diff mbox series

[RFC,13/17] zuf: ioctl implementation

Message ID 20190219115136.29952-14-boaz@plexistor.com (mailing list archive)
State New, archived
Headers show
Series zuf: ZUFS Zero-copy User-mode FileSystem | expand

Commit Message

Boaz Harrosh Feb. 19, 2019, 11:51 a.m. UTC
From: Boaz Harrosh <boazh@netapp.com>

* support for some generic IOCTLs:
  FS_IOC_GETFLAGS, FS_IOC_SETFLAGS, FS_IOC_GETVERSION, FS_IOC_SETVERSION

* Simple support for zusFS defined IOCTLs
  We only support flat structures
  (no emmbedded pointers within the IOCTL structures)
  We try to deduce the size of the IOCTL from the _IOC_SIZE(cmd)
  If zusFS needs a bigger copy it will send a retry with the
  new size. So bad defined IOCTLs always do 2 trips to userland

Signed-off-by: Boaz Harrosh <boazh@netapp.com>
---
 fs/zuf/Makefile    |   2 +-
 fs/zuf/_extern.h   |   5 +
 fs/zuf/directory.c |   4 +
 fs/zuf/file.c      |   4 +
 fs/zuf/ioctl.c     | 282 +++++++++++++++++++++++++++++++++++++++++++++
 fs/zuf/zuf-core.c  |   1 +
 fs/zuf/zus_api.h   |  16 +++
 7 files changed, 313 insertions(+), 1 deletion(-)
 create mode 100644 fs/zuf/ioctl.c
diff mbox series

Patch

diff --git a/fs/zuf/Makefile b/fs/zuf/Makefile
index 970062d6b13f..5304aba901b2 100644
--- a/fs/zuf/Makefile
+++ b/fs/zuf/Makefile
@@ -17,6 +17,6 @@  zuf-y += md.o t1.o t2.o
 zuf-y += zuf-core.o zuf-root.o
 
 # Main FS
-zuf-y += rw.o mmap.o
+zuf-y += rw.o mmap.o ioctl.o
 zuf-y += super.o inode.o directory.o namei.o file.o symlink.o
 zuf-y += module.o
diff --git a/fs/zuf/_extern.h b/fs/zuf/_extern.h
index 5029f865655a..b8e24c6a66d9 100644
--- a/fs/zuf/_extern.h
+++ b/fs/zuf/_extern.h
@@ -46,6 +46,11 @@  bool zuf_dir_emit(struct super_block *sb, struct dir_context *ctx,
 uint zuf_prepare_symname(struct zufs_ioc_new_inode *ioc_new_inode,
 			const char *symname, ulong len, struct page *pages[2]);
 
+/* ioctl.c */
+long zuf_ioctl(struct file *filp, uint cmd, ulong arg);
+#ifdef CONFIG_COMPAT
+long zuf_compat_ioctl(struct file *file, uint cmd, ulong arg);
+#endif
 
 /* mmap.c */
 int zuf_file_mmap(struct file *file, struct vm_area_struct *vma);
diff --git a/fs/zuf/directory.c b/fs/zuf/directory.c
index 645dd367fd8c..11fcbe0ba6ff 100644
--- a/fs/zuf/directory.c
+++ b/fs/zuf/directory.c
@@ -160,4 +160,8 @@  const struct file_operations zuf_dir_operations = {
 	.read		= generic_read_dir,
 	.iterate_shared	= zuf_readdir,
 	.fsync		= noop_fsync,
+	.unlocked_ioctl = zuf_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl	= zuf_compat_ioctl,
+#endif
 };
diff --git a/fs/zuf/file.c b/fs/zuf/file.c
index 392b1a0d5881..48b339cb5f8f 100644
--- a/fs/zuf/file.c
+++ b/fs/zuf/file.c
@@ -507,7 +507,11 @@  const struct file_operations zuf_file_operations = {
 	.fsync			= zuf_fsync,
 	.flush			= zuf_flush,
 	.release		= zuf_file_release,
+	.unlocked_ioctl		= zuf_ioctl,
 	.fallocate		= zuf_fallocate,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl		= zuf_compat_ioctl,
+#endif
 	.copy_file_range	= zuf_copy_file_range,
 	.remap_file_range	= zuf_clone_file_range,
 };
diff --git a/fs/zuf/ioctl.c b/fs/zuf/ioctl.c
new file mode 100644
index 000000000000..13ce65764c38
--- /dev/null
+++ b/fs/zuf/ioctl.c
@@ -0,0 +1,282 @@ 
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * BRIEF DESCRIPTION
+ *
+ * Ioctl operations.
+ *
+ * Copyright (c) 2018 NetApp Inc. All rights reserved.
+ *
+ * ZUFS-License: GPL-2.0. See module.c for LICENSE details.
+ *
+ * Authors:
+ *	Boaz Harrosh <boazh@netapp.com>
+ *	Sagi Manole <sagim@netapp.com>"
+ */
+
+#include <linux/capability.h>
+#include <linux/time.h>
+#include <linux/sched.h>
+#include <linux/compat.h>
+#include <linux/mount.h>
+#include <linux/fadvise.h>
+#include <linux/vmalloc.h>
+#include <linux/capability.h>
+
+#include "zuf.h"
+
+#define ZUFS_SUPPORTED_FS_FLAGS (FS_SYNC_FL | FS_APPEND_FL | FS_IMMUTABLE_FL | \
+				 FS_NOATIME_FL | FS_DIRTY_FL)
+
+#define ZUS_IOCTL_MAX_PAGES	8
+
+static int _ioctl_dispatch(struct inode *inode, uint cmd, ulong arg)
+{
+	struct _ioctl_info {
+		struct zufs_ioc_ioctl ctl;
+		char buf[900];
+	} ctl_alloc = {};
+	enum big_alloc_type bat;
+	struct zufs_ioc_ioctl *ioc_ioctl;
+	size_t ioc_size = _IOC_SIZE(cmd);
+	void __user *parg = (void __user *)arg;
+	struct timespec64 time = current_time(inode);
+	size_t size;
+	bool retry = false;
+	int err;
+
+again:
+	size = sizeof(*ioc_ioctl) + ioc_size;
+
+	zuf_dbg_vfs("[%ld] cmd=0x%x arg=0x%lx size=0x%zx IOC(%d, %d, %zd)\n",
+		    inode->i_ino, cmd, arg, size, _IOC_TYPE(cmd),
+		    _IOC_NR(cmd), ioc_size);
+
+	ioc_ioctl = big_alloc(size, sizeof(ctl_alloc), &ctl_alloc, GFP_KERNEL,
+			      &bat);
+	if (unlikely(!ioc_ioctl))
+		return -ENOMEM;
+
+	memset(ioc_ioctl, 0, sizeof(*ioc_ioctl));
+	ioc_ioctl->hdr.in_len = size;
+	ioc_ioctl->hdr.out_start = offsetof(struct zufs_ioc_ioctl, arg);
+	ioc_ioctl->hdr.out_max = size;
+	ioc_ioctl->hdr.out_len = 0;
+	ioc_ioctl->hdr.operation = ZUFS_OP_IOCTL;
+	ioc_ioctl->zus_ii = ZUII(inode)->zus_ii;
+	ioc_ioctl->cmd = cmd;
+	timespec_to_mt(&ioc_ioctl->time, &time);
+
+	if (arg && ioc_size) {
+		if (copy_from_user(ioc_ioctl->arg, parg, ioc_size)) {
+			err = -EFAULT;
+			goto out;
+		}
+	}
+
+	err = zufc_dispatch(ZUF_ROOT(SBI(inode->i_sb)), &ioc_ioctl->hdr,
+			    NULL, 0);
+
+	if (!retry && err == -EZUFS_RETRY) {
+		ioc_size = ioc_ioctl->new_size - sizeof(*ioc_ioctl);
+		big_free(ioc_ioctl, bat);
+		retry = true;
+		goto again;
+	}
+
+	if (unlikely(err)) {
+		zuf_dbg_err("zufc_dispatch failed => %d IOC(%d, %d, %zd)\n",
+			    err, _IOC_TYPE(cmd), _IOC_NR(cmd), ioc_size);
+		goto out;
+	}
+
+	if (ioc_ioctl->hdr.out_len) {
+		if (copy_to_user(parg, ioc_ioctl->arg,
+		    ioc_ioctl->hdr.out_len)) {
+			err = -EFAULT;
+			goto out;
+		}
+	}
+
+out:
+	big_free(ioc_ioctl, bat);
+
+	return err;
+}
+
+static uint _translate_to_ioc_flags(struct zus_inode *zi)
+{
+	uint zi_flags = le16_to_cpu(zi->i_flags);
+	uint ioc_flags = 0;
+
+	if (zi_flags & S_SYNC)
+		ioc_flags |= FS_SYNC_FL;
+	if (zi_flags & S_APPEND)
+		ioc_flags |= FS_APPEND_FL;
+	if (zi_flags & S_IMMUTABLE)
+		ioc_flags |= FS_IMMUTABLE_FL;
+	if (zi_flags & S_NOATIME)
+		ioc_flags |= FS_NOATIME_FL;
+	if (zi_flags & S_DIRSYNC)
+		ioc_flags |= FS_DIRSYNC_FL;
+
+	return ioc_flags;
+}
+
+static int _ioc_getflags(struct inode *inode, uint __user *parg)
+{
+	struct zus_inode *zi = zus_zi(inode);
+	uint flags = _translate_to_ioc_flags(zi);
+
+	return put_user(flags, parg);
+}
+
+static void _translate_to_zi_flags(struct zus_inode *zi, unsigned int flags)
+{
+	uint zi_flags = le16_to_cpu(zi->i_flags);
+
+	zi_flags &=
+		~(S_SYNC | S_APPEND | S_IMMUTABLE | S_NOATIME | S_DIRSYNC);
+
+	if (flags & FS_SYNC_FL)
+		zi_flags |= S_SYNC;
+	if (flags & FS_APPEND_FL)
+		zi_flags |= S_APPEND;
+	if (flags & FS_IMMUTABLE_FL)
+		zi_flags |= S_IMMUTABLE;
+	if (flags & FS_NOATIME_FL)
+		zi_flags |= S_NOATIME;
+	if (flags & FS_DIRSYNC_FL)
+		zi_flags |= S_DIRSYNC;
+
+	zi->i_flags = cpu_to_le16(zi_flags);
+}
+
+/* use statx ioc to flush zi changes to fs */
+static int __ioc_dispatch_zi_update(struct inode *inode, uint flags)
+{
+	struct zufs_ioc_attr ioc_attr = {
+		.hdr.in_len = sizeof(ioc_attr),
+		.hdr.out_len = sizeof(ioc_attr),
+		.hdr.operation = ZUFS_OP_SETATTR,
+		.zus_ii = ZUII(inode)->zus_ii,
+		.zuf_attr = flags,
+	};
+	int err;
+
+	err = zufc_dispatch(ZUF_ROOT(SBI(inode->i_sb)), &ioc_attr.hdr, NULL, 0);
+	if (unlikely(err && err != -EINTR))
+		zuf_err("zufc_dispatch failed => %d\n", err);
+
+	return err;
+}
+
+static int _ioc_setflags(struct inode *inode, uint __user *parg)
+{
+	struct zus_inode *zi = zus_zi(inode);
+	uint flags, oldflags;
+	int err;
+
+	if (!inode_owner_or_capable(inode))
+		return -EPERM;
+
+	if (get_user(flags, parg))
+		return -EFAULT;
+
+	if (flags & ~ZUFS_SUPPORTED_FS_FLAGS)
+		return -EOPNOTSUPP;
+
+	inode_lock(inode);
+
+	oldflags = le32_to_cpu(zi->i_flags);
+
+	if ((flags ^ oldflags) &
+		(FS_APPEND_FL | FS_IMMUTABLE_FL)) {
+		if (!capable(CAP_LINUX_IMMUTABLE)) {
+			inode_unlock(inode);
+			return -EPERM;
+		}
+	}
+
+	if (!S_ISDIR(inode->i_mode))
+		flags &= ~FS_DIRSYNC_FL;
+
+	flags = flags & FS_FL_USER_MODIFIABLE;
+	flags |= oldflags & ~FS_FL_USER_MODIFIABLE;
+	inode->i_ctime = current_time(inode);
+	timespec_to_mt(&zi->i_ctime, &inode->i_ctime);
+	_translate_to_zi_flags(zi, flags);
+	zuf_set_inode_flags(inode, zi);
+
+	err = __ioc_dispatch_zi_update(inode, ZUFS_STATX_FLAGS | STATX_CTIME);
+
+	inode_unlock(inode);
+	return err;
+}
+
+static int _ioc_setversion(struct inode *inode, uint __user *parg)
+{
+	struct zus_inode *zi = zus_zi(inode);
+	__u32 generation;
+	int err;
+
+	if (!inode_owner_or_capable(inode))
+		return -EPERM;
+
+	if (get_user(generation, parg))
+		return -EFAULT;
+
+	inode_lock(inode);
+
+	inode->i_ctime = current_time(inode);
+	inode->i_generation = generation;
+	timespec_to_mt(&zi->i_ctime, &inode->i_ctime);
+	zi->i_generation = cpu_to_le32(inode->i_generation);
+
+	err = __ioc_dispatch_zi_update(inode, ZUFS_STATX_VERSION | STATX_CTIME);
+
+	inode_unlock(inode);
+	return err;
+}
+
+long zuf_ioctl(struct file *filp, unsigned int cmd, ulong arg)
+{
+	struct inode *inode = filp->f_inode;
+	void __user *parg = (void __user *)arg;
+
+	switch (cmd) {
+	case FS_IOC_GETFLAGS:
+		return _ioc_getflags(inode, parg);
+	case FS_IOC_SETFLAGS:
+		return _ioc_setflags(inode, parg);
+	case FS_IOC_GETVERSION:
+		return put_user(inode->i_generation, (int __user *)arg);
+	case FS_IOC_SETVERSION:
+		return _ioc_setversion(inode, parg);
+	default:
+		return _ioctl_dispatch(inode, cmd, arg);
+	}
+}
+
+#ifdef CONFIG_COMPAT
+long zuf_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	switch (cmd) {
+	case FS_IOC32_GETFLAGS:
+		cmd = FS_IOC_GETFLAGS;
+		break;
+	case FS_IOC32_SETFLAGS:
+		cmd = FS_IOC_SETFLAGS;
+		break;
+	case FS_IOC32_GETVERSION:
+		cmd = FS_IOC_GETVERSION;
+		break;
+	case FS_IOC32_SETVERSION:
+		cmd = FS_IOC_SETVERSION;
+		break;
+	default:
+		return -ENOIOCTLCMD;
+	}
+	return zuf_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
+}
+#endif
+
diff --git a/fs/zuf/zuf-core.c b/fs/zuf/zuf-core.c
index 86f624031d8d..09ad210318f8 100644
--- a/fs/zuf/zuf-core.c
+++ b/fs/zuf/zuf-core.c
@@ -787,6 +787,7 @@  const char *zuf_op_name(enum e_zufs_operation op)
 		CASE_ENUM_NAME(ZUFS_OP_SYNC		);
 		CASE_ENUM_NAME(ZUFS_OP_FALLOCATE	);
 		CASE_ENUM_NAME(ZUFS_OP_LLSEEK		);
+		CASE_ENUM_NAME(ZUFS_OP_IOCTL		);
 		CASE_ENUM_NAME(ZUFS_OP_BREAK		);
 	default:
 		return "UNKNOWN";
diff --git a/fs/zuf/zus_api.h b/fs/zuf/zus_api.h
index 3d6481768308..f32ee615b937 100644
--- a/fs/zuf/zus_api.h
+++ b/fs/zuf/zus_api.h
@@ -350,6 +350,7 @@  enum e_zufs_operation {
 	ZUFS_OP_SYNC,
 	ZUFS_OP_FALLOCATE,
 	ZUFS_OP_LLSEEK,
+	ZUFS_OP_IOCTL,
 
 	ZUFS_OP_BREAK,		/* Kernel telling Server to exit */
 	ZUFS_OP_MAX_OPT,
@@ -586,6 +587,21 @@  struct zufs_ioc_seek {
 	__u64 offset_out;
 };
 
+/* ZUFS_OP_IOCTL */
+struct zufs_ioc_ioctl {
+	struct zufs_ioc_hdr hdr;
+	/* IN */
+	struct zus_inode_info *zus_ii;
+	__u32 cmd;
+	__u64 time;
+
+	/* OUT */
+	union {
+		__u32 new_size;
+		char arg[0];
+	};
+};
+
 /* ~~~~ io_map structures && IOCTL(s) ~~~~ */
 /*
  * These set of structures and helpers are used in return of zufs_ioc_IO and