On Sun, Apr 28, 2024 at 12:46:23AM +0100, Al Viro wrote:
> On Sat, Apr 27, 2024 at 02:40:22PM -0700, Linus Torvalds wrote:
> > On Sat, 27 Apr 2024 at 14:11, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > ... eliminating the need to reopen block devices so they could be
> > > exclusively held.
> >
> > This looks like a good change, but it raises the question of why we
> > did it this odd way to begin with?
> >
> > Is it just because O_EXCL without O_CREAT is kind of odd, and only has
> > meaning for block devices?
> >
> > Or is it just that before we used fiel pointers for block devices, the
> > old model made more sense?
> >
> > Anyway, I like it, it just makes me go "why didn't we do it that way
> > originally?"
>
> Exclusion for swap partitions:
>
> commit 75e9c9e1bffbe4a1767172855296b94ccba28f71
> Author: Alexander Viro <viro@xxxxxxxxxxxx>
> Date: Mon Mar 4 22:56:47 2002 -0800
>
> [PATCH] death of is_mounted() and aother fixes
>
>
> O_EXCL for block devices:
>
> commit c366082d9ed0a0d3c46441d1b3fdf895d8e55ca9
> Author: Andrew Morton <akpm@xxxxxxxx>
> Date: Wed Aug 20 10:26:57 2003 -0700
>
> [PATCH] Allow O_EXCL on a block device to claim exclusive use.
>
> IOW, O_EXCL hadn't been available at the time - it had been implemented
> on top of bd_claim()/bd_release() introduced in the same earlier commit.
>
> Switching swap exclusion to O_EXCL could've been done back in 2003 or
> at any later point; it's just that swapon(2)/swapoff(2) is something that
> rarely gets a look...
FWIW, pretty much the same can be done with zram - open with O_EXCL and to
hell with reopening. Guys, are there any objections to that?
Signed-off-by: Al Viro <viro@xxxxxxxxxxxxxxxxxx>
---
diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index f0639df6cd18..d882a0c7b522 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -426,11 +426,10 @@ static void reset_bdev(struct zram *zram)
if (!zram->backing_dev)
return;
- fput(zram->bdev_file);
/* hope filp_close flush all of IO */
filp_close(zram->backing_dev, NULL);
zram->backing_dev = NULL;
- zram->bdev_file = NULL;
+ zram->bdev = NULL;
zram->disk->fops = &zram_devops;
kvfree(zram->bitmap);
zram->bitmap = NULL;
@@ -473,10 +472,8 @@ static ssize_t backing_dev_store(struct device *dev,
size_t sz;
struct file *backing_dev = NULL;
struct inode *inode;
- struct address_space *mapping;
unsigned int bitmap_sz;
unsigned long nr_pages, *bitmap = NULL;
- struct file *bdev_file = NULL;
int err;
struct zram *zram = dev_to_zram(dev);
@@ -497,15 +494,14 @@ static ssize_t backing_dev_store(struct device *dev,
if (sz > 0 && file_name[sz - 1] == '\n')
file_name[sz - 1] = 0x00;
- backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
+ backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE|O_EXCL, 0);
if (IS_ERR(backing_dev)) {
err = PTR_ERR(backing_dev);
backing_dev = NULL;
goto out;
}
- mapping = backing_dev->f_mapping;
- inode = mapping->host;
+ inode = backing_dev->f_mapping->host;
/* Support only block device in this moment */
if (!S_ISBLK(inode->i_mode)) {
@@ -513,14 +509,6 @@ static ssize_t backing_dev_store(struct device *dev,
goto out;
}
- bdev_file = bdev_file_open_by_dev(inode->i_rdev,
- BLK_OPEN_READ | BLK_OPEN_WRITE, zram, NULL);
- if (IS_ERR(bdev_file)) {
- err = PTR_ERR(bdev_file);
- bdev_file = NULL;
- goto out;
- }
-
nr_pages = i_size_read(inode) >> PAGE_SHIFT;
bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
@@ -531,7 +519,7 @@ static ssize_t backing_dev_store(struct device *dev,
reset_bdev(zram);
- zram->bdev_file = bdev_file;
+ zram->bdev = I_BDEV(inode);
zram->backing_dev = backing_dev;
zram->bitmap = bitmap;
zram->nr_pages = nr_pages;
@@ -544,9 +532,6 @@ static ssize_t backing_dev_store(struct device *dev,
out:
kvfree(bitmap);
- if (bdev_file)
- fput(bdev_file);
-
if (backing_dev)
filp_close(backing_dev, NULL);
@@ -587,7 +572,7 @@ static void read_from_bdev_async(struct zram *zram, struct page *page,
{
struct bio *bio;
- bio = bio_alloc(file_bdev(zram->bdev_file), 1, parent->bi_opf, GFP_NOIO);
+ bio = bio_alloc(zram->bdev, 1, parent->bi_opf, GFP_NOIO);
bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
__bio_add_page(bio, page, PAGE_SIZE, 0);
bio_chain(bio, parent);
@@ -703,7 +688,7 @@ static ssize_t writeback_store(struct device *dev,
continue;
}
- bio_init(&bio, file_bdev(zram->bdev_file), &bio_vec, 1,
+ bio_init(&bio, zram->bdev, &bio_vec, 1,
REQ_OP_WRITE | REQ_SYNC);
bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
__bio_add_page(&bio, page, PAGE_SIZE, 0);
@@ -785,7 +770,7 @@ static void zram_sync_read(struct work_struct *work)
struct bio_vec bv;
struct bio bio;
- bio_init(&bio, file_bdev(zw->zram->bdev_file), &bv, 1, REQ_OP_READ);
+ bio_init(&bio, zw->zram->bdev, &bv, 1, REQ_OP_READ);
bio.bi_iter.bi_sector = zw->entry * (PAGE_SIZE >> 9);
__bio_add_page(&bio, zw->page, PAGE_SIZE, 0);
zw->error = submit_bio_wait(&bio);
diff --git a/drivers/block/zram/zram_drv.h b/drivers/block/zram/zram_drv.h
index 37bf29f34d26..35e322144629 100644
--- a/drivers/block/zram/zram_drv.h
+++ b/drivers/block/zram/zram_drv.h
@@ -132,7 +132,7 @@ struct zram {
spinlock_t wb_limit_lock;
bool wb_limit_enable;
u64 bd_wb_limit;
- struct file *bdev_file;
+ struct block_device *bdev;
unsigned long *bitmap;
unsigned long nr_pages;
#endif
