fanotify: divorce fanotify_path_event and fanotify_fid_event
authorJan Kara <jack@suse.cz>
Tue, 24 Mar 2020 16:04:20 +0000 (17:04 +0100)
committerJan Kara <jack@suse.cz>
Wed, 25 Mar 2020 09:27:16 +0000 (10:27 +0100)
Breakup the union and make them both inherit from abstract fanotify_event.

fanotify_path_event, fanotify_fid_event and fanotify_perm_event inherit
from fanotify_event.

type field in abstract fanotify_event determines the concrete event type.

fanotify_path_event, fanotify_fid_event and fanotify_perm_event are
allocated from separate memcache pools.

Rename fanotify_perm_event casting macro to FANOTIFY_PERM(), so that
FANOTIFY_PE() and FANOTIFY_FE() can be used as casting macros to
fanotify_path_event and fanotify_fid_event.

[JK: Cleanup FANOTIFY_PE() and FANOTIFY_FE() to be proper inline
functions and remove requirement that fanotify_event is the first in
event structures]

Link: https://lore.kernel.org/r/20200319151022.31456-11-amir73il@gmail.com
Suggested-by: Jan Kara <jack@suse.cz>
Signed-off-by: Amir Goldstein <amir73il@gmail.com>
Signed-off-by: Jan Kara <jack@suse.cz>
fs/notify/fanotify/fanotify.c
fs/notify/fanotify/fanotify.h
fs/notify/fanotify/fanotify_user.c

index 64b05be..39eb71f 100644 (file)
@@ -29,7 +29,7 @@ static inline bool fanotify_fsid_equal(__kernel_fsid_t *fsid1,
 }
 
 static bool fanotify_fh_equal(struct fanotify_fh *fh1,
-                            struct fanotify_fh *fh2)
+                             struct fanotify_fh *fh2)
 {
        if (fh1->type != fh2->type || fh1->len != fh2->len)
                return false;
@@ -42,6 +42,17 @@ static bool fanotify_fh_equal(struct fanotify_fh *fh1,
                !memcmp(fanotify_fh_buf(fh1), fanotify_fh_buf(fh2), fh1->len);
 }
 
+static bool fanotify_fid_event_equal(struct fanotify_fid_event *ffe1,
+                                    struct fanotify_fid_event *ffe2)
+{
+       /* Do not merge fid events without object fh */
+       if (!ffe1->object_fh.len)
+               return false;
+
+       return fanotify_fsid_equal(&ffe1->fsid, &ffe2->fsid) &&
+               fanotify_fh_equal(&ffe1->object_fh, &ffe2->object_fh);
+}
+
 static bool should_merge(struct fsnotify_event *old_fsn,
                         struct fsnotify_event *new_fsn)
 {
@@ -51,14 +62,15 @@ static bool should_merge(struct fsnotify_event *old_fsn,
        old = FANOTIFY_E(old_fsn);
        new = FANOTIFY_E(new_fsn);
 
-       if (old_fsn->objectid != new_fsn->objectid || old->pid != new->pid ||
-           old->fh.type != new->fh.type)
+       if (old_fsn->objectid != new_fsn->objectid ||
+           old->type != new->type || old->pid != new->pid)
                return false;
 
-       if (fanotify_event_has_path(old)) {
+       switch (old->type) {
+       case FANOTIFY_EVENT_TYPE_PATH:
                return fanotify_path_equal(fanotify_event_path(old),
                                           fanotify_event_path(new));
-       } else if (fanotify_event_has_fid(old)) {
+       case FANOTIFY_EVENT_TYPE_FID:
                /*
                 * We want to merge many dirent events in the same dir (i.e.
                 * creates/unlinks/renames), but we do not want to merge dirent
@@ -70,11 +82,12 @@ static bool should_merge(struct fsnotify_event *old_fsn,
                if ((old->mask & FS_ISDIR) != (new->mask & FS_ISDIR))
                        return false;
 
-               return fanotify_fsid_equal(&old->fsid, &new->fsid) &&
-                       fanotify_fh_equal(&old->fh, &new->fh);
+               return fanotify_fid_event_equal(FANOTIFY_FE(old),
+                                               FANOTIFY_FE(new));
+       default:
+               WARN_ON_ONCE(1);
        }
 
-       /* Do not merge events if we failed to encode fid */
        return false;
 }
 
@@ -310,6 +323,7 @@ struct fanotify_event *fanotify_alloc_event(struct fsnotify_group *group,
                                            __kernel_fsid_t *fsid)
 {
        struct fanotify_event *event = NULL;
+       struct fanotify_fid_event *ffe = NULL;
        gfp_t gfp = GFP_KERNEL_ACCOUNT;
        struct inode *id = fanotify_fid_inode(inode, mask, data, data_type);
        const struct path *path = fsnotify_data_path(data, data_type);
@@ -334,14 +348,32 @@ struct fanotify_event *fanotify_alloc_event(struct fsnotify_group *group,
                pevent = kmem_cache_alloc(fanotify_perm_event_cachep, gfp);
                if (!pevent)
                        goto out;
+
                event = &pevent->fae;
+               event->type = FANOTIFY_EVENT_TYPE_PATH_PERM;
                pevent->response = 0;
                pevent->state = FAN_EVENT_INIT;
                goto init;
        }
-       event = kmem_cache_alloc(fanotify_event_cachep, gfp);
-       if (!event)
-               goto out;
+
+       if (FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
+               ffe = kmem_cache_alloc(fanotify_fid_event_cachep, gfp);
+               if (!ffe)
+                       goto out;
+
+               event = &ffe->fae;
+               event->type = FANOTIFY_EVENT_TYPE_FID;
+       } else {
+               struct fanotify_path_event *pevent;
+
+               pevent = kmem_cache_alloc(fanotify_path_event_cachep, gfp);
+               if (!pevent)
+                       goto out;
+
+               event = &pevent->fae;
+               event->type = FANOTIFY_EVENT_TYPE_PATH;
+       }
+
 init: __maybe_unused
        /*
         * Use the victim inode instead of the watching inode as the id for
@@ -354,19 +386,23 @@ init: __maybe_unused
                event->pid = get_pid(task_pid(current));
        else
                event->pid = get_pid(task_tgid(current));
-       event->fh.len = 0;
-       if (id && FAN_GROUP_FLAG(group, FAN_REPORT_FID)) {
-               event->fsid = *fsid;
+
+       if (fanotify_event_has_fid(event)) {
+               ffe->object_fh.len = 0;
+               if (fsid)
+                       ffe->fsid = *fsid;
                if (id)
-                       fanotify_encode_fh(&event->fh, id, gfp);
-       } else if (path) {
-               event->fh.type = FILEID_ROOT;
-               event->path = *path;
-               path_get(path);
-       } else {
-               event->fh.type = FILEID_INVALID;
-               event->path.mnt = NULL;
-               event->path.dentry = NULL;
+                       fanotify_encode_fh(&ffe->object_fh, id, gfp);
+       } else if (fanotify_event_has_path(event)) {
+               struct path *p = fanotify_event_path(event);
+
+               if (path) {
+                       *p = *path;
+                       path_get(path);
+               } else {
+                       p->mnt = NULL;
+                       p->dentry = NULL;
+               }
        }
 out:
        memalloc_unuse_memcg();
@@ -486,7 +522,7 @@ static int fanotify_handle_event(struct fsnotify_group *group,
 
                ret = 0;
        } else if (fanotify_is_perm_event(mask)) {
-               ret = fanotify_get_response(group, FANOTIFY_PE(fsn_event),
+               ret = fanotify_get_response(group, FANOTIFY_PERM(event),
                                            iter_info);
        }
 finish:
@@ -505,22 +541,46 @@ static void fanotify_free_group_priv(struct fsnotify_group *group)
        free_uid(user);
 }
 
+static void fanotify_free_path_event(struct fanotify_event *event)
+{
+       path_put(fanotify_event_path(event));
+       kmem_cache_free(fanotify_path_event_cachep, FANOTIFY_PE(event));
+}
+
+static void fanotify_free_perm_event(struct fanotify_event *event)
+{
+       path_put(fanotify_event_path(event));
+       kmem_cache_free(fanotify_perm_event_cachep, FANOTIFY_PERM(event));
+}
+
+static void fanotify_free_fid_event(struct fanotify_event *event)
+{
+       struct fanotify_fid_event *ffe = FANOTIFY_FE(event);
+
+       if (fanotify_fh_has_ext_buf(&ffe->object_fh))
+               kfree(fanotify_fh_ext_buf(&ffe->object_fh));
+       kmem_cache_free(fanotify_fid_event_cachep, ffe);
+}
+
 static void fanotify_free_event(struct fsnotify_event *fsn_event)
 {
        struct fanotify_event *event;
 
        event = FANOTIFY_E(fsn_event);
-       if (fanotify_event_has_path(event))
-               path_put(&event->path);
-       else if (fanotify_fh_has_ext_buf(&event->fh))
-               kfree(fanotify_fh_ext_buf(&event->fh));
        put_pid(event->pid);
-       if (fanotify_is_perm_event(event->mask)) {
-               kmem_cache_free(fanotify_perm_event_cachep,
-                               FANOTIFY_PE(fsn_event));
-               return;
+       switch (event->type) {
+       case FANOTIFY_EVENT_TYPE_PATH:
+               fanotify_free_path_event(event);
+               break;
+       case FANOTIFY_EVENT_TYPE_PATH_PERM:
+               fanotify_free_perm_event(event);
+               break;
+       case FANOTIFY_EVENT_TYPE_FID:
+               fanotify_free_fid_event(event);
+               break;
+       default:
+               WARN_ON_ONCE(1);
        }
-       kmem_cache_free(fanotify_event_cachep, event);
 }
 
 static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
index f9da448..3b50ee4 100644 (file)
@@ -5,7 +5,8 @@
 #include <linux/exportfs.h>
 
 extern struct kmem_cache *fanotify_mark_cache;
-extern struct kmem_cache *fanotify_event_cachep;
+extern struct kmem_cache *fanotify_fid_event_cachep;
+extern struct kmem_cache *fanotify_path_event_cachep;
 extern struct kmem_cache *fanotify_perm_event_cachep;
 
 /* Possible states of the permission event */
@@ -52,43 +53,45 @@ static inline void *fanotify_fh_buf(struct fanotify_fh *fh)
 }
 
 /*
- * Structure for normal fanotify events. It gets allocated in
+ * Common structure for fanotify events. Concrete structs are allocated in
  * fanotify_handle_event() and freed when the information is retrieved by
- * userspace
+ * userspace. The type of event determines how it was allocated, how it will
+ * be freed and which concrete struct it may be cast to.
  */
+enum fanotify_event_type {
+       FANOTIFY_EVENT_TYPE_FID,
+       FANOTIFY_EVENT_TYPE_PATH,
+       FANOTIFY_EVENT_TYPE_PATH_PERM,
+};
+
 struct fanotify_event {
        struct fsnotify_event fse;
        u32 mask;
-       /*
-        * With FAN_REPORT_FID, we do not hold any reference on the
-        * victim object. Instead we store its NFS file handle and its
-        * filesystem's fsid as a unique identifier.
-        */
-       __kernel_fsid_t fsid;
-       struct fanotify_fh fh;
-       /*
-        * We hold ref to this path so it may be dereferenced at any
-        * point during this object's lifetime
-        */
-       struct path path;
+       enum fanotify_event_type type;
        struct pid *pid;
 };
 
-static inline bool fanotify_event_has_path(struct fanotify_event *event)
+struct fanotify_fid_event {
+       struct fanotify_event fae;
+       __kernel_fsid_t fsid;
+       struct fanotify_fh object_fh;
+};
+
+static inline struct fanotify_fid_event *
+FANOTIFY_FE(struct fanotify_event *event)
 {
-       return event->fh.type == FILEID_ROOT;
+       return container_of(event, struct fanotify_fid_event, fae);
 }
 
 static inline bool fanotify_event_has_fid(struct fanotify_event *event)
 {
-       return event->fh.type != FILEID_ROOT &&
-              event->fh.type != FILEID_INVALID;
+       return event->type == FANOTIFY_EVENT_TYPE_FID;
 }
 
 static inline __kernel_fsid_t *fanotify_event_fsid(struct fanotify_event *event)
 {
-       if (fanotify_event_has_fid(event))
-               return &event->fsid;
+       if (event->type == FANOTIFY_EVENT_TYPE_FID)
+               return &FANOTIFY_FE(event)->fsid;
        else
                return NULL;
 }
@@ -96,8 +99,8 @@ static inline __kernel_fsid_t *fanotify_event_fsid(struct fanotify_event *event)
 static inline struct fanotify_fh *fanotify_event_object_fh(
                                                struct fanotify_event *event)
 {
-       if (fanotify_event_has_fid(event))
-               return &event->fh;
+       if (event->type == FANOTIFY_EVENT_TYPE_FID)
+               return &FANOTIFY_FE(event)->object_fh;
        else
                return NULL;
 }
@@ -109,6 +112,17 @@ static inline int fanotify_event_object_fh_len(struct fanotify_event *event)
        return fh ? fh->len : 0;
 }
 
+struct fanotify_path_event {
+       struct fanotify_event fae;
+       struct path path;
+};
+
+static inline struct fanotify_path_event *
+FANOTIFY_PE(struct fanotify_event *event)
+{
+       return container_of(event, struct fanotify_path_event, fae);
+}
+
 /*
  * Structure for permission fanotify events. It gets allocated and freed in
  * fanotify_handle_event() since we wait there for user response. When the
@@ -118,15 +132,16 @@ static inline int fanotify_event_object_fh_len(struct fanotify_event *event)
  */
 struct fanotify_perm_event {
        struct fanotify_event fae;
+       struct path path;
        unsigned short response;        /* userspace answer to the event */
        unsigned short state;           /* state of the event */
        int fd;         /* fd we passed to userspace for this event */
 };
 
 static inline struct fanotify_perm_event *
-FANOTIFY_PE(struct fsnotify_event *fse)
+FANOTIFY_PERM(struct fanotify_event *event)
 {
-       return container_of(fse, struct fanotify_perm_event, fae.fse);
+       return container_of(event, struct fanotify_perm_event, fae);
 }
 
 static inline bool fanotify_is_perm_event(u32 mask)
@@ -140,10 +155,18 @@ static inline struct fanotify_event *FANOTIFY_E(struct fsnotify_event *fse)
        return container_of(fse, struct fanotify_event, fse);
 }
 
+static inline bool fanotify_event_has_path(struct fanotify_event *event)
+{
+       return event->type == FANOTIFY_EVENT_TYPE_PATH ||
+               event->type == FANOTIFY_EVENT_TYPE_PATH_PERM;
+}
+
 static inline struct path *fanotify_event_path(struct fanotify_event *event)
 {
-       if (fanotify_event_has_path(event))
-               return &event->path;
+       if (event->type == FANOTIFY_EVENT_TYPE_PATH)
+               return &FANOTIFY_PE(event)->path;
+       else if (event->type == FANOTIFY_EVENT_TYPE_PATH_PERM)
+               return &FANOTIFY_PERM(event)->path;
        else
                return NULL;
 }
index 0b3b74f..6cb94a6 100644 (file)
@@ -46,7 +46,8 @@
 extern const struct fsnotify_ops fanotify_fsnotify_ops;
 
 struct kmem_cache *fanotify_mark_cache __read_mostly;
-struct kmem_cache *fanotify_event_cachep __read_mostly;
+struct kmem_cache *fanotify_fid_event_cachep __read_mostly;
+struct kmem_cache *fanotify_path_event_cachep __read_mostly;
 struct kmem_cache *fanotify_perm_event_cachep __read_mostly;
 
 #define FANOTIFY_EVENT_ALIGN 4
@@ -64,16 +65,16 @@ static int fanotify_event_info_len(struct fanotify_event *event)
 }
 
 /*
- * Get an fsnotify notification event if one exists and is small
+ * Get an fanotify notification event if one exists and is small
  * enough to fit in "count". Return an error pointer if the count
  * is not large enough. When permission event is dequeued, its state is
  * updated accordingly.
  */
-static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
+static struct fanotify_event *get_one_event(struct fsnotify_group *group,
                                            size_t count)
 {
        size_t event_size = FAN_EVENT_METADATA_LEN;
-       struct fsnotify_event *fsn_event = NULL;
+       struct fanotify_event *event = NULL;
 
        pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
 
@@ -87,15 +88,15 @@ static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
        }
 
        if (event_size > count) {
-               fsn_event = ERR_PTR(-EINVAL);
+               event = ERR_PTR(-EINVAL);
                goto out;
        }
-       fsn_event = fsnotify_remove_first_event(group);
-       if (fanotify_is_perm_event(FANOTIFY_E(fsn_event)->mask))
-               FANOTIFY_PE(fsn_event)->state = FAN_EVENT_REPORTED;
+       event = FANOTIFY_E(fsnotify_remove_first_event(group));
+       if (fanotify_is_perm_event(event->mask))
+               FANOTIFY_PERM(event)->state = FAN_EVENT_REPORTED;
 out:
        spin_unlock(&group->notification_lock);
-       return fsn_event;
+       return event;
 }
 
 static int create_fd(struct fsnotify_group *group, struct path *path,
@@ -252,19 +253,16 @@ static int copy_fid_to_user(struct fanotify_event *event, char __user *buf)
 }
 
 static ssize_t copy_event_to_user(struct fsnotify_group *group,
-                                 struct fsnotify_event *fsn_event,
+                                 struct fanotify_event *event,
                                  char __user *buf, size_t count)
 {
        struct fanotify_event_metadata metadata;
-       struct fanotify_event *event;
-       struct path *path;
+       struct path *path = fanotify_event_path(event);
        struct file *f = NULL;
        int ret, fd = FAN_NOFD;
 
-       pr_debug("%s: group=%p event=%p\n", __func__, group, fsn_event);
+       pr_debug("%s: group=%p event=%p\n", __func__, group, event);
 
-       event = container_of(fsn_event, struct fanotify_event, fse);
-       path = fanotify_event_path(event);
        metadata.event_len = FAN_EVENT_METADATA_LEN;
        metadata.metadata_len = FAN_EVENT_METADATA_LEN;
        metadata.vers = FANOTIFY_METADATA_VERSION;
@@ -293,9 +291,9 @@ static ssize_t copy_event_to_user(struct fsnotify_group *group,
                goto out_close_fd;
 
        if (fanotify_is_perm_event(event->mask))
-               FANOTIFY_PE(fsn_event)->fd = fd;
+               FANOTIFY_PERM(event)->fd = fd;
 
-       if (fanotify_event_has_path(event)) {
+       if (f) {
                fd_install(fd, f);
        } else if (fanotify_event_has_fid(event)) {
                ret = copy_fid_to_user(event, buf + FAN_EVENT_METADATA_LEN);
@@ -332,7 +330,7 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
                             size_t count, loff_t *pos)
 {
        struct fsnotify_group *group;
-       struct fsnotify_event *kevent;
+       struct fanotify_event *event;
        char __user *start;
        int ret;
        DEFINE_WAIT_FUNC(wait, woken_wake_function);
@@ -344,13 +342,13 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
 
        add_wait_queue(&group->notification_waitq, &wait);
        while (1) {
-               kevent = get_one_event(group, count);
-               if (IS_ERR(kevent)) {
-                       ret = PTR_ERR(kevent);
+               event = get_one_event(group, count);
+               if (IS_ERR(event)) {
+                       ret = PTR_ERR(event);
                        break;
                }
 
-               if (!kevent) {
+               if (!event) {
                        ret = -EAGAIN;
                        if (file->f_flags & O_NONBLOCK)
                                break;
@@ -366,7 +364,7 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
                        continue;
                }
 
-               ret = copy_event_to_user(group, kevent, buf, count);
+               ret = copy_event_to_user(group, event, buf, count);
                if (unlikely(ret == -EOPENSTALE)) {
                        /*
                         * We cannot report events with stale fd so drop it.
@@ -381,17 +379,17 @@ static ssize_t fanotify_read(struct file *file, char __user *buf,
                 * Permission events get queued to wait for response.  Other
                 * events can be destroyed now.
                 */
-               if (!fanotify_is_perm_event(FANOTIFY_E(kevent)->mask)) {
-                       fsnotify_destroy_event(group, kevent);
+               if (!fanotify_is_perm_event(event->mask)) {
+                       fsnotify_destroy_event(group, &event->fse);
                } else {
                        if (ret <= 0) {
                                spin_lock(&group->notification_lock);
                                finish_permission_event(group,
-                                       FANOTIFY_PE(kevent), FAN_DENY);
+                                       FANOTIFY_PERM(event), FAN_DENY);
                                wake_up(&group->fanotify_data.access_waitq);
                        } else {
                                spin_lock(&group->notification_lock);
-                               list_add_tail(&kevent->list,
+                               list_add_tail(&event->fse.list,
                                        &group->fanotify_data.access_list);
                                spin_unlock(&group->notification_lock);
                        }
@@ -437,8 +435,6 @@ static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t
 static int fanotify_release(struct inode *ignored, struct file *file)
 {
        struct fsnotify_group *group = file->private_data;
-       struct fanotify_perm_event *event;
-       struct fsnotify_event *fsn_event;
 
        /*
         * Stop new events from arriving in the notification queue. since
@@ -453,6 +449,8 @@ static int fanotify_release(struct inode *ignored, struct file *file)
         */
        spin_lock(&group->notification_lock);
        while (!list_empty(&group->fanotify_data.access_list)) {
+               struct fanotify_perm_event *event;
+
                event = list_first_entry(&group->fanotify_data.access_list,
                                struct fanotify_perm_event, fae.fse.list);
                list_del_init(&event->fae.fse.list);
@@ -466,12 +464,14 @@ static int fanotify_release(struct inode *ignored, struct file *file)
         * response is consumed and fanotify_get_response() returns.
         */
        while (!fsnotify_notify_queue_is_empty(group)) {
-               fsn_event = fsnotify_remove_first_event(group);
-               if (!(FANOTIFY_E(fsn_event)->mask & FANOTIFY_PERM_EVENTS)) {
+               struct fanotify_event *event;
+
+               event = FANOTIFY_E(fsnotify_remove_first_event(group));
+               if (!(event->mask & FANOTIFY_PERM_EVENTS)) {
                        spin_unlock(&group->notification_lock);
-                       fsnotify_destroy_event(group, fsn_event);
+                       fsnotify_destroy_event(group, &event->fse);
                } else {
-                       finish_permission_event(group, FANOTIFY_PE(fsn_event),
+                       finish_permission_event(group, FANOTIFY_PERM(event),
                                                FAN_ALLOW);
                }
                spin_lock(&group->notification_lock);
@@ -1136,7 +1136,10 @@ static int __init fanotify_user_setup(void)
 
        fanotify_mark_cache = KMEM_CACHE(fsnotify_mark,
                                         SLAB_PANIC|SLAB_ACCOUNT);
-       fanotify_event_cachep = KMEM_CACHE(fanotify_event, SLAB_PANIC);
+       fanotify_fid_event_cachep = KMEM_CACHE(fanotify_fid_event,
+                                              SLAB_PANIC);
+       fanotify_path_event_cachep = KMEM_CACHE(fanotify_path_event,
+                                               SLAB_PANIC);
        if (IS_ENABLED(CONFIG_FANOTIFY_ACCESS_PERMISSIONS)) {
                fanotify_perm_event_cachep =
                        KMEM_CACHE(fanotify_perm_event, SLAB_PANIC);