ceph: add truncate size handling support for fscrypt

This will transfer the encrypted last block contents to the MDS
along with the truncate request only when the new size is smaller
and not aligned to the fscrypt BLOCK size. When the last block is
located in the file hole, the truncate request will only contain
the header.

The MDS could fail to do the truncate if there has another client
or process has already updated the RADOS object which contains
the last block, and will return -EAGAIN, then the kclient needs
to retry it. The RMW will take around 50ms, and will let it retry
20 times for now.

Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Reviewed-and-tested-by: Luís Henriques <lhenriques@suse.de>
Reviewed-by: Milind Changire <mchangir@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

1 files changed, 193 insertions, 7 deletions

diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 841f60e8de6a..054fd66609b7 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -596,6 +596,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	ci->i_truncate_seq = 0;
 	ci->i_truncate_size = 0;
 	ci->i_truncate_pending = 0;
+	ci->i_truncate_pagecache_size = 0;
 
 	ci->i_max_size = 0;
 	ci->i_reported_size = 0;
@@ -767,6 +768,10 @@ int ceph_fill_file_size(struct inode *inode, int issued,
 		dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 		     truncate_size);
 		ci->i_truncate_size = truncate_size;
+		if (IS_ENCRYPTED(inode))
+			ci->i_truncate_pagecache_size = size;
+		else
+			ci->i_truncate_pagecache_size = truncate_size;
 	}
 	return queue_trunc;
 }
@@ -2147,7 +2152,7 @@ retry:
 	/* there should be no reader or writer */
 	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
 
-	to = ci->i_truncate_size;
+	to = ci->i_truncate_pagecache_size;
 	wrbuffer_refs = ci->i_wrbuffer_ref;
 	dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
 	     ci->i_truncate_pending, to);
@@ -2157,7 +2162,7 @@ retry:
 	truncate_pagecache(inode, to);
 
 	spin_lock(&ci->i_ceph_lock);
-	if (to == ci->i_truncate_size) {
+	if (to == ci->i_truncate_pagecache_size) {
 		ci->i_truncate_pending = 0;
 		finish = 1;
 	}
@@ -2241,6 +2246,144 @@ static const struct inode_operations ceph_encrypted_symlink_iops = {
 	.listxattr = ceph_listxattr,
 };
 
+/*
+ * Transfer the encrypted last block to the MDS and the MDS
+ * will help update it when truncating a smaller size.
+ *
+ * We don't support a PAGE_SIZE that is smaller than the
+ * CEPH_FSCRYPT_BLOCK_SIZE.
+ */
+static int fill_fscrypt_truncate(struct inode *inode,
+				 struct ceph_mds_request *req,
+				 struct iattr *attr)
+{
+	struct ceph_inode_info *ci = ceph_inode(inode);
+	int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
+	loff_t pos, orig_pos = round_down(attr->ia_size,
+					  CEPH_FSCRYPT_BLOCK_SIZE);
+	u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
+	struct ceph_pagelist *pagelist = NULL;
+	struct kvec iov = {0};
+	struct iov_iter iter;
+	struct page *page = NULL;
+	struct ceph_fscrypt_truncate_size_header header;
+	int retry_op = 0;
+	int len = CEPH_FSCRYPT_BLOCK_SIZE;
+	loff_t i_size = i_size_read(inode);
+	int got, ret, issued;
+	u64 objver;
+
+	ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
+	if (ret < 0)
+		return ret;
+
+	issued = __ceph_caps_issued(ci, NULL);
+
+	dout("%s size %lld -> %lld got cap refs on %s, issued %s\n", __func__,
+	     i_size, attr->ia_size, ceph_cap_string(got),
+	     ceph_cap_string(issued));
+
+	/* Try to writeback the dirty pagecaches */
+	if (issued & (CEPH_CAP_FILE_BUFFER)) {
+		loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SHIFT - 1;
+
+		ret = filemap_write_and_wait_range(inode->i_mapping,
+						   orig_pos, lend);
+		if (ret < 0)
+			goto out;
+	}
+
+	page = __page_cache_alloc(GFP_KERNEL);
+	if (page == NULL) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	pagelist = ceph_pagelist_alloc(GFP_KERNEL);
+	if (!pagelist) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	iov.iov_base = kmap_local_page(page);
+	iov.iov_len = len;
+	iov_iter_kvec(&iter, READ, &iov, 1, len);
+
+	pos = orig_pos;
+	ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
+	if (ret < 0)
+		goto out;
+
+	/* Insert the header first */
+	header.ver = 1;
+	header.compat = 1;
+	header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
+
+	/*
+	 * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
+	 * because in MDS it may need this to do the truncate.
+	 */
+	header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
+
+	/*
+	 * If we hit a hole here, we should just skip filling
+	 * the fscrypt for the request, because once the fscrypt
+	 * is enabled, the file will be split into many blocks
+	 * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
+	 * has a hole, the hole size should be multiple of block
+	 * size.
+	 *
+	 * If the Rados object doesn't exist, it will be set to 0.
+	 */
+	if (!objver) {
+		dout("%s hit hole, ppos %lld < size %lld\n", __func__,
+		     pos, i_size);
+
+		header.data_len = cpu_to_le32(8 + 8 + 4);
+		header.file_offset = 0;
+		ret = 0;
+	} else {
+		header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
+		header.file_offset = cpu_to_le64(orig_pos);
+
+		/* truncate and zero out the extra contents for the last block */
+		memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
+
+		/* encrypt the last block */
+		ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
+						    CEPH_FSCRYPT_BLOCK_SIZE,
+						    0, block,
+						    GFP_KERNEL);
+		if (ret)
+			goto out;
+	}
+
+	/* Insert the header */
+	ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
+	if (ret)
+		goto out;
+
+	if (header.block_size) {
+		/* Append the last block contents to pagelist */
+		ret = ceph_pagelist_append(pagelist, iov.iov_base,
+					   CEPH_FSCRYPT_BLOCK_SIZE);
+		if (ret)
+			goto out;
+	}
+	req->r_pagelist = pagelist;
+out:
+	dout("%s %p size dropping cap refs on %s\n", __func__,
+	     inode, ceph_cap_string(got));
+	ceph_put_cap_refs(ci, got);
+	if (iov.iov_base)
+		kunmap_local(iov.iov_base);
+	if (page)
+		__free_pages(page, 0);
+	if (ret && pagelist)
+		ceph_pagelist_release(pagelist);
+	return ret;
+}
+
 int __ceph_setattr(struct inode *inode, struct iattr *attr,
 		   struct ceph_iattr *cia)
 {
@@ -2249,13 +2392,17 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 	struct ceph_mds_request *req;
 	struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
 	struct ceph_cap_flush *prealloc_cf;
+	loff_t isize = i_size_read(inode);
 	int issued;
 	int release = 0, dirtied = 0;
 	int mask = 0;
 	int err = 0;
 	int inode_dirty_flags = 0;
 	bool lock_snap_rwsem = false;
+	bool fill_fscrypt;
+	int truncate_retry = 20; /* The RMW will take around 50ms */
 
+retry:
 	prealloc_cf = ceph_alloc_cap_flush();
 	if (!prealloc_cf)
 		return -ENOMEM;
@@ -2267,6 +2414,7 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 		return PTR_ERR(req);
 	}
 
+	fill_fscrypt = false;
 	spin_lock(&ci->i_ceph_lock);
 	issued = __ceph_caps_issued(ci, NULL);
 
@@ -2388,10 +2536,27 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 		}
 	}
 	if (ia_valid & ATTR_SIZE) {
-		loff_t isize = i_size_read(inode);
-
 		dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
-		if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
+		/*
+		 * Only when the new size is smaller and not aligned to
+		 * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
+		 */
+		if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
+		    (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
+			mask |= CEPH_SETATTR_SIZE;
+			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
+				   CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
+			set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
+			mask |= CEPH_SETATTR_FSCRYPT_FILE;
+			req->r_args.setattr.size =
+				cpu_to_le64(round_up(attr->ia_size,
+						     CEPH_FSCRYPT_BLOCK_SIZE));
+			req->r_args.setattr.old_size =
+				cpu_to_le64(round_up(isize,
+						     CEPH_FSCRYPT_BLOCK_SIZE));
+			req->r_fscrypt_file = attr->ia_size;
+			fill_fscrypt = true;
+		} else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
 			if (attr->ia_size > isize) {
 				i_size_write(inode, attr->ia_size);
 				inode->i_blocks = calc_inode_blocks(attr->ia_size);
@@ -2414,7 +2579,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 					cpu_to_le64(round_up(isize,
 							     CEPH_FSCRYPT_BLOCK_SIZE));
 				req->r_fscrypt_file = attr->ia_size;
-				/* FIXME: client must zero out any partial blocks! */
 			} else {
 				req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
 				req->r_args.setattr.old_size = cpu_to_le64(isize);
@@ -2481,8 +2645,10 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 
 	release &= issued;
 	spin_unlock(&ci->i_ceph_lock);
-	if (lock_snap_rwsem)
+	if (lock_snap_rwsem) {
 		up_read(&mdsc->snap_rwsem);
+		lock_snap_rwsem = false;
+	}
 
 	if (inode_dirty_flags)
 		__mark_inode_dirty(inode, inode_dirty_flags);
@@ -2494,7 +2660,27 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr,
 		req->r_args.setattr.mask = cpu_to_le32(mask);
 		req->r_num_caps = 1;
 		req->r_stamp = attr->ia_ctime;
+		if (fill_fscrypt) {
+			err = fill_fscrypt_truncate(inode, req, attr);
+			if (err)
+				goto out;
+		}
+
+		/*
+		 * The truncate request will return -EAGAIN when the
+		 * last block has been updated just before the MDS
+		 * successfully gets the xlock for the FILE lock. To
+		 * avoid corrupting the file contents we need to retry
+		 * it.
+		 */
 		err = ceph_mdsc_do_request(mdsc, NULL, req);
+		if (err == -EAGAIN && truncate_retry--) {
+			dout("setattr %p result=%d (%s locally, %d remote), retry it!\n",
+			     inode, err, ceph_cap_string(dirtied), mask);
+			ceph_mdsc_put_request(req);
+			ceph_free_cap_flush(prealloc_cf);
+			goto retry;
+		}
 	}
 out:
 	dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,