Skip to contentSkip to conversation

clean up kernel_{read,write} & friends v2

PATCH In review Christoph Hellwig 3 participants 19 messages May 13, 2020 – May 20, 2020 1 message not in the archive

Subsystems (11)
Networking VFS Block layer netfilter +7 more subsystems
lore
Share & tools
lore
#0 lore raw

Hi Al,

this series fixes a few issues and cleans up the helpers that read from or write to kernel space buffers, and ensures that we don't change the address limit if we are using the ->read_iter and ->write_iter methods that don't need the changed address limit.

Changes since v1:
- __kernel_write must not take sb_writers
- unexported __kernel_write

#1 lore raw

PATCH 1/14 cachefiles: switch to kernel_write

__kernel_write doesn't take a sb_writers references, which we need here.

---
 fs/cachefiles/rdwr.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/cachefiles/rdwr.c +1 -1
index 1dc97f2d62013..c0b99ccd9a79f 100644
@@ -937,7 +937,7 @@ int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 	}
 
 	data = kmap(page);
-	ret = __kernel_write(file, data, len, &pos);
+	ret = kernel_write(file, data, len, &pos);
 	kunmap(page);
 	fput(file);
 	if (ret != len)
Signature · 1 line
2.26.2
#3 lore raw

PATCH 2/14 autofs: switch to kernel_write

In reply to Christoph Hellwig

While pipes don't really need sb_writers projection, __kernel_write is an interface better kept private, and the additional rw_verify_area does not hurt here.

---
 fs/autofs/waitq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/autofs/waitq.c +1 -1
index b04c528b19d34..74c886f7c51cb 100644
@@ -53,7 +53,7 @@ static int autofs_write(struct autofs_sb_info *sbi,
 
 	mutex_lock(&sbi->pipe_mutex);
 	while (bytes) {
-		wr = __kernel_write(file, data, bytes, &file->f_pos);
+		wr = kernel_write(file, data, bytes, &file->f_pos);
 		if (wr <= 0)
 			break;
 		data += wr;
Signature · 1 line
2.26.2
maintainer Acked-by on v1 2/14
#4 lore raw

Re: [PATCH 02/14] autofs: switch to kernel_write

Quoted 7 lines · Christoph Hellwig

While pipes don't really need sb_writers projection, __kernel_write is an
interface better kept private, and the additional rw_verify_area does not
hurt here.

Right, should be fine AFAICS.

Quoted 17 lines
---
 fs/autofs/waitq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/autofs/waitq.c +1 -1
index b04c528b19d34..74c886f7c51cb 100644
@@ -53,7 +53,7 @@ static int autofs_write(struct autofs_sb_info *sbi,
 
 	mutex_lock(&sbi->pipe_mutex);
 	while (bytes) {
-		wr = __kernel_write(file, data, bytes, &file->f_pos);
+		wr = kernel_write(file, data, bytes, &file->f_pos);
 		if (wr <= 0)
 			break;
 		data += wr;
#5 lore raw

PATCH 3/14 bpfilter: switch to kernel_write

In reply to Christoph Hellwig

While pipes don't really need sb_writers projection, __kernel_write is an interface better kept private, and the additional rw_verify_area does not hurt here.

---
 net/bpfilter/bpfilter_kern.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
net/bpfilter/bpfilter_kern.c +1 -1
index c0f0990f30b60..1905e01c3aa9a 100644
@@ -50,7 +50,7 @@ static int __bpfilter_process_sockopt(struct sock *sk, int optname,
 	req.len = optlen;
 	if (!bpfilter_ops.info.pid)
 		goto out;
-	n = __kernel_write(bpfilter_ops.info.pipe_to_umh, &req, sizeof(req),
+	n = kernel_write(bpfilter_ops.info.pipe_to_umh, &req, sizeof(req),
 			   &pos);
 	if (n != sizeof(req)) {
 		pr_err("write fail %zd\n", n);
Signature · 1 line
2.26.2
#6 lore raw

PATCH 4/14 fs: unexport __kernel_write

In reply to Christoph Hellwig

This is a very special interface that skips sb_writes protection, and not used by modules anymore.

---
 fs/read_write.c | 1 -
 1 file changed, 1 deletion(-)
fs/read_write.c +0 -1
index bbfa9b12b15eb..2c601d853ff3d 100644
@@ -522,7 +522,6 @@ ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t
 	inc_syscw(current);
 	return ret;
 }
-EXPORT_SYMBOL(__kernel_write);
 
 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
 			    loff_t *pos)
Signature · 1 line
2.26.2
#7 lore raw

PATCH 5/14 fs: check FMODE_WRITE in __kernel_write

In reply to Christoph Hellwig

We still need to check if the fѕ is open write, even for the low-level helper.

---
 fs/read_write.c | 2 ++
 1 file changed, 2 insertions(+)
fs/read_write.c +2 -0
index 2c601d853ff3d..76be155ad9824 100644
@@ -505,6 +505,8 @@ ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t
 	const char __user *p;
 	ssize_t ret;
 
+	if (!(file->f_mode & FMODE_WRITE))
+		return -EBADF;
 	if (!(file->f_mode & FMODE_CAN_WRITE))
 		return -EINVAL;
 
Signature · 1 line
2.26.2
#8 lore raw

PATCH 6/14 fs: remove the call_{read,write}_iter functions

In reply to Christoph Hellwig

Just open coding the methods calls is a lot easier to follow.

---
 drivers/block/loop.c              |  4 ++--
 drivers/target/target_core_file.c |  4 ++--
 fs/aio.c                          |  4 ++--
 fs/io_uring.c                     |  4 ++--
 fs/read_write.c                   | 12 ++++++------
 fs/splice.c                       |  2 +-
 include/linux/fs.h                | 12 ------------
 7 files changed, 15 insertions(+), 27 deletions(-)
drivers/block/loop.c +2 -2
index da693e6a834e5..ad167050a4ec4 100644
@@ -572,9 +572,9 @@ static int lo_rw_aio(struct loop_device *lo, struct loop_cmd *cmd,
 		kthread_associate_blkcg(cmd->css);
 
 	if (rw == WRITE)
-		ret = call_write_iter(file, &cmd->iocb, &iter);
+		ret = file->f_op->write_iter(&cmd->iocb, &iter);
 	else
-		ret = call_read_iter(file, &cmd->iocb, &iter);
+		ret = file->f_op->read_iter(&cmd->iocb, &iter);
 
 	lo_rw_aio_do_completion(cmd);
 	kthread_associate_blkcg(NULL);
drivers/target/target_core_file.c +2 -2
index 7143d03f0e027..79f0707877917 100644
@@ -303,9 +303,9 @@ fd_execute_rw_aio(struct se_cmd *cmd, struct scatterlist *sgl, u32 sgl_nents,
 		aio_cmd->iocb.ki_flags |= IOCB_DSYNC;
 
 	if (is_write)
-		ret = call_write_iter(file, &aio_cmd->iocb, &iter);
+		ret = file->f_op->write_iter(&aio_cmd->iocb, &iter);
 	else
-		ret = call_read_iter(file, &aio_cmd->iocb, &iter);
+		ret = file->f_op->read_iter(&aio_cmd->iocb, &iter);
 
 	kfree(bvec);
 
fs/aio.c +2 -2
index 5f3d3d8149287..1ccc0efdc357d 100644
@@ -1540,7 +1540,7 @@ static int aio_read(struct kiocb *req, const struct iocb *iocb,
 		return ret;
 	ret = rw_verify_area(READ, file, &req->ki_pos, iov_iter_count(&iter));
 	if (!ret)
-		aio_rw_done(req, call_read_iter(file, req, &iter));
+		aio_rw_done(req, file->f_op->read_iter(req, &iter));
 	kfree(iovec);
 	return ret;
 }
@@ -1580,7 +1580,7 @@ static int aio_write(struct kiocb *req, const struct iocb *iocb,
 			__sb_writers_release(file_inode(file)->i_sb, SB_FREEZE_WRITE);
 		}
 		req->ki_flags |= IOCB_WRITE;
-		aio_rw_done(req, call_write_iter(file, req, &iter));
+		aio_rw_done(req, file->f_op->write_iter(req, &iter));
 	}
 	kfree(iovec);
 	return ret;
fs/io_uring.c +2 -2
index 979d9f977409a..59514051383e4 100644
@@ -2582,7 +2582,7 @@ static int io_read(struct io_kiocb *req, bool force_nonblock)
 		ssize_t ret2;
 
 		if (req->file->f_op->read_iter)
-			ret2 = call_read_iter(req->file, kiocb, &iter);
+			ret2 = req->file->f_op->read_iter(kiocb, &iter);
 		else
 			ret2 = loop_rw_iter(READ, req->file, kiocb, &iter);
 
@@ -2697,7 +2697,7 @@ static int io_write(struct io_kiocb *req, bool force_nonblock)
 			current->signal->rlim[RLIMIT_FSIZE].rlim_cur = req->fsize;
 
 		if (req->file->f_op->write_iter)
-			ret2 = call_write_iter(req->file, kiocb, &iter);
+			ret2 = req->file->f_op->write_iter(kiocb, &iter);
 		else
 			ret2 = loop_rw_iter(WRITE, req->file, kiocb, &iter);
 
fs/read_write.c +6 -6
index 76be155ad9824..f0768313ea010 100644
@@ -412,7 +412,7 @@ static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, lo
 	kiocb.ki_pos = (ppos ? *ppos : 0);
 	iov_iter_init(&iter, READ, &iov, 1, len);
 
-	ret = call_read_iter(filp, &kiocb, &iter);
+	ret = filp->f_op->read_iter(&kiocb, &iter);
 	BUG_ON(ret == -EIOCBQUEUED);
 	if (ppos)
 		*ppos = kiocb.ki_pos;
@@ -481,7 +481,7 @@ static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t
 	kiocb.ki_pos = (ppos ? *ppos : 0);
 	iov_iter_init(&iter, WRITE, &iov, 1, len);
 
-	ret = call_write_iter(filp, &kiocb, &iter);
+	ret = filp->f_op->write_iter(&kiocb, &iter);
 	BUG_ON(ret == -EIOCBQUEUED);
 	if (ret > 0 && ppos)
 		*ppos = kiocb.ki_pos;
@@ -690,9 +690,9 @@ static ssize_t do_iter_readv_writev(struct file *filp, struct iov_iter *iter,
 	kiocb.ki_pos = (ppos ? *ppos : 0);
 
 	if (type == READ)
-		ret = call_read_iter(filp, &kiocb, iter);
+		ret = filp->f_op->read_iter(&kiocb, iter);
 	else
-		ret = call_write_iter(filp, &kiocb, iter);
+		ret = filp->f_op->write_iter(&kiocb, iter);
 	BUG_ON(ret == -EIOCBQUEUED);
 	if (ppos)
 		*ppos = kiocb.ki_pos;
@@ -961,7 +961,7 @@ ssize_t vfs_iocb_iter_read(struct file *file, struct kiocb *iocb,
 	if (ret < 0)
 		return ret;
 
-	ret = call_read_iter(file, iocb, iter);
+	ret = file->f_op->read_iter(iocb, iter);
 out:
 	if (ret >= 0)
 		fsnotify_access(file);
@@ -1025,7 +1025,7 @@ ssize_t vfs_iocb_iter_write(struct file *file, struct kiocb *iocb,
 	if (ret < 0)
 		return ret;
 
-	ret = call_write_iter(file, iocb, iter);
+	ret = file->f_op->write_iter(iocb, iter);
 	if (ret > 0)
 		fsnotify_modify(file);
 
fs/splice.c +1 -1
index fd0a1e7e5959a..a59d4fadf27fe 100644
@@ -310,7 +310,7 @@ ssize_t generic_file_splice_read(struct file *in, loff_t *ppos,
 	i_head = to.head;
 	init_sync_kiocb(&kiocb, in);
 	kiocb.ki_pos = *ppos;
-	ret = call_read_iter(in, &kiocb, &to);
+	ret = in->f_op->read_iter(&kiocb, &to);
 	if (ret > 0) {
 		*ppos = kiocb.ki_pos;
 		file_accessed(in);
include/linux/fs.h +0 -12
index 45cc10cdf6ddd..21f126957c2cf 100644
@@ -1895,18 +1895,6 @@ struct inode_operations {
 	int (*set_acl)(struct inode *, struct posix_acl *, int);
 } ____cacheline_aligned;
 
-static inline ssize_t call_read_iter(struct file *file, struct kiocb *kio,
-				     struct iov_iter *iter)
-{
-	return file->f_op->read_iter(kio, iter);
-}
-
-static inline ssize_t call_write_iter(struct file *file, struct kiocb *kio,
-				      struct iov_iter *iter)
-{
-	return file->f_op->write_iter(kio, iter);
-}
-
 static inline int call_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	return file->f_op->mmap(file, vma);
Signature · 1 line
2.26.2
#9 lore raw

PATCH 7/14 fs: implement kernel_write using __kernel_write

In reply to Christoph Hellwig

Consolidate the two in-kernel write helpers to make upcoming changes easier. The only difference are the missing call to rw_verify_area in kernel_write, and an access_ok check that doesn't make sense for kernel buffers to start with.

---
 fs/read_write.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)
fs/read_write.c +9 -8
index f0768313ea010..abb84391cfbc5 100644
@@ -499,6 +499,7 @@ static ssize_t __vfs_write(struct file *file, const char __user *p,
 		return -EINVAL;
 }
 
+/* caller is responsible for file_start_write/file_end_write */
 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
 {
 	mm_segment_t old_fs;
@@ -528,16 +529,16 @@ ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t
 ssize_t kernel_write(struct file *file, const void *buf, size_t count,
 			    loff_t *pos)
 {
-	mm_segment_t old_fs;
-	ssize_t res;
+	ssize_t ret;
 
-	old_fs = get_fs();
-	set_fs(KERNEL_DS);
-	/* The cast to a user pointer is valid due to the set_fs() */
-	res = vfs_write(file, (__force const char __user *)buf, count, pos);
-	set_fs(old_fs);
+	ret = rw_verify_area(WRITE, file, pos, count);
+	if (ret)
+		return ret;
 
-	return res;
+	file_start_write(file);
+	ret =  __kernel_write(file, buf, count, pos);
+	file_end_write(file);
+	return ret;
 }
 EXPORT_SYMBOL(kernel_write);
 
Signature · 1 line
2.26.2
#10 lore raw

PATCH 8/14 fs: remove __vfs_write

In reply to Christoph Hellwig

Fold it into the two callers.

---
 fs/read_write.c | 46 ++++++++++++++++++++++------------------------
 1 file changed, 22 insertions(+), 24 deletions(-)
fs/read_write.c +22 -24
index abb84391cfbc5..3bcb084f160de 100644
@@ -488,17 +488,6 @@ static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t
 	return ret;
 }
 
-static ssize_t __vfs_write(struct file *file, const char __user *p,
-			   size_t count, loff_t *pos)
-{
-	if (file->f_op->write)
-		return file->f_op->write(file, p, count, pos);
-	else if (file->f_op->write_iter)
-		return new_sync_write(file, p, count, pos);
-	else
-		return -EINVAL;
-}
-
 /* caller is responsible for file_start_write/file_end_write */
 ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
 {
@@ -516,7 +505,12 @@ ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t
 	p = (__force const char __user *)buf;
 	if (count > MAX_RW_COUNT)
 		count =  MAX_RW_COUNT;
-	ret = __vfs_write(file, p, count, pos);
+	if (file->f_op->write)
+		ret = file->f_op->write(file, p, count, pos);
+	else if (file->f_op->write_iter)
+		ret = new_sync_write(file, p, count, pos);
+	else
+		ret = -EINVAL;
 	set_fs(old_fs);
 	if (ret > 0) {
 		fsnotify_modify(file);
@@ -554,19 +548,23 @@ ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_
 		return -EFAULT;
 
 	ret = rw_verify_area(WRITE, file, pos, count);
-	if (!ret) {
-		if (count > MAX_RW_COUNT)
-			count =  MAX_RW_COUNT;
-		file_start_write(file);
-		ret = __vfs_write(file, buf, count, pos);
-		if (ret > 0) {
-			fsnotify_modify(file);
-			add_wchar(current, ret);
-		}
-		inc_syscw(current);
-		file_end_write(file);
+	if (ret)
+		return ret;
+	if (count > MAX_RW_COUNT)
+		count =  MAX_RW_COUNT;
+	file_start_write(file);
+	if (file->f_op->write)
+		ret = file->f_op->write(file, buf, count, pos);
+	else if (file->f_op->write_iter)
+		ret = new_sync_write(file, buf, count, pos);
+	else
+		ret = -EINVAL;
+	if (ret > 0) {
+		fsnotify_modify(file);
+		add_wchar(current, ret);
 	}
-
+	inc_syscw(current);
+	file_end_write(file);
 	return ret;
 }
 
Signature · 1 line
2.26.2
#11 lore raw

PATCH 9/14 fs: don't change the address limit for ->write_iter in __kernel_write

In reply to Christoph Hellwig

If we write to a file that implements ->write_iter there is no need to change the address limit if we send a kvec down. Implement that case, and prefer it over using plain ->write with a changed address limit if available.

---
 fs/read_write.c | 34 ++++++++++++++++++++++------------
 1 file changed, 22 insertions(+), 12 deletions(-)
fs/read_write.c +22 -12
index 3bcb084f160de..8cfca5f8fc3ce 100644
@@ -489,10 +489,9 @@ static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t
 }
 
 /* caller is responsible for file_start_write/file_end_write */
-ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t *pos)
+ssize_t __kernel_write(struct file *file, const void *buf, size_t count,
+		loff_t *pos)
 {
-	mm_segment_t old_fs;
-	const char __user *p;
 	ssize_t ret;
 
 	if (!(file->f_mode & FMODE_WRITE))
@@ -500,18 +499,29 @@ ssize_t __kernel_write(struct file *file, const void *buf, size_t count, loff_t
 	if (!(file->f_mode & FMODE_CAN_WRITE))
 		return -EINVAL;
 
-	old_fs = get_fs();
-	set_fs(KERNEL_DS);
-	p = (__force const char __user *)buf;
 	if (count > MAX_RW_COUNT)
 		count =  MAX_RW_COUNT;
-	if (file->f_op->write)
-		ret = file->f_op->write(file, p, count, pos);
-	else if (file->f_op->write_iter)
-		ret = new_sync_write(file, p, count, pos);
-	else
+	if (file->f_op->write_iter) {
+		struct kvec iov = { .iov_base = (void *)buf, .iov_len = count };
+		struct kiocb kiocb;
+		struct iov_iter iter;
+
+		init_sync_kiocb(&kiocb, file);
+		kiocb.ki_pos = *pos;
+		iov_iter_kvec(&iter, WRITE, &iov, 1, count);
+		ret = file->f_op->write_iter(&kiocb, &iter);
+		if (ret > 0)
+			*pos = kiocb.ki_pos;
+	} else if (file->f_op->write) {
+		mm_segment_t old_fs = get_fs();
+
+		set_fs(KERNEL_DS);
+		ret = file->f_op->write(file, (__force const char __user *)buf,
+				count, pos);
+		set_fs(old_fs);
+	} else {
 		ret = -EINVAL;
-	set_fs(old_fs);
+	}
 	if (ret > 0) {
 		fsnotify_modify(file);
 		add_wchar(current, ret);
Signature · 1 line
2.26.2
#12 lore raw

PATCH 10/14 fs: add a __kernel_read helper

In reply to Christoph Hellwig

This is the counterpart to __kernel_write, and skip the rw_verify_area call compared to kernel_read.

---
 fs/read_write.c    | 21 +++++++++++++++++++++
 include/linux/fs.h |  1 +
 2 files changed, 22 insertions(+)
fs/read_write.c +21 -0
index 8cfca5f8fc3ce..bd12af8a895c8 100644
@@ -430,6 +430,27 @@ ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
 		return -EINVAL;
 }
 
+ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
+{
+	mm_segment_t old_fs = get_fs();
+	ssize_t ret;
+
+	if (!(file->f_mode & FMODE_CAN_READ))
+		return -EINVAL;
+
+	if (count > MAX_RW_COUNT)
+		count =  MAX_RW_COUNT;
+	set_fs(KERNEL_DS);
+	ret = __vfs_read(file, (void __user *)buf, count, pos);
+	set_fs(old_fs);
+	if (ret > 0) {
+		fsnotify_access(file);
+		add_rchar(current, ret);
+	}
+	inc_syscr(current);
+	return ret;
+}
+
 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 {
 	mm_segment_t old_fs;
include/linux/fs.h +1 -0
index 21f126957c2cf..6441aaa25f8f2 100644
@@ -3011,6 +3011,7 @@ extern int kernel_read_file_from_path_initns(const char *, void **, loff_t *, lo
 extern int kernel_read_file_from_fd(int, void **, loff_t *, loff_t,
 				    enum kernel_read_file_id);
 extern ssize_t kernel_read(struct file *, void *, size_t, loff_t *);
+ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos);
 extern ssize_t kernel_write(struct file *, const void *, size_t, loff_t *);
 extern ssize_t __kernel_write(struct file *, const void *, size_t, loff_t *);
 extern struct file * open_exec(const char *);
Signature · 1 line
2.26.2
#13 lore raw

PATCH 11/14 integrity/ima: switch to using __kernel_read

In reply to Christoph Hellwig

__kernel_read has a bunch of additional sanity checks, and this moves the set_fs out of non-core code.

---
 security/integrity/iint.c | 14 +-------------
 1 file changed, 1 insertion(+), 13 deletions(-)
security/integrity/iint.c +1 -13
index e12c4900510f6..1d20003243c3f 100644
@@ -188,19 +188,7 @@ DEFINE_LSM(integrity) = {
 int integrity_kernel_read(struct file *file, loff_t offset,
 			  void *addr, unsigned long count)
 {
-	mm_segment_t old_fs;
-	char __user *buf = (char __user *)addr;
-	ssize_t ret;
-
-	if (!(file->f_mode & FMODE_READ))
-		return -EBADF;
-
-	old_fs = get_fs();
-	set_fs(KERNEL_DS);
-	ret = __vfs_read(file, buf, count, &offset);
-	set_fs(old_fs);
-
-	return ret;
+	return __kernel_read(file, addr, count, &offset);
 }
 
 /*
Signature · 1 line
2.26.2
#14 lore raw

PATCH 12/14 fs: implement kernel_read using __kernel_read

In reply to Christoph Hellwig

Consolidate the two in-kernel read helpers to make upcoming changes easier. The only difference are the missing call to rw_verify_area in kernel_read, and an access_ok check that doesn't make sense for kernel buffers to start with.

---
 fs/read_write.c | 13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)
fs/read_write.c +5 -8
index bd12af8a895c8..4e19152a7efe0 100644
@@ -453,15 +453,12 @@ ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 
 ssize_t kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 {
-	mm_segment_t old_fs;
-	ssize_t result;
+	ssize_t ret;
 
-	old_fs = get_fs();
-	set_fs(KERNEL_DS);
-	/* The cast to a user pointer is valid due to the set_fs() */
-	result = vfs_read(file, (void __user *)buf, count, pos);
-	set_fs(old_fs);
-	return result;
+	ret = rw_verify_area(READ, file, pos, count);
+	if (ret)
+		return ret;
+	return __kernel_read(file, buf, count, pos);
 }
 EXPORT_SYMBOL(kernel_read);
 
Signature · 1 line
2.26.2
#15 lore raw

PATCH 13/14 fs: remove __vfs_read

In reply to Christoph Hellwig

Fold it into the two callers.

---
 fs/read_write.c    | 43 +++++++++++++++++++++----------------------
 include/linux/fs.h |  1 -
 2 files changed, 21 insertions(+), 23 deletions(-)
fs/read_write.c +21 -22
index 4e19152a7efe0..46ddfce17e839 100644
@@ -419,17 +419,6 @@ static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, lo
 	return ret;
 }
 
-ssize_t __vfs_read(struct file *file, char __user *buf, size_t count,
-		   loff_t *pos)
-{
-	if (file->f_op->read)
-		return file->f_op->read(file, buf, count, pos);
-	else if (file->f_op->read_iter)
-		return new_sync_read(file, buf, count, pos);
-	else
-		return -EINVAL;
-}
-
 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 {
 	mm_segment_t old_fs = get_fs();
@@ -441,7 +430,12 @@ ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 	if (count > MAX_RW_COUNT)
 		count =  MAX_RW_COUNT;
 	set_fs(KERNEL_DS);
-	ret = __vfs_read(file, (void __user *)buf, count, pos);
+	if (file->f_op->read)
+		ret = file->f_op->read(file, (void __user *)buf, count, pos);
+	else if (file->f_op->read_iter)
+		ret = new_sync_read(file, (void __user *)buf, count, pos);
+	else
+		ret = -EINVAL;
 	set_fs(old_fs);
 	if (ret > 0) {
 		fsnotify_access(file);
@@ -474,17 +468,22 @@ ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 		return -EFAULT;
 
 	ret = rw_verify_area(READ, file, pos, count);
-	if (!ret) {
-		if (count > MAX_RW_COUNT)
-			count =  MAX_RW_COUNT;
-		ret = __vfs_read(file, buf, count, pos);
-		if (ret > 0) {
-			fsnotify_access(file);
-			add_rchar(current, ret);
-		}
-		inc_syscr(current);
-	}
+	if (ret)
+		return ret;
+	if (count > MAX_RW_COUNT)
+		count =  MAX_RW_COUNT;
 
+	if (file->f_op->read)
+		ret = file->f_op->read(file, buf, count, pos);
+	else if (file->f_op->read_iter)
+		ret = new_sync_read(file, buf, count, pos);
+	else
+		ret = -EINVAL;
+	if (ret > 0) {
+		fsnotify_access(file);
+		add_rchar(current, ret);
+	}
+	inc_syscr(current);
 	return ret;
 }
 
include/linux/fs.h +0 -1
index 6441aaa25f8f2..4c10a07a36178 100644
@@ -1905,7 +1905,6 @@ ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 			      struct iovec *fast_pointer,
 			      struct iovec **ret_pointer);
 
-extern ssize_t __vfs_read(struct file *, char __user *, size_t, loff_t *);
 extern ssize_t vfs_read(struct file *, char __user *, size_t, loff_t *);
 extern ssize_t vfs_write(struct file *, const char __user *, size_t, loff_t *);
 extern ssize_t vfs_readv(struct file *, const struct iovec __user *,
Signature · 1 line
2.26.2
#16 lore raw

PATCH 14/14 fs: don't change the address limit for ->read_iter in __kernel_read

In reply to Christoph Hellwig

If we read to a file that implements ->read_iter there is no need to change the address limit if we send a kvec down. Implement that case, and prefer it over using plain ->read with a changed address limit if available.

---
 fs/read_write.c | 24 +++++++++++++++++-------
 1 file changed, 17 insertions(+), 7 deletions(-)
fs/read_write.c +17 -7
index 46ddfce17e839..c93acbd8bf5a3 100644
@@ -421,7 +421,6 @@ static ssize_t new_sync_read(struct file *filp, char __user *buf, size_t len, lo
 
 ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 {
-	mm_segment_t old_fs = get_fs();
 	ssize_t ret;
 
 	if (!(file->f_mode & FMODE_CAN_READ))
@@ -429,14 +428,25 @@ ssize_t __kernel_read(struct file *file, void *buf, size_t count, loff_t *pos)
 
 	if (count > MAX_RW_COUNT)
 		count =  MAX_RW_COUNT;
-	set_fs(KERNEL_DS);
-	if (file->f_op->read)
+	if (file->f_op->read_iter) {
+		struct kvec iov = { .iov_base = buf, .iov_len = count };
+		struct kiocb kiocb;
+		struct iov_iter iter;
+
+		init_sync_kiocb(&kiocb, file);
+		kiocb.ki_pos = *pos;
+		iov_iter_kvec(&iter, READ, &iov, 1, count);
+		ret = file->f_op->read_iter(&kiocb, &iter);
+		*pos = kiocb.ki_pos;
+	} else if (file->f_op->read) {
+		mm_segment_t old_fs = get_fs();
+
+		set_fs(KERNEL_DS);
 		ret = file->f_op->read(file, (void __user *)buf, count, pos);
-	else if (file->f_op->read_iter)
-		ret = new_sync_read(file, (void __user *)buf, count, pos);
-	else
+		set_fs(old_fs);
+	} else {
 		ret = -EINVAL;
-	set_fs(old_fs);
+	}
 	if (ret > 0) {
 		fsnotify_access(file);
 		add_rchar(current, ret);
Signature · 1 line
2.26.2

A message that is not in the archive (sent off-list or not yet received): [email protected]

#18 lore raw

Re: [PATCH 09/14] fs: don't change the address limit for ->write_iter in __kernel_write

Quoted 10 lines · Hillf Danton
Quoted 10 lines
(quoted diff) +10 -0
+	if (file->f_op->write_iter) {
+		struct kvec iov = { .iov_base = (void *)buf, .iov_len = count };
+		struct kiocb kiocb;
+		struct iov_iter iter;
+
+		init_sync_kiocb(&kiocb, file);
+		kiocb.ki_pos = *pos;
+		iov_iter_kvec(&iter, WRITE, &iov, 1, count);
+		ret = file->f_op->write_iter(&kiocb, &iter);
+		if (ret > 0)
Quoted 8 lines

Review

Quoted 5 lines
(quoted diff) +5 -0
+			*pos = kiocb.ki_pos;
+	} else if (file->f_op->write) {
+		mm_segment_t old_fs = get_fs();
+
+		set_fs(KERNEL_DS);

Would you please shed light on who need it if a workqueue worker does not, given the access to buf?

Can you rephrase the question, I unfortunately do not understand it at all as-is.

#19 lore raw
In reply to Christoph Hellwig

ping?

Quoted 10 lines · Christoph Hellwig

Hi Al,

this series fixes a few issues and cleans up the helpers that read from or write to kernel space buffers, and ensures that we don't change the address limit if we are using the ->read_iter and ->write_iter methods that don't need the changed address limit.

Changes since v1:
- __kernel_write must not take sb_writers
- unexported __kernel_write

---end quoted text---