* [PATCH 1/2] fs: Add iomap_swap_activate
2018-05-01 18:48 [PATCH 0/2] RFC: iomap-based swapfile activation Aleksei Besogonov
@ 2018-05-01 18:48 ` Aleksei Besogonov
2018-05-01 20:44 ` kbuild test robot
2018-05-02 1:02 ` kbuild test robot
2018-05-01 18:48 ` [PATCH 2/2] xfs: add support for iomap-based swapfile activation Aleksei Besogonov
2018-05-01 19:16 ` [PATCH 0/2] RFC: " Eric Sandeen
2 siblings, 2 replies; 6+ messages in thread
From: Aleksei Besogonov @ 2018-05-01 18:48 UTC (permalink / raw)
To: linux-xfs, linux-fsdevel, darrick.wong; +Cc: Aleksei Besogonov
This commit adds an alternative to generic_swapfile_activate function
based on iomap family. The major driver for this is swap file support
on XFS, currently it doesn't work on files with holes due to deficiencies
in the old bmap() interface that is used by generic_swapfile_activate.
---
fs/iomap.c | 106 ++++++++++++++++++++++++++++++++++++++++++++++++++
include/linux/iomap.h | 5 +++
mm/swapfile.c | 1 +
3 files changed, 112 insertions(+)
diff --git a/fs/iomap.c b/fs/iomap.c
index afd163586aa0..8ac6dcbfbeb6 100644
--- a/fs/iomap.c
+++ b/fs/iomap.c
@@ -94,6 +94,7 @@ iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
return written ? written : ret;
}
+EXPORT_SYMBOL_GPL(iomap_apply);
static void
iomap_write_failed(struct inode *inode, loff_t pos, unsigned len)
@@ -1089,3 +1090,108 @@ iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
return ret;
}
EXPORT_SYMBOL_GPL(iomap_dio_rw);
+
+struct iomap_swap_walker_context
+{
+ struct swap_info_struct *sis;
+ unsigned int page_no;
+ unsigned int nr_extents;
+ sector_t *span;
+};
+
+static loff_t iomap_swap_map(struct inode *inode, loff_t pos, loff_t length,
+ void *data, struct iomap *iomap)
+{
+ struct iomap_swap_walker_context *ctx = data;
+ unsigned int blocks_per_page = PAGE_SIZE >> inode->i_blkbits;
+ u64 num_pages;
+ u64 aligned_address;
+ sector_t disk_pos;
+ loff_t eff_length;
+ loff_t ret;
+
+ /* Validate the extent */
+
+ /* Only one bdev per swap file. */
+ if (iomap->bdev != ctx->sis->bdev)
+ goto err;
+ /* Only real or unwritten extents. */
+ if (iomap->type != IOMAP_MAPPED && iomap->type != IOMAP_UNWRITTEN)
+ goto err;
+ /* No uncommitted metadata or shared blocks or inline data. */
+ if (iomap->flags & (IOMAP_F_DIRTY | IOMAP_F_SHARED | IOMAP_F_DATA_INLINE))
+ goto err;
+
+ /*
+ * Swap extents need to be PAGE_SIZE aligned on the disk
+ */
+ aligned_address = ALIGN(iomap->addr, PAGE_SIZE);
+ eff_length = length - (aligned_address - iomap->addr);
+ BUG_ON(eff_length > length);
+ disk_pos = aligned_address >> inode->i_blkbits;
+
+ if (eff_length < 0)
+ return length; /* Continue probing */
+
+ num_pages = eff_length / PAGE_SIZE;
+
+ /* Can't add blocks less than 1 page */
+ if (num_pages == 0)
+ return length; /* Continue probing */
+
+ ret = add_swap_extent(ctx->sis, ctx->page_no, num_pages, disk_pos);
+ if (ret < 0)
+ return ret;
+
+ ctx->nr_extents += ret;
+ ctx->page_no += num_pages;
+ (*ctx->span) += num_pages * blocks_per_page;
+
+ return length; /* continue probing */
+err:
+ pr_err("swapon: swapfile has holes\n");
+ return -EINVAL;
+}
+
+int iomap_swap_activate(struct swap_info_struct *sis, struct file *file,
+ sector_t *span, const struct iomap_ops *ops)
+{
+ struct iomap_swap_walker_context ctx = {
+ .sis = sis,
+ .page_no = 0,
+ .nr_extents = 0,
+ .span = span
+ };
+
+ struct inode * inode = file_inode(file);
+ loff_t len = i_size_read(inode);
+ loff_t pos = 0;
+ loff_t ret = -EINVAL;
+
+
+ ret = filemap_write_and_wait(inode->i_mapping);
+ if (ret)
+ return ret;
+
+ while(len > 0) {
+ ret = iomap_apply(inode, pos, len, IOMAP_REPORT, ops,
+ &ctx, iomap_swap_map);
+ if (ret <= 0)
+ break;
+
+ pos += ret;
+ len -= ret;
+ }
+
+ if (ret < 0) {
+ pr_err("Failed to activate the swap file\n");
+ return -EINVAL;
+ }
+
+ sis->max = ctx.page_no == 0 ? 1 : ctx.page_no;
+ sis->pages = ctx.page_no;
+ sis->highest_bit = ctx.page_no;
+
+ return ctx.nr_extents;
+}
+EXPORT_SYMBOL_GPL(iomap_swap_activate);
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 19a07de28212..a85da0db6c2e 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -10,6 +10,8 @@ struct iov_iter;
struct kiocb;
struct vm_area_struct;
struct vm_fault;
+struct file;
+struct swap_info_struct;
/*
* Types of block ranges for iomap mappings:
@@ -106,4 +108,7 @@ typedef int (iomap_dio_end_io_t)(struct kiocb *iocb, ssize_t ret,
ssize_t iomap_dio_rw(struct kiocb *iocb, struct iov_iter *iter,
const struct iomap_ops *ops, iomap_dio_end_io_t end_io);
+int iomap_swap_activate(struct swap_info_struct *sis, struct file *file,
+ sector_t *span, const struct iomap_ops *ops);
+
#endif /* LINUX_IOMAP_H */
diff --git a/mm/swapfile.c b/mm/swapfile.c
index c7a33717d079..9e6d3c1a882e 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -2381,6 +2381,7 @@ add_swap_extent(struct swap_info_struct *sis, unsigned long start_page,
list_add_tail(&new_se->list, &sis->first_swap_extent.list);
return 1;
}
+EXPORT_SYMBOL_GPL(add_swap_extent);
/*
* A `swap extent' is a simple thing which maps a contiguous range of pages
--
2.14.1
^ permalink raw reply related [flat|nested] 6+ messages in thread* [PATCH 2/2] xfs: add support for iomap-based swapfile activation
2018-05-01 18:48 [PATCH 0/2] RFC: iomap-based swapfile activation Aleksei Besogonov
2018-05-01 18:48 ` [PATCH 1/2] fs: Add iomap_swap_activate Aleksei Besogonov
@ 2018-05-01 18:48 ` Aleksei Besogonov
2018-05-01 19:16 ` [PATCH 0/2] RFC: " Eric Sandeen
2 siblings, 0 replies; 6+ messages in thread
From: Aleksei Besogonov @ 2018-05-01 18:48 UTC (permalink / raw)
To: linux-xfs, linux-fsdevel, darrick.wong; +Cc: Aleksei Besogonov
Use iomap-based swapfile activation instead of the default
generic_swapfile_activate bmap-based infrastructure.
---
fs/xfs/xfs_aops.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 31f1f10eecd1..9c10efaabc6f 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -1491,6 +1491,13 @@ xfs_vm_set_page_dirty(
return newly_dirty;
}
+static int xfs_swap_activate(struct swap_info_struct *sis, struct file *swap_file,
+ sector_t *span)
+{
+ sis->bdev = xfs_find_bdev_for_inode(file_inode(swap_file));
+ return iomap_swap_activate(sis, swap_file, span, &xfs_iomap_ops);
+}
+
const struct address_space_operations xfs_address_space_operations = {
.readpage = xfs_vm_readpage,
.readpages = xfs_vm_readpages,
@@ -1504,4 +1511,5 @@ const struct address_space_operations xfs_address_space_operations = {
.migratepage = buffer_migrate_page,
.is_partially_uptodate = block_is_partially_uptodate,
.error_remove_page = generic_error_remove_page,
+ .swap_activate = xfs_swap_activate,
};
--
2.14.1
^ permalink raw reply related [flat|nested] 6+ messages in thread* Re: [PATCH 0/2] RFC: iomap-based swapfile activation
2018-05-01 18:48 [PATCH 0/2] RFC: iomap-based swapfile activation Aleksei Besogonov
2018-05-01 18:48 ` [PATCH 1/2] fs: Add iomap_swap_activate Aleksei Besogonov
2018-05-01 18:48 ` [PATCH 2/2] xfs: add support for iomap-based swapfile activation Aleksei Besogonov
@ 2018-05-01 19:16 ` Eric Sandeen
2 siblings, 0 replies; 6+ messages in thread
From: Eric Sandeen @ 2018-05-01 19:16 UTC (permalink / raw)
To: Aleksei Besogonov, linux-xfs, linux-fsdevel, darrick.wong
On 5/1/18 1:48 PM, Aleksei Besogonov wrote:
> V2 of iomap-based swapfile support.
>
> Changes from V1: additional validations for file extents, based on
> a patch from Darrick Wong.
Aleksei -
As Darrick pointed out earlier and as you acknowledge here, Darrick already
has a patchset for this in the works, started a couple days before your first
submission.
Generally we don't race to re-implement someone else's patch. If you were unsure
whether Darrick planned to send a V2, it would have been good to ask.
Although I think Darrick does plan to resubmit his patches based on the feedback
provided so far, I'll go ahead & point out a couple things in your series for
future reference:
1) You need a Signed-off-by: line which essentially states that this is your
own work, and that you may submit it for inclusion.
2) If you were building on someone else's patch with their blessing, you'd
probably preceed your Signed-off-by: with theirs (but this is a bit
unusual)
3) There's no reason to export iomap_apply for modules; no module uses it
I'll admit that I didn't look a lot more in depth. ;)
At this point it's probably best to help this along by reviewing & testing
Darrick's (forthcoming?) V2 patchset.
It's not that we don't appreciate the help - it's great that you're motivated
to get this done - it's just that someone else already started the work, and
the normal process there is to review/comment on the first submitter's patches,
unless they have clearly abandoned the work. This keeps the development
process and the mailing list more organized than if several people submit
competing patchsets in parallel.
Thanks,
-Eric
> Aleksei Besogonov (2):
> fs: Add iomap_swap_activate
> xfs: add support for iomap-based swapfile activation
>
> fs/iomap.c | 106 ++++++++++++++++++++++++++++++++++++++++++++++++++
> fs/xfs/xfs_aops.c | 8 ++++
> include/linux/iomap.h | 5 +++
> mm/swapfile.c | 1 +
> 4 files changed, 120 insertions(+)
>
^ permalink raw reply [flat|nested] 6+ messages in thread