I am looking into getting rid of dependency on block device in dax path. One such place is __dax_zero_page_range() which checks if range being zeroed is aligned to block device logical size, then it calls bdev_issue_zeroout() instead of doing memset(). Calling blkdev_issue_zeroout() also clears bad blocks and poison if any in that range. This path is used by iomap_zero_range() which in-turn is being used by filesystems to zero partial filesystem system blocks. For zeroing full filesystem blocks we seem to be calling blkdev_issue_zeroout() which clears bad blocks and poison in that range. So this code currently only seems to help with partial filesystem block zeroing. That too only if truncation/hole_punch happens on device logical block size boundary. To avoid using blkdev_issue_zeroout() in this path, I proposed another patch here which adds another dax operation to zero out a rangex and clear poison. https://lore.kernel.org/linux-fsdevel/20200123165249.GA7664@xxxxxxxxxx/ Thinking more about it, it might be an overkill to solve this problem. How about if we just not clear poison/bad blocks when a partial page is being zeroed. IOW, users will need to hole_punch whole filesystem block worth of data which will free that block and it will be zeroed some other time and clear poison in the process. For partial fs block truncation/punch_hole, we don't clear poison. If above interface is acceptable, then we can get rid of code which tries to call blkdev_issue_zeroout() in iomap_zero_range() path and we don't have to implement another dax operation. Looking for some feedback on this. Vivek Signed-off-by: Vivek Goyal <vgoyal@xxxxxxxxxx> --- fs/dax.c | 50 +++++++++++++++----------------------------------- 1 file changed, 15 insertions(+), 35 deletions(-) Index: redhat-linux/fs/dax.c =================================================================== --- redhat-linux.orig/fs/dax.c 2020-01-29 15:19:18.551902448 -0500 +++ redhat-linux/fs/dax.c 2020-01-29 15:40:56.584824549 -0500 @@ -1044,47 +1044,27 @@ static vm_fault_t dax_load_hole(struct x return ret; } -static bool dax_range_is_aligned(struct block_device *bdev, - unsigned int offset, unsigned int length) -{ - unsigned short sector_size = bdev_logical_block_size(bdev); - - if (!IS_ALIGNED(offset, sector_size)) - return false; - if (!IS_ALIGNED(length, sector_size)) - return false; - - return true; -} - int __dax_zero_page_range(struct block_device *bdev, struct dax_device *dax_dev, sector_t sector, unsigned int offset, unsigned int size) { - if (dax_range_is_aligned(bdev, offset, size)) { - sector_t start_sector = sector + (offset >> 9); - - return blkdev_issue_zeroout(bdev, start_sector, - size >> 9, GFP_NOFS, 0); - } else { - pgoff_t pgoff; - long rc, id; - void *kaddr; - - rc = bdev_dax_pgoff(bdev, sector, PAGE_SIZE, &pgoff); - if (rc) - return rc; - - id = dax_read_lock(); - rc = dax_direct_access(dax_dev, pgoff, 1, &kaddr, NULL); - if (rc < 0) { - dax_read_unlock(id); - return rc; - } - memset(kaddr + offset, 0, size); - dax_flush(dax_dev, kaddr + offset, size); + pgoff_t pgoff; + long rc, id; + void *kaddr; + + rc = bdev_dax_pgoff(bdev, sector, PAGE_SIZE, &pgoff); + if (rc) + return rc; + + id = dax_read_lock(); + rc = dax_direct_access(dax_dev, pgoff, 1, &kaddr, NULL); + if (rc < 0) { dax_read_unlock(id); + return rc; } + memset(kaddr + offset, 0, size); + dax_flush(dax_dev, kaddr + offset, size); + dax_read_unlock(id); return 0; } EXPORT_SYMBOL_GPL(__dax_zero_page_range);
