Dan Carpenter
2010-Mar-20 11:38 UTC
locking problems: Btrfs: be more selective in the defrag ioctl
Hi Chris, There is a locking problem in 940100a4a7b78 "Btrfs: be more selective in the defrag ioctl" There are two places where we break out of the while loop under the lock. fs/btrfs/ioctl.c +708 btrfs_defrag_file(159) error: double lock ''mutex:&inode->i_mutex'' 600 mutex_lock(&inode->i_mutex); 601 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) 602 BTRFS_I(inode)->force_compress = 1; 603 604 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); 605 if (ret) { 606 ret = -ENOSPC; 607 break; Here. 608 } 609 610 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); 611 if (ret) { 612 btrfs_free_reserved_data_space(root, inode, 613 PAGE_CACHE_SIZE); 614 ret = -ENOSPC; 615 break; And here. 616 } Maybe we should have "goto err_reservations;" instead of break? I don''t know the code well enough to say. regards, dan carpenter -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Josef Bacik
2010-Mar-22 13:47 UTC
Re: locking problems: Btrfs: be more selective in the defrag ioctl
On Sat, Mar 20, 2010 at 02:38:51PM +0300, Dan Carpenter wrote:> Hi Chris, > > There is a locking problem in > 940100a4a7b78 "Btrfs: be more selective in the defrag ioctl" > > There are two places where we break out of the while loop under the > lock. > > fs/btrfs/ioctl.c +708 btrfs_defrag_file(159) error: double lock ''mutex:&inode->i_mutex'' > 600 mutex_lock(&inode->i_mutex); > 601 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) > 602 BTRFS_I(inode)->force_compress = 1; > 603 > 604 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); > 605 if (ret) { > 606 ret = -ENOSPC; > 607 break; > > Here. > > 608 } > 609 > 610 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); > 611 if (ret) { > 612 btrfs_free_reserved_data_space(root, inode, > 613 PAGE_CACHE_SIZE); > 614 ret = -ENOSPC; > 615 break; > > And here. > > 616 } > > Maybe we should have "goto err_reservations;" instead of break? I > don''t know the code well enough to say.No, everything is accounted for correctly. If the metadata reservation fails, we free the data space reservation and break. If the data space reservation fails, we''re good to go and can just exit. Thanks, Josef
Dan Carpenter
2010-Mar-22 14:03 UTC
Re: locking problems: Btrfs: be more selective in the defrag ioctl
On Mon, Mar 22, 2010 at 09:47:21AM -0400, Josef Bacik wrote:> On Sat, Mar 20, 2010 at 02:38:51PM +0300, Dan Carpenter wrote: > > Hi Chris, > > > > There is a locking problem in > > 940100a4a7b78 "Btrfs: be more selective in the defrag ioctl" > > > > There are two places where we break out of the while loop under the > > lock. > > > > fs/btrfs/ioctl.c +708 btrfs_defrag_file(159) error: double lock ''mutex:&inode->i_mutex'' > > 600 mutex_lock(&inode->i_mutex); > > 601 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) > > 602 BTRFS_I(inode)->force_compress = 1; > > 603 > > 604 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); > > 605 if (ret) { > > 606 ret = -ENOSPC; > > 607 break; > > > > Here. > > > > 608 } > > 609 > > 610 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); > > 611 if (ret) { > > 612 btrfs_free_reserved_data_space(root, inode, > > 613 PAGE_CACHE_SIZE); > > 614 ret = -ENOSPC; > > 615 break; > > > > And here. > > > > 616 } > > > > Maybe we should have "goto err_reservations;" instead of break? I > > don''t know the code well enough to say. > > No, everything is accounted for correctly. If the metadata reservation fails, > we free the data space reservation and break. If the data space reservation > fails, we''re good to go and can just exit. Thanks, >What about the lock on line 606?> > 600 mutex_lock(&inode->i_mutex);If we break on line 615 or 607 that means that we return with the lock held, or if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) is true then we dead lock. regards, dan carpenter> Josef
Josef Bacik
2010-Mar-22 14:09 UTC
Re: locking problems: Btrfs: be more selective in the defrag ioctl
On Mon, Mar 22, 2010 at 05:03:30PM +0300, Dan Carpenter wrote:> On Mon, Mar 22, 2010 at 09:47:21AM -0400, Josef Bacik wrote: > > On Sat, Mar 20, 2010 at 02:38:51PM +0300, Dan Carpenter wrote: > > > Hi Chris, > > > > > > There is a locking problem in > > > 940100a4a7b78 "Btrfs: be more selective in the defrag ioctl" > > > > > > There are two places where we break out of the while loop under the > > > lock. > > > > > > fs/btrfs/ioctl.c +708 btrfs_defrag_file(159) error: double lock ''mutex:&inode->i_mutex'' > > > 600 mutex_lock(&inode->i_mutex); > > > 601 if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) > > > 602 BTRFS_I(inode)->force_compress = 1; > > > 603 > > > 604 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE); > > > 605 if (ret) { > > > 606 ret = -ENOSPC; > > > 607 break; > > > > > > Here. > > > > > > 608 } > > > 609 > > > 610 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1); > > > 611 if (ret) { > > > 612 btrfs_free_reserved_data_space(root, inode, > > > 613 PAGE_CACHE_SIZE); > > > 614 ret = -ENOSPC; > > > 615 break; > > > > > > And here. > > > > > > 616 } > > > > > > Maybe we should have "goto err_reservations;" instead of break? I > > > don''t know the code well enough to say. > > > > No, everything is accounted for correctly. If the metadata reservation fails, > > we free the data space reservation and break. If the data space reservation > > fails, we''re good to go and can just exit. Thanks, > > > > What about the lock on line 606? > > > > 600 mutex_lock(&inode->i_mutex); > > If we break on line 615 or 607 that means that we return with the lock > held, or if (range->flags & BTRFS_DEFRAG_RANGE_COMPRESS) is true then we > dead lock. >Ahh yeah you are right, should probably just put a mutex_unlock before the break in both cases. Thanks, Josef