Skip to contentSkip to conversation

ceph: remove the inline data support

PATCH Superseded Xiubo Li 2 participants 3 messages Jul 29 – Aug 26

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Xiubo Li

author
#0 lore raw

PATCH v2 1/1 ceph: remove the inline data support

Inline data support has been deprecated in the CephFS MDS and is being formally removed. Existing inline data can be migrated to data pool objects via MDS scrub (see PR #44359).

Add CONFIG_CEPH_FS_INLINE_DATA (default n) to gate all inline data code in the kclient. When disabled, CEPH_FEATURE_MDS_INLINE_DATA is defined as 0 so that the kclient no longer advertises the feature to the MDS. The MDS will then refuse to open files that still have inline data, returning -EPERM, preventing silent data corruption that would occur if the kclient tried to read inline data from OSD objects that do not contain it.

This config option will be kept for a transitional period to allow existing CephFS deployments to migrate inline data via MDS scrub. Once the migration period ends, the option and all inline data code will be permanently removed.

Users upgrading from an older Ceph release should ensure all inline data has been migrated before deploying a kclient with this option disabled:
1. Enable and run MDS scrub to uninline existing data
2. Disable inline_data on the filesystem:

     ceph fs set <fs> inline_data false --yes-i-really-really-mean-it

---
Changes in v2:
- keep CEPH_FEATURE_MDS_INLINE_DATA always defined as (1ULL << 40)
so decode logic can still check MDS feature bits; gate the
supported features set via CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT
- Link to v1: https://patch.msgid.link/[email protected]

---
 fs/ceph/Kconfig                    | 12 ++++++++++++
 fs/ceph/addr.c                     | 16 +++++++++++++++
 fs/ceph/caps.c                     | 40 +++++++++++++++++++++++++++++++++++++-
 fs/ceph/file.c                     | 20 +++++++++++++++++++
 fs/ceph/inode.c                    | 10 ++++++++++
 fs/ceph/mds_client.c               | 30 ++++++++++++++++++++++++++++
 fs/ceph/mds_client.h               |  2 ++
 fs/ceph/snap.c                     |  2 ++
 fs/ceph/super.h                    | 13 +++++++++++++
 include/linux/ceph/ceph_features.h |  7 ++++++-
 10 files changed, 150 insertions(+), 2 deletions(-)
fs/ceph/Kconfig +12 -0
index 3d64a316ca31..be1c8ada5665 100644
@@ -17,6 +17,18 @@ config CEPH_FS
 	  If unsure, say N.
 
 if CEPH_FS
+config CEPH_FS_INLINE_DATA
+	bool "Enable CephFS inline data support (deprecated)"
+	depends on CEPH_FS
+	default n
+	help
+	  Inline data support has been deprecated in the CephFS MDS
+	  and is being formally removed.  Existing inline data can be
+	  migrated to data pool objects via MDS scrub.  Choose N
+	  unless you have a legacy filesystem that requires it,
+	  and ensure all inline data has been migrated (see
+	  ceph fs set <fs> inline_data false) before enabling.
+
 config CEPH_FSCACHE
 	bool "Enable Ceph client caching support"
 	depends on CEPH_FS=m && FSCACHE || CEPH_FS=y && FSCACHE=y
fs/ceph/addr.c +16 -0
index e2da3ab9f808..5211320a50ac 100644
@@ -274,6 +274,7 @@ static void finish_netfs_read(struct ceph_osd_request *req)
 	ceph_dec_osd_stopping_blocker(fsc->mdsc);
 }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 {
 	struct netfs_io_request *rreq = subreq->rreq;
@@ -335,6 +336,7 @@ static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 	netfs_read_subreq_terminated(subreq);
 	return true;
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
 static int ceph_netfs_prepare_read(struct netfs_io_subrequest *subreq)
 {
@@ -372,8 +374,10 @@ static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq)
 		goto out;
 	}
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
 		return;
+#endif
 
 	// TODO: This rounding here is slightly dodgy.  It *should* work, for
 	// now, as the cache only deals in blocks that are a multiple of
@@ -2006,7 +2010,11 @@ static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
 	      off, ceph_cap_string(got));
 
 	if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	    !ceph_has_inline_data(ci)) {
+#else
+	    true) {
+#endif
 		CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
 		ceph_add_rw_context(fi, &rw_ctx);
 		ret = filemap_fault(vmf);
@@ -2037,7 +2045,11 @@ static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
 			goto out_inline;
 		}
 		err = __ceph_do_getattr(inode, page,
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 					 CEPH_STAT_CAP_INLINE_DATA, true);
+#else
+					 0, true);
+#endif
 		if (err < 0 || off >= i_size_read(inode)) {
 			unlock_page(page);
 			put_page(page);
@@ -2168,6 +2180,7 @@ static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
 	return ret;
 }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
 			   char	*data, size_t len)
 {
@@ -2214,7 +2227,9 @@ void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
 		put_page(page);
 	}
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 int ceph_uninline_data(struct file *file)
 {
 	struct inode *inode = file_inode(file);
@@ -2370,6 +2385,7 @@ int ceph_uninline_data(struct file *file)
 	      ceph_vinop(inode), inline_version, err);
 	return err;
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
 static const struct vm_operations_struct ceph_vmops = {
 	.fault		= ceph_filemap_fault,
fs/ceph/caps.c +39 -1
index f8d898ad091e..2fd2116a3308 100644
@@ -1266,7 +1266,9 @@ struct cap_msg_args {
 	kuid_t			uid;
 	kgid_t			gid;
 	umode_t			mode;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	bool			inline_data;
+#endif
 	bool			wake;
 	bool			encrypted;
 	u32			fscrypt_auth_len;
@@ -1336,8 +1338,16 @@ static void encode_cap_msg(struct ceph_msg *msg, struct cap_msg_args *arg)
 	p = fc + 1;
 	/* flock buffer size (version 2) */
 	ceph_encode_32(&p, 0);
-	/* inline version (version 4) */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE);
+#else
+	/*
+	 * Inline data support has been removed.  The field is still
+	 * encoded for protocol compatibility; always advertise
+	 * CEPH_INLINE_NONE to indicate no inline data is present.
+	 */
+	ceph_encode_64(&p, CEPH_INLINE_NONE);
+#endif
 	/* inline data size */
 	ceph_encode_32(&p, 0);
 	/*
@@ -1499,7 +1509,9 @@ static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
 	arg->gid = inode->i_gid;
 	arg->mode = inode->i_mode;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	arg->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
+#endif
 	if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) &&
 	    !list_empty(&ci->i_cap_snaps)) {
 		struct ceph_cap_snap *capsnap;
@@ -1620,7 +1632,9 @@ static inline int __send_flush_snap(struct inode *inode,
 	arg.gid = capsnap->gid;
 	arg.mode = capsnap->mode;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	arg.inline_data = capsnap->inline_data;
+#endif
 	arg.flags = 0;
 	arg.wake = false;
 	arg.encrypted = IS_ENCRYPTED(inode);
@@ -3192,6 +3206,7 @@ int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi, int need,
 			return ret;
 		}
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 		if (S_ISREG(ci->netfs.inode.i_mode) &&
 		    ceph_has_inline_data(ci) &&
 		    (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&
@@ -3223,6 +3238,7 @@ int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi, int need,
 				return ret;
 			continue;
 		}
+#endif
 		break;
 	}
 	*got = _got;
@@ -3518,10 +3534,12 @@ static void invalidate_aliases(struct inode *inode)
 
 struct cap_extra_info {
 	struct ceph_string *pool_ns;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	/* inline data */
 	u64 inline_version;
 	void *inline_data;
 	u32 inline_len;
+#endif
 	/* dirstat */
 	bool dirstat_valid;
 	u64 nfiles;
@@ -3564,7 +3582,9 @@ static void handle_cap_grant(struct inode *inode,
 	bool queue_trunc = false;
 	bool queue_invalidate = false;
 	bool deleted_inode = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	bool fill_inline = false;
+#endif
 	bool revoke_wait = false;
 	int flags = 0;
 
@@ -3830,6 +3850,7 @@ static void handle_cap_grant(struct inode *inode,
 			check_caps = 2; /* check all caps */
 	}
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (extra_info->inline_version > 0 &&
 	    extra_info->inline_version >= ci->i_inline_version) {
 		ci->i_inline_version = extra_info->inline_version;
@@ -3837,6 +3858,7 @@ static void handle_cap_grant(struct inode *inode,
 		    (newcaps & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
 			fill_inline = true;
 	}
+#endif
 
 	if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
 		if (ci->i_auth_cap == cap) {
@@ -3856,9 +3878,11 @@ static void handle_cap_grant(struct inode *inode,
 	}
 	spin_unlock(&ci->i_ceph_lock);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (fill_inline)
 		ceph_fill_inline_data(inode, NULL, extra_info->inline_data,
 				      extra_info->inline_len);
+#endif
 
 	if (queue_trunc)
 		ceph_queue_vmtruncate(inode);
@@ -4476,12 +4500,26 @@ void ceph_handle_caps(struct ceph_mds_session *session,
 	}
 
 	if (msg_version >= 4) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 		ceph_decode_64_safe(&p, end, extra_info.inline_version, bad);
 		ceph_decode_32_safe(&p, end, extra_info.inline_len, bad);
 		if (p + extra_info.inline_len > end)
 			goto bad;
 		extra_info.inline_data = p;
 		p += extra_info.inline_len;
+#else
+		/*
+		 * Inline data support has been removed, but cap grant
+		 * messages (msg_version >= 4) always carry inline_version
+		 * + inline_data.  We must consume the fields to keep the
+		 * decode pointer in sync.
+		 */
+		u32 __len;
+
+		ceph_decode_skip_64(&p, end, bad);
+		ceph_decode_32_safe(&p, end, __len, bad);
+		ceph_decode_skip_n(&p, end, __len, bad);
+#endif
 	}
 
 	if (msg_version >= 5) {
fs/ceph/file.c +20 -0
index 3823671ab95a..901ed97b1963 100644
@@ -224,7 +224,9 @@ static int ceph_init_file_info(struct inode *inode, struct file *file,
 		ceph_inode_to_fs_client(&ci->netfs.inode)->mount_options;
 	struct ceph_client *cl = ceph_inode_to_client(inode);
 	struct ceph_file_info *fi;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	int ret;
+#endif
 
 	doutc(cl, "%p %llx.%llx %p 0%o (%s)\n", inode, ceph_vinop(inode),
 	      file, inode->i_mode, isdir ? "dir" : "regular");
@@ -258,14 +260,17 @@ static int ceph_init_file_info(struct inode *inode, struct file *file,
 	INIT_LIST_HEAD(&fi->rw_contexts);
 	fi->filp_gen = READ_ONCE(ceph_inode_to_fs_client(inode)->filp_gen);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if ((file->f_mode & FMODE_WRITE) && ceph_has_inline_data(ci)) {
 		ret = ceph_uninline_data(file);
 		if (ret < 0)
 			goto error;
 	}
+#endif
 
 	return 0;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 error:
 	ceph_fscache_unuse_cookie(inode, file->f_mode & FMODE_WRITE);
 	ceph_put_fmode(ci, fi->fmode, 1);
@@ -273,6 +278,7 @@ static int ceph_init_file_info(struct inode *inode, struct file *file,
 	/* wake up anyone waiting for caps on this inode */
 	wake_up_all(&ci->i_cap_wq);
 	return ret;
+#endif
 }
 
 /*
@@ -702,7 +708,9 @@ static int ceph_finish_async_create(struct inode *dir, struct inode *inode,
 
 	ktime_get_real_ts64(&now);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	iinfo.inline_version = CEPH_INLINE_NONE;
+#endif
 	iinfo.change_attr = 1;
 	ceph_encode_timespec64(&iinfo.btime, &now);
 
@@ -2224,7 +2232,11 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
 		      inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len,
 		      ceph_cap_string(got));
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 		if (!ceph_has_inline_data(ci)) {
+#else
+		if (true) {
+#endif
 			if (!retry_op &&
 			    (iocb->ki_flags & IOCB_DIRECT) &&
 			    !IS_ENCRYPTED(inode)) {
@@ -2267,7 +2279,11 @@ static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to)
 			if (!page)
 				return -ENOMEM;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 			mask = CEPH_STAT_CAP_INLINE_DATA;
+#else
+			mask = 0;
+#endif
 		}
 
 		statret = __ceph_do_getattr(inode, page, mask, !!page);
@@ -2349,8 +2365,12 @@ static ssize_t ceph_splice_read(struct file *in, loff_t *ppos,
 	if (ceph_inode_is_shutdown(inode))
 		return -ESTALE;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (ceph_has_inline_data(ci) ||
 	    (fi->flags & CEPH_F_SYNC))
+#else
+	if (fi->flags & CEPH_F_SYNC)
+#endif
 		return copy_splice_read(in, ppos, pipe, len, flags);
 
 	ret = ceph_start_io_read(inode);
fs/ceph/inode.c +10 -0
index 61d7c0b8161f..20abf7dc9c64 100644
@@ -627,7 +627,9 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	spin_lock_init(&ci->i_ceph_lock);
 
 	ci->i_version = 0;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	ci->i_inline_version = 0;
+#endif
 	ci->i_time_warp_seq = 0;
 	ci->i_ceph_flags = 0;
 	atomic64_set(&ci->i_ordered_count, 1);
@@ -1030,7 +1032,9 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 	bool wake = false;
 	bool queue_trunc = false;
 	bool new_version = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	bool fill_inline = false;
+#endif
 	umode_t mode = le32_to_cpu(info->mode);
 	dev_t rdev = le32_to_cpu(info->rdev);
 
@@ -1362,6 +1366,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 		}
 	}
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (iinfo->inline_version > 0 &&
 	    iinfo->inline_version >= ci->i_inline_version) {
 		int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
@@ -1370,6 +1375,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 		    (locked_page || (info_caps & cache_caps)))
 			fill_inline = true;
 	}
+#endif
 
 	if (cap_fmode >= 0) {
 		if (!info_caps)
@@ -1382,9 +1388,11 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 
 	ceph_fscache_register_inode_cookie(inode);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (fill_inline)
 		ceph_fill_inline_data(inode, locked_page,
 				      iinfo->inline_data, iinfo->inline_len);
+#endif
 
 	if (wake)
 		wake_up_all(&ci->i_cap_wq);
@@ -3022,6 +3030,7 @@ int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
 	req->r_args.getattr.mask = cpu_to_le32(mask);
 	req->r_locked_page = locked_page;
 	err = ceph_mdsc_do_request(mdsc, NULL, req);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	if (locked_page && err == 0) {
 		u64 inline_version = req->r_reply_info.targeti.inline_version;
 		if (inline_version == 0) {
@@ -3034,6 +3043,7 @@ int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
 			err = req->r_reply_info.targeti.inline_len;
 		}
 	}
+#endif
 	ceph_mdsc_put_request(req);
 	doutc(cl, "result=%d\n", err);
 	return err;
fs/ceph/mds_client.c +30 -0
index 18be36c2e9e8..6a439da76616 100644
@@ -156,12 +156,26 @@ static int parse_reply_info_in(void **p, void *end,
 	*p += info->xattr_len;
 
 	if (features == (u64)-1) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 		/* inline data */
 		ceph_decode_64_safe(p, end, info->inline_version, bad);
 		ceph_decode_32_safe(p, end, info->inline_len, bad);
 		ceph_decode_need(p, end, info->inline_len, bad);
 		info->inline_data = *p;
 		*p += info->inline_len;
+#else
+		/*
+		 * Inline data support has been removed, but the MDS always
+		 * encodes inline_version + inline_data in versioned replies.
+		 * We must consume these fields from the message to keep the
+		 * decode pointer in sync, even though we discard them.
+		 */
+		u32 __len;
+
+		ceph_decode_skip_64(p, end, bad);
+		ceph_decode_32_safe(p, end, __len, bad);
+		ceph_decode_skip_n(p, end, __len, bad);
+#endif
 		/* quota */
 		err = parse_reply_info_quota(p, end, info);
 		if (err < 0)
@@ -276,6 +290,7 @@ static int parse_reply_info_in(void **p, void *end,
 		*p = end;
 	} else {
 		/* legacy (unversioned) struct */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 		if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
 			ceph_decode_64_safe(p, end, info->inline_version, bad);
 			ceph_decode_32_safe(p, end, info->inline_len, bad);
@@ -284,6 +299,21 @@ static int parse_reply_info_in(void **p, void *end,
 			*p += info->inline_len;
 		} else
 			info->inline_version = CEPH_INLINE_NONE;
+#else
+		/*
+		 * Inline data support has been removed.  When the MDS
+		 * advertises CEPH_FEATURE_MDS_INLINE_DATA it includes
+		 * inline_version + inline_data in legacy replies.  We
+		 * must consume them to keep the decode pointer in sync.
+		 */
+		if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
+			u32 __len;
+
+			ceph_decode_skip_64(p, end, bad);
+			ceph_decode_32_safe(p, end, __len, bad);
+			ceph_decode_skip_n(p, end, __len, bad);
+		}
+#endif
 
 		if (features & CEPH_FEATURE_MDS_QUOTA) {
 			err = parse_reply_info_quota(p, end, info);
fs/ceph/mds_client.h +2 -0
index 731d6ad04956..9bcf1a9d5ba8 100644
@@ -149,9 +149,11 @@ struct ceph_mds_reply_info_in {
 	char *symlink;
 	u32 xattr_len;
 	char *xattr_data;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	u64 inline_version;
 	u32 inline_len;
 	char *inline_data;
+#endif
 	u32 pool_ns_len;
 	char *pool_ns_data;
 	u64 max_bytes;
fs/ceph/snap.c +2 -0
index e5d8d59ffc17..3bc3a2ce72a2 100644
@@ -613,7 +613,9 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci,
 		capsnap->xattr_version = 0;
 	}
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
+#endif
 
 	/* dirty page count moved from _head to this cap_snap;
 	   all subsequent writes page dirties occur _after_ this
fs/ceph/super.h +13 -0
index 3737ea7ed88b..eced08f52b79 100644
@@ -274,7 +274,9 @@ struct ceph_cap_snap {
 	u32 truncate_seq;
 	int writing;   /* a sync write is still in progress */
 	int dirty_pages;     /* dirty pages awaiting writeback */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	bool inline_data;
+#endif
 	bool need_flush;
 };
 
@@ -381,7 +383,9 @@ struct ceph_inode_info {
 	spinlock_t i_ceph_lock;
 
 	u64 i_version;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 	u64 i_inline_version;
+#endif
 	u32 i_time_warp_seq;
 
 	unsigned long i_ceph_flags;
@@ -1335,11 +1339,14 @@ extern void __ceph_touch_fmode(struct ceph_inode_info *ci,
 extern const struct address_space_operations ceph_aops;
 extern const struct netfs_request_ops ceph_netfs_ops;
 int ceph_mmap_prepare(struct vm_area_desc *desc);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 extern int ceph_uninline_data(struct file *file);
+#endif
 extern int ceph_pool_perm_check(struct inode *inode, int need);
 extern void ceph_pool_perm_destroy(struct ceph_mds_client* mdsc);
 int ceph_purge_inode_cap(struct inode *inode, struct ceph_cap *cap, bool *invalidate);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
 {
 	if (ci->i_inline_version == CEPH_INLINE_NONE ||
@@ -1347,6 +1354,12 @@ static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
 		return false;
 	return true;
 }
+#else
+static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
+{
+	return false;
+}
+#endif
 
 /* file.c */
 extern const struct file_operations ceph_file_fops;
include/linux/ceph/ceph_features.h +6 -1
index 3a47acd9cc14..207b93787a4f 100644
@@ -132,6 +132,11 @@ DEFINE_CEPH_FEATURE(38, 1, OSD_ERASURE_CODES)
 DEFINE_CEPH_FEATURE(38, 1, OSD_OSD_TMAP2OMAP) // overlap
 DEFINE_CEPH_FEATURE(39, 1, OSDMAP_ENC)
 DEFINE_CEPH_FEATURE(40, 1, MDS_INLINE_DATA)
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
+#define CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT	CEPH_FEATURE_MDS_INLINE_DATA
+#else
+#define CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT	0
+#endif
 DEFINE_CEPH_FEATURE(41, 1, CRUSH_TUNABLES3)
 DEFINE_CEPH_FEATURE(41, 1, OSD_PRIMARY_AFFINITY) // overlap
 DEFINE_CEPH_FEATURE(42, 1, MSGR_KEEPALIVE2)
@@ -204,7 +209,7 @@ DEFINE_CEPH_FEATURE_DEPRECATED(63, 1, RESERVED_BROKEN, LUMINOUS) // client-facin
 	 CEPH_FEATURE_CRUSH_V2 |		\
 	 CEPH_FEATURE_EXPORT_PEER |		\
 	 CEPH_FEATURE_OSDMAP_ENC |		\
-	 CEPH_FEATURE_MDS_INLINE_DATA |		\
+	 CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT |	\
 	 CEPH_FEATURE_CRUSH_TUNABLES3 |		\
 	 CEPH_FEATURE_OSD_PRIMARY_AFFINITY |	\
 	 CEPH_FEATURE_MSGR_KEEPALIVE2 |		\

---
base-commit: 602bb4525c8146865b7f6e44c81d54bf97861aed
change-id: 20260728-b4-ceph-remove-inline-data-d55c6466dd7e

Best regards,

Signature · 1 line
Xiubo Li <[email protected]>
#1 lore raw
Quoted 119 lines · On Wed, 2026-07-29 at 09:28 +0800, Xiubo Li via B4 Relay wrote:

From: Xiubo Li <[email protected]>

Inline data support has been deprecated in the CephFS MDS and is being formally removed.  Existing inline data can be migrated to data pool objects via MDS scrub (see PR #44359).

Add CONFIG_CEPH_FS_INLINE_DATA (default n) to gate all inline data code in the kclient.  When disabled, CEPH_FEATURE_MDS_INLINE_DATA is defined as 0 so that the kclient no longer advertises the feature to the MDS.  The MDS will then refuse to open files that still have inline data, returning -EPERM, preventing silent data corruption that would occur if the kclient tried to read inline data from OSD objects that do not contain it.

This config option will be kept for a transitional period to allow existing CephFS deployments to migrate inline data via MDS scrub.  Once the migration period ends, the option and all inline data code will be permanently removed.

Users upgrading from an older Ceph release should ensure all inline data has been migrated before deploying a kclient with this option disabled:
  1. Enable and run MDS scrub to uninline existing data
  2. Disable inline_data on the filesystem:
     ceph fs set <fs> inline_data false --yes-i-really-really-mean-it

---
Changes in v2:
- keep CEPH_FEATURE_MDS_INLINE_DATA always defined as (1ULL << 40) so decode logic can still check MDS feature bits; gate the supported features set via CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT
- Link to v1:
https://patch.msgid.link/[email protected]
---
 fs/ceph/Kconfig                    | 12 ++++++++++++ fs/ceph/addr.c                     | 16 +++++++++++++++ fs/ceph/caps.c                     | 40
+++++++++++++++++++++++++++++++++++++-
 fs/ceph/file.c                     | 20 +++++++++++++++++++ fs/ceph/inode.c                    | 10 ++++++++++ fs/ceph/mds_client.c               | 30 ++++++++++++++++++++++++++++ fs/ceph/mds_client.h               |  2 ++
 fs/ceph/snap.c                     |  2 ++ fs/ceph/super.h                    | 13 +++++++++++++ include/linux/ceph/ceph_features.h |  7 ++++++-

 10 files changed, 150 insertions(+), 2 deletions(-)
fs/ceph/Kconfig +0 -0
index 3d64a316ca31..be1c8ada5665 100644
@@ -17,6 +17,18 @@ config CEPH_FS

    If unsure, say N.

 if CEPH_FS

fs/ceph/Kconfig +7 -0
+config CEPH_FS_INLINE_DATA
+	bool "Enable CephFS inline data support (deprecated)"
+	depends on CEPH_FS
+	default n
+	help
+	  Inline data support has been deprecated in the CephFS MDS
+	  and is being formally removed.  Existing inline data can

be

fs/ceph/Kconfig +5 -0
+	  migrated to data pool objects via MDS scrub.  Choose N
+	  unless you have a legacy filesystem that requires it,
+	  and ensure all inline data has been migrated (see
+	  ceph fs set <fs> inline_data false) before enabling.
+

 config CEPH_FSCACHE
  bool "Enable Ceph client caching support" depends on CEPH_FS=m && FSCACHE || CEPH_FS=y && FSCACHE=y

fs/ceph/addr.c +0 -0
index e2da3ab9f808..5211320a50ac 100644
@@ -274,6 +274,7 @@ static void finish_netfs_read(struct
 ceph_osd_request *req)
  	ceph_dec_osd_stopping_blocker(fsc->mdsc);
  }


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 {
  struct netfs_io_request *rreq = subreq->rreq;

fs/ceph/addr.c +1 -0
@@ -335,6 +336,7 @@ static bool ceph_netfs_issue_op_inline(struct
 netfs_io_subrequest *subreq)
  	netfs_read_subreq_terminated(subreq);
  	return true;
  }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */


 static int ceph_netfs_prepare_read(struct netfs_io_subrequest *subreq)
 {

fs/ceph/addr.c +0 -0
@@ -372,8 +374,10 @@ static void ceph_netfs_issue_read(struct
 netfs_io_subrequest *subreq)
  		goto out;
  	}


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
  return;
+#endif

  // TODO: This rounding here is slightly dodgy.  It *should* work, for
  // now, as the cache only deals in blocks that are a multiple of

fs/ceph/addr.c +0 -0
@@ -2006,7 +2010,11 @@ static vm_fault_t ceph_filemap_fault(struct
 vm_fault *vmf)
  	      off, ceph_cap_string(got));


  if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) || +#ifdef CONFIG_CEPH_FS_INLINE_DATA !ceph_has_inline_data(ci)) {

fs/ceph/addr.c static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf) · line 2017

Quoted 3 lines
fs/ceph/addr.c +3 -0
+#else
+	    true) {
+#endif

Maybe, some static inline function here?

Quoted 4 lines

  CEPH_DEFINE_RW_CONTEXT(rw_ctx, got); ceph_add_rw_context(fi, &rw_ctx); ret = filemap_fault(vmf);

fs/ceph/addr.c static vm_fault_t ceph_filemap_fault(struct · line 2054

Quoted 10 lines
fs/ceph/addr.c +4 -0
@@ -2037,7 +2045,11 @@ static vm_fault_t ceph_filemap_fault(struct
 vm_fault *vmf)
  			goto out_inline;
  		}
  		err = __ceph_do_getattr(inode, page,
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  					 CEPH_STAT_CAP_INLINE_DATA,
 true);
+#else
+					 0, true);
+#endif

Maybe some macro that can hide this?

Quoted 98 lines
fs/ceph/addr.c +0 -0
  		if (err < 0 || off >= i_size_read(inode)) {

  unlock_page(page);
  put_page(page);

fs/ceph/addr.c +0 -0
@@ -2168,6 +2180,7 @@ static vm_fault_t ceph_page_mkwrite(struct
 vm_fault *vmf)
  	return ret;
  }


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
     char *data, size_t len)
 {

fs/ceph/addr.c +1 -0
@@ -2214,7 +2227,9 @@ void ceph_fill_inline_data(struct inode *inode,
 struct page *locked_page,
  		put_page(page);
  	}
  }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 int ceph_uninline_data(struct file *file)
 {
  struct inode *inode = file_inode(file);

fs/ceph/addr.c +1 -0
@@ -2370,6 +2385,7 @@ int ceph_uninline_data(struct file *file)
  	      ceph_vinop(inode), inline_version, err);
  	return err;
  }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */


 static const struct vm_operations_struct ceph_vmops = {
  .fault = ceph_filemap_fault,

fs/ceph/caps.c +11 -1
index f8d898ad091e..2fd2116a3308 100644
@@ -1266,7 +1266,9 @@ struct cap_msg_args {
  	kuid_t			uid;
  	kgid_t			gid;
  	umode_t			mode;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	bool			inline_data;
+#endif
  	bool			wake;
  	bool			encrypted;
  	u32			fscrypt_auth_len;
@@ -1336,8 +1338,16 @@ static void encode_cap_msg(struct ceph_msg
 *msg, struct cap_msg_args *arg)
  	p = fc + 1;
  	/* flock buffer size (version 2) */
  	ceph_encode_32(&p, 0);
-	/* inline version (version 4) */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE);
+#else
+	/*
+	 * Inline data support has been removed.  The field is still
+	 * encoded for protocol compatibility; always advertise
+	 * CEPH_INLINE_NONE to indicate no inline data is present.
+	 */
+	ceph_encode_64(&p, CEPH_INLINE_NONE);
+#endif
  	/* inline data size */
  	ceph_encode_32(&p, 0);

  /*

fs/ceph/caps.c +0 -0
@@ -1499,7 +1509,9 @@ static void __prep_cap(struct cap_msg_args
 *arg, struct ceph_cap *cap,
  	arg->gid = inode->i_gid;
  	arg->mode = inode->i_mode;


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  arg->inline_data = ci->i_inline_version != CEPH_INLINE_NONE; +#endif
  if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) && !list_empty(&ci->i_cap_snaps)) {
  struct ceph_cap_snap *capsnap;

fs/ceph/caps.c +0 -0
@@ -1620,7 +1632,9 @@ static inline int __send_flush_snap(struct
 inode *inode,
  	arg.gid = capsnap->gid;
  	arg.mode = capsnap->mode;


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  arg.inline_data = capsnap->inline_data;
+#endif
  arg.flags = 0;
  arg.wake = false;
  arg.encrypted = IS_ENCRYPTED(inode);

fs/ceph/caps.c +0 -0
@@ -3192,6 +3206,7 @@ int __ceph_get_caps(struct inode *inode, struct
 ceph_file_info *fi, int need,
  			return ret;
  		}


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (S_ISREG(ci->netfs.inode.i_mode) && ceph_has_inline_data(ci) && (_got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&

fs/ceph/caps.c int __ceph_get_caps(struct inode *inode, struct · line 3242

Quoted 5 lines
fs/ceph/caps.c +1 -0
@@ -3223,6 +3238,7 @@ int __ceph_get_caps(struct inode *inode, struct
 ceph_file_info *fi, int need,
  				return ret;
  			continue;
  		}
+#endif

Ditto. Maybe, some static inline function?

Quoted 39 lines
fs/ceph/caps.c +0 -0
  		break;
  	}

  *got = _got;

fs/ceph/caps.c +0 -0
@@ -3518,10 +3534,12 @@ static void invalidate_aliases(struct inode
 *inode)


 struct cap_extra_info {
  struct ceph_string *pool_ns;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  /* inline data */
  u64 inline_version;
  void *inline_data;
  u32 inline_len;
+#endif
  /* dirstat */
  bool dirstat_valid;
  u64 nfiles;

fs/ceph/caps.c +2 -0
@@ -3564,7 +3582,9 @@ static void handle_cap_grant(struct inode
 *inode,
  	bool queue_trunc = false;
  	bool queue_invalidate = false;
  	bool deleted_inode = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	bool fill_inline = false;
+#endif
  	bool revoke_wait = false;
  	int flags = 0;

fs/ceph/caps.c +0 -0
@@ -3830,6 +3850,7 @@ static void handle_cap_grant(struct inode
 *inode,

  check_caps = 2; /* check all caps */ }

+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (extra_info->inline_version > 0 && extra_info->inline_version >= ci->i_inline_version) { ci->i_inline_version = extra_info->inline_version;

fs/ceph/caps.c +0 -0
@@ -3837,6 +3858,7 @@ static void handle_cap_grant(struct inode
 *inode,
Quoted 5 lines

      (newcaps &
(CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
  fill_inline = true;
  }
+#endif

Ditto.

Quoted 39 lines


  if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) { if (ci->i_auth_cap == cap) {

fs/ceph/caps.c +0 -0
@@ -3856,9 +3878,11 @@ static void handle_cap_grant(struct inode
 *inode,
  	}
  	spin_unlock(&ci->i_ceph_lock);


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (fill_inline) ceph_fill_inline_data(inode, NULL, extra_info-

Quoted 1 line

inline_data,

        extra_info->inline_len); +#endif

  if (queue_trunc) ceph_queue_vmtruncate(inode);

fs/ceph/caps.c +0 -0
@@ -4476,12 +4500,26 @@ void ceph_handle_caps(struct ceph_mds_session
 *session,
  	}


  if (msg_version >= 4) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA ceph_decode_64_safe(&p, end, extra_info.inline_version, bad);
  ceph_decode_32_safe(&p, end, extra_info.inline_len, bad);
  if (p + extra_info.inline_len > end) goto bad;
  extra_info.inline_data = p;
  p += extra_info.inline_len;

fs/ceph/caps.c +3 -0
+#else
+		/*
+		 * Inline data support has been removed, but cap

grant

fs/ceph/caps.c +1 -0
+		 * messages (msg_version >= 4) always carry

inline_version

fs/ceph/caps.c +1 -0
+		 * + inline_data.  We must consume the fields to

keep the

fs/ceph/caps.c void ceph_handle_caps(struct ceph_mds_session *session, · line 4517

Quoted 3 lines
fs/ceph/caps.c +3 -0
+		 * decode pointer in sync.
+		 */
+		u32 __len;

The leading double-underscore is reserved namespace in C. Maybe, simply len?

Quoted 30 lines
fs/ceph/caps.c +5 -0
+
+		ceph_decode_skip_64(&p, end, bad);
+		ceph_decode_32_safe(&p, end, __len, bad);
+		ceph_decode_skip_n(&p, end, __len, bad);
+#endif

  }

  if (msg_version >= 5) {

fs/ceph/file.c +0 -0
index 3823671ab95a..901ed97b1963 100644
@@ -224,7 +224,9 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,

  ceph_inode_to_fs_client(&ci->netfs.inode)-

Quoted 1 line

mount_options;

  struct ceph_client *cl = ceph_inode_to_client(inode); struct ceph_file_info *fi;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  int ret;
+#endif

  doutc(cl, "%p %llx.%llx %p 0%o (%s)\n", inode, ceph_vinop(inode),
        file, inode->i_mode, isdir ? "dir" : "regular");

fs/ceph/file.c +0 -0
@@ -258,14 +260,17 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,
  	INIT_LIST_HEAD(&fi->rw_contexts);

  fi->filp_gen = READ_ONCE(ceph_inode_to_fs_client(inode)-

Quoted 1 line

filp_gen);

Quoted 9 lines


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if ((file->f_mode & FMODE_WRITE) &&
ceph_has_inline_data(ci)) {
  ret = ceph_uninline_data(file);
  if (ret < 0)
  goto error;
  }
+#endif

Maybe, static inline function?

Quoted 36 lines


  return 0;

+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 error:
  ceph_fscache_unuse_cookie(inode, file->f_mode & FMODE_WRITE);
  ceph_put_fmode(ci, fi->fmode, 1);

fs/ceph/file.c +1 -0
@@ -273,6 +278,7 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,
  	/* wake up anyone waiting for caps on this inode */
  	wake_up_all(&ci->i_cap_wq);
  	return ret;
+#endif
  }


 /*

fs/ceph/file.c +0 -0
@@ -702,7 +708,9 @@ static int ceph_finish_async_create(struct inode
 *dir, struct inode *inode,


  ktime_get_real_ts64(&now);

+#ifdef CONFIG_CEPH_FS_INLINE_DATA iinfo.inline_version = CEPH_INLINE_NONE; +#endif
  iinfo.change_attr = 1; ceph_encode_timespec64(&iinfo.btime, &now);

fs/ceph/file.c +0 -0
@@ -2224,7 +2232,11 @@ static ssize_t ceph_read_iter(struct kiocb
 *iocb, struct iov_iter *to)
  		      inode, ceph_vinop(inode), iocb->ki_pos,
 (unsigned)len,
  		      ceph_cap_string(got));


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (!ceph_has_inline_data(ci)) {

fs/ceph/file.c static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) · line 2239

Quoted 3 lines
fs/ceph/file.c +3 -0
+#else
+		if (true) {
+#endif

Ditto.

fs/ceph/file.c static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) · line 2286

Quoted 13 lines

  if (!retry_op && (iocb->ki_flags & IOCB_DIRECT) && !IS_ENCRYPTED(inode)) {

fs/ceph/file.c +0 -0
@@ -2267,7 +2279,11 @@ static ssize_t ceph_read_iter(struct kiocb
 *iocb, struct iov_iter *to)
  			if (!page)
  				return -ENOMEM;


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  mask = CEPH_STAT_CAP_INLINE_DATA;

fs/ceph/file.c +3 -0
+#else
+			mask = 0;
+#endif

Ditto.

Quoted 88 lines

  }

  statret = __ceph_do_getattr(inode, page, mask, !!page);

fs/ceph/file.c +0 -0
@@ -2349,8 +2365,12 @@ static ssize_t ceph_splice_read(struct file
 *in, loff_t *ppos,
  	if (ceph_inode_is_shutdown(inode))
  		return -ESTALE;


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (ceph_has_inline_data(ci) ||
      (fi->flags & CEPH_F_SYNC))

fs/ceph/file.c +3 -0
+#else
+	if (fi->flags & CEPH_F_SYNC)
+#endif

  return copy_splice_read(in, ppos, pipe, len, flags);
  ret = ceph_start_io_read(inode);

fs/ceph/inode.c +0 -0
index 61d7c0b8161f..20abf7dc9c64 100644
@@ -627,7 +627,9 @@ struct inode *ceph_alloc_inode(struct super_block
 *sb)
  	spin_lock_init(&ci->i_ceph_lock);


  ci->i_version = 0;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  ci->i_inline_version = 0;
+#endif
  ci->i_time_warp_seq = 0;
  ci->i_ceph_flags = 0;
  atomic64_set(&ci->i_ordered_count, 1);

fs/ceph/inode.c +2 -0
@@ -1030,7 +1032,9 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
  	bool wake = false;
  	bool queue_trunc = false;
  	bool new_version = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	bool fill_inline = false;
+#endif
  	umode_t mode = le32_to_cpu(info->mode);
  	dev_t rdev = le32_to_cpu(info->rdev);

fs/ceph/inode.c +0 -0
@@ -1362,6 +1366,7 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
  		}
  	}


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (iinfo->inline_version > 0 && iinfo->inline_version >= ci->i_inline_version) { int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;

fs/ceph/inode.c +1 -0
@@ -1370,6 +1375,7 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
  		    (locked_page || (info_caps & cache_caps)))
  			fill_inline = true;
  	}
+#endif


  if (cap_fmode >= 0) {
  if (!info_caps)

fs/ceph/inode.c +0 -0
@@ -1382,9 +1388,11 @@ int ceph_fill_inode(struct inode *inode,
 struct page *locked_page,


  ceph_fscache_register_inode_cookie(inode);

+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  if (fill_inline)
  ceph_fill_inline_data(inode, locked_page, iinfo->inline_data, iinfo-

Quoted 1 line

inline_len);

+#endif

  if (wake)
  wake_up_all(&ci->i_cap_wq);

fs/ceph/inode.c +1 -0
@@ -3022,6 +3030,7 @@ int __ceph_do_getattr(struct inode *inode,
 struct page *locked_page,
  	req->r_args.getattr.mask = cpu_to_le32(mask);
  	req->r_locked_page = locked_page;
  	err = ceph_mdsc_do_request(mdsc, NULL, req);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	if (locked_page && err == 0) {

  u64 inline_version = req-

Quoted 1 line

r_reply_info.targeti.inline_version;

  if (inline_version == 0) {

fs/ceph/inode.c int __ceph_do_getattr(struct inode *inode, · line 3047

Quoted 5 lines
fs/ceph/inode.c +1 -0
@@ -3034,6 +3043,7 @@ int __ceph_do_getattr(struct inode *inode,
 struct page *locked_page,
  			err = req->r_reply_info.targeti.inline_len;
  		}
  	}
+#endif

Ditto.

Quoted 30 lines
fs/ceph/inode.c +0 -0
  	ceph_mdsc_put_request(req);
  	doutc(cl, "result=%d\n", err);

  return err;

fs/ceph/mds_client.c +0 -0
index 18be36c2e9e8..6a439da76616 100644
@@ -156,12 +156,26 @@ static int parse_reply_info_in(void **p, void
 *end,
  	*p += info->xattr_len;


  if (features == (u64)-1) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  /* inline data */
  ceph_decode_64_safe(p, end, info->inline_version, bad);
  ceph_decode_32_safe(p, end, info->inline_len, bad); ceph_decode_need(p, end, info->inline_len, bad); info->inline_data = *p;
  *p += info->inline_len;

fs/ceph/mds_client.c +3 -0
+#else
+		/*
+		 * Inline data support has been removed, but the MDS

always

fs/ceph/mds_client.c +1 -0
+		 * encodes inline_version + inline_data in versioned

replies.

fs/ceph/mds_client.c +1 -0
+		 * We must consume these fields from the message to

keep the

fs/ceph/mds_client.c +1 -0
+		 * decode pointer in sync, even though we discard

them.

fs/ceph/mds_client.c

Quoted 2 lines
fs/ceph/mds_client.c +2 -0
+		 */
+		u32 __len;

Ditto, Maybe, simply len?

Quoted 34 lines
fs/ceph/mds_client.c +5 -0
+
+		ceph_decode_skip_64(p, end, bad);
+		ceph_decode_32_safe(p, end, __len, bad);
+		ceph_decode_skip_n(p, end, __len, bad);
+#endif

  /* quota */
  err = parse_reply_info_quota(p, end, info); if (err < 0)

fs/ceph/mds_client.c +1 -0
@@ -276,6 +290,7 @@ static int parse_reply_info_in(void **p, void
 *end,
  		*p = end;
  	} else {
  		/* legacy (unversioned) struct */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  		if (features & CEPH_FEATURE_MDS_INLINE_DATA) {

  ceph_decode_64_safe(p, end, info-

Quoted 1 line

inline_version, bad);

  ceph_decode_32_safe(p, end, info-

Quoted 1 line

inline_len, bad);

fs/ceph/mds_client.c +3 -0
@@ -284,6 +299,21 @@ static int parse_reply_info_in(void **p, void
 *end,
  			*p += info->inline_len;
  		} else
  			info->inline_version = CEPH_INLINE_NONE;
+#else
+		/*
+		 * Inline data support has been removed.  When the

MDS

fs/ceph/mds_client.c +1 -0
+		 * advertises CEPH_FEATURE_MDS_INLINE_DATA it

includes

fs/ceph/mds_client.c +1 -0
+		 * inline_version + inline_data in legacy replies. 

We

fs/ceph/mds_client.c +1 -0
+		 * must consume them to keep the decode pointer in

sync.

fs/ceph/mds_client.c static int parse_reply_info_in(void **p, void *end, · line 310

Quoted 3 lines
fs/ceph/mds_client.c +3 -0
+		 */
+		if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
+			u32 __len;

Ditto. Maybe, len?

fs/ceph/mds_client.c static int parse_reply_info_in(void **p, void *end, · line 316

Quoted 6 lines
fs/ceph/mds_client.c +6 -0
+
+			ceph_decode_skip_64(p, end, bad);
+			ceph_decode_32_safe(p, end, __len, bad);
+			ceph_decode_skip_n(p, end, __len, bad);
+		}
+#endif

Ditto. Maybe, static inline function?

Quoted 119 lines


  if (features & CEPH_FEATURE_MDS_QUOTA) { err = parse_reply_info_quota(p, end, info);

fs/ceph/mds_client.h +2 -0
index 731d6ad04956..9bcf1a9d5ba8 100644
@@ -149,9 +149,11 @@ struct ceph_mds_reply_info_in {
  	char *symlink;
  	u32 xattr_len;
  	char *xattr_data;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  	u64 inline_version;
  	u32 inline_len;
  	char *inline_data;
+#endif
  	u32 pool_ns_len;
  	char *pool_ns_data;
  	u64 max_bytes;
fs/ceph/snap.c +0 -0
index e5d8d59ffc17..3bc3a2ce72a2 100644
@@ -613,7 +613,9 @@ static void ceph_queue_cap_snap(struct
 ceph_inode_info *ci,
  		capsnap->xattr_version = 0;
  	}


+#ifdef CONFIG_CEPH_FS_INLINE_DATA capsnap->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
+#endif

  /* dirty page count moved from _head to this cap_snap; all subsequent writes page dirties occur _after_ this

fs/ceph/super.h +0 -0
index 3737ea7ed88b..eced08f52b79 100644
@@ -274,7 +274,9 @@ struct ceph_cap_snap {
  	u32 truncate_seq;

  int writing;   /* a sync write is still in progress */ int dirty_pages;     /* dirty pages awaiting writeback */ +#ifdef CONFIG_CEPH_FS_INLINE_DATA
  bool inline_data;
+#endif
  bool need_flush;
 };

fs/ceph/super.h +0 -0
@@ -381,7 +383,9 @@ struct ceph_inode_info {
  	spinlock_t i_ceph_lock;


  u64 i_version;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  u64 i_inline_version;
+#endif
  u32 i_time_warp_seq;

  unsigned long i_ceph_flags;

fs/ceph/super.h +2 -0
@@ -1335,11 +1339,14 @@ extern void __ceph_touch_fmode(struct
 ceph_inode_info *ci,
  extern const struct address_space_operations ceph_aops;
  extern const struct netfs_request_ops ceph_netfs_ops;
  int ceph_mmap_prepare(struct vm_area_desc *desc);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
  extern int ceph_uninline_data(struct file *file);
+#endif
  extern int ceph_pool_perm_check(struct inode *inode, int need);
  extern void ceph_pool_perm_destroy(struct ceph_mds_client* mdsc);
  int ceph_purge_inode_cap(struct inode *inode, struct ceph_cap *cap,
 bool *invalidate);


+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static inline bool ceph_has_inline_data(struct ceph_inode_info *ci) {
  if (ci->i_inline_version == CEPH_INLINE_NONE ||

fs/ceph/super.h +6 -0
@@ -1347,6 +1354,12 @@ static inline bool ceph_has_inline_data(struct
 ceph_inode_info *ci)
  		return false;
  	return true;
  }
+#else
+static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
+{
+	return false;
+}
+#endif


 /* file.c */
 extern const struct file_operations ceph_file_fops;

include/linux/ceph/ceph_features.h +0 -0
index 3a47acd9cc14..207b93787a4f 100644
@@ -132,6 +132,11 @@ DEFINE_CEPH_FEATURE(38, 1, OSD_ERASURE_CODES)

 DEFINE_CEPH_FEATURE(38, 1, OSD_OSD_TMAP2OMAP) // overlap DEFINE_CEPH_FEATURE(39, 1, OSDMAP_ENC) DEFINE_CEPH_FEATURE(40, 1, MDS_INLINE_DATA)

include/linux/ceph/ceph_features.h +2 -0
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
+#define

CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT CEPH_FEATURE_MDS_INLINE_DATA

include/linux/ceph/ceph_features.h +3 -0
+#else
+#define CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT	0
+#endif

 DEFINE_CEPH_FEATURE(41, 1, CRUSH_TUNABLES3) DEFINE_CEPH_FEATURE(41, 1, OSD_PRIMARY_AFFINITY) // overlap DEFINE_CEPH_FEATURE(42, 1, MSGR_KEEPALIVE2)

include/linux/ceph/ceph_features.h +0 -0
@@ -204,7 +209,7 @@ DEFINE_CEPH_FEATURE_DEPRECATED(63, 1,

RESERVED_BROKEN, LUMINOUS) // client-facin
  CEPH_FEATURE_CRUSH_V2 | \
  CEPH_FEATURE_EXPORT_PEER | \
  CEPH_FEATURE_OSDMAP_ENC | \

include/linux/ceph/ceph_features.h +1 -1
-	 CEPH_FEATURE_MDS_INLINE_DATA |		\
+	 CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT |	\

  CEPH_FEATURE_CRUSH_TUNABLES3 | \
  CEPH_FEATURE_OSD_PRIMARY_AFFINITY | \
  CEPH_FEATURE_MSGR_KEEPALIVE2 | \

Quoted 5 lines

---
base-commit: 602bb4525c8146865b7f6e44c81d54bf97861aed
change-id: 20260728-b4-ceph-remove-inline-data-d55c6466dd7e

git apply --check ./\[PATCH\ v2\]\ ceph\:\ remove\ the\ inline\ data\
support.mbox
error: patch failed: fs/ceph/Kconfig:17
error: fs/ceph/Kconfig: patch does not apply
error: patch failed: fs/ceph/addr.c:274
error: fs/ceph/addr.c: patch does not apply
error: patch failed: fs/ceph/caps.c:1266
error: fs/ceph/caps.c: patch does not apply
error: patch failed: fs/ceph/file.c:224
error: fs/ceph/file.c: patch does not apply
error: patch failed: fs/ceph/inode.c:627
error: fs/ceph/inode.c: patch does not apply
error: patch failed: fs/ceph/mds_client.c:156
error: fs/ceph/mds_client.c: patch does not apply
error: patch failed: fs/ceph/mds_client.h:149
error: fs/ceph/mds_client.h: patch does not apply
error: patch failed: fs/ceph/snap.c:613
error: fs/ceph/snap.c: patch does not apply
error: patch failed: fs/ceph/super.h:274
error: fs/ceph/super.h: patch does not apply
error: patch failed: include/linux/ceph/ceph_features.h:132
error: include/linux/ceph/ceph_features.h: patch does not apply

Quoted 1 line

./scripts/checkpatch.pl --strict ./\[PATCH\ v2\]\ ceph\:\ remove\ the\
inline\ data\ support.mbox
ERROR: DOS line endings
#145: FILE: fs/ceph/Kconfig:20:
+config CEPH_FS_INLINE_DATA^M$

ERROR: DOS line endings
#146: FILE: fs/ceph/Kconfig:21:
+^Ibool "Enable CephFS inline data support (deprecated)"^M$

Every new line in this patch is indented with spaces, not tabs.

Thanks,
Slava.

author
#2 lore raw

Hi Viacheslav,

Sorry for late reply. I just missed this email.

Thanks for your feedback and finding. I will fix it later.

BRs
Xiubo

Quoted 785 lines · Viacheslav Dubeyko
Quoted 113 lines · On Wed, 2026-07-29 at 09:28 +0800, Xiubo Li via B4 Relay wrote:

From: Xiubo Li <[email protected]>

Inline data support has been deprecated in the CephFS MDS and is being formally removed. Existing inline data can be migrated to data pool objects via MDS scrub (see PR #44359).

Add CONFIG_CEPH_FS_INLINE_DATA (default n) to gate all inline data code in the kclient. When disabled, CEPH_FEATURE_MDS_INLINE_DATA is defined as 0 so that the kclient no longer advertises the feature to the MDS. The MDS will then refuse to open files that still have inline data, returning -EPERM, preventing silent data corruption that would occur if the kclient tried to read inline data from OSD objects that do not contain it.

This config option will be kept for a transitional period to allow existing CephFS deployments to migrate inline data via MDS scrub. Once the migration period ends, the option and all inline data code will be permanently removed.

Users upgrading from an older Ceph release should ensure all inline data has been migrated before deploying a kclient with this option disabled:
1. Enable and run MDS scrub to uninline existing data
2. Disable inline_data on the filesystem:

     ceph fs set <fs> inline_data false --yes-i-really-really-mean-it

---
Changes in v2:
- keep CEPH_FEATURE_MDS_INLINE_DATA always defined as (1ULL << 40) so decode logic can still check MDS feature bits; gate the supported features set via CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT
- Link to v1:
https://patch.msgid.link/[email protected]

---
 fs/ceph/Kconfig                    | 12 ++++++++++++
 fs/ceph/addr.c                     | 16 +++++++++++++++
 fs/ceph/caps.c                     | 40
+++++++++++++++++++++++++++++++++++++-
 fs/ceph/file.c                     | 20 +++++++++++++++++++
 fs/ceph/inode.c                    | 10 ++++++++++
 fs/ceph/mds_client.c               | 30 ++++++++++++++++++++++++++++
 fs/ceph/mds_client.h               |  2 ++
 fs/ceph/snap.c                     |  2 ++
 fs/ceph/super.h                    | 13 +++++++++++++
 include/linux/ceph/ceph_features.h |  7 ++++++-
 10 files changed, 150 insertions(+), 2 deletions(-)
fs/ceph/Kconfig +7 -0
index 3d64a316ca31..be1c8ada5665 100644
@@ -17,6 +17,18 @@ config CEPH_FS
        If unsure, say N.
 
 if CEPH_FS
+config CEPH_FS_INLINE_DATA
+     bool "Enable CephFS inline data support (deprecated)"
+     depends on CEPH_FS
+     default n
+     help
+       Inline data support has been deprecated in the CephFS MDS
+       and is being formally removed.  Existing inline data can

be

fs/ceph/Kconfig +5 -0
+       migrated to data pool objects via MDS scrub.  Choose N
+       unless you have a legacy filesystem that requires it,
+       and ensure all inline data has been migrated (see
+       ceph fs set <fs> inline_data false) before enabling.
+
 config CEPH_FSCACHE
      bool "Enable Ceph client caching support"
      depends on CEPH_FS=m && FSCACHE || CEPH_FS=y && FSCACHE=y
fs/ceph/addr.c +1 -0
index e2da3ab9f808..5211320a50ac 100644
@@ -274,6 +274,7 @@ static void finish_netfs_read(struct
 ceph_osd_request *req)
      ceph_dec_osd_stopping_blocker(fsc->mdsc);
 }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest
 *subreq)
 {
      struct netfs_io_request *rreq = subreq->rreq;
fs/ceph/addr.c +1 -0
@@ -335,6 +336,7 @@ static bool ceph_netfs_issue_op_inline(struct
 netfs_io_subrequest *subreq)
      netfs_read_subreq_terminated(subreq);
      return true;
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
 static int ceph_netfs_prepare_read(struct netfs_io_subrequest

*subreq)
{

fs/ceph/addr.c +2 -0
@@ -372,8 +374,10 @@ static void ceph_netfs_issue_read(struct
 netfs_io_subrequest *subreq)
              goto out;
      }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (ceph_has_inline_data(ci) &&
 ceph_netfs_issue_op_inline(subreq))
              return;
+#endif
 
      // TODO: This rounding here is slightly dodgy.  It *should*
work, for
      // now, as the cache only deals in blocks that are a

multiple of

fs/ceph/addr.c static vm_fault_t ceph_filemap_fault(struct · line 2018

Quoted 9 lines
fs/ceph/addr.c +4 -0
@@ -2006,7 +2010,11 @@ static vm_fault_t ceph_filemap_fault(struct
 vm_fault *vmf)
            off, ceph_cap_string(got));
 
      if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
          !ceph_has_inline_data(ci)) {
+#else
+         true) {
+#endif

Maybe, some static inline function here?

Quoted 4 lines
fs/ceph/addr.c +0 -0
              CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
              ceph_add_rw_context(fi, &rw_ctx);
              ret = filemap_fault(vmf);

fs/ceph/addr.c static vm_fault_t ceph_filemap_fault(struct · line 2054

Quoted 10 lines
fs/ceph/addr.c +4 -0
@@ -2037,7 +2045,11 @@ static vm_fault_t ceph_filemap_fault(struct
 vm_fault *vmf)
                      goto out_inline;
              }
              err = __ceph_do_getattr(inode, page,
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
                                       CEPH_STAT_CAP_INLINE_DATA,
 true);
+#else
+                                      0, true);
+#endif

Maybe some macro that can hide this?

Quoted 98 lines
fs/ceph/addr.c +0 -0
              if (err < 0 || off >= i_size_read(inode)) {
                      unlock_page(page);
                      put_page(page);
fs/ceph/addr.c +1 -0
@@ -2168,6 +2180,7 @@ static vm_fault_t ceph_page_mkwrite(struct
 vm_fault *vmf)
      return ret;
 }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 void ceph_fill_inline_data(struct inode *inode, struct page
 *locked_page,

char *data, size_t len)
{

fs/ceph/addr.c +2 -0
@@ -2214,7 +2227,9 @@ void ceph_fill_inline_data(struct inode *inode,
 struct page *locked_page,
              put_page(page);
      }
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 int ceph_uninline_data(struct file *file)
 {
      struct inode *inode = file_inode(file);
fs/ceph/addr.c +1 -0
@@ -2370,6 +2385,7 @@ int ceph_uninline_data(struct file *file)
            ceph_vinop(inode), inline_version, err);
      return err;
 }
+#endif /* CONFIG_CEPH_FS_INLINE_DATA */
 
 static const struct vm_operations_struct ceph_vmops = {
      .fault          = ceph_filemap_fault,
fs/ceph/caps.c +11 -1
index f8d898ad091e..2fd2116a3308 100644
@@ -1266,7 +1266,9 @@ struct cap_msg_args {
      kuid_t                  uid;
      kgid_t                  gid;
      umode_t                 mode;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      bool                    inline_data;
+#endif
      bool                    wake;
      bool                    encrypted;
      u32                     fscrypt_auth_len;
@@ -1336,8 +1338,16 @@ static void encode_cap_msg(struct ceph_msg
 *msg, struct cap_msg_args *arg)
      p = fc + 1;
      /* flock buffer size (version 2) */
      ceph_encode_32(&p, 0);
-     /* inline version (version 4) */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      ceph_encode_64(&p, arg->inline_data ? 0 : CEPH_INLINE_NONE);
+#else
+     /*
+      * Inline data support has been removed.  The field is still
+      * encoded for protocol compatibility; always advertise
+      * CEPH_INLINE_NONE to indicate no inline data is present.
+      */
+     ceph_encode_64(&p, CEPH_INLINE_NONE);
+#endif
      /* inline data size */
      ceph_encode_32(&p, 0);
      /*
fs/ceph/caps.c +2 -0
@@ -1499,7 +1509,9 @@ static void __prep_cap(struct cap_msg_args
 *arg, struct ceph_cap *cap,
      arg->gid = inode->i_gid;
      arg->mode = inode->i_mode;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      arg->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
+#endif
      if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) &&
          !list_empty(&ci->i_cap_snaps)) {
              struct ceph_cap_snap *capsnap;
fs/ceph/caps.c +2 -0
@@ -1620,7 +1632,9 @@ static inline int __send_flush_snap(struct
 inode *inode,
      arg.gid = capsnap->gid;
      arg.mode = capsnap->mode;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      arg.inline_data = capsnap->inline_data;
+#endif
      arg.flags = 0;
      arg.wake = false;
      arg.encrypted = IS_ENCRYPTED(inode);
fs/ceph/caps.c +1 -0
@@ -3192,6 +3206,7 @@ int __ceph_get_caps(struct inode *inode, struct
 ceph_file_info *fi, int need,
                      return ret;
              }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
              if (S_ISREG(ci->netfs.inode.i_mode) &&
                  ceph_has_inline_data(ci) &&
                  (_got &

(CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) &&

fs/ceph/caps.c int __ceph_get_caps(struct inode *inode, struct · line 3242

Quoted 5 lines
fs/ceph/caps.c +1 -0
@@ -3223,6 +3238,7 @@ int __ceph_get_caps(struct inode *inode, struct
 ceph_file_info *fi, int need,
                              return ret;
                      continue;
              }
+#endif

Ditto. Maybe, some static inline function?

Quoted 38 lines
fs/ceph/caps.c +0 -0
              break;
      }
      *got = _got;
fs/ceph/caps.c +2 -0
@@ -3518,10 +3534,12 @@ static void invalidate_aliases(struct inode
 *inode)
 
 struct cap_extra_info {
      struct ceph_string *pool_ns;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      /* inline data */
      u64 inline_version;
      void *inline_data;
      u32 inline_len;
+#endif
      /* dirstat */
      bool dirstat_valid;
      u64 nfiles;
fs/ceph/caps.c +3 -0
@@ -3564,7 +3582,9 @@ static void handle_cap_grant(struct inode
 *inode,
      bool queue_trunc = false;
      bool queue_invalidate = false;
      bool deleted_inode = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      bool fill_inline = false;
+#endif
      bool revoke_wait = false;
      int flags = 0;
@@ -3830,6 +3850,7 @@ static void handle_cap_grant(struct inode
 *inode,
                      check_caps = 2; /* check all caps */
      }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (extra_info->inline_version > 0 &&
          extra_info->inline_version >= ci->i_inline_version) {
              ci->i_inline_version = extra_info->inline_version;

fs/ceph/caps.c static void handle_cap_grant(struct inode · line 3863

Quoted 6 lines
fs/ceph/caps.c +1 -0
@@ -3837,6 +3858,7 @@ static void handle_cap_grant(struct inode
 *inode,
                  (newcaps &
 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)))
                      fill_inline = true;
      }
+#endif

Ditto.

Quoted 39 lines
fs/ceph/caps.c +0 -0
      if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
              if (ci->i_auth_cap == cap) {
fs/ceph/caps.c +2 -0
@@ -3856,9 +3878,11 @@ static void handle_cap_grant(struct inode
 *inode,
      }
      spin_unlock(&ci->i_ceph_lock);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (fill_inline)
              ceph_fill_inline_data(inode, NULL, extra_info-
 >inline_data,
                                    extra_info->inline_len);
+#endif
 
      if (queue_trunc)
              ceph_queue_vmtruncate(inode);
fs/ceph/caps.c +4 -0
@@ -4476,12 +4500,26 @@ void ceph_handle_caps(struct ceph_mds_session
 *session,
      }
 
      if (msg_version >= 4) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
              ceph_decode_64_safe(&p, end,
 extra_info.inline_version, bad);
              ceph_decode_32_safe(&p, end, extra_info.inline_len,
 bad);
              if (p + extra_info.inline_len > end)
                      goto bad;
              extra_info.inline_data = p;
              p += extra_info.inline_len;
+#else
+             /*
+              * Inline data support has been removed, but cap

grant
+ * messages (msg_version >= 4) always carry
inline_version
+ * + inline_data. We must consume the fields to
keep the

fs/ceph/caps.c void ceph_handle_caps(struct ceph_mds_session *session, · line 4517

Quoted 3 lines
fs/ceph/caps.c +3 -0
+              * decode pointer in sync.
+              */
+             u32 __len;

The leading double-underscore is reserved namespace in C. Maybe, simply len?

Quoted 29 lines
fs/ceph/caps.c +5 -0
+
+             ceph_decode_skip_64(&p, end, bad);
+             ceph_decode_32_safe(&p, end, __len, bad);
+             ceph_decode_skip_n(&p, end, __len, bad);
+#endif
      }
 
      if (msg_version >= 5) {
fs/ceph/file.c +2 -0
index 3823671ab95a..901ed97b1963 100644
@@ -224,7 +224,9 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,
              ceph_inode_to_fs_client(&ci->netfs.inode)-
 >mount_options;
      struct ceph_client *cl = ceph_inode_to_client(inode);
      struct ceph_file_info *fi;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      int ret;
+#endif
 
      doutc(cl, "%p %llx.%llx %p 0%o (%s)\n", inode,
ceph_vinop(inode),
            file, inode->i_mode, isdir ? "dir" : "regular");
fs/ceph/file.c +0 -0
@@ -258,14 +260,17 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,
      INIT_LIST_HEAD(&fi->rw_contexts);
      fi->filp_gen = READ_ONCE(ceph_inode_to_fs_client(inode)-

fs/ceph/file.c static int ceph_init_file_info(struct inode · line 272

Quoted 10 lines
fs/ceph/file.c +2 -0
 >filp_gen);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if ((file->f_mode & FMODE_WRITE) &&
 ceph_has_inline_data(ci)) {
              ret = ceph_uninline_data(file);
              if (ret < 0)
                      goto error;
      }
+#endif

Maybe, static inline function?

Quoted 29 lines
fs/ceph/file.c +1 -0
      return 0;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 error:
      ceph_fscache_unuse_cookie(inode, file->f_mode &
FMODE_WRITE);
      ceph_put_fmode(ci, fi->fmode, 1);
fs/ceph/file.c +1 -0
@@ -273,6 +278,7 @@ static int ceph_init_file_info(struct inode
 *inode, struct file *file,
      /* wake up anyone waiting for caps on this inode */
      wake_up_all(&ci->i_cap_wq);
      return ret;
+#endif
 }
 

/*

fs/ceph/file.c +2 -0
@@ -702,7 +708,9 @@ static int ceph_finish_async_create(struct inode
 *dir, struct inode *inode,
 
      ktime_get_real_ts64(&now);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      iinfo.inline_version = CEPH_INLINE_NONE;
+#endif
      iinfo.change_attr = 1;
      ceph_encode_timespec64(&iinfo.btime, &now);

fs/ceph/file.c static ssize_t ceph_read_iter(struct kiocb · line 2241

Quoted 10 lines
fs/ceph/file.c +4 -0
@@ -2224,7 +2232,11 @@ static ssize_t ceph_read_iter(struct kiocb
 *iocb, struct iov_iter *to)
                    inode, ceph_vinop(inode), iocb->ki_pos,
 (unsigned)len,
                    ceph_cap_string(got));
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
              if (!ceph_has_inline_data(ci)) {
+#else
+             if (true) {
+#endif

Ditto.

fs/ceph/file.c static ssize_t ceph_read_iter(struct kiocb · line 2287

Quoted 13 lines
fs/ceph/file.c +0 -0
                      if (!retry_op &&
                          (iocb->ki_flags & IOCB_DIRECT) &&
                          !IS_ENCRYPTED(inode)) {
fs/ceph/file.c +4 -0
@@ -2267,7 +2279,11 @@ static ssize_t ceph_read_iter(struct kiocb
 *iocb, struct iov_iter *to)
                      if (!page)
                              return -ENOMEM;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
                      mask = CEPH_STAT_CAP_INLINE_DATA;
+#else
+                     mask = 0;
+#endif

Ditto.

Quoted 88 lines
fs/ceph/file.c +0 -0
              }
 

statret = __ceph_do_getattr(inode, page, mask,
!!page);

fs/ceph/file.c +4 -0
@@ -2349,8 +2365,12 @@ static ssize_t ceph_splice_read(struct file
 *in, loff_t *ppos,
      if (ceph_inode_is_shutdown(inode))
              return -ESTALE;
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (ceph_has_inline_data(ci) ||
          (fi->flags & CEPH_F_SYNC))
+#else
+     if (fi->flags & CEPH_F_SYNC)
+#endif
              return copy_splice_read(in, ppos, pipe, len, flags);
 
      ret = ceph_start_io_read(inode);
fs/ceph/inode.c +2 -0
index 61d7c0b8161f..20abf7dc9c64 100644
@@ -627,7 +627,9 @@ struct inode *ceph_alloc_inode(struct super_block
 *sb)
      spin_lock_init(&ci->i_ceph_lock);
 
      ci->i_version = 0;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      ci->i_inline_version = 0;
+#endif
      ci->i_time_warp_seq = 0;
      ci->i_ceph_flags = 0;
      atomic64_set(&ci->i_ordered_count, 1);
fs/ceph/inode.c +3 -0
@@ -1030,7 +1032,9 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
      bool wake = false;
      bool queue_trunc = false;
      bool new_version = false;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      bool fill_inline = false;
+#endif
      umode_t mode = le32_to_cpu(info->mode);
      dev_t rdev = le32_to_cpu(info->rdev);
@@ -1362,6 +1366,7 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
              }
      }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (iinfo->inline_version > 0 &&
          iinfo->inline_version >= ci->i_inline_version) {

int cache_caps = CEPH_CAP_FILE_CACHE |
CEPH_CAP_FILE_LAZYIO;

fs/ceph/inode.c +1 -0
@@ -1370,6 +1375,7 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,
                  (locked_page || (info_caps & cache_caps)))
                      fill_inline = true;
      }
+#endif
 
      if (cap_fmode >= 0) {
              if (!info_caps)
fs/ceph/inode.c +2 -0
@@ -1382,9 +1388,11 @@ int ceph_fill_inode(struct inode *inode,
 struct page *locked_page,
 
      ceph_fscache_register_inode_cookie(inode);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (fill_inline)
              ceph_fill_inline_data(inode, locked_page,
                                    iinfo->inline_data, iinfo-
 >inline_len);
+#endif
 
      if (wake)
              wake_up_all(&ci->i_cap_wq);
fs/ceph/inode.c +1 -0
@@ -3022,6 +3030,7 @@ int __ceph_do_getattr(struct inode *inode,
 struct page *locked_page,
      req->r_args.getattr.mask = cpu_to_le32(mask);
      req->r_locked_page = locked_page;
      err = ceph_mdsc_do_request(mdsc, NULL, req);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      if (locked_page && err == 0) {
              u64 inline_version = req-
Quoted 1 line

r_reply_info.targeti.inline_version;

              if (inline_version == 0) {

fs/ceph/inode.c int __ceph_do_getattr(struct inode *inode, · line 3047

Quoted 5 lines
fs/ceph/inode.c +1 -0
@@ -3034,6 +3043,7 @@ int __ceph_do_getattr(struct inode *inode,
 struct page *locked_page,
                      err = req->r_reply_info.targeti.inline_len;
              }
      }
+#endif

Ditto.

Quoted 30 lines
fs/ceph/inode.c +0 -0
      ceph_mdsc_put_request(req);
      doutc(cl, "result=%d\n", err);
      return err;
fs/ceph/mds_client.c +4 -0
index 18be36c2e9e8..6a439da76616 100644
@@ -156,12 +156,26 @@ static int parse_reply_info_in(void **p, void
 *end,
      *p += info->xattr_len;
 
      if (features == (u64)-1) {
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
              /* inline data */
              ceph_decode_64_safe(p, end, info->inline_version,
 bad);
              ceph_decode_32_safe(p, end, info->inline_len, bad);
              ceph_decode_need(p, end, info->inline_len, bad);
              info->inline_data = *p;
              *p += info->inline_len;
+#else
+             /*
+              * Inline data support has been removed, but the MDS

always
+ * encodes inline_version + inline_data in versioned
replies.
+ * We must consume these fields from the message to
keep the
+ * decode pointer in sync, even though we discard
them.

fs/ceph/mds_client.c

Quoted 2 lines
fs/ceph/mds_client.c +2 -0
+              */
+             u32 __len;

Ditto, Maybe, simply len?

Quoted 34 lines
fs/ceph/mds_client.c +6 -0
+
+             ceph_decode_skip_64(p, end, bad);
+             ceph_decode_32_safe(p, end, __len, bad);
+             ceph_decode_skip_n(p, end, __len, bad);
+#endif
              /* quota */
              err = parse_reply_info_quota(p, end, info);
              if (err < 0)
@@ -276,6 +290,7 @@ static int parse_reply_info_in(void **p, void
 *end,
              *p = end;
      } else {
              /* legacy (unversioned) struct */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
              if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
                      ceph_decode_64_safe(p, end, info-
Quoted 1 line

inline_version, bad);

                      ceph_decode_32_safe(p, end, info-
Quoted 1 line

inline_len, bad);

fs/ceph/mds_client.c +3 -0
@@ -284,6 +299,21 @@ static int parse_reply_info_in(void **p, void
 *end,
                      *p += info->inline_len;
              } else
                      info->inline_version = CEPH_INLINE_NONE;
+#else
+             /*
+              * Inline data support has been removed.  When the

MDS
+ * advertises CEPH_FEATURE_MDS_INLINE_DATA it
includes
+ * inline_version + inline_data in legacy replies.
We
+ * must consume them to keep the decode pointer in
sync.

fs/ceph/mds_client.c static int parse_reply_info_in(void **p, void *end, · line 310

Quoted 3 lines
fs/ceph/mds_client.c +3 -0
+              */
+             if (features & CEPH_FEATURE_MDS_INLINE_DATA) {
+                     u32 __len;

Ditto. Maybe, len?

fs/ceph/mds_client.c static int parse_reply_info_in(void **p, void *end, · line 316

Quoted 6 lines
fs/ceph/mds_client.c +6 -0
+
+                     ceph_decode_skip_64(p, end, bad);
+                     ceph_decode_32_safe(p, end, __len, bad);
+                     ceph_decode_skip_n(p, end, __len, bad);
+             }
+#endif

Ditto. Maybe, static inline function?

Quoted 120 lines
              if (features & CEPH_FEATURE_MDS_QUOTA) {
                      err = parse_reply_info_quota(p, end, info);
fs/ceph/mds_client.h +2 -0
index 731d6ad04956..9bcf1a9d5ba8 100644
@@ -149,9 +149,11 @@ struct ceph_mds_reply_info_in {
      char *symlink;
      u32 xattr_len;
      char *xattr_data;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      u64 inline_version;
      u32 inline_len;
      char *inline_data;
+#endif
      u32 pool_ns_len;
      char *pool_ns_data;
      u64 max_bytes;
fs/ceph/snap.c +2 -0
index e5d8d59ffc17..3bc3a2ce72a2 100644
@@ -613,7 +613,9 @@ static void ceph_queue_cap_snap(struct
 ceph_inode_info *ci,
              capsnap->xattr_version = 0;
      }
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      capsnap->inline_data = ci->i_inline_version !=
 CEPH_INLINE_NONE;
+#endif
 

/* dirty page count moved from _head to this cap_snap;
all subsequent writes page dirties occur _after_ this

fs/ceph/super.h +7 -0
index 3737ea7ed88b..eced08f52b79 100644
@@ -274,7 +274,9 @@ struct ceph_cap_snap {
      u32 truncate_seq;
      int writing;   /* a sync write is still in progress */
      int dirty_pages;     /* dirty pages awaiting writeback */
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      bool inline_data;
+#endif
      bool need_flush;
 };
 
@@ -381,7 +383,9 @@ struct ceph_inode_info {
      spinlock_t i_ceph_lock;
 
      u64 i_version;
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
      u64 i_inline_version;
+#endif
      u32 i_time_warp_seq;
 
      unsigned long i_ceph_flags;
@@ -1335,11 +1339,14 @@ extern void __ceph_touch_fmode(struct
 ceph_inode_info *ci,
 extern const struct address_space_operations ceph_aops;
 extern const struct netfs_request_ops ceph_netfs_ops;
 int ceph_mmap_prepare(struct vm_area_desc *desc);
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 extern int ceph_uninline_data(struct file *file);
+#endif
 extern int ceph_pool_perm_check(struct inode *inode, int need);
 extern void ceph_pool_perm_destroy(struct ceph_mds_client* mdsc);
 int ceph_purge_inode_cap(struct inode *inode, struct ceph_cap *cap,
 bool *invalidate);
 
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
 static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
 {
      if (ci->i_inline_version == CEPH_INLINE_NONE ||
fs/ceph/super.h +6 -0
@@ -1347,6 +1354,12 @@ static inline bool ceph_has_inline_data(struct
 ceph_inode_info *ci)
              return false;
      return true;
 }
+#else
+static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
+{
+     return false;
+}
+#endif
 
 /* file.c */

extern const struct file_operations ceph_file_fops;

include/linux/ceph/ceph_features.h +2 -0
index 3a47acd9cc14..207b93787a4f 100644
@@ -132,6 +132,11 @@ DEFINE_CEPH_FEATURE(38, 1, OSD_ERASURE_CODES)
 DEFINE_CEPH_FEATURE(38, 1, OSD_OSD_TMAP2OMAP) // overlap
 DEFINE_CEPH_FEATURE(39, 1, OSDMAP_ENC)
 DEFINE_CEPH_FEATURE(40, 1, MDS_INLINE_DATA)
+#ifdef CONFIG_CEPH_FS_INLINE_DATA
+#define

CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT CEPH_FEATURE_MDS_INLINE_DATA

include/linux/ceph/ceph_features.h +3 -0
+#else
+#define CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT 0
+#endif
 DEFINE_CEPH_FEATURE(41, 1, CRUSH_TUNABLES3)
 DEFINE_CEPH_FEATURE(41, 1, OSD_PRIMARY_AFFINITY) // overlap
 DEFINE_CEPH_FEATURE(42, 1, MSGR_KEEPALIVE2)
@@ -204,7 +209,7 @@ DEFINE_CEPH_FEATURE_DEPRECATED(63, 1,

RESERVED_BROKEN, LUMINOUS) // client-facin

       CEPH_FEATURE_CRUSH_V2 |                \
       CEPH_FEATURE_EXPORT_PEER |             \
       CEPH_FEATURE_OSDMAP_ENC |              \
include/linux/ceph/ceph_features.h +1 -2
-      CEPH_FEATURE_MDS_INLINE_DATA |         \
+      CEPH_FEATURE_MDS_INLINE_DATA_DEFAULT | \
       CEPH_FEATURE_CRUSH_TUNABLES3 |         \
       CEPH_FEATURE_OSD_PRIMARY_AFFINITY |    \
       CEPH_FEATURE_MSGR_KEEPALIVE2 |         \
 
---
Quoted 4 lines

base-commit: 602bb4525c8146865b7f6e44c81d54bf97861aed
change-id: 20260728-b4-ceph-remove-inline-data-d55c6466dd7e

git apply --check ./\[PATCH\ v2\]\ ceph\:\ remove\ the\ inline\ data\
support.mbox
error: patch failed: fs/ceph/Kconfig:17
error: fs/ceph/Kconfig: patch does not apply
error: patch failed: fs/ceph/addr.c:274
error: fs/ceph/addr.c: patch does not apply
error: patch failed: fs/ceph/caps.c:1266
error: fs/ceph/caps.c: patch does not apply
error: patch failed: fs/ceph/file.c:224
error: fs/ceph/file.c: patch does not apply
error: patch failed: fs/ceph/inode.c:627
error: fs/ceph/inode.c: patch does not apply
error: patch failed: fs/ceph/mds_client.c:156
error: fs/ceph/mds_client.c: patch does not apply
error: patch failed: fs/ceph/mds_client.h:149
error: fs/ceph/mds_client.h: patch does not apply
error: patch failed: fs/ceph/snap.c:613
error: fs/ceph/snap.c: patch does not apply
error: patch failed: fs/ceph/super.h:274
error: fs/ceph/super.h: patch does not apply
error: patch failed: include/linux/ceph/ceph_features.h:132
error: include/linux/ceph/ceph_features.h: patch does not apply

Quoted 1 line

./scripts/checkpatch.pl --strict ./\[PATCH\ v2\]\ ceph\:\ remove\ the\
inline\ data\ support.mbox
ERROR: DOS line endings
#145: FILE: fs/ceph/Kconfig:20:
+config CEPH_FS_INLINE_DATA^M$

ERROR: DOS line endings
#146: FILE: fs/ceph/Kconfig:21:
+^Ibool "Enable CephFS inline data support (deprecated)"^M$

Every new line in this patch is indented with spaces, not tabs.

Thanks,
Slava.