Improve support for exporting btrfs subvolumes.
Subsystems (2)
PATCH 1/1 Improve support for exporting btrfs subvolumes.
If you export two subvolumes of a btrfs filesystem, they will both be given the same uuid so lookups will be confused.
blkid cannot differentiate the two, so we must use the fsid from statfs64 to identify the filesystem.
We cannot tell if blkid or statfs is best without knowing internal details of the filesystem in question, so we need to encode specific knowledge of btrfs in mountd. This is unfortunate.
To ensure smooth handling of this and possible future changes in uuid generation, we add infrastructure for multiple different uuids to be recognised on old filehandles, but only the preferred on is used on new filehandles.
--
This is a substantially revised version of a patch I posted a while ago.
I tried to find a way to do it would hard coding knowledge of btrfs in nfs-utils, but it isn't possible. For some filesystems, f_fsid is best, for some it is worst. No way to tell the difference.
This patch add infrastructure so that if we find a better way to get a good uuid (e.g. a new syscall), we can slot it in for new filehandles, but old filehandles using the old uuid will still work.
I believe this is ready for inclusion upstream.
Thanks,
NeilBrown
utils/mountd/cache.c +78 -29
@@ -170,13 +170,16 @@ void auth_unix_gid(FILE *f)#if USE_BLKID static const char *get_uuid_blkdev (char *path) { +/* We set *safe if we know that we need the +* fsid from statfs too. +*/ static blkid_cache cache =NULL ;struct stat stb;char *devname;blkid_tag_iterate iter;blkid_dev dev;const char *type; -const char *val =NULL ; +const char *val, *uuid =NULL ;if (cache ==NULL )blkid_get_cache (&cache,NULL );
@@ -193,42 +196,29 @@ static const char *get_uuid_blkdev(char *path) iter =blkid_tag_iterate_begin (dev);if (!iter)return NULL ; -while (blkid_tag_next (iter, &type, &val) ==0 ) +while (blkid_tag_next (iter, &type, &val) ==0 ) {if (strcmp (type,"UUID" ) ==0 ) + uuid = val; +if (strcmp (type,"TYPE" ) ==0 && +strcmp (val,"btrfs" ) ==0 ) { + uuid =NULL ;break ; + } + }blkid_tag_iterate_end (iter); -return val; + return uuid; }#else #define get_uuid_blkdev (path) (NULL)#endif -int get_uuid (char *path,char *uuid,int uuidlen,char *u) +int get_uuid (const char *val,int uuidlen,char *u) {/* extract hex digits from uuidstr and compose a uuid * of the given length (max 16), xoring bytes to make -* a smaller uuid. Then compare with uuid +* a smaller uuid. */ int i =0 ; -const char *val =NULL ; -char fsid_val[17 ]; - -if (path) { - val =get_uuid_blkdev (path); -if (!val) { -struct statfs64 st; - -if (statfs64 (path, &st)) -return 0 ; -if (!st.f_fsid.__val[0 ] && !st.f_fsid.__val[1 ]) -return 0 ; -snprintf (fsid_val,17 ,"%08x%08x" , - st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); - val = fsid_val; - } - }else { - val = uuid; - }memset (u,0 , uuidlen);for ( ; *val ; val++) {
@@ -252,6 +242,60 @@ int get_uuid(char *path, char *uuid, int uuidlen, char *u)return 1 ; } +int uuid_by_path (char *path,int type,int uuidlen,char *uuid) +{ +/* get a uuid for the filesystem found at 'path'. +* There are several possible ways of generating the +* uuids (types). +* Type 0 is used for new filehandles, while other types +* may be used to interpret old filehandle - to ensure smooth +* forward migration. +* We return 1 if a uuid was found (and it might be worth +* trying the next type) or 0 if no more uuid types can be +* extracted. +*/ + +/* Possible sources of uuid are +* - blkid uuid +* - statfs64 uuid +* +* On some filesystems (e.g. vfat) the statfs64 uuid is simply an +* encoding of the device that the filesystem is mounted from, so +* it we be very bad to use that (as device numbers change). blkid +* must be preferred. +* On other filesystems (e.g. btrfs) the statfs64 uuid contains +* important info that the blkid uuid cannot contain: This happens +* when multiple subvolumes are exported (they have the same +* blkid uuid but different statfs64 uuids). +* We rely on get_uuid_blkdev *knowing* which is which and not returning +* a uuid for filesystems where the statfs64 uuid is better. +* +*/ +struct statfs64 st; +char fsid_val[17 ]; +const char *blkid_val; +const char *val; + + blkid_val =get_uuid_blkdev (path); + +if (statfs64 (path, &st) ==0 && + (st.f_fsid.__val[0 ] || st.f_fsid.__val[1 ])) +snprintf (fsid_val,17 ,"%08x%08x" , + st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); +else + fsid_val[0 ] =0 ; + +if (blkid_val && (type--) ==0 ) + val = blkid_val; +else if (fsid_val[0 ] && (type--) ==0 ) + val = fsid_val; +else +return 0 ; + +get_uuid (val, uuidlen, uuid); +return 1 ; +} +/* Iterate through /etc/mtab, finding mountpoints * at or below a given path */
@@ -398,6 +442,7 @@ void nfsd_fh(FILE *f)struct stat stb;char u[16 ];char *path; +int type;if (exp->m_export.e_flags &NFSEXP_CROSSMOUNT ) {static nfs_export *prev =NULL ;
@@ -461,10 +506,14 @@ void nfsd_fh(FILE *f)continue ; check_uuid:if (exp->m_export.e_uuid) -get_uuid (NULL , exp->m_export.e_uuid, +get_uuid (exp->m_export.e_uuid, uuidlen, u); -else if (get_uuid (path,NULL , uuidlen, u) ==0 ) -continue ; +else +for (type =0 ; +uuid_by_path (path, type, uuidlen, u); + type++) +if (memcmp (u, fhuuid, uuidlen) !=0 ) +break ;if (memcmp (u, fhuuid, uuidlen) !=0 )continue ;
@@ -600,13 +649,13 @@ static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exwrite_secinfo (f, exp, flag_mask);if (exp->e_uuid ==NULL || different_fs) {char u[16 ]; -if (get_uuid (path,NULL ,16 , u)) { +if (uuid_by_path (path,0 ,16 , u)) {qword_print (f,"uuid" );qword_printhex (f, u,16 ); } }else {char u[16 ]; -get_uuid (NULL , exp->e_uuid,16 , u); +get_uuid (exp->e_uuid,16 , u);qword_print (f,"uuid" );qword_printhex (f, u,16 ); }
A message that is not in the archive (sent off-list or not yet received): 19481.43625.476833.275104-wvvUuzkyo1EYVZTmpyfIwg@public.gmane.org
Quoted 24 lines · Neil Brown
If you export two subvolumes of a btrfs filesystem, they will both be given the same uuid so lookups will be confused.
blkid cannot differentiate the two, so we must use the fsid from statfs64 to identify the filesystem.We cannot tell if blkid or statfs is best without knowing internal details of the filesystem in question, so we need to encode specific knowledge of btrfs in mountd. This is unfortunate.
To ensure smooth handling of this and possible future changes in uuid generation, we add infrastructure for multiple different uuids to be recognised on old filehandles, but only the preferred on is used on new filehandles.
Signed-off-by NeilBrown <[email protected]>--
This is a substantially revised version of a patch I posted a while ago.
I tried to find a way to do it would hard coding knowledge of btrfs in nfs-utils, but it isn't possible. For some filesystems,f_fsidis best, for some it is worst. No way to tell the difference.
Quoted 5 lines
This patch add infrastructure so that if we find a better way to get a good uuid (e.g. a new syscall), we can slot it in for new filehandles, but old filehandles using the old uuid will still work.
I believe this is ready for inclusion upstream.
Committed...
steved.
Quoted 6 lines · Neil Brown
If you export two subvolumes of a btrfs filesystem, they will both be given the same uuid so lookups will be confused.
blkid cannot differentiate the two, so we must use the fsid from statfs64 to identify the filesystem.
Quoted 8 lines
We cannot tell if blkid or statfs is best without knowing internal details of the filesystem in question, so we need to encode specific knowledge of btrfs in mountd. This is unfortunate.
To ensure smooth handling of this and possible future changes in uuid generation, we add infrastructure for multiple different uuids to be recognised on old filehandles, but only the preferred on is used on new filehandles.
Could you just contatenate the two (or hash them somehow)? Or does that just use up too much space in the filehandle?
--b.
Quoted 201 lines
Signed-off-by NeilBrown <[email protected]>--
This is a substantially revised version of a patch I posted a while ago.
I tried to find a way to do it would hard coding knowledge of btrfs in nfs-utils, but it isn't possible. For some filesystems,f_fsidis best, for some it is worst. No way to tell the difference.This patch add infrastructure so that if we find a better way to get a good uuid (e.g. a new syscall), we can slot it in for new filehandles, but old filehandles using the old uuid will still work.
I believe this is ready for inclusion upstream.
Thanks,
NeilBrownutils/mountd/cache.c +78
-30@@ -170,13 +170,16 @@ void auth_unix_gid(FILE *f)#if USE_BLKID static const char *get_uuid_blkdev (char *path) { +/* We set *safe if we know that we need the +* fsid from statfs too. +*/ static blkid_cache cache =NULL ;struct stat stb;char *devname;blkid_tag_iterate iter;blkid_dev dev;const char *type; -const char *val =NULL ; +const char *val, *uuid =NULL ;if (cache ==NULL )blkid_get_cache (&cache,NULL );@@ -193,42 +196,29 @@ static const char *get_uuid_blkdev(char *path) iter =blkid_tag_iterate_begin (dev);if (!iter)return NULL ; -while (blkid_tag_next (iter, &type, &val) ==0 ) +while (blkid_tag_next (iter, &type, &val) ==0 ) {if (strcmp (type,"UUID" ) ==0 ) + uuid = val; +if (strcmp (type,"TYPE" ) ==0 && +strcmp (val,"btrfs" ) ==0 ) { + uuid =NULL ;break ; + } + }blkid_tag_iterate_end (iter); -return val; + return uuid; }#else #define get_uuid_blkdev (path) (NULL)#endif -int get_uuid (char *path,char *uuid,int uuidlen,char *u) +int get_uuid (const char *val,int uuidlen,char *u) {/* extract hex digits from uuidstr and compose a uuid * of the given length (max 16), xoring bytes to make -* a smaller uuid. Then compare with uuid +* a smaller uuid. */ int i =0 ; -const char *val =NULL ; -char fsid_val[17 ]; - -if (path) { - val =get_uuid_blkdev (path); -if (!val) { -struct statfs64 st; - -if (statfs64 (path, &st)) -return 0 ; -if (!st.f_fsid.__val[0 ] && !st.f_fsid.__val[1 ]) -return 0 ; -snprintf (fsid_val,17 ,"%08x%08x" , - st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); - val = fsid_val; - } - }else { - val = uuid; - }memset (u,0 , uuidlen);for ( ; *val ; val++) {@@ -252,6 +242,60 @@ int get_uuid(char *path, char *uuid, int uuidlen, char *u)return 1 ; } +int uuid_by_path (char *path,int type,int uuidlen,char *uuid) +{ +/* get a uuid for the filesystem found at 'path'. +* There are several possible ways of generating the +* uuids (types). +* Type 0 is used for new filehandles, while other types +* may be used to interpret old filehandle - to ensure smooth +* forward migration. +* We return 1 if a uuid was found (and it might be worth +* trying the next type) or 0 if no more uuid types can be +* extracted. +*/ + +/* Possible sources of uuid are +* - blkid uuid +* - statfs64 uuid +* +* On some filesystems (e.g. vfat) the statfs64 uuid is simply an +* encoding of the device that the filesystem is mounted from, so +* it we be very bad to use that (as device numbers change). blkid +* must be preferred. +* On other filesystems (e.g. btrfs) the statfs64 uuid contains +* important info that the blkid uuid cannot contain: This happens +* when multiple subvolumes are exported (they have the same +* blkid uuid but different statfs64 uuids). +* We rely on get_uuid_blkdev *knowing* which is which and not returning +* a uuid for filesystems where the statfs64 uuid is better. +* +*/ +struct statfs64 st; +char fsid_val[17 ]; +const char *blkid_val; +const char *val; + + blkid_val =get_uuid_blkdev (path); + +if (statfs64 (path, &st) ==0 && + (st.f_fsid.__val[0 ] || st.f_fsid.__val[1 ])) +snprintf (fsid_val,17 ,"%08x%08x" , + st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); +else + fsid_val[0 ] =0 ; + +if (blkid_val && (type--) ==0 ) + val = blkid_val; +else if (fsid_val[0 ] && (type--) ==0 ) + val = fsid_val; +else +return 0 ; + +get_uuid (val, uuidlen, uuid); +return 1 ; +} +/* Iterate through /etc/mtab, finding mountpoints * at or below a given path */ @@ -398,6 +442,7 @@ void nfsd_fh(FILE *f)struct stat stb;char u[16 ];char *path; +int type;if (exp->m_export.e_flags &NFSEXP_CROSSMOUNT ) {static nfs_export *prev =NULL ;@@ -461,10 +506,14 @@ void nfsd_fh(FILE *f)continue ; check_uuid:if (exp->m_export.e_uuid) -get_uuid (NULL , exp->m_export.e_uuid, +get_uuid (exp->m_export.e_uuid, uuidlen, u); -else if (get_uuid (path,NULL , uuidlen, u) ==0 ) -continue ; +else +for (type =0 ; +uuid_by_path (path, type, uuidlen, u); + type++) +if (memcmp (u, fhuuid, uuidlen) !=0 ) +break ;if (memcmp (u, fhuuid, uuidlen) !=0 )continue ;@@ -600,13 +649,13 @@ static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exwrite_secinfo (f, exp, flag_mask);if (exp->e_uuid ==NULL || different_fs) {char u[16 ]; -if (get_uuid (path,NULL ,16 , u)) { +if (uuid_by_path (path,0 ,16 , u)) {qword_print (f,"uuid" );qword_printhex (f, u,16 ); } }else {char u[16 ]; -get_uuid (NULL , exp->e_uuid,16 , u); +get_uuid (exp->e_uuid,16 , u);qword_print (f,"uuid" );qword_printhex (f, u,16 ); } --To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 11 lines · On Wed, 23 Jun 2010 14:28:38 -0400 "J. Bruce Fields" <[email protected]>…
Quoted 10 lines · Neil Brown
If you export two subvolumes of a btrfs filesystem, they will both be given the same uuid so lookups will be confused.
blkid cannot differentiate the two, so we must use the fsid from statfs64 to identify the filesystem.We cannot tell if blkid or statfs is best without knowing internal details of the filesystem in question, so we need to encode specific knowledge of btrfs in mountd. This is unfortunate.
Quoted 7 lines
Quoted 4 lines
To ensure smooth handling of this and possible future changes in uuid generation, we add infrastructure for multiple different uuids to be recognised on old filehandles, but only the preferred on is used on new filehandles.
Could you just contatenate the two (or hash them somehow)? Or does that just use up too much space in the filehandle?
I did consider xoring them together but came to the conclusion that would actually be a regression.
If you look down at the comment that I included in uuid_by_path, you will see that some filesystems (e.g. VFAT) just use the major/minor device number for the f_fsid from statfs. As you know that is not necessarily stable over reboots, while the UUID that blkid gives is.
So if we always adding the two uuids somehow, it would be an improvement for btrfs, no change for e.g. ext3/XFS, and a regression for VFAT (and others I think). Not good.
Thanks,
NeilBrown
Quoted 204 lines
--b.
Quoted 201 lines
Signed-off-by NeilBrown <[email protected]>--
This is a substantially revised version of a patch I posted a while ago.
I tried to find a way to do it would hard coding knowledge of btrfs in nfs-utils, but it isn't possible. For some filesystems,f_fsidis best, for some it is worst. No way to tell the difference.This patch add infrastructure so that if we find a better way to get a good uuid (e.g. a new syscall), we can slot it in for new filehandles, but old filehandles using the old uuid will still work.
I believe this is ready for inclusion upstream.
Thanks,
NeilBrownutils/mountd/cache.c +78
-30@@ -170,13 +170,16 @@ void auth_unix_gid(FILE *f)#if USE_BLKID static const char *get_uuid_blkdev (char *path) { +/* We set *safe if we know that we need the +* fsid from statfs too. +*/ static blkid_cache cache =NULL ;struct stat stb;char *devname;blkid_tag_iterate iter;blkid_dev dev;const char *type; -const char *val =NULL ; +const char *val, *uuid =NULL ;if (cache ==NULL )blkid_get_cache (&cache,NULL );@@ -193,42 +196,29 @@ static const char *get_uuid_blkdev(char *path) iter =blkid_tag_iterate_begin (dev);if (!iter)return NULL ; -while (blkid_tag_next (iter, &type, &val) ==0 ) +while (blkid_tag_next (iter, &type, &val) ==0 ) {if (strcmp (type,"UUID" ) ==0 ) + uuid = val; +if (strcmp (type,"TYPE" ) ==0 && +strcmp (val,"btrfs" ) ==0 ) { + uuid =NULL ;break ; + } + }blkid_tag_iterate_end (iter); -return val; + return uuid; }#else #define get_uuid_blkdev (path) (NULL)#endif -int get_uuid (char *path,char *uuid,int uuidlen,char *u) +int get_uuid (const char *val,int uuidlen,char *u) {/* extract hex digits from uuidstr and compose a uuid * of the given length (max 16), xoring bytes to make -* a smaller uuid. Then compare with uuid +* a smaller uuid. */ int i =0 ; -const char *val =NULL ; -char fsid_val[17 ]; - -if (path) { - val =get_uuid_blkdev (path); -if (!val) { -struct statfs64 st; - -if (statfs64 (path, &st)) -return 0 ; -if (!st.f_fsid.__val[0 ] && !st.f_fsid.__val[1 ]) -return 0 ; -snprintf (fsid_val,17 ,"%08x%08x" , - st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); - val = fsid_val; - } - }else { - val = uuid; - }memset (u,0 , uuidlen);for ( ; *val ; val++) {@@ -252,6 +242,60 @@ int get_uuid(char *path, char *uuid, int uuidlen, char *u)return 1 ; } +int uuid_by_path (char *path,int type,int uuidlen,char *uuid) +{ +/* get a uuid for the filesystem found at 'path'. +* There are several possible ways of generating the +* uuids (types). +* Type 0 is used for new filehandles, while other types +* may be used to interpret old filehandle - to ensure smooth +* forward migration. +* We return 1 if a uuid was found (and it might be worth +* trying the next type) or 0 if no more uuid types can be +* extracted. +*/ + +/* Possible sources of uuid are +* - blkid uuid +* - statfs64 uuid +* +* On some filesystems (e.g. vfat) the statfs64 uuid is simply an +* encoding of the device that the filesystem is mounted from, so +* it we be very bad to use that (as device numbers change). blkid +* must be preferred. +* On other filesystems (e.g. btrfs) the statfs64 uuid contains +* important info that the blkid uuid cannot contain: This happens +* when multiple subvolumes are exported (they have the same +* blkid uuid but different statfs64 uuids). +* We rely on get_uuid_blkdev *knowing* which is which and not returning +* a uuid for filesystems where the statfs64 uuid is better. +* +*/ +struct statfs64 st; +char fsid_val[17 ]; +const char *blkid_val; +const char *val; + + blkid_val =get_uuid_blkdev (path); + +if (statfs64 (path, &st) ==0 && + (st.f_fsid.__val[0 ] || st.f_fsid.__val[1 ])) +snprintf (fsid_val,17 ,"%08x%08x" , + st.f_fsid.__val[0 ], st.f_fsid.__val[1 ]); +else + fsid_val[0 ] =0 ; + +if (blkid_val && (type--) ==0 ) + val = blkid_val; +else if (fsid_val[0 ] && (type--) ==0 ) + val = fsid_val; +else +return 0 ; + +get_uuid (val, uuidlen, uuid); +return 1 ; +} +/* Iterate through /etc/mtab, finding mountpoints * at or below a given path */ @@ -398,6 +442,7 @@ void nfsd_fh(FILE *f)struct stat stb;char u[16 ];char *path; +int type;if (exp->m_export.e_flags &NFSEXP_CROSSMOUNT ) {static nfs_export *prev =NULL ;@@ -461,10 +506,14 @@ void nfsd_fh(FILE *f)continue ; check_uuid:if (exp->m_export.e_uuid) -get_uuid (NULL , exp->m_export.e_uuid, +get_uuid (exp->m_export.e_uuid, uuidlen, u); -else if (get_uuid (path,NULL , uuidlen, u) ==0 ) -continue ; +else +for (type =0 ; +uuid_by_path (path, type, uuidlen, u); + type++) +if (memcmp (u, fhuuid, uuidlen) !=0 ) +break ;if (memcmp (u, fhuuid, uuidlen) !=0 )continue ;@@ -600,13 +649,13 @@ static int dump_to_cache(FILE *f, char *domain, char *path, struct exportent *exwrite_secinfo (f, exp, flag_mask);if (exp->e_uuid ==NULL || different_fs) {char u[16 ]; -if (get_uuid (path,NULL ,16 , u)) { +if (uuid_by_path (path,0 ,16 , u)) {qword_print (f,"uuid" );qword_printhex (f, u,16 ); } }else {char u[16 ]; -get_uuid (NULL , exp->e_uuid,16 , u); +get_uuid (exp->e_uuid,16 , u);qword_print (f,"uuid" );qword_printhex (f, u,16 ); } --To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
A message that is not in the archive (sent off-list or not yet received): [email protected]
Quoted 22 lines · Neil Brown
Quoted 11 lines · On Wed, 23 Jun 2010 14:28:38 -0400 "J. Bruce Fields" <[email protected]>…
Quoted 10 lines · Neil Brown
If you export two subvolumes of a btrfs filesystem, they will both be given the same uuid so lookups will be confused.
blkid cannot differentiate the two, so we must use the fsid from statfs64 to identify the filesystem.We cannot tell if blkid or statfs is best without knowing internal details of the filesystem in question, so we need to encode specific knowledge of btrfs in mountd. This is unfortunate.
Quoted 7 lines
Quoted 4 lines
To ensure smooth handling of this and possible future changes in uuid generation, we add infrastructure for multiple different uuids to be recognised on old filehandles, but only the preferred on is used on new filehandles.
Could you just contatenate the two (or hash them somehow)? Or does that just use up too much space in the filehandle?
Quoted 9 lines
I did consider xoring them together but came to the conclusion that would actually be a regression.
If you look down at the comment that I included inuuid_by_path, you will see that some filesystems (e.g. VFAT) just use the major/minor device number for thef_fsidfrom statfs. As you know that is not necessarily stable over reboots, while the UUID that blkid gives is.
So if we always adding the two uuids somehow, it would be an improvement for btrfs, no change for e.g. ext3/XFS, and a regression for VFAT (and others I think). Not good.
OK, got it.
--b.