diff options
Diffstat (limited to 'fs/configfs/dir.c')
-rw-r--r-- | fs/configfs/dir.c | 360 |
1 files changed, 277 insertions, 83 deletions
diff --git a/fs/configfs/dir.c b/fs/configfs/dir.c index a48dc7dd8765..8e93341f3e82 100644 --- a/fs/configfs/dir.c +++ b/fs/configfs/dir.c @@ -30,11 +30,25 @@ #include <linux/mount.h> #include <linux/module.h> #include <linux/slab.h> +#include <linux/err.h> #include <linux/configfs.h> #include "configfs_internal.h" DECLARE_RWSEM(configfs_rename_sem); +/* + * Protects mutations of configfs_dirent linkage together with proper i_mutex + * Also protects mutations of symlinks linkage to target configfs_dirent + * Mutators of configfs_dirent linkage must *both* have the proper inode locked + * and configfs_dirent_lock locked, in that order. + * This allows one to safely traverse configfs_dirent trees and symlinks without + * having to lock inodes. + * + * Protects setting of CONFIGFS_USET_DROPPING: checking the flag + * unlocked is not reliable unless in detach_groups() called from + * rmdir()/unregister() and from configfs_attach_group() + */ +DEFINE_SPINLOCK(configfs_dirent_lock); static void configfs_d_iput(struct dentry * dentry, struct inode * inode) @@ -74,13 +88,20 @@ static struct configfs_dirent *configfs_new_dirent(struct configfs_dirent * pare sd = kmem_cache_zalloc(configfs_dir_cachep, GFP_KERNEL); if (!sd) - return NULL; + return ERR_PTR(-ENOMEM); atomic_set(&sd->s_count, 1); INIT_LIST_HEAD(&sd->s_links); INIT_LIST_HEAD(&sd->s_children); - list_add(&sd->s_sibling, &parent_sd->s_children); sd->s_element = element; + spin_lock(&configfs_dirent_lock); + if (parent_sd->s_type & CONFIGFS_USET_DROPPING) { + spin_unlock(&configfs_dirent_lock); + kmem_cache_free(configfs_dir_cachep, sd); + return ERR_PTR(-ENOENT); + } + list_add(&sd->s_sibling, &parent_sd->s_children); + spin_unlock(&configfs_dirent_lock); return sd; } @@ -118,8 +139,8 @@ int configfs_make_dirent(struct configfs_dirent * parent_sd, struct configfs_dirent * sd; sd = configfs_new_dirent(parent_sd, element); - if (!sd) - return -ENOMEM; + if (IS_ERR(sd)) + return PTR_ERR(sd); sd->s_mode = mode; sd->s_type = type; @@ -164,7 +185,7 @@ static int create_dir(struct config_item * k, struct dentry * p, error = configfs_dirent_exists(p->d_fsdata, d->d_name.name); if (!error) error = configfs_make_dirent(p->d_fsdata, d, k, mode, - CONFIGFS_DIR); + CONFIGFS_DIR | CONFIGFS_USET_CREATING); if (!error) { error = configfs_create(d, mode, init_dir); if (!error) { @@ -173,7 +194,9 @@ static int create_dir(struct config_item * k, struct dentry * p, } else { struct configfs_dirent *sd = d->d_fsdata; if (sd) { + spin_lock(&configfs_dirent_lock); list_del_init(&sd->s_sibling); + spin_unlock(&configfs_dirent_lock); configfs_put(sd); } } @@ -186,6 +209,9 @@ static int create_dir(struct config_item * k, struct dentry * p, * configfs_create_dir - create a directory for an config_item. * @item: config_itemwe're creating directory for. * @dentry: config_item's dentry. + * + * Note: user-created entries won't be allowed under this new directory + * until it is validated by configfs_dir_set_ready() */ static int configfs_create_dir(struct config_item * item, struct dentry *dentry) @@ -208,6 +234,44 @@ static int configfs_create_dir(struct config_item * item, struct dentry *dentry) return error; } +/* + * Allow userspace to create new entries under a new directory created with + * configfs_create_dir(), and under all of its chidlren directories recursively. + * @sd configfs_dirent of the new directory to validate + * + * Caller must hold configfs_dirent_lock. + */ +static void configfs_dir_set_ready(struct configfs_dirent *sd) +{ + struct configfs_dirent *child_sd; + + sd->s_type &= ~CONFIGFS_USET_CREATING; + list_for_each_entry(child_sd, &sd->s_children, s_sibling) + if (child_sd->s_type & CONFIGFS_USET_CREATING) + configfs_dir_set_ready(child_sd); +} + +/* + * Check that a directory does not belong to a directory hierarchy being + * attached and not validated yet. + * @sd configfs_dirent of the directory to check + * + * @return non-zero iff the directory was validated + * + * Note: takes configfs_dirent_lock, so the result may change from false to true + * in two consecutive calls, but never from true to false. + */ +int configfs_dirent_is_ready(struct configfs_dirent *sd) +{ + int ret; + + spin_lock(&configfs_dirent_lock); + ret = !(sd->s_type & CONFIGFS_USET_CREATING); + spin_unlock(&configfs_dirent_lock); + + return ret; +} + int configfs_create_link(struct configfs_symlink *sl, struct dentry *parent, struct dentry *dentry) @@ -224,7 +288,9 @@ int configfs_create_link(struct configfs_symlink *sl, else { struct configfs_dirent *sd = dentry->d_fsdata; if (sd) { + spin_lock(&configfs_dirent_lock); list_del_init(&sd->s_sibling); + spin_unlock(&configfs_dirent_lock); configfs_put(sd); } } @@ -238,7 +304,9 @@ static void remove_dir(struct dentry * d) struct configfs_dirent * sd; sd = d->d_fsdata; + spin_lock(&configfs_dirent_lock); list_del_init(&sd->s_sibling); + spin_unlock(&configfs_dirent_lock); configfs_put(sd); if (d->d_inode) simple_rmdir(parent->d_inode,d); @@ -256,6 +324,8 @@ static void remove_dir(struct dentry * d) * The only thing special about this is that we remove any files in * the directory before we remove the directory, and we've inlined * what used to be configfs_rmdir() below, instead of calling separately. + * + * Caller holds the mutex of the item's inode */ static void configfs_remove_dir(struct config_item * item) @@ -303,7 +373,19 @@ static struct dentry * configfs_lookup(struct inode *dir, struct configfs_dirent * parent_sd = dentry->d_parent->d_fsdata; struct configfs_dirent * sd; int found = 0; - int err = 0; + int err; + + /* + * Fake invisibility if dir belongs to a group/default groups hierarchy + * being attached + * + * This forbids userspace to read/write attributes of items which may + * not complete their initialization, since the dentries of the + * attributes won't be instantiated. + */ + err = -ENOENT; + if (!configfs_dirent_is_ready(parent_sd)) + goto out; list_for_each_entry(sd, &parent_sd->s_children, s_sibling) { if (sd->s_type & CONFIGFS_NOT_PINNED) { @@ -326,41 +408,49 @@ static struct dentry * configfs_lookup(struct inode *dir, return simple_lookup(dir, dentry, nd); } +out: return ERR_PTR(err); } /* * Only subdirectories count here. Files (CONFIGFS_NOT_PINNED) are - * attributes and are removed by rmdir(). We recurse, taking i_mutex - * on all children that are candidates for default detach. If the - * result is clean, then configfs_detach_group() will handle dropping - * i_mutex. If there is an error, the caller will clean up the i_mutex - * holders via configfs_detach_rollback(). + * attributes and are removed by rmdir(). We recurse, setting + * CONFIGFS_USET_DROPPING on all children that are candidates for + * default detach. + * If there is an error, the caller will reset the flags via + * configfs_detach_rollback(). */ -static int configfs_detach_prep(struct dentry *dentry) +static int configfs_detach_prep(struct dentry *dentry, struct mutex **wait_mutex) { struct configfs_dirent *parent_sd = dentry->d_fsdata; struct configfs_dirent *sd; int ret; + /* Mark that we're trying to drop the group */ + parent_sd->s_type |= CONFIGFS_USET_DROPPING; + ret = -EBUSY; if (!list_empty(&parent_sd->s_links)) goto out; ret = 0; list_for_each_entry(sd, &parent_sd->s_children, s_sibling) { - if (sd->s_type & CONFIGFS_NOT_PINNED) + if (!sd->s_element || + (sd->s_type & CONFIGFS_NOT_PINNED)) continue; if (sd->s_type & CONFIGFS_USET_DEFAULT) { - mutex_lock(&sd->s_dentry->d_inode->i_mutex); - /* Mark that we've taken i_mutex */ - sd->s_type |= CONFIGFS_USET_DROPPING; + /* Abort if racing with mkdir() */ + if (sd->s_type & CONFIGFS_USET_IN_MKDIR) { + if (wait_mutex) + *wait_mutex = &sd->s_dentry->d_inode->i_mutex; + return -EAGAIN; + } /* * Yup, recursive. If there's a problem, blame * deep nesting of default_groups */ - ret = configfs_detach_prep(sd->s_dentry); + ret = configfs_detach_prep(sd->s_dentry, wait_mutex); if (!ret) continue; } else @@ -374,7 +464,7 @@ out: } /* - * Walk the tree, dropping i_mutex wherever CONFIGFS_USET_DROPPING is + * Walk the tree, resetting CONFIGFS_USET_DROPPING wherever it was * set. */ static void configfs_detach_rollback(struct dentry *dentry) @@ -382,16 +472,11 @@ static void configfs_detach_rollback(struct dentry *dentry) struct configfs_dirent *parent_sd = dentry->d_fsdata; struct configfs_dirent *sd; - list_for_each_entry(sd, &parent_sd->s_children, s_sibling) { - if (sd->s_type & CONFIGFS_USET_DEFAULT) { - configfs_detach_rollback(sd->s_dentry); + parent_sd->s_type &= ~CONFIGFS_USET_DROPPING; - if (sd->s_type & CONFIGFS_USET_DROPPING) { - sd->s_type &= ~CONFIGFS_USET_DROPPING; - mutex_unlock(&sd->s_dentry->d_inode->i_mutex); - } - } - } + list_for_each_entry(sd, &parent_sd->s_children, s_sibling) + if (sd->s_type & CONFIGFS_USET_DEFAULT) + configfs_detach_rollback(sd->s_dentry); } static void detach_attrs(struct config_item * item) @@ -410,7 +495,9 @@ static void detach_attrs(struct config_item * item) list_for_each_entry_safe(sd, tmp, &parent_sd->s_children, s_sibling) { if (!sd->s_element || !(sd->s_type & CONFIGFS_NOT_PINNED)) continue; + spin_lock(&configfs_dirent_lock); list_del_init(&sd->s_sibling); + spin_unlock(&configfs_dirent_lock); configfs_drop_dentry(sd, dentry); configfs_put(sd); } @@ -466,16 +553,12 @@ static void detach_groups(struct config_group *group) child = sd->s_dentry; + mutex_lock(&child->d_inode->i_mutex); + configfs_detach_group(sd->s_element); child->d_inode->i_flags |= S_DEAD; - /* - * From rmdir/unregister, a configfs_detach_prep() pass - * has taken our i_mutex for us. Drop it. - * From mkdir/register cleanup, there is no sem held. - */ - if (sd->s_type & CONFIGFS_USET_DROPPING) - mutex_unlock(&child->d_inode->i_mutex); + mutex_unlock(&child->d_inode->i_mutex); d_delete(child); dput(child); @@ -532,36 +615,21 @@ static int create_default_group(struct config_group *parent_group, static int populate_groups(struct config_group *group) { struct config_group *new_group; - struct dentry *dentry = group->cg_item.ci_dentry; int ret = 0; int i; if (group->default_groups) { - /* - * FYI, we're faking mkdir here - * I'm not sure we need this semaphore, as we're called - * from our parent's mkdir. That holds our parent's - * i_mutex, so afaik lookup cannot continue through our - * parent to find us, let alone mess with our tree. - * That said, taking our i_mutex is closer to mkdir - * emulation, and shouldn't hurt. - */ - mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD); - for (i = 0; group->default_groups[i]; i++) { new_group = group->default_groups[i]; ret = create_default_group(group, new_group); - if (ret) + if (ret) { + detach_groups(group); break; + } } - - mutex_unlock(&dentry->d_inode->i_mutex); } - if (ret) - detach_groups(group); - return ret; } @@ -676,7 +744,15 @@ static int configfs_attach_item(struct config_item *parent_item, if (!ret) { ret = populate_attrs(item); if (ret) { + /* + * We are going to remove an inode and its dentry but + * the VFS may already have hit and used them. Thus, + * we must lock them as rmdir() would. + */ + mutex_lock(&dentry->d_inode->i_mutex); configfs_remove_dir(item); + dentry->d_inode->i_flags |= S_DEAD; + mutex_unlock(&dentry->d_inode->i_mutex); d_delete(dentry); } } @@ -684,6 +760,7 @@ static int configfs_attach_item(struct config_item *parent_item, return ret; } +/* Caller holds the mutex of the item's inode */ static void configfs_detach_item(struct config_item *item) { detach_attrs(item); @@ -702,16 +779,30 @@ static int configfs_attach_group(struct config_item *parent_item, sd = dentry->d_fsdata; sd->s_type |= CONFIGFS_USET_DIR; + /* + * FYI, we're faking mkdir in populate_groups() + * We must lock the group's inode to avoid races with the VFS + * which can already hit the inode and try to add/remove entries + * under it. + * + * We must also lock the inode to remove it safely in case of + * error, as rmdir() would. + */ + mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD); ret = populate_groups(to_config_group(item)); if (ret) { configfs_detach_item(item); - d_delete(dentry); + dentry->d_inode->i_flags |= S_DEAD; } + mutex_unlock(&dentry->d_inode->i_mutex); + if (ret) + d_delete(dentry); } return ret; } +/* Caller holds the mutex of the group's inode */ static void configfs_detach_group(struct config_item *item) { detach_groups(to_config_group(item)); @@ -1001,14 +1092,15 @@ EXPORT_SYMBOL(configfs_undepend_item); static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) { - int ret, module_got = 0; - struct config_group *group; - struct config_item *item; + int ret = 0; + int module_got = 0; + struct config_group *group = NULL; + struct config_item *item = NULL; struct config_item *parent_item; struct configfs_subsystem *subsys; struct configfs_dirent *sd; struct config_item_type *type; - struct module *owner = NULL; + struct module *subsys_owner = NULL, *new_item_owner = NULL; char *name; if (dentry->d_parent == configfs_sb->s_root) { @@ -1017,6 +1109,16 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) } sd = dentry->d_parent->d_fsdata; + + /* + * Fake invisibility if dir belongs to a group/default groups hierarchy + * being attached + */ + if (!configfs_dirent_is_ready(sd)) { + ret = -ENOENT; + goto out; + } + if (!(sd->s_type & CONFIGFS_USET_DIR)) { ret = -EPERM; goto out; @@ -1035,38 +1137,57 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) goto out_put; } + /* + * The subsystem may belong to a different module than the item + * being created. We don't want to safely pin the new item but + * fail to pin the subsystem it sits under. + */ + if (!subsys->su_group.cg_item.ci_type) { + ret = -EINVAL; + goto out_put; + } + subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner; + if (!try_module_get(subsys_owner)) { + ret = -EINVAL; + goto out_put; + } + name = kmalloc(dentry->d_name.len + 1, GFP_KERNEL); if (!name) { ret = -ENOMEM; - goto out_put; + goto out_subsys_put; } snprintf(name, dentry->d_name.len + 1, "%s", dentry->d_name.name); mutex_lock(&subsys->su_mutex); - group = NULL; - item = NULL; if (type->ct_group_ops->make_group) { group = type->ct_group_ops->make_group(to_config_group(parent_item), name); - if (group) { + if (!group) + group = ERR_PTR(-ENOMEM); + if (!IS_ERR(group)) { link_group(to_config_group(parent_item), group); item = &group->cg_item; - } + } else + ret = PTR_ERR(group); } else { item = type->ct_group_ops->make_item(to_config_group(parent_item), name); - if (item) + if (!item) + item = ERR_PTR(-ENOMEM); + if (!IS_ERR(item)) link_obj(parent_item, item); + else + ret = PTR_ERR(item); } mutex_unlock(&subsys->su_mutex); kfree(name); - if (!item) { + if (ret) { /* - * If item == NULL, then link_obj() was never called. + * If ret != 0, then link_obj() was never called. * There are no extra references to clean up. */ - ret = -ENOMEM; - goto out_put; + goto out_subsys_put; } /* @@ -1080,8 +1201,8 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) goto out_unlink; } - owner = type->ct_owner; - if (!try_module_get(owner)) { + new_item_owner = type->ct_owner; + if (!try_module_get(new_item_owner)) { ret = -EINVAL; goto out_unlink; } @@ -1093,11 +1214,28 @@ static int configfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) */ module_got = 1; + /* + * Make racing rmdir() fail if it did not tag parent with + * CONFIGFS_USET_DROPPING + * Note: if CONFIGFS_USET_DROPPING is already set, attach_group() will + * fail and let rmdir() terminate correctly + */ + spin_lock(&configfs_dirent_lock); + /* This will make configfs_detach_prep() fail */ + sd->s_type |= CONFIGFS_USET_IN_MKDIR; + spin_unlock(&configfs_dirent_lock); + if (group) ret = configfs_attach_group(parent_item, item, dentry); else ret = configfs_attach_item(parent_item, item, dentry); + spin_lock(&configfs_dirent_lock); + sd->s_type &= ~CONFIGFS_USET_IN_MKDIR; + if (!ret) + configfs_dir_set_ready(dentry->d_fsdata); + spin_unlock(&configfs_dirent_lock); + out_unlink: if (ret) { /* Tear down everything we built up */ @@ -1113,9 +1251,13 @@ out_unlink: mutex_unlock(&subsys->su_mutex); if (module_got) - module_put(owner); + module_put(new_item_owner); } +out_subsys_put: + if (ret) + module_put(subsys_owner); + out_put: /* * link_obj()/link_group() took a reference from child->parent, @@ -1134,7 +1276,7 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry) struct config_item *item; struct configfs_subsystem *subsys; struct configfs_dirent *sd; - struct module *owner = NULL; + struct module *subsys_owner = NULL, *dead_item_owner = NULL; int ret; if (dentry->d_parent == configfs_sb->s_root) @@ -1161,12 +1303,36 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry) return -EINVAL; } - ret = configfs_detach_prep(dentry); - if (ret) { - configfs_detach_rollback(dentry); - config_item_put(parent_item); - return ret; - } + /* configfs_mkdir() shouldn't have allowed this */ + BUG_ON(!subsys->su_group.cg_item.ci_type); + subsys_owner = subsys->su_group.cg_item.ci_type->ct_owner; + + /* + * Ensure that no racing symlink() will make detach_prep() fail while + * the new link is temporarily attached + */ + do { + struct mutex *wait_mutex; + + mutex_lock(&configfs_symlink_mutex); + spin_lock(&configfs_dirent_lock); + ret = configfs_detach_prep(dentry, &wait_mutex); + if (ret) + configfs_detach_rollback(dentry); + spin_unlock(&configfs_dirent_lock); + mutex_unlock(&configfs_symlink_mutex); + + if (ret) { + if (ret != -EAGAIN) { + config_item_put(parent_item); + return ret; + } + + /* Wait until the racing operation terminates */ + mutex_lock(wait_mutex); + mutex_unlock(wait_mutex); + } + } while (ret == -EAGAIN); /* Get a working ref for the duration of this function */ item = configfs_get_config_item(dentry); @@ -1175,7 +1341,7 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry) config_item_put(parent_item); if (item->ci_type) - owner = item->ci_type->ct_owner; + dead_item_owner = item->ci_type->ct_owner; if (sd->s_type & CONFIGFS_USET_DIR) { configfs_detach_group(item); @@ -1197,7 +1363,8 @@ static int configfs_rmdir(struct inode *dir, struct dentry *dentry) /* Drop our reference from above */ config_item_put(item); - module_put(owner); + module_put(dead_item_owner); + module_put(subsys_owner); return 0; } @@ -1253,13 +1420,24 @@ static int configfs_dir_open(struct inode *inode, struct file *file) { struct dentry * dentry = file->f_path.dentry; struct configfs_dirent * parent_sd = dentry->d_fsdata; + int err; mutex_lock(&dentry->d_inode->i_mutex); - file->private_data = configfs_new_dirent(parent_sd, NULL); + /* + * Fake invisibility if dir belongs to a group/default groups hierarchy + * being attached + */ + err = -ENOENT; + if (configfs_dirent_is_ready(parent_sd)) { + file->private_data = configfs_new_dirent(parent_sd, NULL); + if (IS_ERR(file->private_data)) + err = PTR_ERR(file->private_data); + else + err = 0; + } mutex_unlock(&dentry->d_inode->i_mutex); - return file->private_data ? 0 : -ENOMEM; - + return err; } static int configfs_dir_close(struct inode *inode, struct file *file) @@ -1268,7 +1446,9 @@ static int configfs_dir_close(struct inode *inode, struct file *file) struct configfs_dirent * cursor = file->private_data; mutex_lock(&dentry->d_inode->i_mutex); + spin_lock(&configfs_dirent_lock); list_del_init(&cursor->s_sibling); + spin_unlock(&configfs_dirent_lock); mutex_unlock(&dentry->d_inode->i_mutex); release_configfs_dirent(cursor); @@ -1308,7 +1488,9 @@ static int configfs_readdir(struct file * filp, void * dirent, filldir_t filldir /* fallthrough */ default: if (filp->f_pos == 2) { + spin_lock(&configfs_dirent_lock); list_move(q, &parent_sd->s_children); + spin_unlock(&configfs_dirent_lock); } for (p=q->next; p!= &parent_sd->s_children; p=p->next) { struct configfs_dirent *next; @@ -1331,7 +1513,9 @@ static int configfs_readdir(struct file * filp, void * dirent, filldir_t filldir dt_type(next)) < 0) return 0; + spin_lock(&configfs_dirent_lock); list_move(q, p); + spin_unlock(&configfs_dirent_lock); p = q; filp->f_pos++; } @@ -1362,6 +1546,7 @@ static loff_t configfs_dir_lseek(struct file * file, loff_t offset, int origin) struct list_head *p; loff_t n = file->f_pos - 2; + spin_lock(&configfs_dirent_lock); list_del(&cursor->s_sibling); p = sd->s_children.next; while (n && p != &sd->s_children) { @@ -1373,6 +1558,7 @@ static loff_t configfs_dir_lseek(struct file * file, loff_t offset, int origin) p = p->next; } list_add_tail(&cursor->s_sibling, p); + spin_unlock(&configfs_dirent_lock); } } mutex_unlock(&dentry->d_inode->i_mutex); @@ -1422,6 +1608,10 @@ int configfs_register_subsystem(struct configfs_subsystem *subsys) if (err) { d_delete(dentry); dput(dentry); + } else { + spin_lock(&configfs_dirent_lock); + configfs_dir_set_ready(dentry->d_fsdata); + spin_unlock(&configfs_dirent_lock); } } @@ -1448,9 +1638,13 @@ void configfs_unregister_subsystem(struct configfs_subsystem *subsys) mutex_lock_nested(&configfs_sb->s_root->d_inode->i_mutex, I_MUTEX_PARENT); mutex_lock_nested(&dentry->d_inode->i_mutex, I_MUTEX_CHILD); - if (configfs_detach_prep(dentry)) { + mutex_lock(&configfs_symlink_mutex); + spin_lock(&configfs_dirent_lock); + if (configfs_detach_prep(dentry, NULL)) { printk(KERN_ERR "configfs: Tried to unregister non-empty subsystem!\n"); } + spin_unlock(&configfs_dirent_lock); + mutex_unlock(&configfs_symlink_mutex); configfs_detach_group(&group->cg_item); dentry->d_inode->i_flags |= S_DEAD; mutex_unlock(&dentry->d_inode->i_mutex); |