Skip to contentSkip to conversation

fs/ceph: optimize struct layouts

PATCH Superseded Max Kellermann 3 participants 23 messages Jun 12 – Aug 12

Subsystems (2)

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

cover fs/ceph: optimize struct layouts

This patch set aims to reduce the memory usage of the Ceph filesystems by optimizing the layout of structs, most importantly struct ceph_inode_info:

- use bool instead of int for booelans
- use u32 instead of u64 where possible
- reorder fields to eliminate padding holes
- eliminate redundant fields

Without this patch set, ceph_inode_info looks like this with pahole:

/* size: 1456, cachelines: 23, members: 78 */ /* sum members: 1424, holes: 8, sum holes: 32 */ /* member types with holes: 4, total: 4 */

With this patch set, struct ceph_inode_info got smaller by 96 bytes:

/* size: 1360, cachelines: 22, members: 75 */
/* member types with holes: 2, total: 2 */

Max Kellermann (12):
fs/ceph/super: remove unused field i_cap_migration_resv fs/ceph/super: make field i_truncate_pagecache_size optional include/ceph/ceph_fs.h: convert pool_id to u32 fs/ceph/super.h: convert ceph_inode_xattr fields to bool fs/ceph/super.h: convert ceph_cap_snap.writing fields to bool fs/ceph: consistently use u32 for time_warp_seq fs/ceph/super: reorder fields to eliminate padding holes fs/ceph: remove i_truncate_mutex, use i_fragtree_mutex for both fs/ceph/super.h: add const to helpers fs/ceph/super.h: add helper ceph_in_snap()
fs/ceph: use ceph_vino() etc. instead of accessing i_vino directly fs/ceph: remove redundant inode number from ceph_inode_info

 fs/ceph/acl.c                |   2 +-
 fs/ceph/addr.c               |  23 ++++----
 fs/ceph/cache.c              |   4 +-
 fs/ceph/caps.c               |   8 +--
 fs/ceph/dir.c                |  20 +++----
 fs/ceph/export.c             |   8 +--
 fs/ceph/file.c               |  28 +++++-----
 fs/ceph/inode.c              |  38 ++++++++-----
 fs/ceph/mds_client.c         |  12 ++---
 fs/ceph/mds_client.h         |   2 +-
 fs/ceph/quota.c              |   6 +--
 fs/ceph/snap.c               |   8 +--
 fs/ceph/super.h              | 102 +++++++++++++++++++++++++----------
 fs/ceph/util.c               |   8 +--
 fs/ceph/xattr.c              |  18 +++----
 include/linux/ceph/ceph_fs.h |   2 +-
 16 files changed, 171 insertions(+), 118 deletions(-)
Signature · 1 line
2.47.3
#1 lore raw

PATCH 1/12 fs/ceph/super: remove unused field `i_cap_migration_resv`

This field was added 17 years ago by commit de57606c23afd ("ceph:
client types") but it was never used.

This reduces the size of struct ceph_inode_info by 8 bytes.

Fixes de57606c23afd ("ceph: client types")
---
 fs/ceph/super.h | 1 -
 1 file changed, 1 deletion(-)
fs/ceph/super.h +0 -1
index afc89ce91804..6aec40955387 100644
@@ -448,7 +448,6 @@ struct ceph_inode_info {
 	wait_queue_head_t i_cap_wq;      /* threads waiting on a capability */
 	unsigned long i_hold_caps_max; /* jiffies */
 	struct list_head i_cap_delay_list;  /* for delayed cap release to mds */
-	struct ceph_cap_reservation i_cap_migration_resv;
 	struct list_head i_cap_snaps;   /* snapped state pending flush to mds */
 	struct ceph_snap_context *i_head_snapc;  /* set if wr_buffer_head > 0 or
 						    dirty|flushing caps */
Signature · 1 line
2.47.3
maintainer Reviewed-by on v1 1/12
#2 lore raw

Re: [PATCH 01/12] fs/ceph/super: remove unused field `i_cap_migration_resv`

Quoted 22 lines · Max Kellermann

This field was added 17 years ago by commit de57606c23afd ("ceph:
client types") but it was never used.

This reduces the size of struct ceph_inode_info by 8 bytes.

Fixes de57606c23afd ("ceph: client types")

---
 fs/ceph/super.h | 1 -

 1 file changed, 1 deletion(-)
fs/ceph/super.h +0 -0
index afc89ce91804..6aec40955387 100644
@@ -448,7 +448,6 @@ struct ceph_inode_info {

  wait_queue_head_t i_cap_wq;      /* threads waiting on a capability */
  unsigned long i_hold_caps_max; /* jiffies */ struct list_head i_cap_delay_list;  /* for delayed cap release to mds */

fs/ceph/super.h +0 -1
-	struct ceph_cap_reservation i_cap_migration_resv;
Quoted 6 lines

  struct list_head i_cap_snaps;   /* snapped state pending flush to mds */
  struct ceph_snap_context *i_head_snapc;  /* set if wr_buffer_head > 0 or
      dirty|flushing caps */

Removing the dead field makes sense to me.

Thanks,
Slava.

#3 lore raw

PATCH 2/12 fs/ceph/super: make field `i_truncate_pagecache_size` optional

In reply to Max Kellermann

This field is only used with CONFIG_FS_ENCRYPTION; without it, it always equals i_truncate_size.

This reduces the size of struct ceph_inode_info by 8 bytes on kernels that have filesystem encryption disabled.

---
 fs/ceph/inode.c |  8 ++++++--
 fs/ceph/super.h | 11 +++++++++++
 2 files changed, 17 insertions(+), 2 deletions(-)
fs/ceph/inode.c +6 -2
index 22c7da1ea61c..1504b889622c 100644
@@ -680,7 +680,9 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	ci->i_truncate_seq = 0;
 	ci->i_truncate_size = 0;
 	ci->i_truncate_pending = 0;
+#ifdef CONFIG_FS_ENCRYPTION
 	ci->i_truncate_pagecache_size = 0;
+#endif
 
 	ci->i_max_size = 0;
 	ci->i_reported_size = 0;
@@ -865,6 +867,7 @@ int ceph_fill_file_size(struct inode *inode, int issued,
 
 		ci->i_truncate_size = truncate_size;
 
+#ifdef CONFIG_FS_ENCRYPTION
 		if (IS_ENCRYPTED(inode)) {
 			doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
 			      ci->i_truncate_pagecache_size, size);
@@ -872,6 +875,7 @@ int ceph_fill_file_size(struct inode *inode, int issued,
 		} else {
 			ci->i_truncate_pagecache_size = truncate_size;
 		}
+#endif
 	}
 	return queue_trunc;
 }
@@ -2331,7 +2335,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
 	/* there should be no reader or writer */
 	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
 
-	to = ci->i_truncate_pagecache_size;
+	to = ceph_get_truncate_pagecache_size(ci);
 	wrbuffer_refs = ci->i_wrbuffer_ref;
 	doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
 	      ci->i_truncate_pending, to);
@@ -2341,7 +2345,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
 	truncate_pagecache(inode, to);
 
 	spin_lock(&ci->i_ceph_lock);
-	if (to == ci->i_truncate_pagecache_size) {
+	if (to == ceph_get_truncate_pagecache_size(ci)) {
 		ci->i_truncate_pending = 0;
 		finish = 1;
 	}
fs/ceph/super.h +11 -0
index 6aec40955387..f0f03aeb5b19 100644
@@ -461,11 +461,14 @@ struct ceph_inode_info {
 	u32 i_truncate_seq;        /* last truncate to smaller size */
 	u64 i_truncate_size;       /*  and the size we last truncated down to */
 	int i_truncate_pending;    /*  still need to call vmtruncate */
+
+#ifdef CONFIG_FS_ENCRYPTION
 	/*
 	 * For none fscrypt case it equals to i_truncate_size or it will
 	 * equals to fscrypt_file_size
 	 */
 	u64 i_truncate_pagecache_size;
+#endif
 
 	u64 i_max_size;            /* max file size authorized by mds */
 	u64 i_reported_size; /* (max_)size reported to or requested of mds */
@@ -660,6 +663,14 @@ static inline struct inode *ceph_find_inode(struct super_block *sb,
 	return ilookup5(sb, (unsigned long)vino.ino, ceph_ino_compare, &vino);
 }
 
+static inline u64 ceph_get_truncate_pagecache_size(const struct ceph_inode_info *ci)
+{
+#ifdef CONFIG_CEPH_FS_SNAPSHOTS
+	return ci->i_truncate_pagecache_size;
+#else
+	return ci->i_truncate_size;
+#endif
+}
 
 /*
  * Ceph inode.
Signature · 1 line
2.47.3
#4 lore raw

Re: [PATCH 02/12] fs/ceph/super: make field `i_truncate_pagecache_size` optional

Quoted 101 lines · Max Kellermann

This field is only used with CONFIG_FS_ENCRYPTION; without it, it always equals i_truncate_size.

This reduces the size of struct ceph_inode_info by 8 bytes on kernels that have filesystem encryption disabled.

---
 fs/ceph/inode.c |  8 ++++++--
 fs/ceph/super.h | 11 +++++++++++

 2 files changed, 17 insertions(+), 2 deletions(-)
fs/ceph/inode.c +2 -0
index 22c7da1ea61c..1504b889622c 100644
@@ -680,7 +680,9 @@ struct inode *ceph_alloc_inode(struct super_block
 *sb)
  	ci->i_truncate_seq = 0;
  	ci->i_truncate_size = 0;
  	ci->i_truncate_pending = 0;
+#ifdef CONFIG_FS_ENCRYPTION
  	ci->i_truncate_pagecache_size = 0;
+#endif


  ci->i_max_size = 0;
  ci->i_reported_size = 0;

fs/ceph/inode.c +0 -0
@@ -865,6 +867,7 @@ int ceph_fill_file_size(struct inode *inode, int
 issued,


  ci->i_truncate_size = truncate_size;

+#ifdef CONFIG_FS_ENCRYPTION
  if (IS_ENCRYPTED(inode)) {
  doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
        ci->i_truncate_pagecache_size, size);

fs/ceph/inode.c +0 -0
@@ -872,6 +875,7 @@ int ceph_fill_file_size(struct inode *inode, int
 issued,
  		} else {

  ci->i_truncate_pagecache_size = truncate_size;
  }
+#endif
  }
  return queue_trunc;
 }

fs/ceph/inode.c +0 -0
@@ -2331,7 +2335,7 @@ void __ceph_do_pending_vmtruncate(struct inode
 *inode)
  	/* there should be no reader or writer */
  	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);

fs/ceph/inode.c +1 -1
-	to = ci->i_truncate_pagecache_size;
+	to = ceph_get_truncate_pagecache_size(ci);

  wrbuffer_refs = ci->i_wrbuffer_ref; doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
        ci->i_truncate_pending, to);

fs/ceph/inode.c +0 -0
@@ -2341,7 +2345,7 @@ void __ceph_do_pending_vmtruncate(struct inode
 *inode)
  	truncate_pagecache(inode, to);


  spin_lock(&ci->i_ceph_lock);

fs/ceph/inode.c +1 -1
-	if (to == ci->i_truncate_pagecache_size) {
+	if (to == ceph_get_truncate_pagecache_size(ci)) {

  ci->i_truncate_pending = 0;
  finish = 1;
  }

fs/ceph/super.h +0 -0
index 6aec40955387..f0f03aeb5b19 100644
@@ -461,11 +461,14 @@ struct ceph_inode_info {

  u32 i_truncate_seq;        /* last truncate to smaller size */
  u64 i_truncate_size;       /*  and the size we last truncated down to */
  int i_truncate_pending;    /*  still need to call vmtruncate */

fs/ceph/super.h +2 -0
+
+#ifdef CONFIG_FS_ENCRYPTION

  /*
  * For none fscrypt case it equals to i_truncate_size or it will
  * equals to fscrypt_file_size
  */
  u64 i_truncate_pagecache_size;
+#endif

  u64 i_max_size;            /* max file size authorized by mds */
  u64 i_reported_size; /* (max_)size reported to or requested of mds */

fs/ceph/super.h +0 -0
@@ -660,6 +663,14 @@ static inline struct inode
 *ceph_find_inode(struct super_block *sb,
  	return ilookup5(sb, (unsigned long)vino.ino,
 ceph_ino_compare, &vino);
  }


+static inline u64 ceph_get_truncate_pagecache_size(const struct ceph_inode_info *ci)

fs/ceph/super.h static inline struct inode *ceph_find_inode(struct super_block *sb, · line 668

Quoted 2 lines
fs/ceph/super.h +2 -0
+{
+#ifdef CONFIG_CEPH_FS_SNAPSHOTS

Maybe, I am missing something. But where CONFIG_CEPH_FS_SNAPSHOTS is defined? Currently, we don't have it in the fs/ceph/Kconfig and the patch has no declaration of this compilation option.

Thanks,
Slava.

Quoted 8 lines
fs/ceph/super.h +5 -0
+	return ci->i_truncate_pagecache_size;
+#else
+	return ci->i_truncate_size;
+#endif
+}


 /*
  * Ceph inode.

#5 lore raw

Re: [PATCH 02/12] fs/ceph/super: make field `i_truncate_pagecache_size` optional

Quoted 2 lines · Viacheslav Dubeyko

Maybe, I am missing something. But where CONFIG_CEPH_FS_SNAPSHOTS is defined?

No, I was missing something, I pasted the wrong macro name; it should have been CONFIG_FS_ENCRYPTION. I was about to send v2 today with the fix (but got distracted).

(The CONFIG_CEPH_FS_SNAPSHOTS is from another draft patch I was experimenting with that would allow shrinking the structs further for people who were not interested in using Ceph snapshots.)

Signature · 8 lines
Max Kellermann
Principal Architect
Hosting Technology

cm4all | Im Mediapark 6a | 50670 Köln | Germany
General information about the company can be found here:
https://www.cm4all.com/impressum
A member of the IONOS Group
#6 lore raw

PATCH 3/12 include/ceph/ceph_fs.h: convert `pool_id` to u32

In reply to Max Kellermann

While the Ceph OSD protocol transmits OSD pool ids as 64 bit integers, the MDS protocol is limited to 32 bits. This is a protocol limitation we cannot fix, but it gives us the chance to reduce the struct sizes by only using the integer size we really need.

There is one caveat: previously, -1 was used to indicate "invalid pool id". This patch changes this magic value to 0 (i.e. removes the special-case code from ceph_file_layout_{from,to}_legacy()). This is fine because on the wire, this magic value is 0, too.

This reduces the size of struct ceph_inode_info by 16 bytes (because it contains struct ceph_file_layout twice, and that struct shrinks by 8 bytes; the small pool_id now fits in the existing padding hole).

---
 fs/ceph/addr.c               | 16 ++++++++--------
 fs/ceph/caps.c               |  2 +-
 fs/ceph/inode.c              |  2 +-
 fs/ceph/mds_client.h         |  2 +-
 fs/ceph/util.c               |  8 +-------
 fs/ceph/xattr.c              | 10 +++++-----
 include/linux/ceph/ceph_fs.h |  2 +-
 7 files changed, 18 insertions(+), 24 deletions(-)
fs/ceph/addr.c +8 -8
index 0a86f672cc09..a11b5c633358 100644
@@ -2381,7 +2381,7 @@ enum {
 };
 
 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
-				s64 pool, struct ceph_string *pool_ns)
+				u32 pool, struct ceph_string *pool_ns)
 {
 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(&ci->netfs.inode);
 	struct ceph_mds_client *mdsc = fsc->mdsc;
@@ -2420,10 +2420,10 @@ static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
 		goto out;
 
 	if (pool_ns)
-		doutc(cl, "pool %lld ns %.*s no perm cached\n", pool,
+		doutc(cl, "pool %u ns %.*s no perm cached\n", pool,
 		      (int)pool_ns->len, pool_ns->str);
 	else
-		doutc(cl, "pool %lld no perm cached\n", pool);
+		doutc(cl, "pool %u no perm cached\n", pool);
 
 	down_write(&mdsc->pool_perm_rwsem);
 	p = &mdsc->pool_perm_tree.rb_node;
@@ -2548,10 +2548,10 @@ static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
 	if (!err)
 		err = have;
 	if (pool_ns)
-		doutc(cl, "pool %lld ns %.*s result = %d\n", pool,
+		doutc(cl, "pool %u ns %.*s result = %d\n", pool,
 		      (int)pool_ns->len, pool_ns->str, err);
 	else
-		doutc(cl, "pool %lld result = %d\n", pool, err);
+		doutc(cl, "pool %u result = %d\n", pool, err);
 	return err;
 }
 
@@ -2560,7 +2560,7 @@ int ceph_pool_perm_check(struct inode *inode, int need)
 	struct ceph_client *cl = ceph_inode_to_client(inode);
 	struct ceph_inode_info *ci = ceph_inode(inode);
 	struct ceph_string *pool_ns;
-	s64 pool;
+	u32 pool;
 	int ret, flags;
 
 	/* Only need to do this for regular files */
@@ -2587,11 +2587,11 @@ int ceph_pool_perm_check(struct inode *inode, int need)
 check:
 	if (flags & CEPH_I_POOL_PERM) {
 		if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
-			doutc(cl, "pool %lld no read perm\n", pool);
+			doutc(cl, "pool %u no read perm\n", pool);
 			return -EPERM;
 		}
 		if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
-			doutc(cl, "pool %lld no write perm\n", pool);
+			doutc(cl, "pool %u no write perm\n", pool);
 			return -EPERM;
 		}
 		return 0;
fs/ceph/caps.c +1 -1
index d51454e995a8..cf9916608dc2 100644
@@ -3638,7 +3638,7 @@ static void handle_cap_grant(struct inode *inode,
 
 	if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
 		/* file layout may have changed */
-		s64 old_pool = ci->i_layout.pool_id;
+		u32 old_pool = ci->i_layout.pool_id;
 		struct ceph_string *old_ns;
 
 		ceph_file_layout_from_legacy(&ci->i_layout, &grant->layout);
fs/ceph/inode.c +1 -1
index 1504b889622c..c26217ed2034 100644
@@ -1175,7 +1175,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 	if (new_version ||
 	    (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 		u64 size = le64_to_cpu(info->size);
-		s64 old_pool = ci->i_layout.pool_id;
+		u32 old_pool = ci->i_layout.pool_id;
 		struct ceph_string *old_ns;
 
 		ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
fs/ceph/mds_client.h +1 -1
index 4e6c87f8414c..7b848fdd6f70 100644
@@ -396,7 +396,7 @@ struct ceph_mds_request {
 struct ceph_pool_perm {
 	struct rb_node node;
 	int perm;
-	s64 pool;
+	u32 pool;
 	size_t pool_ns_len;
 	char pool_ns[];
 };
fs/ceph/util.c +1 -7
index 2c34875675bf..5bb40af08b1c 100644
@@ -35,9 +35,6 @@ void ceph_file_layout_from_legacy(struct ceph_file_layout *fl,
 	fl->stripe_count = le32_to_cpu(legacy->fl_stripe_count);
 	fl->object_size = le32_to_cpu(legacy->fl_object_size);
 	fl->pool_id = le32_to_cpu(legacy->fl_pg_pool);
-	if (fl->pool_id == 0 && fl->stripe_unit == 0 &&
-	    fl->stripe_count == 0 && fl->object_size == 0)
-		fl->pool_id = -1;
 }
 
 void ceph_file_layout_to_legacy(struct ceph_file_layout *fl,
@@ -46,10 +43,7 @@ void ceph_file_layout_to_legacy(struct ceph_file_layout *fl,
 	legacy->fl_stripe_unit = cpu_to_le32(fl->stripe_unit);
 	legacy->fl_stripe_count = cpu_to_le32(fl->stripe_count);
 	legacy->fl_object_size = cpu_to_le32(fl->object_size);
-	if (fl->pool_id >= 0)
-		legacy->fl_pg_pool = cpu_to_le32(fl->pool_id);
-	else
-		legacy->fl_pg_pool = 0;
+	legacy->fl_pg_pool = cpu_to_le32(fl->pool_id);
 }
 
 int ceph_flags_to_mode(int flags)
fs/ceph/xattr.c +5 -5
index e773be07f767..9808eb10625d 100644
@@ -50,7 +50,7 @@ static bool ceph_vxattrcb_layout_exists(struct ceph_inode_info *ci)
 {
 	struct ceph_file_layout *fl = &ci->i_layout;
 	return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||
-		fl->object_size > 0 || fl->pool_id >= 0 ||
+		fl->object_size > 0 || fl->pool_id > 0 ||
 		rcu_dereference_raw(fl->pool_ns) != NULL);
 }
 
@@ -61,7 +61,7 @@ static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
 	struct ceph_client *cl = fsc->client;
 	struct ceph_osd_client *osdc = &fsc->client->osdc;
 	struct ceph_string *pool_ns;
-	s64 pool = ci->i_layout.pool_id;
+	u32 pool = ci->i_layout.pool_id;
 	const char *pool_name;
 	const char *ns_field = " pool_namespace=";
 	char buf[128];
@@ -81,7 +81,7 @@ static ssize_t ceph_vxattrcb_layout(struct ceph_inode_info *ci, char *val,
 		total_len = len + strlen(pool_name);
 	} else {
 		len = snprintf(buf, sizeof(buf),
-		"stripe_unit=%u stripe_count=%u object_size=%u pool=%lld",
+		"stripe_unit=%u stripe_count=%u object_size=%u pool=%u",
 		ci->i_layout.stripe_unit, ci->i_layout.stripe_count,
 		ci->i_layout.object_size, pool);
 		total_len = len;
@@ -164,7 +164,7 @@ static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
 	ssize_t ret;
 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci->netfs.inode.i_sb);
 	struct ceph_osd_client *osdc = &fsc->client->osdc;
-	s64 pool = ci->i_layout.pool_id;
+	u32 pool = ci->i_layout.pool_id;
 	const char *pool_name;
 
 	down_read(&osdc->lock);
@@ -174,7 +174,7 @@ static ssize_t ceph_vxattrcb_layout_pool(struct ceph_inode_info *ci,
 		if (ret <= size)
 			memcpy(val, pool_name, ret);
 	} else {
-		ret = ceph_fmt_xattr(val, size, "%lld", pool);
+		ret = ceph_fmt_xattr(val, size, "%u", pool);
 	}
 	up_read(&osdc->lock);
 	return ret;
include/linux/ceph/ceph_fs.h +1 -1
index 69ac3e55a3fe..7faa18493a07 100644
@@ -69,7 +69,7 @@ struct ceph_file_layout {
 	u32 stripe_unit;   /* stripe unit, in bytes */
 	u32 stripe_count;  /* over this many objects */
 	u32 object_size;   /* until objects are this big */
-	s64 pool_id;        /* rados pool id */
+	u32 pool_id;        /* rados pool id */
 	struct ceph_string __rcu *pool_ns; /* rados pool namespace */
 };
 
Signature · 1 line
2.47.3
#7 lore raw

Re: [PATCH 03/12] include/ceph/ceph_fs.h: convert `pool_id` to u32

Quoted 6 lines · Max Kellermann

While the Ceph OSD protocol transmits OSD pool ids as 64 bit integers,
the MDS protocol is limited to 32 bits.  This is a protocol limitation
we cannot fix, but it gives us the chance to reduce the struct sizes by only using the integer size we really need.

We have tendency of growing capacity of file system volumes. I would prefer to increase this value to u64 instead of decreasing to u32.

Quoted 6 lines

There is one caveat: previously, -1 was used to indicate "invalid pool
id".  This patch changes this magic value to 0 (i.e. removes the special-case code from ceph_file_layout_{from,to}_legacy()).  This is fine because on the wire, this magic value is 0, too.

Potentially, we can use U64_MAX as invalid pool id.

Thanks,
Slava.

Quoted 235 lines

This reduces the size of struct ceph_inode_info by 16 bytes (because
it contains struct ceph_file_layout twice, and that struct shrinks by 8 bytes; the small pool_id now fits in the existing padding hole).

---
 fs/ceph/addr.c               | 16 ++++++++-------- fs/ceph/caps.c               |  2 +- fs/ceph/inode.c              |  2 +- fs/ceph/mds_client.h         |  2 +- fs/ceph/util.c               |  8 +------- fs/ceph/xattr.c              | 10 +++++----- include/linux/ceph/ceph_fs.h |  2 +-

 7 files changed, 18 insertions(+), 24 deletions(-)
fs/ceph/addr.c +0 -0
index 0a86f672cc09..a11b5c633358 100644
@@ -2381,7 +2381,7 @@ enum {
  };


 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,

fs/ceph/addr.c +0 -1
-				s64 pool, struct ceph_string

*pool_ns)

fs/ceph/addr.c +1 -0
+				u32 pool, struct ceph_string

*pool_ns)
 {
  struct ceph_fs_client *fsc = ceph_inode_to_fs_client(&ci-

Quoted 1 line

netfs.inode);

  struct ceph_mds_client *mdsc = fsc->mdsc;

fs/ceph/addr.c +0 -0
@@ -2420,10 +2420,10 @@ static int __ceph_pool_perm_get(struct
 ceph_inode_info *ci,
  		goto out;


  if (pool_ns)

fs/ceph/addr.c +0 -1
-		doutc(cl, "pool %lld ns %.*s no perm cached\n",

pool,

fs/ceph/addr.c +1 -0
+		doutc(cl, "pool %u ns %.*s no perm cached\n", pool,

        (int)pool_ns->len, pool_ns->str); else

fs/ceph/addr.c +1 -1
-		doutc(cl, "pool %lld no perm cached\n", pool);
+		doutc(cl, "pool %u no perm cached\n", pool);


  down_write(&mdsc->pool_perm_rwsem);
  p = &mdsc->pool_perm_tree.rb_node;

fs/ceph/addr.c +1 -1
@@ -2548,10 +2548,10 @@ static int __ceph_pool_perm_get(struct
 ceph_inode_info *ci,
  	if (!err)
  		err = have;
  	if (pool_ns)
-		doutc(cl, "pool %lld ns %.*s result = %d\n", pool,
+		doutc(cl, "pool %u ns %.*s result = %d\n", pool,
  		      (int)pool_ns->len, pool_ns->str, err);

  else

fs/ceph/addr.c +1 -1
-		doutc(cl, "pool %lld result = %d\n", pool, err);
+		doutc(cl, "pool %u result = %d\n", pool, err);

  return err;
 }

fs/ceph/addr.c +1 -1
@@ -2560,7 +2560,7 @@ int ceph_pool_perm_check(struct inode *inode,
 int need)
  	struct ceph_client *cl = ceph_inode_to_client(inode);
  	struct ceph_inode_info *ci = ceph_inode(inode);
  	struct ceph_string *pool_ns;
-	s64 pool;
+	u32 pool;
  	int ret, flags;


  /* Only need to do this for regular files */

fs/ceph/addr.c +0 -0
@@ -2587,11 +2587,11 @@ int ceph_pool_perm_check(struct inode *inode,
 int need)

 check:
  if (flags & CEPH_I_POOL_PERM) {
  if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {

fs/ceph/addr.c +1 -1
-			doutc(cl, "pool %lld no read perm\n", pool);
+			doutc(cl, "pool %u no read perm\n", pool);

  return -EPERM;
  }
  if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {

fs/ceph/addr.c +0 -1
-			doutc(cl, "pool %lld no write perm\n",

pool);

fs/ceph/addr.c +1 -0
+			doutc(cl, "pool %u no write perm\n", pool);

  return -EPERM;
  }
  return 0;

fs/ceph/caps.c +0 -0
index d51454e995a8..cf9916608dc2 100644
@@ -3638,7 +3638,7 @@ static void handle_cap_grant(struct inode
 *inode,


  if (newcaps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) {
  /* file layout may have changed */

fs/ceph/caps.c +1 -1
-		s64 old_pool = ci->i_layout.pool_id;
+		u32 old_pool = ci->i_layout.pool_id;

  struct ceph_string *old_ns;

  ceph_file_layout_from_legacy(&ci->i_layout, &grant-

Quoted 1 line

layout);

fs/ceph/inode.c +0 -0
index 1504b889622c..c26217ed2034 100644
@@ -1175,7 +1175,7 @@ int ceph_fill_inode(struct inode *inode, struct
 page *locked_page,

  if (new_version ||
      (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
  u64 size = le64_to_cpu(info->size);

fs/ceph/inode.c +1 -1
-		s64 old_pool = ci->i_layout.pool_id;
+		u32 old_pool = ci->i_layout.pool_id;

  struct ceph_string *old_ns;

  ceph_file_layout_from_legacy(&ci->i_layout, &info-

Quoted 1 line

layout);

fs/ceph/mds_client.h +1 -1
index 4e6c87f8414c..7b848fdd6f70 100644
@@ -396,7 +396,7 @@ struct ceph_mds_request {
  struct ceph_pool_perm {
  	struct rb_node node;
  	int perm;
-	s64 pool;
+	u32 pool;
  	size_t pool_ns_len;
  	char pool_ns[];
  };
fs/ceph/util.c +0 -3
index 2c34875675bf..5bb40af08b1c 100644
@@ -35,9 +35,6 @@ void ceph_file_layout_from_legacy(struct
 ceph_file_layout *fl,
  	fl->stripe_count = le32_to_cpu(legacy->fl_stripe_count);
  	fl->object_size = le32_to_cpu(legacy->fl_object_size);
  	fl->pool_id = le32_to_cpu(legacy->fl_pg_pool);
-	if (fl->pool_id == 0 && fl->stripe_unit == 0 &&
-	    fl->stripe_count == 0 && fl->object_size == 0)
-		fl->pool_id = -1;
  }


 void ceph_file_layout_to_legacy(struct ceph_file_layout *fl,

fs/ceph/util.c +1 -4
@@ -46,10 +43,7 @@ void ceph_file_layout_to_legacy(struct
 ceph_file_layout *fl,
  	legacy->fl_stripe_unit = cpu_to_le32(fl->stripe_unit);
  	legacy->fl_stripe_count = cpu_to_le32(fl->stripe_count);
  	legacy->fl_object_size = cpu_to_le32(fl->object_size);
-	if (fl->pool_id >= 0)
-		legacy->fl_pg_pool = cpu_to_le32(fl->pool_id);
-	else
-		legacy->fl_pg_pool = 0;
+	legacy->fl_pg_pool = cpu_to_le32(fl->pool_id);
  }


 int ceph_flags_to_mode(int flags)

fs/ceph/xattr.c +0 -0
index e773be07f767..9808eb10625d 100644
@@ -50,7 +50,7 @@ static bool ceph_vxattrcb_layout_exists(struct
 ceph_inode_info *ci)
  {
  	struct ceph_file_layout *fl = &ci->i_layout;

  return (fl->stripe_unit > 0 || fl->stripe_count > 0 ||

fs/ceph/xattr.c +1 -1
-		fl->object_size > 0 || fl->pool_id >= 0 ||
+		fl->object_size > 0 || fl->pool_id > 0 ||

  rcu_dereference_raw(fl->pool_ns) != NULL); }

fs/ceph/xattr.c +1 -1
@@ -61,7 +61,7 @@ static ssize_t ceph_vxattrcb_layout(struct
 ceph_inode_info *ci, char *val,
  	struct ceph_client *cl = fsc->client;
  	struct ceph_osd_client *osdc = &fsc->client->osdc;
  	struct ceph_string *pool_ns;
-	s64 pool = ci->i_layout.pool_id;
+	u32 pool = ci->i_layout.pool_id;
  	const char *pool_name;
  	const char *ns_field = " pool_namespace=";

  char buf[128];

fs/ceph/xattr.c +1 -1
@@ -81,7 +81,7 @@ static ssize_t ceph_vxattrcb_layout(struct
 ceph_inode_info *ci, char *val,
  		total_len = len + strlen(pool_name);
  	} else {
  		len = snprintf(buf, sizeof(buf),
-		"stripe_unit=%u stripe_count=%u object_size=%u
 pool=%lld",
+		"stripe_unit=%u stripe_count=%u object_size=%u
 pool=%u",

  ci->i_layout.stripe_unit, ci->i_layout.stripe_count, ci->i_layout.object_size, pool); total_len = len;

fs/ceph/xattr.c +0 -0
@@ -164,7 +164,7 @@ static ssize_t ceph_vxattrcb_layout_pool(struct
 ceph_inode_info *ci,
  	ssize_t ret;

  struct ceph_fs_client *fsc = ceph_sb_to_fs_client(ci-

Quoted 1 line

netfs.inode.i_sb);

  struct ceph_osd_client *osdc = &fsc->client->osdc;

fs/ceph/xattr.c +1 -1
-	s64 pool = ci->i_layout.pool_id;
+	u32 pool = ci->i_layout.pool_id;

  const char *pool_name;

  down_read(&osdc->lock);

fs/ceph/xattr.c +1 -1
@@ -174,7 +174,7 @@ static ssize_t ceph_vxattrcb_layout_pool(struct
 ceph_inode_info *ci,
  		if (ret <= size)
  			memcpy(val, pool_name, ret);
  	} else {
-		ret = ceph_fmt_xattr(val, size, "%lld", pool);
+		ret = ceph_fmt_xattr(val, size, "%u", pool);
  	}
  	up_read(&osdc->lock);

  return ret;

include/linux/ceph/ceph_fs.h +0 -0
index 69ac3e55a3fe..7faa18493a07 100644
@@ -69,7 +69,7 @@ struct ceph_file_layout {

  u32 stripe_unit;   /* stripe unit, in bytes */ u32 stripe_count;  /* over this many objects */ u32 object_size;   /* until objects are this big */

include/linux/ceph/ceph_fs.h +1 -1
-	s64 pool_id;        /* rados pool id */
+	u32 pool_id;        /* rados pool id */

  struct ceph_string __rcu *pool_ns; /* rados pool namespace */
 };

#8 lore raw

Re: [PATCH 03/12] include/ceph/ceph_fs.h: convert `pool_id` to u32

Quoted 2 lines · Viacheslav Dubeyko

We have tendency of growing capacity of file system volumes. I would prefer to increase this value to u64 instead of decreasing to u32.

Which would requires a new incompatible MDS protocol. As long as that does not exist, it makes no sense to cast the wire-level u32 to s64.

Signature · 8 lines
Max Kellermann
Principal Architect
Hosting Technology

cm4all | Im Mediapark 6a | 50670 Köln | Germany
General information about the company can be found here:
https://www.cm4all.com/impressum
A member of the IONOS Group
#9 lore raw

PATCH 4/12 fs/ceph/super.h: convert ceph_inode_xattr fields to `bool`

In reply to Max Kellermann

This reduces the size of struct ceph_inode_xattr by 8 bytes.

---
 fs/ceph/super.h | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
fs/ceph/super.h +3 -3
index f0f03aeb5b19..18960a9e735a 100644
@@ -318,10 +318,10 @@ struct ceph_inode_xattr {
 	int name_len;
 	const char *val;
 	int val_len;
-	int dirty;
+	bool dirty;
 
-	int should_free_name;
-	int should_free_val;
+	bool should_free_name;
+	bool should_free_val;
 };
 
 /*
Signature · 1 line
2.47.3
maintainer Reviewed-by on v1 4/12
#10 lore raw

Re: [PATCH 04/12] fs/ceph/super.h: convert ceph_inode_xattr fields to `bool`

Quoted 22 lines · Max Kellermann

This reduces the size of struct ceph_inode_xattr by 8 bytes.

---
 fs/ceph/super.h | 6 +++---

 1 file changed, 3 insertions(+), 3 deletions(-)
fs/ceph/super.h +1 -1
index f0f03aeb5b19..18960a9e735a 100644
@@ -318,10 +318,10 @@ struct ceph_inode_xattr {
  	int name_len;
  	const char *val;
  	int val_len;
-	int dirty;
+	bool dirty;

fs/ceph/super.h +2 -2
-	int should_free_name;
-	int should_free_val;
+	bool should_free_name;
+	bool should_free_val;
Quoted 3 lines

 };

 /*

To use the bool instead of int data type looks good.

Thanks,
Slava.

#11 lore raw

PATCH 5/12 fs/ceph/super.h: convert ceph_cap_snap.writing fields to `bool`

In reply to Max Kellermann

This will allow better struct packing.

---
 fs/ceph/caps.c  | 2 +-
 fs/ceph/snap.c  | 4 ++--
 fs/ceph/super.h | 2 +-
 3 files changed, 4 insertions(+), 4 deletions(-)
fs/ceph/caps.c +1 -1
index cf9916608dc2..e50bcb393c8f 100644
@@ -3288,7 +3288,7 @@ static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
 					struct ceph_cap_snap,
 					ci_item);
 
-		capsnap->writing = 0;
+		capsnap->writing = false;
 		if (ceph_try_drop_cap_snap(ci, capsnap))
 			/* put the ref held by ceph_queue_cap_snap() */
 			put++;
fs/ceph/snap.c +2 -2
index 52b4c2684f92..6dbc7859ff49 100644
@@ -521,7 +521,7 @@ static bool has_new_snaps(struct ceph_snap_context *o,
  * However, if a (sync) write is currently in-progress when we apply
  * the snapshot, we have to wait until the write succeeds or fails
  * (and a final size/mtime is known).  In this case the
- * cap_snap->writing = 1, and is said to be "pending."  When the write
+ * cap_snap->writing = true, and is said to be "pending."  When the write
  * finishes, we __ceph_finish_cap_snap().
  *
  * Caller must hold snap_rwsem for read (i.e., the realm topology won't
@@ -627,7 +627,7 @@ static void ceph_queue_cap_snap(struct ceph_inode_info *ci,
 		doutc(cl, "%p %llx.%llx cap_snap %p snapc %p seq %llu used WR,"
 		      " now pending\n", inode, ceph_vinop(inode), capsnap,
 		      old_snapc, old_snapc->seq);
-		capsnap->writing = 1;
+		capsnap->writing = true;
 	} else {
 		/* note mtime, size NOW. */
 		__ceph_finish_cap_snap(ci, capsnap);
fs/ceph/super.h +1 -1
index 18960a9e735a..505a1552f355 100644
@@ -269,7 +269,7 @@ struct ceph_cap_snap {
 	u64 time_warp_seq;
 	u64 truncate_size;
 	u32 truncate_seq;
-	int writing;   /* a sync write is still in progress */
+	bool writing;   /* a sync write is still in progress */
 	int dirty_pages;     /* dirty pages awaiting writeback */
 	bool inline_data;
 	bool need_flush;
Signature · 1 line
2.47.3
maintainer Reviewed-by on v1 5/12
#12 lore raw

Re: [PATCH 05/12] fs/ceph/super.h: convert ceph_cap_snap.writing fields to `bool`

Quoted 63 lines · Max Kellermann

This will allow better struct packing.

---
 fs/ceph/caps.c  | 2 +-
 fs/ceph/snap.c  | 4 ++--
 fs/ceph/super.h | 2 +-

 3 files changed, 4 insertions(+), 4 deletions(-)
fs/ceph/caps.c +0 -0
index cf9916608dc2..e50bcb393c8f 100644
@@ -3288,7 +3288,7 @@ static void __ceph_put_cap_refs(struct
 ceph_inode_info *ci, int had,
  					struct ceph_cap_snap,
  					ci_item);

fs/ceph/caps.c +1 -1
-		capsnap->writing = 0;
+		capsnap->writing = false;

  if (ceph_try_drop_cap_snap(ci, capsnap)) /* put the ref held by ceph_queue_cap_snap() */
  put++;

fs/ceph/snap.c +0 -1
index 52b4c2684f92..6dbc7859ff49 100644
@@ -521,7 +521,7 @@ static bool has_new_snaps(struct
 ceph_snap_context *o,
   * However, if a (sync) write is currently in-progress when we apply
   * the snapshot, we have to wait until the write succeeds or fails
   * (and a final size/mtime is known).  In this case the
- * cap_snap->writing = 1, and is said to be "pending."  When the

write
+ * cap_snap->writing = true, and is said to be "pending."  When the write
  * finishes, we __ceph_finish_cap_snap().
  *
  * Caller must hold snap_rwsem for read (i.e., the realm topology won't

fs/ceph/snap.c +0 -0
@@ -627,7 +627,7 @@ static void ceph_queue_cap_snap(struct
 ceph_inode_info *ci,

  doutc(cl, "%p %llx.%llx cap_snap %p snapc %p seq %llu used WR,"
        " now pending\n", inode, ceph_vinop(inode), capsnap,
        old_snapc, old_snapc->seq);

fs/ceph/snap.c +1 -1
-		capsnap->writing = 1;
+		capsnap->writing = true;

  } else {
  /* note mtime, size NOW. */ __ceph_finish_cap_snap(ci, capsnap);

fs/ceph/super.h +1 -1
index 18960a9e735a..505a1552f355 100644
@@ -269,7 +269,7 @@ struct ceph_cap_snap {
  	u64 time_warp_seq;
  	u64 truncate_size;
  	u32 truncate_seq;
-	int writing;   /* a sync write is still in progress */
+	bool writing;   /* a sync write is still in progress */
Quoted 3 lines

  int dirty_pages;     /* dirty pages awaiting writeback */ bool inline_data;
  bool need_flush;

Looks good.

Thanks,
Slava.

#13 lore raw

PATCH 6/12 fs/ceph: consistently use `u32` for `time_warp_seq`

In reply to Max Kellermann

Previously, both u32 and u64 was used with implicit casts. All this does is add useless overhead.

This will allow better struct packing in struct ceph_cap_snap.

---
 fs/ceph/inode.c | 6 +++---
 fs/ceph/super.h | 4 ++--
 2 files changed, 5 insertions(+), 5 deletions(-)
fs/ceph/inode.c +3 -3
index c26217ed2034..17df694fe978 100644
@@ -915,7 +915,7 @@ void ceph_inode_set_subvolume(struct inode *inode, u64 subvolume_id)
 }
 
 void ceph_fill_file_time(struct inode *inode, int issued,
-			 u64 time_warp_seq, struct timespec64 *ctime,
+			 u32 time_warp_seq, struct timespec64 *ctime,
 			 struct timespec64 *mtime, struct timespec64 *atime)
 {
 	struct ceph_client *cl = ceph_inode_to_client(inode);
@@ -939,7 +939,7 @@ void ceph_fill_file_time(struct inode *inode, int issued,
 		    ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 			/* the MDS did a utimes() */
 			doutc(cl, "mtime %ptSp -> %ptSp tw %d -> %d\n", &imtime, mtime,
-			      ci->i_time_warp_seq, (int)time_warp_seq);
+			      ci->i_time_warp_seq, time_warp_seq);
 
 			inode_set_mtime_to_ts(inode, *mtime);
 			inode_set_atime_to_ts(inode, *atime);
@@ -971,7 +971,7 @@ void ceph_fill_file_time(struct inode *inode, int issued,
 		}
 	}
 	if (warn) /* time_warp_seq shouldn't go backwards */
-		doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
+		doutc(cl, "%p mds time_warp_seq %u < %u\n", inode,
 		      time_warp_seq, ci->i_time_warp_seq);
 }
 
fs/ceph/super.h +2 -2
index 505a1552f355..9b5119b1d0a9 100644
@@ -266,7 +266,7 @@ struct ceph_cap_snap {
 	u64 size;
 	u64 change_attr;
 	struct timespec64 mtime, atime, ctime, btime;
-	u64 time_warp_seq;
+	u32 time_warp_seq;
 	u64 truncate_size;
 	u32 truncate_seq;
 	bool writing;   /* a sync write is still in progress */
@@ -1091,7 +1091,7 @@ extern int ceph_fill_file_size(struct inode *inode, int issued,
 			       u32 truncate_seq, u64 truncate_size, u64 size);
 extern void ceph_inode_set_subvolume(struct inode *inode, u64 subvolume_id);
 extern void ceph_fill_file_time(struct inode *inode, int issued,
-				u64 time_warp_seq, struct timespec64 *ctime,
+				u32 time_warp_seq, struct timespec64 *ctime,
 				struct timespec64 *mtime,
 				struct timespec64 *atime);
 extern int ceph_fill_inode(struct inode *inode, struct page *locked_page,
Signature · 1 line
2.47.3
maintainer Reviewed-by on v1 6/12
#14 lore raw

Re: [PATCH 06/12] fs/ceph: consistently use `u32` for `time_warp_seq`

Quoted 74 lines · Max Kellermann

Previously, both u32 and u64 was used with implicit casts.  All this does is add useless overhead.

This will allow better struct packing in struct ceph_cap_snap.

---
 fs/ceph/inode.c | 6 +++---
 fs/ceph/super.h | 4 ++--

 2 files changed, 5 insertions(+), 5 deletions(-)
fs/ceph/inode.c +0 -0
index c26217ed2034..17df694fe978 100644
@@ -915,7 +915,7 @@ void ceph_inode_set_subvolume(struct inode
 *inode, u64 subvolume_id)
  }


 void ceph_fill_file_time(struct inode *inode, int issued,

fs/ceph/inode.c +0 -1
-			 u64 time_warp_seq, struct timespec64

*ctime,

fs/ceph/inode.c +1 -0
+			 u32 time_warp_seq, struct timespec64

*ctime,
  struct timespec64 *mtime, struct timespec64 *atime)
 {
  struct ceph_client *cl = ceph_inode_to_client(inode);

fs/ceph/inode.c +0 -0
@@ -939,7 +939,7 @@ void ceph_fill_file_time(struct inode *inode, int
 issued,
  		    ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq)
 > 0) {
  			/* the MDS did a utimes() */

  doutc(cl, "mtime %ptSp -> %ptSp tw %d -> %d\n", &imtime, mtime,

fs/ceph/inode.c +0 -1
-			      ci->i_time_warp_seq,

(int)time_warp_seq);

fs/ceph/inode.c +1 -0
+			      ci->i_time_warp_seq, time_warp_seq);


  inode_set_mtime_to_ts(inode, *mtime); inode_set_atime_to_ts(inode, *atime);

fs/ceph/inode.c +0 -0
@@ -971,7 +971,7 @@ void ceph_fill_file_time(struct inode *inode, int
 issued,
  		}
  	}

  if (warn) /* time_warp_seq shouldn't go backwards */

fs/ceph/inode.c +1 -1
-		doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
+		doutc(cl, "%p mds time_warp_seq %u < %u\n", inode,

        time_warp_seq, ci->i_time_warp_seq); }

fs/ceph/super.h +1 -1
index 505a1552f355..9b5119b1d0a9 100644
@@ -266,7 +266,7 @@ struct ceph_cap_snap {
  	u64 size;
  	u64 change_attr;
  	struct timespec64 mtime, atime, ctime, btime;
-	u64 time_warp_seq;
+	u32 time_warp_seq;
  	u64 truncate_size;
  	u32 truncate_seq;

  bool writing;   /* a sync write is still in progress */

fs/ceph/super.h +0 -0
@@ -1091,7 +1091,7 @@ extern int ceph_fill_file_size(struct inode
 *inode, int issued,
  			       u32 truncate_seq, u64 truncate_size,
 u64 size);

 extern void ceph_inode_set_subvolume(struct inode *inode, u64 subvolume_id);
 extern void ceph_fill_file_time(struct inode *inode, int issued,

fs/ceph/super.h +0 -1
-				u64 time_warp_seq, struct timespec64

*ctime,

fs/ceph/super.h +1 -0
+				u32 time_warp_seq, struct timespec64
Quoted 5 lines

*ctime,
  struct timespec64 *mtime, struct timespec64 *atime); extern int ceph_fill_inode(struct inode *inode, struct page *locked_page,

Makes sense.

Thanks,
Slava.

#15 lore raw

PATCH 7/12 fs/ceph/super: reorder fields to eliminate padding holes

In reply to Max Kellermann

This reduces the size of struct ceph_inode_info by 32 bytes.

---
 fs/ceph/super.h | 20 ++++++++++++--------
 1 file changed, 12 insertions(+), 8 deletions(-)
fs/ceph/super.h +12 -8
index 9b5119b1d0a9..646ebc03d263 100644
@@ -248,6 +248,9 @@ struct ceph_cap_flush {
  */
 struct ceph_cap_snap {
 	refcount_t nref;
+
+	umode_t mode;
+
 	struct list_head ci_item;
 
 	struct ceph_cap_flush cap_flush;
@@ -256,7 +259,6 @@ struct ceph_cap_snap {
 	int issued, dirty;
 	struct ceph_snap_context *context;
 
-	umode_t mode;
 	kuid_t uid;
 	kgid_t gid;
 
@@ -266,11 +268,12 @@ struct ceph_cap_snap {
 	u64 size;
 	u64 change_attr;
 	struct timespec64 mtime, atime, ctime, btime;
-	u32 time_warp_seq;
 	u64 truncate_size;
 	u32 truncate_seq;
-	bool writing;   /* a sync write is still in progress */
+	u32 time_warp_seq;
 	int dirty_pages;     /* dirty pages awaiting writeback */
+
+	bool writing;   /* a sync write is still in progress */
 	bool inline_data;
 	bool need_flush;
 };
@@ -315,8 +318,8 @@ struct ceph_inode_xattr {
 	struct rb_node node;
 
 	const char *name;
-	int name_len;
 	const char *val;
+	int name_len;
 	int val_len;
 	bool dirty;
 
@@ -377,9 +380,9 @@ struct ceph_inode_info {
 
 	spinlock_t i_ceph_lock;
 
+	u32 i_time_warp_seq;
 	u64 i_version;
 	u64 i_inline_version;
-	u32 i_time_warp_seq;
 
 	unsigned long i_ceph_flags;
 	atomic64_t i_release_count;
@@ -410,8 +413,8 @@ struct ceph_inode_info {
 
 	s32 i_dir_pin;
 
-	struct rb_root i_fragtree;
 	int i_fragtree_nsplits;
+	struct rb_root i_fragtree;
 	struct mutex i_fragtree_mutex;
 
 	struct ceph_inode_xattrs_info i_xattrs;
@@ -451,15 +454,14 @@ struct ceph_inode_info {
 	struct list_head i_cap_snaps;   /* snapped state pending flush to mds */
 	struct ceph_snap_context *i_head_snapc;  /* set if wr_buffer_head > 0 or
 						    dirty|flushing caps */
-	unsigned i_snap_caps;           /* cap bits for snapped files */
 
 	unsigned long i_last_rd;
 	unsigned long i_last_wr;
 	int i_nr_by_mode[CEPH_FILE_MODE_BITS];  /* open file counts */
 
 	struct mutex i_truncate_mutex;
-	u32 i_truncate_seq;        /* last truncate to smaller size */
 	u64 i_truncate_size;       /*  and the size we last truncated down to */
+	u32 i_truncate_seq;        /* last truncate to smaller size */
 	int i_truncate_pending;    /*  still need to call vmtruncate */
 
 #ifdef CONFIG_FS_ENCRYPTION
@@ -488,6 +490,8 @@ struct ceph_inode_info {
 	struct list_head i_unsafe_iops;   /* uncommitted mds inode ops */
 	spinlock_t i_unsafe_lock;
 
+	unsigned int i_snap_caps;           /* cap bits for snapped files */
+
 	union {
 		struct ceph_snap_realm *i_snap_realm; /* snap realm (if caps) */
 		struct ceph_snapid_map *i_snapid_map; /* snapid -> dev_t */
Signature · 1 line
2.47.3
#16 lore raw

PATCH 8/12 fs/ceph: remove i_truncate_mutex, use i_fragtree_mutex for both

In reply to Max Kellermann

i_fragtree_mutex is only used for directories and i_truncate_mutex is only used for regular files. Since these two fields will never be both be used on any given object, let's merge them into one.

(This is a minimal diff that defines a preprocessor macro, to avoid refactoring all code lines using these two mutexes.)

This reduces the size of struct ceph_inode_info by 8 bytes.

---
 fs/ceph/inode.c | 1 -
 fs/ceph/super.h | 6 +++++-
 2 files changed, 5 insertions(+), 2 deletions(-)
fs/ceph/inode.c +0 -1
index 17df694fe978..c0d88cf1f6e3 100644
@@ -676,7 +676,6 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
 	for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 		ci->i_nr_by_mode[i] = 0;
 
-	mutex_init(&ci->i_truncate_mutex);
 	ci->i_truncate_seq = 0;
 	ci->i_truncate_size = 0;
 	ci->i_truncate_pending = 0;
fs/ceph/super.h +5 -1
index 646ebc03d263..19f26724b285 100644
@@ -459,7 +459,11 @@ struct ceph_inode_info {
 	unsigned long i_last_wr;
 	int i_nr_by_mode[CEPH_FILE_MODE_BITS];  /* open file counts */
 
-	struct mutex i_truncate_mutex;
+/* since i_fragtree_mutex is only used for directories and
+   i_truncate_mutex is only used for regular files, we use the same
+   field for both */
+#define i_truncate_mutex i_fragtree_mutex
+
 	u64 i_truncate_size;       /*  and the size we last truncated down to */
 	u32 i_truncate_seq;        /* last truncate to smaller size */
 	int i_truncate_pending;    /*  still need to call vmtruncate */
Signature · 1 line
2.47.3
#17 lore raw

PATCH 9/12 fs/ceph/super.h: add `const` to helpers

In reply to Max Kellermann

This documents that these functions do not modify the parameters.

---
 fs/ceph/super.h | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)
fs/ceph/super.h +9 -9
index 19f26724b285..b1d38fa6647f 100644
@@ -581,7 +581,7 @@ static inline u32 ceph_ino_to_ino32(u64 vino)
  * we do want to set it to something, so that generic vfs code has an
  * appropriate value for tracepoints and the like.
  */
-static inline ino_t ceph_vino_to_ino_t(struct ceph_vino vino)
+static inline ino_t ceph_vino_to_ino_t(const struct ceph_vino vino)
 {
 	if (sizeof(ino_t) == sizeof(u32))
 		return ceph_ino_to_ino32(vino.ino);
@@ -591,12 +591,12 @@ static inline ino_t ceph_vino_to_ino_t(struct ceph_vino vino)
 /* for printf-style formatting */
 #define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap
 
-static inline u64 ceph_ino(struct inode *inode)
+static inline u64 ceph_ino(const struct inode *inode)
 {
 	return ceph_inode(inode)->i_vino.ino;
 }
 
-static inline u64 ceph_snap(struct inode *inode)
+static inline u64 ceph_snap(const struct inode *inode)
 {
 	return ceph_inode(inode)->i_vino.snap;
 }
@@ -810,7 +810,7 @@ static inline struct ceph_dentry_info *ceph_dentry(const struct dentry *dentry)
 /*
  * caps helpers
  */
-static inline bool __ceph_is_any_real_caps(struct ceph_inode_info *ci)
+static inline bool __ceph_is_any_real_caps(const struct ceph_inode_info *ci)
 {
 	return !RB_EMPTY_ROOT(&ci->i_caps);
 }
@@ -1068,7 +1068,7 @@ void ceph_umount_begin(struct super_block *sb);
  * a cap_snap is "pending" if it is still awaiting an in-progress
  * sync write (that may/may not still update size, mtime, etc.).
  */
-static inline bool __ceph_have_pending_cap_snap(struct ceph_inode_info *ci)
+static inline bool __ceph_have_pending_cap_snap(const struct ceph_inode_info *ci)
 {
 	return !list_empty(&ci->i_cap_snaps) &&
 	       list_last_entry(&ci->i_cap_snaps, struct ceph_cap_snap,
@@ -1165,7 +1165,7 @@ extern int ceph_getattr(struct mnt_idmap *idmap,
 			u32 request_mask, unsigned int flags);
 void ceph_inode_shutdown(struct inode *inode);
 
-static inline bool ceph_inode_is_shutdown(struct inode *inode)
+static inline bool ceph_inode_is_shutdown(const struct inode *inode)
 {
 	unsigned long flags = READ_ONCE(ceph_inode(inode)->i_ceph_flags);
 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
@@ -1349,7 +1349,7 @@ 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);
 
-static inline bool ceph_has_inline_data(struct ceph_inode_info *ci)
+static inline bool ceph_has_inline_data(const struct ceph_inode_info *ci)
 {
 	if (ci->i_inline_version == CEPH_INLINE_NONE ||
 	    ci->i_inline_version == 1) /* initial version, no data */
@@ -1424,7 +1424,7 @@ enum quota_get_realm {
 	QUOTA_GET_ANY
 };
 
-static inline bool __ceph_has_quota(struct ceph_inode_info *ci,
+static inline bool __ceph_has_quota(const struct ceph_inode_info *ci,
 				    enum quota_get_realm which)
 {
 	bool has_quota = false;
@@ -1457,7 +1457,7 @@ static inline void __ceph_update_quota(struct ceph_inode_info *ci,
 		ceph_adjust_quota_realms_count(&ci->netfs.inode, has_quota);
 }
 
-static inline int __ceph_sparse_read_ext_count(struct inode *inode, u64 len)
+static inline int __ceph_sparse_read_ext_count(const struct inode *inode, u64 len)
 {
 	int cnt = 0;
 
Signature · 1 line
2.47.3
#18 lore raw

PATCH 10/12 fs/ceph/super.h: add helper ceph_in_snap()

In reply to Max Kellermann

Some code simplification.

---
 fs/ceph/acl.c        |  2 +-
 fs/ceph/addr.c       |  2 +-
 fs/ceph/dir.c        | 20 ++++++++++----------
 fs/ceph/export.c     |  8 ++++----
 fs/ceph/file.c       | 14 +++++++-------
 fs/ceph/inode.c      | 14 +++++++-------
 fs/ceph/mds_client.c | 10 +++++-----
 fs/ceph/quota.c      |  4 ++--
 fs/ceph/super.h      |  8 ++++++++
 fs/ceph/xattr.c      |  4 ++--
 10 files changed, 47 insertions(+), 39 deletions(-)
fs/ceph/acl.c +1 -1
index 85d3dd48b167..dbc64ab60308 100644
@@ -99,7 +99,7 @@ int ceph_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
 	struct timespec64 old_ctime = inode_get_ctime(inode);
 	umode_t new_mode = inode->i_mode, old_mode = inode->i_mode;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP) {
+	if (ceph_in_snap(inode)) {
 		ret = -EROFS;
 		goto out;
 	}
fs/ceph/addr.c +1 -1
index a11b5c633358..b114ef483937 100644
@@ -2567,7 +2567,7 @@ int ceph_pool_perm_check(struct inode *inode, int need)
 	if (!S_ISREG(inode->i_mode))
 		return 0;
 
-	if (ci->i_vino.snap != CEPH_NOSNAP) {
+	if (ceph_in_snap(inode)) {
 		/*
 		 * Pool permission check needs to write to the first object.
 		 * But for snapshot, head of the first object may have already
fs/ceph/dir.c +10 -10
index 27ce9e55e947..c297cc57f88c 100644
@@ -727,7 +727,7 @@ struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
 	struct ceph_client *cl = ceph_inode_to_client(parent);
 
 	/* .snap dir? */
-	if (ceph_snap(parent) == CEPH_NOSNAP &&
+	if (!ceph_in_snap(parent) &&
 	    strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) {
 		struct dentry *res;
 		struct inode *inode = ceph_get_snapdir(parent);
@@ -919,7 +919,7 @@ static int ceph_mknod(struct mnt_idmap *idmap, struct inode *dir,
 	struct ceph_acl_sec_ctx as_ctx = {};
 	int err;
 
-	if (ceph_snap(dir) != CEPH_NOSNAP)
+	if (ceph_in_snap(dir))
 		return -EROFS;
 
 	err = ceph_wait_on_conflict_unlink(dentry);
@@ -1031,7 +1031,7 @@ static int ceph_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	umode_t mode = S_IFLNK | 0777;
 	int err;
 
-	if (ceph_snap(dir) != CEPH_NOSNAP)
+	if (ceph_in_snap(dir))
 		return -EROFS;
 
 	err = ceph_wait_on_conflict_unlink(dentry);
@@ -1115,7 +1115,7 @@ static struct dentry *ceph_mkdir(struct mnt_idmap *idmap, struct inode *dir,
 		op = CEPH_MDS_OP_MKSNAP;
 		doutc(cl, "mksnap %llx.%llx/'%pd' dentry %p\n",
 		      ceph_vinop(dir), dentry, dentry);
-	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
+	} else if (!ceph_in_snap(dir)) {
 		doutc(cl, "mkdir %llx.%llx/'%pd' dentry %p mode 0%ho\n",
 		      ceph_vinop(dir), dentry, dentry, mode);
 		op = CEPH_MDS_OP_MKDIR;
@@ -1202,7 +1202,7 @@ static int ceph_link(struct dentry *old_dentry, struct inode *dir,
 	if (err)
 		return err;
 
-	if (ceph_snap(dir) != CEPH_NOSNAP)
+	if (ceph_in_snap(dir))
 		return -EROFS;
 
 	err = fscrypt_prepare_link(old_dentry, dir, dentry);
@@ -1354,7 +1354,7 @@ static int ceph_unlink(struct inode *dir, struct dentry *dentry)
 		doutc(cl, "rmsnap %llx.%llx/'%pd' dn\n", ceph_vinop(dir),
 		      dentry);
 		op = CEPH_MDS_OP_RMSNAP;
-	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
+	} else if (!ceph_in_snap(dir)) {
 		doutc(cl, "unlink/rmdir %llx.%llx/'%pd' inode %llx.%llx\n",
 		      ceph_vinop(dir), dentry, ceph_vinop(inode));
 		op = d_is_dir(dentry) ?
@@ -1483,7 +1483,7 @@ static int ceph_rename(struct mnt_idmap *idmap, struct inode *old_dir,
 
 	if (ceph_snap(old_dir) != ceph_snap(new_dir))
 		return -EXDEV;
-	if (ceph_snap(old_dir) != CEPH_NOSNAP) {
+	if (ceph_in_snap(old_dir)) {
 		if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
 			op = CEPH_MDS_OP_RENAMESNAP;
 		else
@@ -1981,7 +1981,7 @@ static int ceph_d_revalidate(struct inode *dir, const struct qstr *name,
 	mdsc = ceph_sb_to_fs_client(dir->i_sb)->mdsc;
 
 	/* always trust cached snapped dentries, snapdir dentry */
-	if (ceph_snap(dir) != CEPH_NOSNAP) {
+	if (ceph_in_snap(dir)) {
 		doutc(cl, "%p '%pd' inode %p is SNAPPED\n", dentry,
 		      dentry, inode);
 		valid = 1;
@@ -2065,7 +2065,7 @@ static int ceph_d_delete(const struct dentry *dentry)
 	/* won't release caps */
 	if (d_really_is_negative(dentry))
 		return 0;
-	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
+	if (ceph_in_snap(d_inode(dentry)))
 		return 0;
 	/* valid lease? */
 	di = ceph_dentry(dentry);
@@ -2120,7 +2120,7 @@ static void ceph_d_prune(struct dentry *dentry)
 
 	/* we hold d_lock, so d_parent is stable */
 	dir_ci = ceph_inode(d_inode(dentry->d_parent));
-	if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
+	if (ceph_in_snap(&dir_ci->netfs.inode))
 		return;
 
 	/* who calls d_delete() should also disable dcache readdir */
fs/ceph/export.c +4 -4
index b2f2af104679..d132db40ba00 100644
@@ -99,7 +99,7 @@ static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
 	static const int connected_handle_length = CEPH_FH_WITH_PARENT_SIZE;
 	int type;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
 
 	if (parent_inode && (*max_len < connected_handle_length)) {
@@ -372,7 +372,7 @@ static struct dentry *ceph_get_parent(struct dentry *child)
 	struct ceph_client *cl = ceph_inode_to_client(inode);
 	struct dentry *dn;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP) {
+	if (ceph_in_snap(inode)) {
 		struct inode* dir;
 		bool unlinked = false;
 		/* do not support non-directory */
@@ -456,7 +456,7 @@ static int __get_snap_name(struct dentry *parent, char *name,
 	if (ceph_ino(inode) != ceph_ino(dir))
 		goto out;
 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
-		if (ceph_snap(dir) == CEPH_NOSNAP) {
+		if (!ceph_in_snap(dir)) {
 			/*
 			 * .get_name() from struct export_operations
 			 * assumes that its 'name' parameter is pointing
@@ -555,7 +555,7 @@ static int ceph_get_name(struct dentry *parent, char *name,
 	struct ceph_mds_reply_info_parsed *rinfo;
 	int err;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return __get_snap_name(parent, name, child);
 
 	mdsc = ceph_inode_to_fs_client(inode)->mdsc;
fs/ceph/file.c +7 -7
index d54d71669176..00e62991a416 100644
@@ -437,7 +437,7 @@ int ceph_open(struct inode *inode, struct file *file)
 	}
 
 	/* snapped files are read-only */
-	if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE))
+	if (ceph_in_snap(inode) && (file->f_mode & FMODE_WRITE))
 		return -EROFS;
 
 	/* trivially open snapdir */
@@ -469,7 +469,7 @@ int ceph_open(struct inode *inode, struct file *file)
 			ceph_check_caps(ci, 0);
 
 		return ceph_init_file(inode, file, fmode);
-	} else if (!do_sync && ceph_snap(inode) != CEPH_NOSNAP &&
+	} else if (!do_sync && ceph_in_snap(inode) &&
 		   (ci->i_snap_caps & wanted) == wanted) {
 		__ceph_touch_fmode(ci, mdsc, fmode);
 		spin_unlock(&ci->i_ceph_lock);
@@ -1534,7 +1534,7 @@ ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter,
 	bool should_dirty = !write && user_backed_iter(iter);
 	bool sparse = ceph_test_mount_opt(fsc, SPARSEREAD);
 
-	if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP)
+	if (write && ceph_in_snap(file_inode(file)))
 		return -EROFS;
 
 	doutc(cl, "sync_direct_%s on file %p %lld~%u snapc %p seq %lld\n",
@@ -1770,7 +1770,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
 	struct timespec64 mtime = current_time(inode);
 	size_t count = iov_iter_count(from);
 
-	if (ceph_snap(file_inode(file)) != CEPH_NOSNAP)
+	if (ceph_in_snap(file_inode(file)))
 		return -EROFS;
 
 	doutc(cl, "on file %p %lld~%u snapc %p seq %lld\n", file, pos,
@@ -2405,7 +2405,7 @@ static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	if (ceph_inode_is_shutdown(inode))
 		return -ESTALE;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return -EROFS;
 
 	prealloc_cf = ceph_alloc_cap_flush();
@@ -2757,7 +2757,7 @@ static long ceph_fallocate(struct file *file, int mode,
 
 	inode_lock(inode);
 
-	if (ceph_snap(inode) != CEPH_NOSNAP) {
+	if (ceph_in_snap(inode)) {
 		ret = -EROFS;
 		goto unlock;
 	}
@@ -3046,7 +3046,7 @@ static ssize_t __ceph_copy_file_range(struct file *src_file, loff_t src_off,
 			return -EXDEV;
 		}
 	}
-	if (ceph_snap(dst_inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(dst_inode))
 		return -EROFS;
 
 	/*
fs/ceph/inode.c +7 -7
index c0d88cf1f6e3..dcdbc24d1e21 100644
@@ -765,7 +765,7 @@ void ceph_evict_inode(struct inode *inode)
 	 * caps in i_snap_caps.
 	 */
 	if (ci->i_snap_realm) {
-		if (ceph_snap(inode) == CEPH_NOSNAP) {
+		if (!ceph_in_snap(inode)) {
 			doutc(cl, " dropping residual ref to snap realm %p\n",
 			      ci->i_snap_realm);
 			ceph_change_snap_realm(inode, NULL);
@@ -1065,7 +1065,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 	info_caps = le32_to_cpu(info->cap.caps);
 
 	/* prealloc new cap struct */
-	if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
+	if (info_caps && !ceph_in_snap(inode)) {
 		new_cap = ceph_get_cap(mdsc, caps_reservation);
 		if (!new_cap)
 			return -ENOMEM;
@@ -1087,7 +1087,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 		pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 						     iinfo->pool_ns_len);
 
-	if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
+	if (ceph_in_snap(inode) && !ci->i_snapid_map)
 		ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 
 	spin_lock(&ci->i_ceph_lock);
@@ -1332,7 +1332,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 
 	/* were we issued a capability? */
 	if (info_caps) {
-		if (ceph_snap(inode) == CEPH_NOSNAP) {
+		if (!ceph_in_snap(inode)) {
 			ceph_add_cap(inode, session,
 				     le64_to_cpu(info->cap.cap_id),
 				     info_caps,
@@ -1432,7 +1432,7 @@ static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
 	doutc(cl, "%p duration %lu ms ttl %lu\n", dentry, duration, ttl);
 
 	/* only track leases on regular dentries */
-	if (ceph_snap(dir) != CEPH_NOSNAP)
+	if (ceph_in_snap(dir))
 		return;
 
 	if (mask & CEPH_LEASE_PRIMARY_LINK)
@@ -2929,7 +2929,7 @@ int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
 	int err;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return -EROFS;
 
 	if (ceph_inode_is_shutdown(inode))
@@ -3184,7 +3184,7 @@ int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
 		valid_mask |= STATX_CHANGE_COOKIE;
 	}
 
-	if (ceph_snap(inode) == CEPH_NOSNAP)
+	if (!ceph_in_snap(inode))
 		stat->dev = sb->s_dev;
 	else
 		stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
fs/ceph/mds_client.c +5 -5
index ed17e0023705..1534a91342a7 100644
@@ -1308,7 +1308,7 @@ static struct inode *get_nonsnap_parent(struct dentry *dentry)
 
 	while (dentry && !IS_ROOT(dentry)) {
 		inode = d_inode_rcu(dentry);
-		if (!inode || ceph_snap(inode) == CEPH_NOSNAP)
+		if (!inode || !ceph_in_snap(inode))
 			break;
 		dentry = dentry->d_parent;
 	}
@@ -1382,7 +1382,7 @@ static int __choose_mds(struct ceph_mds_client *mdsc,
 			inode = d_inode(req->r_dentry);
 			if (inode)
 				ihold(inode);
-		} else if (ceph_snap(dir) != CEPH_NOSNAP) {
+		} else if (ceph_in_snap(dir)) {
 			/* direct snapped/virtual snapdir requests
 			 * based on parent dir inode */
 			inode = get_nonsnap_parent(parent);
@@ -2782,7 +2782,7 @@ char *ceph_mdsc_build_path(struct ceph_mds_client *mdsc, struct dentry *dentry,
 			spin_unlock(&cur->d_lock);
 			parent = dget_parent(cur);
 		} else if (for_wire && inode && dentry != cur &&
-			   ceph_snap(inode) == CEPH_NOSNAP) {
+			   !ceph_in_snap(inode)) {
 			spin_unlock(&cur->d_lock);
 			pos++; /* get rid of any prepended '/' */
 			break;
@@ -2891,7 +2891,7 @@ static int build_dentry_path(struct ceph_mds_client *mdsc, struct dentry *dentry
 	rcu_read_lock();
 	if (!dir)
 		dir = d_inode_rcu(dentry->d_parent);
-	if (dir && parent_locked && ceph_snap(dir) == CEPH_NOSNAP &&
+	if (dir && parent_locked && !ceph_in_snap(dir) &&
 	    !IS_ENCRYPTED(dir)) {
 		path_info->vino.ino = ceph_ino(dir);
 		path_info->vino.snap = ceph_snap(dir);
@@ -2917,7 +2917,7 @@ static int build_inode_path(struct inode *inode, struct ceph_path_info *path_inf
 	struct dentry *dentry;
 	char *path;
 
-	if (ceph_snap(inode) == CEPH_NOSNAP) {
+	if (!ceph_in_snap(inode)) {
 		path_info->vino.ino = ceph_ino(inode);
 		path_info->vino.snap = ceph_snap(inode);
 		path_info->pathlen = 0;
fs/ceph/quota.c +2 -2
index 053d5bf0c9f0..08641d578a0b 100644
@@ -223,7 +223,7 @@ static int get_quota_realm(struct ceph_mds_client *mdsc, struct inode *inode,
 
 	if (realmp)
 		*realmp = NULL;
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return 0;
 
 restart:
@@ -341,7 +341,7 @@ static bool check_quota_exceeded(struct inode *inode, enum quota_check_op op,
 	u64 max, rvalue;
 	bool exceeded = false;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return false;
 
 	down_read(&mdsc->snap_rwsem);
fs/ceph/super.h +8 -0
index b1d38fa6647f..abdbffbdf0e6 100644
@@ -601,6 +601,14 @@ static inline u64 ceph_snap(const struct inode *inode)
 	return ceph_inode(inode)->i_vino.snap;
 }
 
+/**
+ * Is this inode in a ".snap" directory?
+ */
+static inline bool ceph_in_snap(const struct inode *inode)
+{
+	return ceph_snap(inode) != CEPH_NOSNAP;
+}
+
 /**
  * ceph_present_ino - format an inode number for presentation to userland
  * @sb: superblock where the inode lives
fs/ceph/xattr.c +2 -2
index 9808eb10625d..c5e9ef656b5f 100644
@@ -270,7 +270,7 @@ static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
 	bool ret = false;
 	spin_lock(&ci->i_ceph_lock);
 	if ((ci->i_max_files || ci->i_max_bytes) &&
-	    ci->i_vino.snap == CEPH_NOSNAP &&
+	    !ceph_in_snap(&ci->netfs.inode) &&
 	    ci->i_snap_realm &&
 	    ci->i_snap_realm->ino == ci->i_vino.ino)
 		ret = true;
@@ -1193,7 +1193,7 @@ int __ceph_setxattr(struct inode *inode, const char *name,
 	bool check_realm = false;
 	bool lock_snap_rwsem = false;
 
-	if (ceph_snap(inode) != CEPH_NOSNAP)
+	if (ceph_in_snap(inode))
 		return -EROFS;
 
 	vxattr = ceph_match_vxattr(inode, name);
Signature · 1 line
2.47.3
#19 lore raw

PATCH 11/12 fs/ceph: use ceph_vino() etc. instead of accessing i_vino directly

In reply to Max Kellermann

Prepare to remove (optimize) i_vino.

---
 fs/ceph/addr.c       |  5 ++---
 fs/ceph/caps.c       |  4 ++--
 fs/ceph/file.c       | 14 +++++++-------
 fs/ceph/mds_client.c |  2 +-
 fs/ceph/quota.c      |  2 +-
 fs/ceph/snap.c       |  4 ++--
 fs/ceph/super.h      |  7 +++----
 fs/ceph/xattr.c      |  4 ++--
 8 files changed, 20 insertions(+), 22 deletions(-)
fs/ceph/addr.c +2 -3
index b114ef483937..d9a32897f4f0 100644
@@ -282,7 +282,6 @@ static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 	struct ceph_mds_reply_info_in *iinfo;
 	struct ceph_mds_request *req;
 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
-	struct ceph_inode_info *ci = ceph_inode(inode);
 	ssize_t err = 0;
 	size_t len;
 	int mode;
@@ -302,7 +301,7 @@ static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
 		err = PTR_ERR(req);
 		goto out;
 	}
-	req->r_ino1 = ci->i_vino;
+	req->r_ino1 = ceph_vino(inode);
 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA);
 	req->r_num_caps = 2;
 
@@ -2466,7 +2465,7 @@ static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
 	rd_req->r_base_oloc.pool = pool;
 	if (pool_ns)
 		rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
-	ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
+	ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ceph_ino(&ci->netfs.inode));
 
 	err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
 	if (err)
fs/ceph/caps.c +2 -2
index e50bcb393c8f..151baed53721 100644
@@ -730,7 +730,7 @@ void ceph_add_cap(struct inode *inode,
 			ceph_change_snap_realm(inode, realm);
 		else
 			WARN(1, "%s: couldn't find snap realm 0x%llx (ino 0x%llx oldrealm 0x%llx)\n",
-			     __func__, realmino, ci->i_vino.ino,
+			     __func__, realmino, ceph_ino(inode),
 			     ci->i_snap_realm ? ci->i_snap_realm->ino : 0);
 	}
 
@@ -1176,7 +1176,7 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
 	} else {
 		cap->queue_release = 0;
 	}
-	cap->cap_ino = ci->i_vino.ino;
+	cap->cap_ino = ceph_ino(inode);
 
 	spin_unlock(&session->s_cap_lock);
 
fs/ceph/file.c +7 -7
index 00e62991a416..2cc5d8db40ac 100644
@@ -1113,7 +1113,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
 		      read_off, read_len);
 
 		req = ceph_osdc_new_request(osdc, &ci->i_layout,
-					ci->i_vino, read_off, &read_len, 0, 1,
+					ceph_vino(inode), read_off, &read_len, 0, 1,
 					sparse ? CEPH_OSD_OP_SPARSE_READ :
 						 CEPH_OSD_OP_READ,
 					CEPH_OSD_FLAG_READ,
@@ -1819,7 +1819,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
 		rmw = first || last;
 
 		doutc(cl, "ino %llx %lld~%llu adjusted %lld~%llu -- %srmw\n",
-		      ci->i_vino.ino, pos, len, write_pos, write_len,
+		      ceph_ino(inode), pos, len, write_pos, write_len,
 		      rmw ? "" : "no ");
 
 		/*
@@ -1852,7 +1852,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
 			 * depending on how the request was aligned.
 			 */
 			req = ceph_osdc_new_request(osdc, &ci->i_layout,
-					ci->i_vino, first ? first_pos : last_pos,
+					ceph_vino(inode), first ? first_pos : last_pos,
 					&read_len, 0, (first && last) ? 2 : 1,
 					CEPH_OSD_OP_SPARSE_READ, CEPH_OSD_FLAG_READ,
 					NULL, ci->i_truncate_seq,
@@ -2050,7 +2050,7 @@ ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos,
 		}
 
 		req = ceph_osdc_new_request(osdc, &ci->i_layout,
-					    ci->i_vino, write_pos, &write_len,
+					    ceph_vino(inode), write_pos, &write_len,
 					    rmw ? 1 : 0, rmw ? 2 : 1,
 					    CEPH_OSD_OP_WRITE,
 					    CEPH_OSD_FLAG_WRITE,
@@ -2975,12 +2975,12 @@ static ssize_t ceph_do_objects_copy(struct ceph_inode_info *src_ci, u64 *src_off
 					      &dst_objoff, &dst_objlen);
 		ceph_oid_init(&src_oid);
 		ceph_oid_printf(&src_oid, "%llx.%08llx",
-				src_ci->i_vino.ino, src_objnum);
+				ceph_ino(&src_ci->netfs.inode), src_objnum);
 		ceph_oid_init(&dst_oid);
 		ceph_oid_printf(&dst_oid, "%llx.%08llx",
-				dst_ci->i_vino.ino, dst_objnum);
+				ceph_ino(&dst_ci->netfs.inode), dst_objnum);
 		/* Do an object remote copy */
-		req = ceph_alloc_copyfrom_request(osdc, src_ci->i_vino.snap,
+		req = ceph_alloc_copyfrom_request(osdc, ceph_snap(&src_ci->netfs.inode),
 						  &src_oid, &src_oloc,
 						  &dst_oid, &dst_oloc,
 						  dst_ci->i_truncate_seq,
fs/ceph/mds_client.c +1 -1
index 1534a91342a7..ebe2b79eafca 100644
@@ -1986,7 +1986,7 @@ static void remove_session_caps(struct ceph_mds_session *session)
 			if (cap == prev)
 				break;
 			prev = cap;
-			vino = cap->ci->i_vino;
+			vino = ceph_vino(&cap->ci->netfs.inode);
 			spin_unlock(&session->s_cap_lock);
 
 			inode = ceph_find_inode(sb, vino);
fs/ceph/quota.c +1 -1
index 08641d578a0b..02d7a3062f2d 100644
@@ -31,7 +31,7 @@ static inline bool ceph_has_realms_with_quotas(struct inode *inode)
 	if (root && ceph_ino(root) == CEPH_INO_ROOT)
 		return false;
 	/* MDS stray dirs have no quota realms */
-	if (ceph_vino_is_reserved(ceph_inode(inode)->i_vino))
+	if (ceph_vino_is_reserved(ceph_vino(inode)))
 		return false;
 	/* otherwise, we can't know for sure */
 	return true;
fs/ceph/snap.c +2 -2
index 6dbc7859ff49..44200eca9af4 100644
@@ -985,7 +985,7 @@ void ceph_change_snap_realm(struct inode *inode, struct ceph_snap_realm *realm)
 	if (oldrealm) {
 		spin_lock(&oldrealm->inodes_with_caps_lock);
 		list_del_init(&ci->i_snap_realm_item);
-		if (oldrealm->ino == ci->i_vino.ino)
+		if (oldrealm->ino == ceph_ino(inode))
 			oldrealm->inode = NULL;
 		spin_unlock(&oldrealm->inodes_with_caps_lock);
 		ceph_put_snap_realm(mdsc, oldrealm);
@@ -996,7 +996,7 @@ void ceph_change_snap_realm(struct inode *inode, struct ceph_snap_realm *realm)
 	if (realm) {
 		spin_lock(&realm->inodes_with_caps_lock);
 		list_add(&ci->i_snap_realm_item, &realm->inodes_with_caps);
-		if (realm->ino == ci->i_vino.ino)
+		if (realm->ino == ceph_ino(inode))
 			realm->inode = inode;
 		spin_unlock(&realm->inodes_with_caps_lock);
 	}
fs/ceph/super.h +3 -4
index abdbffbdf0e6..5e4addbfc0f2 100644
@@ -589,7 +589,7 @@ static inline ino_t ceph_vino_to_ino_t(const struct ceph_vino vino)
 }
 
 /* for printf-style formatting */
-#define ceph_vinop(i) ceph_inode(i)->i_vino.ino, ceph_inode(i)->i_vino.snap
+#define ceph_vinop(i) ceph_ino(i), ceph_snap(i)
 
 static inline u64 ceph_ino(const struct inode *inode)
 {
@@ -634,9 +634,8 @@ static inline u64 ceph_present_inode(struct inode *inode)
 static inline int ceph_ino_compare(struct inode *inode, void *data)
 {
 	struct ceph_vino *pvino = (struct ceph_vino *)data;
-	struct ceph_inode_info *ci = ceph_inode(inode);
-	return ci->i_vino.ino == pvino->ino &&
-		ci->i_vino.snap == pvino->snap;
+	return ceph_ino(inode) == pvino->ino &&
+		ceph_snap(inode) == pvino->snap;
 }
 
 /*
fs/ceph/xattr.c +2 -2
index c5e9ef656b5f..cffb9236d4e0 100644
@@ -272,7 +272,7 @@ static bool ceph_vxattrcb_quota_exists(struct ceph_inode_info *ci)
 	if ((ci->i_max_files || ci->i_max_bytes) &&
 	    !ceph_in_snap(&ci->netfs.inode) &&
 	    ci->i_snap_realm &&
-	    ci->i_snap_realm->ino == ci->i_vino.ino)
+	    ci->i_snap_realm->ino == ceph_ino(&ci->netfs.inode))
 		ret = true;
 	spin_unlock(&ci->i_ceph_lock);
 	return ret;
@@ -1328,7 +1328,7 @@ int __ceph_setxattr(struct inode *inode, const char *name,
 			spin_lock(&ci->i_ceph_lock);
 			if ((ci->i_max_files || ci->i_max_bytes) &&
 			    !(ci->i_snap_realm &&
-			      ci->i_snap_realm->ino == ci->i_vino.ino))
+			      ci->i_snap_realm->ino == ceph_ino(&ci->netfs.inode)))
 				err = -EOPNOTSUPP;
 			spin_unlock(&ci->i_ceph_lock);
 		}
Signature · 1 line
2.47.3
#20 lore raw

PATCH 12/12 fs/ceph: remove redundant inode number from ceph_inode_info

In reply to Max Kellermann

On 64 bit systems, ceph_vino_to_ino_t() is a no-op, and inode.i_ino is the same as ceph_vino.ino.

This reduces the size of struct ceph_inode_info by 8 bytes.

---
 fs/ceph/cache.c |  4 +++-
 fs/ceph/inode.c |  7 +++++++
 fs/ceph/super.h | 23 +++++++++++++++++++++++
 3 files changed, 33 insertions(+), 1 deletion(-)
fs/ceph/cache.c +3 -1
index f678bab189d8..289615e58196 100644
@@ -16,6 +16,7 @@ void ceph_fscache_register_inode_cookie(struct inode *inode)
 {
 	struct ceph_inode_info *ci = ceph_inode(inode);
 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
+	struct ceph_vino vino;
 
 	/* No caching for filesystem? */
 	if (!fsc->fscache)
@@ -31,9 +32,10 @@ void ceph_fscache_register_inode_cookie(struct inode *inode)
 
 	WARN_ON_ONCE(ci->netfs.cache);
 
+	vino = ceph_vino(inode);
 	ci->netfs.cache =
 		fscache_acquire_cookie(fsc->fscache, 0,
-				       &ci->i_vino, sizeof(ci->i_vino),
+				       &vino, sizeof(vino),
 				       &ci->i_version, sizeof(ci->i_version),
 				       i_size_read(inode));
 	if (ci->netfs.cache)
fs/ceph/inode.c +7 -0
index dcdbc24d1e21..c8ee7d735861 100644
@@ -44,11 +44,18 @@ static void ceph_inode_work(struct work_struct *work);
  */
 static int ceph_set_ino_cb(struct inode *inode, void *data)
 {
+	const struct ceph_vino *vino = data;
 	struct ceph_inode_info *ci = ceph_inode(inode);
 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 
+#if BITS_PER_LONG >= 64
+	inode->i_ino = vino->ino;
+	ci->i_snap = vino->snap;
+#else
 	ci->i_vino = *(struct ceph_vino *)data;
 	inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
+#endif
+
 	inode_set_iversion_raw(inode, 0);
 	percpu_counter_inc(&mdsc->metric.total_inodes);
 
fs/ceph/super.h +23 -0
index 5e4addbfc0f2..5cb535b126eb 100644
@@ -376,7 +376,15 @@ struct ceph_inode_xattrs_info {
  */
 struct ceph_inode_info {
 	struct netfs_inode netfs; /* Netfslib context and vfs inode */
+
+#if BITS_PER_LONG >= 64
+	/* on 64 bit systems, the full Ceph inode number is stored in
+	 * netfs.inode.i_ino
+	 */
+	u64 i_snap;
+#else
 	struct ceph_vino i_vino;   /* ceph ino + snap */
+#endif
 
 	spinlock_t i_ceph_lock;
 
@@ -563,7 +571,14 @@ ceph_inode_to_client(const struct inode *inode)
 static inline struct ceph_vino
 ceph_vino(const struct inode *inode)
 {
+#if BITS_PER_LONG >= 64
+	return (struct ceph_vino){
+		.ino = inode->i_ino,
+		.snap = ceph_inode(inode)->i_snap,
+	};
+#else
 	return ceph_inode(inode)->i_vino;
+#endif
 }
 
 static inline u32 ceph_ino_to_ino32(u64 vino)
@@ -593,12 +608,20 @@ static inline ino_t ceph_vino_to_ino_t(const struct ceph_vino vino)
 
 static inline u64 ceph_ino(const struct inode *inode)
 {
+#if BITS_PER_LONG >= 64
+	return inode->i_ino;
+#else
 	return ceph_inode(inode)->i_vino.ino;
+#endif
 }
 
 static inline u64 ceph_snap(const struct inode *inode)
 {
+#if BITS_PER_LONG >= 64
+	return ceph_inode(inode)->i_snap;
+#else
 	return ceph_inode(inode)->i_vino.snap;
+#endif
 }
 
 /**
Signature · 1 line
2.47.3
maintainer
#21 lore raw
In reply to Max Kellermann

Hi Max,

This series does not apply cleanly — patch 3 (convert pool_id to u32) fails on both ceph-client/testing and linux/master after Ilya's latest rebase. Patches 1 and 2 apply fine, but patch 3 can't build a fake ancestor for fs/ceph/inode.c because the blob SHA has diverged.

Could you rebase and resend?

--
Alex Markuze

#22 lore raw
Quoted 1 line · Alex Markuze

Could you rebase and resend?

Done, sent v3. (Note that you replied to v1, but I had already posted v2 a while ago.)

Signature · 8 lines
Max Kellermann
Principal Architect
Hosting Technology

cm4all | Im Mediapark 6a | 50670 Köln | Germany
General information about the company can be found here:
https://www.cm4all.com/impressum
A member of the IONOS Group