Revert "[fuse] fix deadlock between fuse_put_super() and request_end()"
[safe/jmp/linux-2.6] / fs / fuse / inode.c
index fa03f80..fd34037 100644 (file)
@@ -1,6 +1,6 @@
 /*
   FUSE: Filesystem in Userspace
-  Copyright (C) 2001-2005  Miklos Szeredi <miklos@szeredi.hu>
+  Copyright (C) 2001-2006  Miklos Szeredi <miklos@szeredi.hu>
 
   This program can be distributed under the terms of the GNU GPL.
   See the file COPYING.
@@ -15,7 +15,6 @@
 #include <linux/seq_file.h>
 #include <linux/init.h>
 #include <linux/module.h>
-#include <linux/moduleparam.h>
 #include <linux/parser.h>
 #include <linux/statfs.h>
 
@@ -23,13 +22,14 @@ MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
 MODULE_DESCRIPTION("Filesystem in Userspace");
 MODULE_LICENSE("GPL");
 
-spinlock_t fuse_lock;
 static kmem_cache_t *fuse_inode_cachep;
-static int mount_count;
+static struct subsystem connections_subsys;
 
-static int mount_max = 1000;
-module_param(mount_max, int, 0644);
-MODULE_PARM_DESC(mount_max, "Maximum number of FUSE mounts allowed, if -1 then unlimited (default: 1000)");
+struct fuse_conn_attr {
+       struct attribute attr;
+       ssize_t (*show)(struct fuse_conn *, char *);
+       ssize_t (*store)(struct fuse_conn *, const char *, size_t);
+};
 
 #define FUSE_SUPER_MAGIC 0x65735546
 
@@ -37,6 +37,13 @@ struct fuse_mount_data {
        int fd;
        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 flags;
+       unsigned max_read;
 };
 
 static struct inode *fuse_alloc_inode(struct super_block *sb)
@@ -89,8 +96,8 @@ void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
 
 static void fuse_clear_inode(struct inode *inode)
 {
-       struct fuse_conn *fc = get_fuse_conn(inode);
-       if (fc) {
+       if (inode->i_sb->s_flags & MS_ACTIVE) {
+               struct fuse_conn *fc = get_fuse_conn(inode);
                struct fuse_inode *fi = get_fuse_inode(inode);
                fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
                fi->forget_req = NULL;
@@ -124,6 +131,7 @@ static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
        i_size_write(inode, attr->size);
        if (S_ISREG(inode->i_mode)) {
                fuse_init_common(inode);
+               fuse_init_file_inode(inode);
        } else if (S_ISDIR(inode->i_mode))
                fuse_init_dir(inode);
        else if (S_ISLNK(inode->i_mode))
@@ -133,11 +141,8 @@ static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
                fuse_init_common(inode);
                init_special_inode(inode, inode->i_mode,
                                   new_decode_dev(attr->rdev));
-       } else {
-               /* Don't let user create weird files */
-               inode->i_mode = S_IFREG;
-               fuse_init_common(inode);
-       }
+       } else
+               BUG();
 }
 
 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
@@ -170,6 +175,7 @@ struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
                return NULL;
 
        if ((inode->i_state & I_NEW)) {
+               inode->i_flags |= S_NOATIME|S_NOCMTIME;
                inode->i_generation = generation;
                inode->i_data.backing_dev_info = &fc->bdi;
                fuse_init_inode(inode, attr);
@@ -189,25 +195,38 @@ struct inode *fuse_iget(struct super_block *sb, unsigned long nodeid,
        return inode;
 }
 
+static void fuse_umount_begin(struct super_block *sb)
+{
+       fuse_abort_conn(get_fuse_conn_super(sb));
+}
+
 static void fuse_put_super(struct super_block *sb)
 {
        struct fuse_conn *fc = get_fuse_conn_super(sb);
 
-       spin_lock(&fuse_lock);
-       mount_count --;
-       fc->sb = NULL;
-       fc->user_id = 0;
+       down_write(&fc->sbput_sem);
+       while (!list_empty(&fc->background))
+               fuse_release_background(fc,
+                                       list_entry(fc->background.next,
+                                                  struct fuse_req, bg_entry));
+
+       spin_lock(&fc->lock);
+       fc->mounted = 0;
+       fc->connected = 0;
+       spin_unlock(&fc->lock);
+       up_write(&fc->sbput_sem);
        /* Flush all readers on this fs */
+       kill_fasync(&fc->fasync, SIGIO, POLL_IN);
        wake_up_all(&fc->waitq);
-       fuse_release_conn(fc);
-       *get_fuse_conn_super_p(sb) = NULL;
-       spin_unlock(&fuse_lock);
+       kobject_del(&fc->kobj);
+       kobject_put(&fc->kobj);
 }
 
 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 {
        stbuf->f_type    = FUSE_SUPER_MAGIC;
        stbuf->f_bsize   = attr->bsize;
+       stbuf->f_frsize  = attr->frsize;
        stbuf->f_blocks  = attr->blocks;
        stbuf->f_bfree   = attr->bfree;
        stbuf->f_bavail  = attr->bavail;
@@ -224,14 +243,16 @@ static int fuse_statfs(struct super_block *sb, struct kstatfs *buf)
        struct fuse_statfs_out outarg;
        int err;
 
-        req = fuse_get_request(fc);
-       if (!req)
-               return -ERESTARTSYS;
+       req = fuse_get_req(fc);
+       if (IS_ERR(req))
+               return PTR_ERR(req);
 
+       memset(&outarg, 0, sizeof(outarg));
        req->in.numargs = 0;
        req->in.h.opcode = FUSE_STATFS;
        req->out.numargs = 1;
-       req->out.args[0].size = sizeof(outarg);
+       req->out.args[0].size =
+               fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
        req->out.args[0].value = &outarg;
        request_send(fc, req);
        err = req->out.h.error;
@@ -245,10 +266,10 @@ enum {
        OPT_FD,
        OPT_ROOTMODE,
        OPT_USER_ID,
+       OPT_GROUP_ID,
        OPT_DEFAULT_PERMISSIONS,
        OPT_ALLOW_OTHER,
-       OPT_ALLOW_ROOT,
-       OPT_KERNEL_CACHE,
+       OPT_MAX_READ,
        OPT_ERR
 };
 
@@ -256,10 +277,10 @@ static match_table_t tokens = {
        {OPT_FD,                        "fd=%u"},
        {OPT_ROOTMODE,                  "rootmode=%o"},
        {OPT_USER_ID,                   "user_id=%u"},
+       {OPT_GROUP_ID,                  "group_id=%u"},
        {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
        {OPT_ALLOW_OTHER,               "allow_other"},
-       {OPT_ALLOW_ROOT,                "allow_root"},
-       {OPT_KERNEL_CACHE,              "kernel_cache"},
+       {OPT_MAX_READ,                  "max_read=%u"},
        {OPT_ERR,                       NULL}
 };
 
@@ -267,7 +288,7 @@ static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
 {
        char *p;
        memset(d, 0, sizeof(struct fuse_mount_data));
-       d->fd = -1;
+       d->max_read = ~0;
 
        while ((p = strsep(&opt, ",")) != NULL) {
                int token;
@@ -282,25 +303,51 @@ static int parse_fuse_opt(char *opt, struct fuse_mount_data *d)
                        if (match_int(&args[0], &value))
                                return 0;
                        d->fd = value;
+                       d->fd_present = 1;
                        break;
 
                case OPT_ROOTMODE:
                        if (match_octal(&args[0], &value))
                                return 0;
                        d->rootmode = value;
+                       d->rootmode_present = 1;
                        break;
 
                case OPT_USER_ID:
                        if (match_int(&args[0], &value))
                                return 0;
                        d->user_id = value;
+                       d->user_id_present = 1;
+                       break;
+
+               case OPT_GROUP_ID:
+                       if (match_int(&args[0], &value))
+                               return 0;
+                       d->group_id = value;
+                       d->group_id_present = 1;
+                       break;
+
+               case OPT_DEFAULT_PERMISSIONS:
+                       d->flags |= FUSE_DEFAULT_PERMISSIONS;
+                       break;
+
+               case OPT_ALLOW_OTHER:
+                       d->flags |= FUSE_ALLOW_OTHER;
+                       break;
+
+               case OPT_MAX_READ:
+                       if (match_int(&args[0], &value))
+                               return 0;
+                       d->max_read = value;
                        break;
 
                default:
                        return 0;
                }
        }
-       if (d->fd == -1)
+
+       if (!d->fd_present || !d->rootmode_present ||
+           !d->user_id_present || !d->group_id_present)
                return 0;
 
        return 1;
@@ -311,80 +358,46 @@ static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
        struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 
        seq_printf(m, ",user_id=%u", fc->user_id);
+       seq_printf(m, ",group_id=%u", fc->group_id);
+       if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
+               seq_puts(m, ",default_permissions");
+       if (fc->flags & FUSE_ALLOW_OTHER)
+               seq_puts(m, ",allow_other");
+       if (fc->max_read != ~0)
+               seq_printf(m, ",max_read=%u", fc->max_read);
        return 0;
 }
 
-static void free_conn(struct fuse_conn *fc)
+static void fuse_conn_release(struct kobject *kobj)
 {
-       while (!list_empty(&fc->unused_list)) {
-               struct fuse_req *req;
-               req = list_entry(fc->unused_list.next, struct fuse_req, list);
-               list_del(&req->list);
-               fuse_request_free(req);
-       }
-       kfree(fc);
-}
-
-/* Must be called with the fuse lock held */
-void fuse_release_conn(struct fuse_conn *fc)
-{
-       if (!fc->sb && !fc->file)
-               free_conn(fc);
+       kfree(get_fuse_conn_kobj(kobj));
 }
 
 static struct fuse_conn *new_conn(void)
 {
        struct fuse_conn *fc;
 
-       fc = kmalloc(sizeof(*fc), GFP_KERNEL);
-       if (fc != NULL) {
-               int i;
-               memset(fc, 0, sizeof(*fc));
-               fc->sb = NULL;
-               fc->file = NULL;
-               fc->user_id = 0;
+       fc = kzalloc(sizeof(*fc), GFP_KERNEL);
+       if (fc) {
+               spin_lock_init(&fc->lock);
                init_waitqueue_head(&fc->waitq);
+               init_waitqueue_head(&fc->blocked_waitq);
                INIT_LIST_HEAD(&fc->pending);
                INIT_LIST_HEAD(&fc->processing);
-               INIT_LIST_HEAD(&fc->unused_list);
-               sema_init(&fc->outstanding_sem, 0);
-               for (i = 0; i < FUSE_MAX_OUTSTANDING; i++) {
-                       struct fuse_req *req = fuse_request_alloc();
-                       if (!req) {
-                               free_conn(fc);
-                               return NULL;
-                       }
-                       list_add(&req->list, &fc->unused_list);
-               }
+               INIT_LIST_HEAD(&fc->io);
+               INIT_LIST_HEAD(&fc->background);
+               init_rwsem(&fc->sbput_sem);
+               kobj_set_kset_s(fc, connections_subsys);
+               kobject_init(&fc->kobj);
+               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;
                fc->reqctr = 0;
+               fc->blocked = 1;
        }
        return fc;
 }
 
-static struct fuse_conn *get_conn(struct file *file, struct super_block *sb)
-{
-       struct fuse_conn *fc;
-
-       if (file->f_op != &fuse_dev_operations)
-               return ERR_PTR(-EINVAL);
-       fc = new_conn();
-       if (fc == NULL)
-               return ERR_PTR(-ENOMEM);
-       spin_lock(&fuse_lock);
-       if (file->private_data) {
-               free_conn(fc);
-               fc = ERR_PTR(-EINVAL);
-       } else {
-               file->private_data = fc;
-               fc->sb = sb;
-               fc->file = file;
-       }
-       spin_unlock(&fuse_lock);
-       return fc;
-}
-
 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
 {
        struct fuse_attr attr;
@@ -401,19 +414,64 @@ static struct super_operations fuse_super_operations = {
        .read_inode     = fuse_read_inode,
        .clear_inode    = fuse_clear_inode,
        .put_super      = fuse_put_super,
+       .umount_begin   = fuse_umount_begin,
        .statfs         = fuse_statfs,
        .show_options   = fuse_show_options,
 };
 
-static int inc_mount_count(void)
+static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 {
-       int success = 0;
-       spin_lock(&fuse_lock);
-       mount_count ++;
-       if (mount_max == -1 || mount_count <= mount_max)
-               success = 1;
-       spin_unlock(&fuse_lock);
-       return success;
+       struct fuse_init_out *arg = &req->misc.init_out;
+
+       if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
+               fc->conn_error = 1;
+       else {
+               unsigned long ra_pages;
+
+               if (arg->minor >= 6) {
+                       ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
+                       if (arg->flags & FUSE_ASYNC_READ)
+                               fc->async_read = 1;
+               } else
+                       ra_pages = fc->max_read / PAGE_CACHE_SIZE;
+
+               fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
+               fc->minor = arg->minor;
+               fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
+       }
+       fuse_put_request(fc, req);
+       fc->blocked = 0;
+       wake_up_all(&fc->blocked_waitq);
+}
+
+static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
+{
+       struct fuse_init_in *arg = &req->misc.init_in;
+
+       arg->major = FUSE_KERNEL_VERSION;
+       arg->minor = FUSE_KERNEL_MINOR_VERSION;
+       arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
+       arg->flags |= FUSE_ASYNC_READ;
+       req->in.h.opcode = FUSE_INIT;
+       req->in.numargs = 1;
+       req->in.args[0].size = sizeof(*arg);
+       req->in.args[0].value = arg;
+       req->out.numargs = 1;
+       /* Variable length arguement used for backward compatibility
+          with interface version < 7.5.  Rest of init_out is zeroed
+          by do_get_request(), so a short reply is not a problem */
+       req->out.argvar = 1;
+       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);
+}
+
+static unsigned long long conn_id(void)
+{
+       /* BKL is held for ->get_sb() */
+       static unsigned long long ctr = 1;
+       return ctr++;
 }
 
 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
@@ -422,6 +480,8 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        struct inode *root;
        struct fuse_mount_data d;
        struct file *file;
+       struct dentry *root_dentry;
+       struct fuse_req *init_req;
        int err;
 
        if (!parse_fuse_opt((char *) data, &d))
@@ -437,39 +497,72 @@ static int fuse_fill_super(struct super_block *sb, void *data, int silent)
        if (!file)
                return -EINVAL;
 
-       fc = get_conn(file, sb);
-       fput(file);
-       if (IS_ERR(fc))
-               return PTR_ERR(fc);
+       if (file->f_op != &fuse_dev_operations)
+               return -EINVAL;
 
-       fc->user_id = d.user_id;
+       /* Setting file->private_data can't race with other mount()
+          instances, since BKL is held for ->get_sb() */
+       if (file->private_data)
+               return -EINVAL;
+
+       fc = new_conn();
+       if (!fc)
+               return -ENOMEM;
 
-       *get_fuse_conn_super_p(sb) = fc;
+       fc->flags = d.flags;
+       fc->user_id = d.user_id;
+       fc->group_id = d.group_id;
+       fc->max_read = d.max_read;
 
-       err = -ENFILE;
-       if (!inc_mount_count() && current->uid != 0)
-               goto err;
+       /* Used by get_root_inode() */
+       sb->s_fs_info = fc;
 
        err = -ENOMEM;
        root = get_root_inode(sb, d.rootmode);
-       if (root == NULL)
+       if (!root)
                goto err;
 
-       sb->s_root = d_alloc_root(root);
-       if (!sb->s_root) {
+       root_dentry = d_alloc_root(root);
+       if (!root_dentry) {
                iput(root);
                goto err;
        }
-       fuse_send_init(fc);
+
+       init_req = fuse_request_alloc();
+       if (!init_req)
+               goto err_put_root;
+
+       err = kobject_set_name(&fc->kobj, "%llu", conn_id());
+       if (err)
+               goto err_free_req;
+
+       err = kobject_add(&fc->kobj);
+       if (err)
+               goto err_free_req;
+
+       sb->s_root = root_dentry;
+       fc->mounted = 1;
+       fc->connected = 1;
+       kobject_get(&fc->kobj);
+       file->private_data = fc;
+       /*
+        * atomic_dec_and_test() in fput() provides the necessary
+        * memory barrier for file->private_data to be visible on all
+        * CPUs after this
+        */
+       fput(file);
+
+       fuse_send_init(fc, init_req);
+
        return 0;
 
+ err_free_req:
+       fuse_request_free(init_req);
+ err_put_root:
+       dput(root_dentry);
  err:
-       spin_lock(&fuse_lock);
-       mount_count --;
-       fc->sb = NULL;
-       fuse_release_conn(fc);
-       spin_unlock(&fuse_lock);
-       *get_fuse_conn_super_p(sb) = NULL;
+       fput(file);
+       kobject_put(&fc->kobj);
        return err;
 }
 
@@ -487,6 +580,69 @@ static struct file_system_type fuse_fs_type = {
        .kill_sb        = kill_anon_super,
 };
 
+static ssize_t fuse_conn_waiting_show(struct fuse_conn *fc, char *page)
+{
+       return sprintf(page, "%i\n", atomic_read(&fc->num_waiting));
+}
+
+static ssize_t fuse_conn_abort_store(struct fuse_conn *fc, const char *page,
+                                    size_t count)
+{
+       fuse_abort_conn(fc);
+       return count;
+}
+
+static struct fuse_conn_attr fuse_conn_waiting =
+       __ATTR(waiting, 0400, fuse_conn_waiting_show, NULL);
+static struct fuse_conn_attr fuse_conn_abort =
+       __ATTR(abort, 0600, NULL, fuse_conn_abort_store);
+
+static struct attribute *fuse_conn_attrs[] = {
+       &fuse_conn_waiting.attr,
+       &fuse_conn_abort.attr,
+       NULL,
+};
+
+static ssize_t fuse_conn_attr_show(struct kobject *kobj,
+                                  struct attribute *attr,
+                                  char *page)
+{
+       struct fuse_conn_attr *fca =
+               container_of(attr, struct fuse_conn_attr, attr);
+
+       if (fca->show)
+               return fca->show(get_fuse_conn_kobj(kobj), page);
+       else
+               return -EACCES;
+}
+
+static ssize_t fuse_conn_attr_store(struct kobject *kobj,
+                                   struct attribute *attr,
+                                   const char *page, size_t count)
+{
+       struct fuse_conn_attr *fca =
+               container_of(attr, struct fuse_conn_attr, attr);
+
+       if (fca->store)
+               return fca->store(get_fuse_conn_kobj(kobj), page, count);
+       else
+               return -EACCES;
+}
+
+static struct sysfs_ops fuse_conn_sysfs_ops = {
+       .show   = &fuse_conn_attr_show,
+       .store  = &fuse_conn_attr_store,
+};
+
+static struct kobj_type ktype_fuse_conn = {
+       .release        = fuse_conn_release,
+       .sysfs_ops      = &fuse_conn_sysfs_ops,
+       .default_attrs  = fuse_conn_attrs,
+};
+
+static decl_subsys(fuse, NULL, NULL);
+static decl_subsys(connections, &ktype_fuse_conn, NULL);
+
 static void fuse_inode_init_once(void *foo, kmem_cache_t *cachep,
                                 unsigned long flags)
 {
@@ -524,6 +680,34 @@ static void fuse_fs_cleanup(void)
        kmem_cache_destroy(fuse_inode_cachep);
 }
 
+static int fuse_sysfs_init(void)
+{
+       int err;
+
+       kset_set_kset_s(&fuse_subsys, fs_subsys);
+       err = subsystem_register(&fuse_subsys);
+       if (err)
+               goto out_err;
+
+       kset_set_kset_s(&connections_subsys, fuse_subsys);
+       err = subsystem_register(&connections_subsys);
+       if (err)
+               goto out_fuse_unregister;
+
+       return 0;
+
+ out_fuse_unregister:
+       subsystem_unregister(&fuse_subsys);
+ out_err:
+       return err;
+}
+
+static void fuse_sysfs_cleanup(void)
+{
+       subsystem_unregister(&connections_subsys);
+       subsystem_unregister(&fuse_subsys);
+}
+
 static int __init fuse_init(void)
 {
        int res;
@@ -531,7 +715,6 @@ static int __init fuse_init(void)
        printk("fuse init (API version %i.%i)\n",
               FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
 
-       spin_lock_init(&fuse_lock);
        res = fuse_fs_init();
        if (res)
                goto err;
@@ -540,8 +723,14 @@ static int __init fuse_init(void)
        if (res)
                goto err_fs_cleanup;
 
+       res = fuse_sysfs_init();
+       if (res)
+               goto err_dev_cleanup;
+
        return 0;
 
+ err_dev_cleanup:
+       fuse_dev_cleanup();
  err_fs_cleanup:
        fuse_fs_cleanup();
  err:
@@ -552,6 +741,7 @@ static void __exit fuse_exit(void)
 {
        printk(KERN_DEBUG "fuse exit\n");
 
+       fuse_sysfs_cleanup();
        fuse_fs_cleanup();
        fuse_dev_cleanup();
 }