIST_HEAD(&fc->interrupts);
-               INIT_LIST_HEAD(&fc->bg_queue);
-               atomic_set(&fc->num_waiting, 0);
-               fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
-               fc->bdi.unplug_io_fn = default_unplug_io_fn;
-               /* fuse does it's own writeback accounting */
-               fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
-               fc->dev = sb->s_dev;
-               err = bdi_init(&fc->bdi);
-               if (err)
-                       goto error_kfree;
-               if (sb->s_bdev) {
-                       err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
-                                          MAJOR(fc->dev), MINOR(fc->dev));
-               } else {
-                       err = bdi_register_dev(&fc->bdi, fc->dev);
-               }
-               if (err)
-                       goto error_bdi_destroy;
-               /*
-                * For a single fuse filesystem use max 1% of dirty +
-                * writeback threshold.
-                *
-                * This gives about 1M of write buffer for memory maps on a
-                * machine with 1G and 10% dirty_ratio, which should be more
-                * than enough.
-                *
-                * Privileged users can raise it by writing to
-                *
-                *    /sys/class/bdi/<bdi>/max_ratio
-                */
-               bdi_set_max_ratio(&fc->bdi, 1);
-               fc->reqctr = 0;
-               fc->blocked = 1;
-               fc->attr_version = 1;
-               get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
+       memset(fc, 0, sizeof(*fc));
+       spin_lock_init(&fc->lock);
+       mutex_init(&fc->inst_mutex);
+       atomic_set(&fc->count, 1);
+       init_waitqueue_head(&fc->waitq);
+       init_waitqueue_head(&fc->blocked_waitq);
+       init_waitqueue_head(&fc->reserved_req_waitq);
+       INIT_LIST_HEAD(&fc->pending);
+       INIT_LIST_HEAD(&fc->processing);
+       INIT_LIST_HEAD(&fc->io);
+       INIT_LIST_HEAD(&fc->interrupts);
+       INIT_LIST_HEAD(&fc->bg_queue);
+       INIT_LIST_HEAD(&fc->entry);
+       atomic_set(&fc->num_waiting, 0);
+       fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
+       fc->bdi.unplug_io_fn = default_unplug_io_fn;
+       /* fuse does it's own writeback accounting */
+       fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
+       fc->khctr = 0;
+       fc->polled_files = RB_ROOT;
+       fc->dev = sb->s_dev;
+       err = bdi_init(&fc->bdi);
+       if (err)
+               goto error_mutex_destroy;
+       if (sb->s_bdev) {
+               err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
+                                  MAJOR(fc->dev), MINOR(fc->dev));
+       } else {
+               err = bdi_register_dev(&fc->bdi, fc->dev);
        }
-       return fc;
+       if (err)
+               goto error_bdi_destroy;
+       /*
+        * For a single fuse filesystem use max 1% of dirty +
+        * writeback threshold.
+        *
+        * This gives about 1M of write buffer for memory maps on a
+        * machine with 1G and 10% dirty_ratio, which should be more
+        * than enough.
+        *
+        * Privileged users can raise it by writing to
+        *
+        *    /sys/class/bdi/<bdi>/max_ratio
+        */
+       bdi_set_max_ratio(&fc->bdi, 1);
+       fc->reqctr = 0;
+       fc->blocked = 1;
+       fc->attr_version = 1;
+       get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 
-error_bdi_destroy:
+       return 0;
+
+ error_bdi_destroy:
        bdi_destroy(&fc->bdi);
-error_kfree:
+ error_mutex_destroy:
        mutex_destroy(&fc->inst_mutex);
-       kfree(fc);
-       return NULL;
+       return err;
 }
+EXPORT_SYMBOL_GPL(fuse_conn_init);
 
 void fuse_conn_put(struct fuse_conn *fc)
 {
@@ -530,8 +533,7 @@ void fuse_conn_put(struct fuse_conn *fc)
                if (fc->destroy_req)
                        fuse_request_free(fc->destroy_req);
                mutex_destroy(&fc->inst_mutex);
-               bdi_destroy(&fc->bdi);
-               kfree(fc);
+               fc->release(fc);
        }
 }
 
@@ -541,7 +543,7 @@ struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
        return fc;
 }
 
-static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
+static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 {
        struct fuse_attr attr;
        memset(&attr, 0, sizeof(attr));
@@ -552,6 +554,166 @@ static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
        return fuse_iget(sb, 1, 0, &attr, 0, 0);
 }
 
+struct fuse_inode_handle {
+       u64 nodeid;
+       u32 generation;
+};
+
+static struct dentry *fuse_get_dentry(struct super_block *sb,
+                                     struct fuse_inode_handle *handle)
+{
+       struct fuse_conn *fc = get_fuse_conn_super(sb);
+       struct inode *inode;
+       struct dentry *entry;
+       int err = -ESTALE;
+
+       if (handle->nodeid == 0)
+               goto out_err;
+
+       inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
+       if (!inode) {
+               struct fuse_entry_out outarg;
+               struct qstr name;
+
+               if (!fc->export_support)
+                       goto out_err;
+
+               name.len = 1;
+               name.name = ".";
+               err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
+                                      &inode);
+               if (err && err != -ENOENT)
+                       goto out_err;
+               if (err || !inode) {
+                       err = -ESTALE;
+                       goto out_err;
+               }
+               err = -EIO;
+               if (get_node_id(inode) != handle->nodeid)
+                       goto out_iput;
+       }
+       err = -ESTALE;
+       if (inode->i_generation != handle->generation)
+               goto out_iput;
+
+       entry = d_obtain_alias(inode);
+       if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
+               entry->d_op = &fuse_dentry_operations;
+               fuse_invalidate_entry_cache(entry);
+       }
+
+       return entry;
+
+ out_iput:
+       iput(inode);
+ out_err:
+       return ERR_PTR(err);
+}
+
+static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
+                          int connectable)
+{
+       struct inode *inode = dentry->d_inode;
+       bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
+       int len = encode_parent ? 6 : 3;
+       u64 nodeid;
+       u32 generation;
+
+       if (*max_len < len)
+               return  255;
+
+       nodeid = get_fuse_inode(inode)->nodeid;
+       generation = inode->i_generation;
+
+       fh[0] = (u32)(nodeid >> 32);
+       fh[1] = (u32)(nodeid & 0xffffffff);
+       fh[2] = generation;
+
+       if (encode_parent) {
+               struct inode *parent;
+
+               spin_lock(&dentry->d_lock);
+               parent = dentry->d_parent->d_inode;
+               nodeid = get_fuse_inode(parent)->nodeid;
+               generation = parent->i_generation;
+               spin_unlock(&dentry->d_lock);
+
+               fh[3] = (u32)(nodeid >> 32);
+               fh[4] = (u32)(nodeid & 0xffffffff);
+               fh[5] = generation;
+       }
+
+       *max_len = len;
+       return encode_parent ? 0x82 : 0x81;
+}
+
+static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
+               struct fid *fid, int fh_len, int fh_type)
+{
+       struct fuse_inode_handle handle;
+
+       if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
+               return NULL;
+
+       handle.nodeid = (u64) fid->raw[0] << 32;
+       handle.nodeid |= (u64) fid->raw[1];
+       handle.generation = fid->raw[2];
+       return fuse_get_dentry(sb, &handle);
+}
+
+static struct dentry *fuse_fh_to_parent(struct super_block *sb,
+               struct fid *fid, int fh_len, int fh_type)
+{
+       struct fuse_inode_handle parent;
+
+       if (fh_type != 0x82 || fh_len < 6)
+               return NULL;
+
+       parent.nodeid = (u64) fid->raw[3] << 32;
+       parent.nodeid |= (u64) fid->raw[4];
+       parent.generation = fid->raw[5];
+       return fuse_get_dentry(sb, &parent);
+}
+
+static struct dentry *fuse_get_parent(struct dentry *child)
+{
+       struct inode *child_inode = child->d_inode;
+       struct fuse_conn *fc = get_fuse_conn(child_inode);
+       struct inode *inode;
+       struct dentry *parent;
+       struct fuse_entry_out outarg;
+       struct qstr name;
+       int err;
+
+       if (!fc->export_support)
+               return ERR_PTR(-ESTALE);
+
+       name.len = 2;
+       name.name = "..";
+       err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
+                              &name, &outarg, &inode);
+       if (err) {
+               if (err == -ENOENT)
+                       return ERR_PTR(-ESTALE);
+               return ERR_PTR(err);
+       }
+
+       parent = d_obtain_alias(inode);
+       if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
+               parent->d_op = &fuse_dentry_operations;
+               fuse_invalidate_entry_cache(parent);
+       }
+
+       return parent;
+}
+
+static const struct export_operations fuse_export_operations = {
+       .fh_to_dentry   = fuse_fh_to_dentry,
+       .fh_to_parent   = fuse_fh_to_parent,
+       .encode_fh      = fuse_encode_fh,
+       .get_parent     = fuse_get_parent,
+};
+
 static const struct super_operations fuse_super_operations = {
        .alloc_inode    = fuse_alloc_inode,
        .destroy_inode  = fuse_destroy_inode,
@@ -581,6 +743,11 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
                                fc->no_lock = 1;
                        if (arg->flags & FUSE_ATOMIC_O_TRUNC)
                                fc->atomic_o_trunc = 1;
+                       if (arg->minor >= 9) {
+                               /* LOOKUP has dependency on proto version */
+                               if (arg->flags & FUSE_EXPORT_SUPPORT)
+                                       fc->export_support = 1;
+                       }
                        if (arg->flags & FUSE_BIG_WRITES)
                                fc->big_writes = 1;
                } else {
@@ -594,7 +761,6 @@ static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
                fc->max_write = max_t(unsigned, 4096, fc->max_write);
                fc->conn_init = 1;
        }
-       fuse_put_request(fc, req);
        fc->blocked = 0;
        wake_up_all(&fc->blocked_waitq);
 }
@@ -607,7 +773,7 @@ static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
        arg->minor = FUSE_KERNEL_MINOR_VERSION;
        arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
        arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
-               FUSE_BIG_WRITES;
+               FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
        req->in.h.opcode = FUSE_INIT;
        req->in.numargs = 1;
        req->in.args[0].size = sizeof(*arg);
@@ -620,7 +786,12 @@ static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
        req->out.args[0].size = sizeof(struct fuse_init_out);
        req->out.args[0].value = &req->misc.init_out;
        req->end = process_init_reply;
-       request_send_background(fc, req);
+       fuse_request_send_background(fc, req);
+}
+
+static void fuse_free_conn(struct fuse_conn *fc)
+{
+       kfree(fc);
 }
 
 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
@@ -634,16 +805,18 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        int err;
        int is_bdev = sb->s_bdev != NULL;
 
+       err = -EINVAL;
        if (sb->s_flags & MS_MANDLOCK)
-               return -EINVAL;
+               goto err;
 
        if (!parse_fuse_opt((char *) data, &d, is_bdev))
-               return -EINVAL;
+               goto err;
 
        if (is_bdev) {
 #ifdef CONFIG_BLOCK
+               err = -EINVAL;
                if (!sb_set_blocksize(sb, d.blksize))
-                       return -EINVAL;
+                       goto err;
 #endif
        } else {
                sb->s_blocksize = PAGE_CACHE_SIZE;
@@ -652,18 +825,28 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        sb->s_magic = FUSE_SUPER_MAGIC;
        sb->s_op = &fuse_super_operations;
        sb->s_maxbytes = MAX_LFS_FILESIZE;
+       sb->s_export_op = &fuse_export_operations;
 
        file = fget(d.fd);
+       err = -EINVAL;
        if (!file)
-               return -EINVAL;
+               goto err;
 
        if (file->f_op != &fuse_dev_operations)
-               return -EINVAL;
+               goto err_fput;
 
-       fc = new_conn(sb);
+       fc = kmalloc(sizeof(*fc), GFP_KERNEL);
+       err = -ENOMEM;
        if (!fc)
-               return -ENOMEM;
+               goto err_fput;
 
+       err = fuse_conn_init(fc, sb);
+       if (err) {
+               kfree(fc);
+               goto err_fput;
+       }
+
+       fc->release = fuse_free_conn;
        fc->flags = d.flags;
        fc->user_id = d.user_id;
        fc->group_id = d.group_id;
@@ -673,14 +856,14 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        sb->s_fs_info = fc;
 
        err = -ENOMEM;
-       root = get_root_inode(sb, d.rootmode);
+       root = fuse_get_root_inode(sb, d.rootmode);
        if (!root)
-               goto err;
+               goto err_put_conn;
 
        root_dentry = d_alloc_root(root);
        if (!root_dentry) {
                iput(root);
-               goto err;
+               goto err_put_conn;
        }
 
        init_req = fuse_request_alloc();
@@ -690,7 +873,7 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        if (is_bdev) {
                fc->destroy_req = fuse_request_alloc();
                if (!fc->destroy_req)
-                       goto err_put_root;
+                       goto err_free_init_req;
        }
 
        mutex_lock(&fuse_mutex);
@@ -720,12 +903,15 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 
  err_unlock:
        mutex_unlock(&fuse_mutex);
+ err_free_init_req:
        fuse_request_free(init_req);
  err_put_root:
        dput(root_dentry);
- err:
-       fput(file);
+ err_put_conn:
        fuse_conn_put(fc);
+ err_fput:
+       fput(file);
+ err:
        return err;
 }
 
@@ -781,9 +967,9 @@ static inline void unregister_fuseblk(void)
 }
 #endif
 
-static void fuse_inode_init_once(struct kmem_cache *cachep, void *foo)
+static void fuse_inode_init_once(void *foo)
 {
-       struct inode * inode = foo;
+       struct inode *inode = foo;
 
        inode_init_once(inode);
 }
@@ -862,7 +1048,7 @@ static int __init fuse_init(void)
 {
        int res;
 
-       printk("fuse init (API version %i.%i)\n",
+       printk(KERN_INFO "fuse init (API version %i.%i)\n",
               FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
 
        INIT_LIST_HEAD(&fuse_conn_list);