From 91f68c89d8f35fe98ea04159b9a3b42d0149478f Mon Sep 17 00:00:00 2001 From: Jeff Moyer Date: Thu, 12 Jul 2012 09:43:14 -0400 Subject: block: fix infinite loop in __getblk_slow Commit 080399aaaf35 ("block: don't mark buffers beyond end of disk as mapped") exposed a bug in __getblk_slow that causes mount to hang as it loops infinitely waiting for a buffer that lies beyond the end of the disk to become uptodate. The problem was initially reported by Torsten Hilbrich here: https://lkml.org/lkml/2012/6/18/54 and also reported independently here: http://www.sysresccd.org/forums/viewtopic.php?f=13&t=4511 and then Richard W.M. Jones and Marcos Mello noted a few separate bugzillas also associated with the same issue. This patch has been confirmed to fix: https://bugzilla.redhat.com/show_bug.cgi?id=835019 The main problem is here, in __getblk_slow: for (;;) { struct buffer_head * bh; int ret; bh = __find_get_block(bdev, block, size); if (bh) return bh; ret = grow_buffers(bdev, block, size); if (ret < 0) return NULL; if (ret == 0) free_more_memory(); } __find_get_block does not find the block, since it will not be marked as mapped, and so grow_buffers is called to fill in the buffers for the associated page. I believe the for (;;) loop is there primarily to retry in the case of memory pressure keeping grow_buffers from succeeding. However, we also continue to loop for other cases, like the block lying beond the end of the disk. So, the fix I came up with is to only loop when grow_buffers fails due to memory allocation issues (return value of 0). The attached patch was tested by myself, Torsten, and Rich, and was found to resolve the problem in call cases. Signed-off-by: Jeff Moyer Reported-and-Tested-by: Torsten Hilbrich Tested-by: Richard W.M. Jones Reviewed-by: Josh Boyer Cc: Stable # 3.0+ [ Jens is on vacation, taking this directly - Linus ] -- Stable Notes: this patch requires backport to 3.0, 3.2 and 3.3. Signed-off-by: Linus Torvalds --- fs/buffer.c | 22 +++++++++++++--------- 1 file changed, 13 insertions(+), 9 deletions(-) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index 838a9cf246bd..c7062c896d7c 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -1036,6 +1036,9 @@ grow_buffers(struct block_device *bdev, sector_t block, int size) static struct buffer_head * __getblk_slow(struct block_device *bdev, sector_t block, int size) { + int ret; + struct buffer_head *bh; + /* Size must be multiple of hard sectorsize */ if (unlikely(size & (bdev_logical_block_size(bdev)-1) || (size < 512 || size > PAGE_SIZE))) { @@ -1048,20 +1051,21 @@ __getblk_slow(struct block_device *bdev, sector_t block, int size) return NULL; } - for (;;) { - struct buffer_head * bh; - int ret; +retry: + bh = __find_get_block(bdev, block, size); + if (bh) + return bh; + ret = grow_buffers(bdev, block, size); + if (ret == 0) { + free_more_memory(); + goto retry; + } else if (ret > 0) { bh = __find_get_block(bdev, block, size); if (bh) return bh; - - ret = grow_buffers(bdev, block, size); - if (ret < 0) - return NULL; - if (ret == 0) - free_more_memory(); } + return NULL; } /* -- cgit v1.2.3 From 5e8830dc85d0a6258132977381430b327cf553f2 Mon Sep 17 00:00:00 2001 From: Jan Kara Date: Tue, 12 Jun 2012 16:20:23 +0200 Subject: fs: Push file_update_time() into __block_page_mkwrite() Tested-by: Kamal Mostafa Tested-by: Peter M. Petrakis Tested-by: Dann Frazier Tested-by: Massimo Morana Signed-off-by: Jan Kara Signed-off-by: Al Viro --- fs/buffer.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index c7062c896d7c..d5ec360e332d 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -2318,6 +2318,12 @@ int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, loff_t size; int ret; + /* + * Update file times before taking page lock. We may end up failing the + * fault so this update may be superfluous but who really cares... + */ + file_update_time(vma->vm_file); + lock_page(page); size = i_size_read(inode); if ((page->mapping != inode->i_mapping) || -- cgit v1.2.3 From 14da9200140f8d722ad1767dfabadebd8b34f2ad Mon Sep 17 00:00:00 2001 From: Jan Kara Date: Tue, 12 Jun 2012 16:20:37 +0200 Subject: fs: Protect write paths by sb_start_write - sb_end_write There are several entry points which dirty pages in a filesystem. mmap (handled by block_page_mkwrite()), buffered write (handled by __generic_file_aio_write()), splice write (generic_file_splice_write), truncate, and fallocate (these can dirty last partial page - handled inside each filesystem separately). Protect these places with sb_start_write() and sb_end_write(). ->page_mkwrite() calls are particularly complex since they are called with mmap_sem held and thus we cannot use standard sb_start_write() due to lock ordering constraints. We solve the problem by using a special freeze protection sb_start_pagefault() which ranks below mmap_sem. BugLink: https://bugs.launchpad.net/bugs/897421 Tested-by: Kamal Mostafa Tested-by: Peter M. Petrakis Tested-by: Dann Frazier Tested-by: Massimo Morana Signed-off-by: Jan Kara Signed-off-by: Al Viro --- fs/buffer.c | 22 ++++------------------ 1 file changed, 4 insertions(+), 18 deletions(-) (limited to 'fs/buffer.c') diff --git a/fs/buffer.c b/fs/buffer.c index d5ec360e332d..9f6d2e41281d 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -2306,8 +2306,8 @@ EXPORT_SYMBOL(block_commit_write); * beyond EOF, then the page is guaranteed safe against truncation until we * unlock the page. * - * Direct callers of this function should call vfs_check_frozen() so that page - * fault does not busyloop until the fs is thawed. + * Direct callers of this function should protect against filesystem freezing + * using sb_start_write() - sb_end_write() functions. */ int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, get_block_t get_block) @@ -2345,18 +2345,7 @@ int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, if (unlikely(ret < 0)) goto out_unlock; - /* - * Freezing in progress? We check after the page is marked dirty and - * with page lock held so if the test here fails, we are sure freezing - * code will wait during syncing until the page fault is done - at that - * point page will be dirty and unlocked so freezing code will write it - * and writeprotect it again. - */ set_page_dirty(page); - if (inode->i_sb->s_frozen != SB_UNFROZEN) { - ret = -EAGAIN; - goto out_unlock; - } wait_on_page_writeback(page); return 0; out_unlock: @@ -2371,12 +2360,9 @@ int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, int ret; struct super_block *sb = vma->vm_file->f_path.dentry->d_inode->i_sb; - /* - * This check is racy but catches the common case. The check in - * __block_page_mkwrite() is reliable. - */ - vfs_check_frozen(sb, SB_FREEZE_WRITE); + sb_start_pagefault(sb); ret = __block_page_mkwrite(vma, vmf, get_block); + sb_end_pagefault(sb); return block_page_mkwrite_return(ret); } EXPORT_SYMBOL(block_page_mkwrite); -- cgit v1.2.3