On 10/02/2013 07:07 PM, Miklos Szeredi wrote:
On Wed, Oct 2, 2013 at 1:17 PM, Maxim Patlasov <MPatlasov@xxxxxxxxxxxxx> wrote:If writeback happens while fuse is in FUSE_NOWRITE condition, the request will be queued but not processed immediately (see fuse_flush_writepages()). Until FUSE_NOWRITE becomes relaxed, more writebacks can happen. They will be queued as "secondary" requests to that first ("primary") request. When FUSE_NOWRITE is relaxed and fuse_send_writepage() is called, it must crop both primary and secondary requests according to the actual i_size. Otherwise, if only primary is cropped, an extending write(2) may increase i_size soon and then secondary requests won't be cropped properly. The result would be stale data written to the server to a file offset where zeros must be. Changed in v2: - avoid NULL pointer dereference in fuse_drop_writepage(). Signed-off-by: Maxim Patlasov <MPatlasov@xxxxxxxxxxxxx> --- fs/fuse/file.c | 64 ++++++++++++++++++++++++++++++++++++++++++++++++-------- 1 file changed, 55 insertions(+), 9 deletions(-) diff --git a/fs/fuse/file.c b/fs/fuse/file.c index 575e44f..89a2e76 100644 --- a/fs/fuse/file.c +++ b/fs/fuse/file.c @@ -1435,6 +1435,51 @@ static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) wake_up(&fi->page_waitq); } +/* Drop list of secondary writepage requests */ +static void fuse_drop_writepage(struct fuse_conn *fc, struct fuse_req *req) +{ + struct backing_dev_info *bdi = req ? + req->inode->i_mapping->backing_dev_info : NULL; + + while (req) { + struct fuse_req *next = req->misc.write.next; + dec_bdi_stat(bdi, BDI_WRITEBACK); + dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP); + fuse_writepage_free(fc, req); + fuse_put_request(fc, req); + req = next; + } +} + +/* Crop the misc.write.in.size of parent and secondary writepage requests */ +static bool fuse_crop_writepage(struct fuse_conn *fc, struct fuse_req *req, + loff_t size, struct fuse_req **drop_list) +{ + if (req->misc.write.in.offset >= size) + return true; + + while (req) { + struct fuse_req *next = req->misc.write.next; + struct fuse_write_in *inarg = &req->misc.write.in; + __u64 data_size = inarg->size ? : + req->num_pages * PAGE_CACHE_SIZE; + + if (inarg->offset + data_size <= size) { + inarg->size = data_size; + } else if (inarg->offset < size) { + inarg->size = size - inarg->offset; + } else { + /* Got truncated off completely */ + req->misc.write.next = *drop_list; + *drop_list = req;This corrupts the list (the req is not taken off the list before being added to another). It could be fixed, but why not check this instead in fuse_writepage_end() before queuing the next request?
Nice idea. This will make next patch ("crop on attach") redundant. I'll resend updated patch-set.
Thanks, Maxim -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
