From ec56b1f1fdc69599963574ce94cc5693d535dd64 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 4 Jun 2015 09:18:18 +1000 Subject: xfs: mmap lock needs to be inside freeze protection Lock ordering for the new mmap lock needs to be: mmap_sem sb_start_pagefault i_mmap_lock page lock Right now xfs_vm_page_mkwrite gets this the wrong way around, While technically it cannot deadlock due to the current freeze ordering, it's still a landmine that might explode if we change anything in future. Hence we need to nest the locks correctly. Signed-off-by: Dave Chinner Reviewed-by: Jan Kara Reviewed-by: Brian Foster Signed-off-by: Dave Chinner --- fs/xfs/xfs_file.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'fs/xfs/xfs_file.c') diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index 8121e75352ee..0b4e79fd8d05 100644 --- a/fs/xfs/xfs_file.c +++ b/fs/xfs/xfs_file.c @@ -1487,15 +1487,20 @@ xfs_filemap_page_mkwrite( struct vm_fault *vmf) { struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host); - int error; + int ret; trace_xfs_filemap_page_mkwrite(ip); + sb_start_pagefault(VFS_I(ip)->i_sb); + file_update_time(vma->vm_file); xfs_ilock(ip, XFS_MMAPLOCK_SHARED); - error = block_page_mkwrite(vma, vmf, xfs_get_blocks); + + ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks); + xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); + sb_end_pagefault(VFS_I(ip)->i_sb); - return error; + return block_page_mkwrite_return(ret); } const struct file_operations xfs_file_operations = { -- cgit v1.2.3 From 6b698edeeef00c127d73501b386590299f01327a Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 4 Jun 2015 09:18:53 +1000 Subject: xfs: add DAX file operations support Add the initial support for DAX file operations to XFS. This includes the necessary block allocation and mmap page fault hooks for DAX to function. Note that there are changes to the splice interfaces to ensure that for DAX splice avoids direct page cache manipulations and instead takes the DAX IO paths for read/write operations. Signed-off-by: Dave Chinner Reviewed-by: Brian Foster Signed-off-by: Dave Chinner --- fs/xfs/xfs_aops.c | 116 ++++++++++++++++++++++++++++++++++++++--------------- fs/xfs/xfs_aops.h | 7 +++- fs/xfs/xfs_file.c | 118 +++++++++++++++++++++++++++++++----------------------- 3 files changed, 158 insertions(+), 83 deletions(-) (limited to 'fs/xfs/xfs_file.c') diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c index a56960dd1684..1d195e80d62e 100644 --- a/fs/xfs/xfs_aops.c +++ b/fs/xfs/xfs_aops.c @@ -1349,7 +1349,7 @@ __xfs_get_blocks( sector_t iblock, struct buffer_head *bh_result, int create, - int direct) + bool direct) { struct xfs_inode *ip = XFS_I(inode); struct xfs_mount *mp = ip->i_mount; @@ -1414,6 +1414,7 @@ __xfs_get_blocks( if (error) return error; new = 1; + } else { /* * Delalloc reservations do not require a transaction, @@ -1508,49 +1509,29 @@ xfs_get_blocks( struct buffer_head *bh_result, int create) { - return __xfs_get_blocks(inode, iblock, bh_result, create, 0); + return __xfs_get_blocks(inode, iblock, bh_result, create, false); } -STATIC int +int xfs_get_blocks_direct( struct inode *inode, sector_t iblock, struct buffer_head *bh_result, int create) { - return __xfs_get_blocks(inode, iblock, bh_result, create, 1); + return __xfs_get_blocks(inode, iblock, bh_result, create, true); } -/* - * Complete a direct I/O write request. - * - * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. - * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite - * wholly within the EOF and so there is nothing for us to do. Note that in this - * case the completion can be called in interrupt context, whereas if we have an - * ioend we will always be called in task context (i.e. from a workqueue). - */ -STATIC void -xfs_end_io_direct_write( - struct kiocb *iocb, +static void +__xfs_end_io_direct_write( + struct inode *inode, + struct xfs_ioend *ioend, loff_t offset, - ssize_t size, - void *private) + ssize_t size) { - struct inode *inode = file_inode(iocb->ki_filp); - struct xfs_inode *ip = XFS_I(inode); - struct xfs_mount *mp = ip->i_mount; - struct xfs_ioend *ioend = private; + struct xfs_mount *mp = XFS_I(inode)->i_mount; - trace_xfs_gbmap_direct_endio(ip, offset, size, - ioend ? ioend->io_type : 0, NULL); - - if (!ioend) { - ASSERT(offset + size <= i_size_read(inode)); - return; - } - - if (XFS_FORCED_SHUTDOWN(mp)) + if (XFS_FORCED_SHUTDOWN(mp) || ioend->io_error) goto out_end_io; /* @@ -1587,10 +1568,10 @@ xfs_end_io_direct_write( * here can result in EOF moving backwards and Bad Things Happen when * that occurs. */ - spin_lock(&ip->i_flags_lock); + spin_lock(&XFS_I(inode)->i_flags_lock); if (offset + size > i_size_read(inode)) i_size_write(inode, offset + size); - spin_unlock(&ip->i_flags_lock); + spin_unlock(&XFS_I(inode)->i_flags_lock); /* * If we are doing an append IO that needs to update the EOF on disk, @@ -1607,6 +1588,75 @@ out_end_io: return; } +/* + * Complete a direct I/O write request. + * + * The ioend structure is passed from __xfs_get_blocks() to tell us what to do. + * If no ioend exists (i.e. @private == NULL) then the write IO is an overwrite + * wholly within the EOF and so there is nothing for us to do. Note that in this + * case the completion can be called in interrupt context, whereas if we have an + * ioend we will always be called in task context (i.e. from a workqueue). + */ +STATIC void +xfs_end_io_direct_write( + struct kiocb *iocb, + loff_t offset, + ssize_t size, + void *private) +{ + struct inode *inode = file_inode(iocb->ki_filp); + struct xfs_ioend *ioend = private; + + trace_xfs_gbmap_direct_endio(XFS_I(inode), offset, size, + ioend ? ioend->io_type : 0, NULL); + + if (!ioend) { + ASSERT(offset + size <= i_size_read(inode)); + return; + } + + __xfs_end_io_direct_write(inode, ioend, offset, size); +} + +/* + * For DAX we need a mapping buffer callback for unwritten extent conversion + * when page faults allocate blocks and then zero them. Note that in this + * case the mapping indicated by the ioend may extend beyond EOF. We most + * definitely do not want to extend EOF here, so we trim back the ioend size to + * EOF. + */ +#ifdef CONFIG_FS_DAX +void +xfs_end_io_dax_write( + struct buffer_head *bh, + int uptodate) +{ + struct xfs_ioend *ioend = bh->b_private; + struct inode *inode = ioend->io_inode; + ssize_t size = ioend->io_size; + + ASSERT(IS_DAX(ioend->io_inode)); + + /* if there was an error zeroing, then don't convert it */ + if (!uptodate) + ioend->io_error = -EIO; + + /* + * Trim update to EOF, so we don't extend EOF during unwritten extent + * conversion of partial EOF blocks. + */ + spin_lock(&XFS_I(inode)->i_flags_lock); + if (ioend->io_offset + size > i_size_read(inode)) + size = i_size_read(inode) - ioend->io_offset; + spin_unlock(&XFS_I(inode)->i_flags_lock); + + __xfs_end_io_direct_write(inode, ioend, ioend->io_offset, size); + +} +#else +void xfs_end_io_dax_write(struct buffer_head *bh, int uptodate) { } +#endif + STATIC ssize_t xfs_vm_direct_IO( struct kiocb *iocb, diff --git a/fs/xfs/xfs_aops.h b/fs/xfs/xfs_aops.h index ac644e0137a4..86afd1ac7895 100644 --- a/fs/xfs/xfs_aops.h +++ b/fs/xfs/xfs_aops.h @@ -53,7 +53,12 @@ typedef struct xfs_ioend { } xfs_ioend_t; extern const struct address_space_operations xfs_address_space_operations; -extern int xfs_get_blocks(struct inode *, sector_t, struct buffer_head *, int); + +int xfs_get_blocks(struct inode *inode, sector_t offset, + struct buffer_head *map_bh, int create); +int xfs_get_blocks_direct(struct inode *inode, sector_t offset, + struct buffer_head *map_bh, int create); +void xfs_end_io_dax_write(struct buffer_head *bh, int uptodate); extern void xfs_count_page_state(struct page *, int *, int *); diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index 0b4e79fd8d05..a629dce4903e 100644 --- a/fs/xfs/xfs_file.c +++ b/fs/xfs/xfs_file.c @@ -284,7 +284,7 @@ xfs_file_read_iter( if (file->f_mode & FMODE_NOCMTIME) ioflags |= XFS_IO_INVIS; - if (unlikely(ioflags & XFS_IO_ISDIRECT)) { + if ((ioflags & XFS_IO_ISDIRECT) && !IS_DAX(inode)) { xfs_buftarg_t *target = XFS_IS_REALTIME_INODE(ip) ? mp->m_rtdev_targp : mp->m_ddev_targp; @@ -378,7 +378,11 @@ xfs_file_splice_read( trace_xfs_file_splice_read(ip, count, *ppos, ioflags); - ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); + /* for dax, we need to avoid the page cache */ + if (IS_DAX(VFS_I(ip))) + ret = default_file_splice_read(infilp, ppos, pipe, count, flags); + else + ret = generic_file_splice_read(infilp, ppos, pipe, count, flags); if (ret > 0) XFS_STATS_ADD(xs_read_bytes, ret); @@ -672,7 +676,7 @@ xfs_file_dio_aio_write( mp->m_rtdev_targp : mp->m_ddev_targp; /* DIO must be aligned to device logical sector size */ - if ((pos | count) & target->bt_logical_sectormask) + if (!IS_DAX(inode) && ((pos | count) & target->bt_logical_sectormask)) return -EINVAL; /* "unaligned" here means not aligned to a filesystem block */ @@ -758,8 +762,11 @@ xfs_file_dio_aio_write( out: xfs_rw_iunlock(ip, iolock); - /* No fallback to buffered IO on errors for XFS. */ - ASSERT(ret < 0 || ret == count); + /* + * No fallback to buffered IO on errors for XFS. DAX can result in + * partial writes, but direct IO will either complete fully or fail. + */ + ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip))); return ret; } @@ -842,7 +849,7 @@ xfs_file_write_iter( if (XFS_FORCED_SHUTDOWN(ip->i_mount)) return -EIO; - if (unlikely(iocb->ki_flags & IOCB_DIRECT)) + if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode)) ret = xfs_file_dio_aio_write(iocb, from); else ret = xfs_file_buffered_aio_write(iocb, from); @@ -1063,17 +1070,6 @@ xfs_file_readdir( return xfs_readdir(ip, ctx, bufsize); } -STATIC int -xfs_file_mmap( - struct file *filp, - struct vm_area_struct *vma) -{ - vma->vm_ops = &xfs_file_vm_ops; - - file_accessed(filp); - return 0; -} - /* * This type is designed to indicate the type of offset we would like * to search from page cache for xfs_seek_hole_data(). @@ -1454,26 +1450,11 @@ xfs_file_llseek( * ordering of: * * mmap_sem (MM) - * i_mmap_lock (XFS - truncate serialisation) - * page_lock (MM) - * i_lock (XFS - extent map serialisation) + * sb_start_pagefault(vfs, freeze) + * i_mmap_lock (XFS - truncate serialisation) + * page_lock (MM) + * i_lock (XFS - extent map serialisation) */ -STATIC int -xfs_filemap_fault( - struct vm_area_struct *vma, - struct vm_fault *vmf) -{ - struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host); - int error; - - trace_xfs_filemap_fault(ip); - - xfs_ilock(ip, XFS_MMAPLOCK_SHARED); - error = filemap_fault(vma, vmf); - xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); - - return error; -} /* * mmap()d file has taken write protection fault and is being made writable. We @@ -1486,21 +1467,66 @@ xfs_filemap_page_mkwrite( struct vm_area_struct *vma, struct vm_fault *vmf) { - struct xfs_inode *ip = XFS_I(vma->vm_file->f_mapping->host); + struct inode *inode = file_inode(vma->vm_file); int ret; - trace_xfs_filemap_page_mkwrite(ip); + trace_xfs_filemap_page_mkwrite(XFS_I(inode)); - sb_start_pagefault(VFS_I(ip)->i_sb); + sb_start_pagefault(inode->i_sb); file_update_time(vma->vm_file); - xfs_ilock(ip, XFS_MMAPLOCK_SHARED); + xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED); + + if (IS_DAX(inode)) { + ret = __dax_mkwrite(vma, vmf, xfs_get_blocks_direct, + xfs_end_io_dax_write); + } else { + ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks); + ret = block_page_mkwrite_return(ret); + } + + xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED); + sb_end_pagefault(inode->i_sb); + + return ret; +} + +STATIC int +xfs_filemap_fault( + struct vm_area_struct *vma, + struct vm_fault *vmf) +{ + struct xfs_inode *ip = XFS_I(file_inode(vma->vm_file)); + int ret; - ret = __block_page_mkwrite(vma, vmf, xfs_get_blocks); + trace_xfs_filemap_fault(ip); + + /* DAX can shortcut the normal fault path on write faults! */ + if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(VFS_I(ip))) + return xfs_filemap_page_mkwrite(vma, vmf); + xfs_ilock(ip, XFS_MMAPLOCK_SHARED); + ret = filemap_fault(vma, vmf); xfs_iunlock(ip, XFS_MMAPLOCK_SHARED); - sb_end_pagefault(VFS_I(ip)->i_sb); - return block_page_mkwrite_return(ret); + return ret; +} + +static const struct vm_operations_struct xfs_file_vm_ops = { + .fault = xfs_filemap_fault, + .map_pages = filemap_map_pages, + .page_mkwrite = xfs_filemap_page_mkwrite, +}; + +STATIC int +xfs_file_mmap( + struct file *filp, + struct vm_area_struct *vma) +{ + file_accessed(filp); + vma->vm_ops = &xfs_file_vm_ops; + if (IS_DAX(file_inode(filp))) + vma->vm_flags |= VM_MIXEDMAP; + return 0; } const struct file_operations xfs_file_operations = { @@ -1531,9 +1557,3 @@ const struct file_operations xfs_dir_file_operations = { #endif .fsync = xfs_dir_fsync, }; - -static const struct vm_operations_struct xfs_file_vm_ops = { - .fault = xfs_filemap_fault, - .map_pages = filemap_map_pages, - .page_mkwrite = xfs_filemap_page_mkwrite, -}; -- cgit v1.2.3 From 4f69f578a87d39c20b1ff70005a125e4594c3de8 Mon Sep 17 00:00:00 2001 From: Dave Chinner Date: Thu, 4 Jun 2015 09:19:08 +1000 Subject: xfs: add DAX block zeroing support Add initial support for DAX block zeroing operations to XFS. DAX cannot use buffered IO through the page cache for zeroing, nor do we need to issue IO for uncached block zeroing. In both cases, we can simply call out to the dax block zeroing function. Signed-off-by: Dave Chinner Reviewed-by: Brian Foster Signed-off-by: Dave Chinner --- fs/xfs/xfs_bmap_util.c | 23 +++++++++++++++++++---- fs/xfs/xfs_file.c | 45 +++++++++++++++++++++++++++------------------ 2 files changed, 46 insertions(+), 22 deletions(-) (limited to 'fs/xfs/xfs_file.c') diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c index a52bbd3abc7d..4a2965515ca8 100644 --- a/fs/xfs/xfs_bmap_util.c +++ b/fs/xfs/xfs_bmap_util.c @@ -1133,14 +1133,29 @@ xfs_zero_remaining_bytes( break; ASSERT(imap.br_blockcount >= 1); ASSERT(imap.br_startoff == offset_fsb); + ASSERT(imap.br_startblock != DELAYSTARTBLOCK); + + if (imap.br_startblock == HOLESTARTBLOCK || + imap.br_state == XFS_EXT_UNWRITTEN) { + /* skip the entire extent */ + lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + + imap.br_blockcount) - 1; + continue; + } + lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1; if (lastoffset > endoff) lastoffset = endoff; - if (imap.br_startblock == HOLESTARTBLOCK) - continue; - ASSERT(imap.br_startblock != DELAYSTARTBLOCK); - if (imap.br_state == XFS_EXT_UNWRITTEN) + + /* DAX can just zero the backing device directly */ + if (IS_DAX(VFS_I(ip))) { + error = dax_zero_page_range(VFS_I(ip), offset, + lastoffset - offset + 1, + xfs_get_blocks_direct); + if (error) + return error; continue; + } error = xfs_buf_read_uncached(XFS_IS_REALTIME_INODE(ip) ? mp->m_rtdev_targp : mp->m_ddev_targp, diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index a629dce4903e..cfd9b4f5ad6e 100644 --- a/fs/xfs/xfs_file.c +++ b/fs/xfs/xfs_file.c @@ -79,14 +79,15 @@ xfs_rw_ilock_demote( } /* - * xfs_iozero + * xfs_iozero clears the specified range supplied via the page cache (except in + * the DAX case). Writes through the page cache will allocate blocks over holes, + * though the callers usually map the holes first and avoid them. If a block is + * not completely zeroed, then it will be read from disk before being partially + * zeroed. * - * xfs_iozero clears the specified range of buffer supplied, - * and marks all the affected blocks as valid and modified. If - * an affected block is not allocated, it will be allocated. If - * an affected block is not completely overwritten, and is not - * valid before the operation, it will be read from disk before - * being partially zeroed. + * In the DAX case, we can just directly write to the underlying pages. This + * will not allocate blocks, but will avoid holes and unwritten extents and so + * not do unnecessary work. */ int xfs_iozero( @@ -96,7 +97,8 @@ xfs_iozero( { struct page *page; struct address_space *mapping; - int status; + int status = 0; + mapping = VFS_I(ip)->i_mapping; do { @@ -108,20 +110,27 @@ xfs_iozero( if (bytes > count) bytes = count; - status = pagecache_write_begin(NULL, mapping, pos, bytes, - AOP_FLAG_UNINTERRUPTIBLE, - &page, &fsdata); - if (status) - break; + if (IS_DAX(VFS_I(ip))) { + status = dax_zero_page_range(VFS_I(ip), pos, bytes, + xfs_get_blocks_direct); + if (status) + break; + } else { + status = pagecache_write_begin(NULL, mapping, pos, bytes, + AOP_FLAG_UNINTERRUPTIBLE, + &page, &fsdata); + if (status) + break; - zero_user(page, offset, bytes); + zero_user(page, offset, bytes); - status = pagecache_write_end(NULL, mapping, pos, bytes, bytes, - page, fsdata); - WARN_ON(status <= 0); /* can't return less than zero! */ + status = pagecache_write_end(NULL, mapping, pos, bytes, + bytes, page, fsdata); + WARN_ON(status <= 0); /* can't return less than zero! */ + status = 0; + } pos += bytes; count -= bytes; - status = 0; } while (count); return (-status); -- cgit v1.2.3