fuse, virtiofs: Do not alloc/install fuse device in fuse_fill_super_common()
Subsystems (3)
Vivek Goyal
maintainerPATCH 1/1 fuse, virtiofs: Do not alloc/install fuse device in fuse_fill_super_common()
As of now fuse_fill_super_common() allocates and installs one fuse device. Filesystems like virtiofs can have more than one filesystem queues and can have one fuse device per queue. Given, fuse_fill_super_common() only handles one device, virtiofs allocates and installes fuse devices for all queues except one.
This makes logic little twisted and hard to understand. It probably is better to not do any device allocation/installation in fuse_fill_super_common() and let caller take care of it instead.
--- fs/fuse/fuse_i.h | 3 --- fs/fuse/inode.c | 19 ++++++++++++------- fs/fuse/virtio_fs.c | 9 +-------- 3 files changed, 13 insertions(+), 18 deletions(-)
fs/fuse/fuse_i.h +0 -3
@@ -485,9 +485,6 @@ struct fuse_fs_context {unsigned int max_read;unsigned int blksize;const char *subtype; - -/* fuse_dev pointer to fill in, should contain NULL on entry */ -void **fudptr; };/**
fs/fuse/inode.c +12 -7
@@ -1191,16 +1191,12 @@ int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)mutex_lock (&fuse_mutex); err = -EINVAL ; -if (*ctx->fudptr) - goto err_unlock; - err =fuse_ctl_add_conn (fc);if (err) goto err_unlock;list_add_tail (&fc->entry, &fuse_conn_list); sb->s_root = root_dentry; - *ctx->fudptr = fud;mutex_unlock (&fuse_mutex);return 0 ;
@@ -1220,6 +1216,7 @@ static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)struct file *file;int err;struct fuse_conn *fc; +struct fuse_dev *fud; err = -EINVAL ; file =fget (ctx->fd);
@@ -1233,13 +1230,16 @@ static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)if ((file->f_op != &fuse_dev_operations) || (file->f_cred->user_ns != sb->s_user_ns)) goto err_fput; - ctx->fudptr = &file->private_data; - fc =kmalloc (sizeof (*fc),GFP_KERNEL ); err = -ENOMEM ; -if (!fc) + fud =fuse_dev_alloc (); +if (!fud) goto err_fput; + fc =kmalloc (sizeof (*fc),GFP_KERNEL ); +if (!fc) + goto err_free_dev; +fuse_conn_init (fc, sb->s_user_ns, &fuse_dev_fiq_ops,NULL ); fc->release = fuse_free_conn; sb->s_fs_info = fc;
@@ -1247,6 +1247,9 @@ static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc) err =fuse_fill_super_common (sb, ctx);if (err) goto err_put_conn; + +fuse_dev_install (fud, fc); + file->private_data = fud;/* * atomic_dec_and_test() in fput() provides the necessary * memory barrier for file->private_data to be visible on all
@@ -1259,6 +1262,8 @@ static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc) err_put_conn:fuse_conn_put (fc); sb->s_fs_info =NULL ; + err_free_dev: +fuse_dev_free (fud); err_fput:fput (file); err:
fs/fuse/virtio_fs.c +1 -8
@@ -1066,8 +1066,7 @@ static int virtio_fs_fill_super(struct super_block *sb) } err = -ENOMEM ; -/* Allocate fuse_dev for hiprio and notification queues */ -for (i =0 ; i <VQ_REQUEST ; i++) { +for (i =0 ; i < fs->nvqs; i++) {struct virtio_fs_vq *fsvq = &fs->vqs[i]; fsvq->fud =fuse_dev_alloc ();
@@ -1075,18 +1074,12 @@ static int virtio_fs_fill_super(struct super_block *sb) goto err_free_fuse_devs; } - ctx.fudptr = (void **)&fs->vqs[VQ_REQUEST ].fud; err =fuse_fill_super_common (sb, &ctx);if (err <0 ) goto err_free_fuse_devs; - fc = fs->vqs[VQ_REQUEST ].fud->fc; -for (i =0 ; i < fs->nvqs; i++) {struct virtio_fs_vq *fsvq = &fs->vqs[i]; - -if (i ==VQ_REQUEST ) -continue ;/* already initialized */ fuse_dev_install (fsvq->fud, fc); }
Signature · 1 line
2.25.4
Quoted 10 lines · Vivek Goyal
As of now
fuse_fill_super_common()allocates and installs one fuse device. Filesystems like virtiofs can have more than one filesystem queues and can have one fuse device per queue. Given,fuse_fill_super_common()only handles one device, virtiofs allocates and installes fuse devices for all queues except one.This makes logic little twisted and hard to understand. It probably is better to not do any device allocation/installation in
fuse_fill_super_common()and let caller take care of it instead.
Quoted 6 lines
Signed-off-by Vivek Goyal <[email protected]>--- fs/fuse/fuse_i.h | 3 --- fs/fuse/inode.c | 19 ++++++++++++------- fs/fuse/virtio_fs.c | 9 +-------- 3 files changed, 13 insertions(+), 18 deletions(-)
Miklos Szeredi
maintainerQuoted 10 lines · Vivek Goyal
As of now
fuse_fill_super_common()allocates and installs one fuse device. Filesystems like virtiofs can have more than one filesystem queues and can have one fuse device per queue. Given,fuse_fill_super_common()only handles one device, virtiofs allocates and installes fuse devices for all queues except one.This makes logic little twisted and hard to understand. It probably is better to not do any device allocation/installation in
fuse_fill_super_common()and let caller take care of it instead.
I don't have the details about the fuse super block setup in my head, but leaving the fuse_dev_alloc_install() call inside fuse_fill_super_common() and adding new fuse_dev_alloc()/fuse_dev_install() calls looks highly suspicious.
Thanks,
Miklos
Vivek Goyal
maintainerQuoted 8 lines · Miklos Szeredi
Quoted 7 lines · Vivek Goyal
As of now
fuse_fill_super_common()allocates and installs one fuse device. Filesystems like virtiofs can have more than one filesystem queues and can have one fuse device per queue. Given,fuse_fill_super_common()only handles one device, virtiofs allocates and installes fuse devices for all queues except one.
Quoted 8 lines
Quoted 3 lines
This makes logic little twisted and hard to understand. It probably is better to not do any device allocation/installation in
fuse_fill_super_common()and let caller take care of it instead.I don't have the details about the fuse super block setup in my head, but leaving the
fuse_dev_alloc_install()call insidefuse_fill_super_common()and adding new fuse_dev_alloc()/fuse_dev_install() calls looks highly suspicious.
Good catch. My bad. I forgot to remove that fuse_dev_alloc_install() call from fuse_fill_super_common(). I tested it and it worked. So I guess I just ended up installing extra fud device in fc which was never used.
I will fix this and post V2 of the patch.
Thanks
Vivek