diff options
Diffstat (limited to 'rsync.h')
| -rw-r--r-- | rsync.h | 40 |
1 files changed, 31 insertions, 9 deletions
@@ -272,12 +272,30 @@ enum logcode {FNONE=0, FERROR=1, FINFO=2, FLOG=3 }; #define NO_INT64 #endif -/* We want to manipulate 64-bit inums. On some systems - * STRUCT_STAT.st_ino can be bigger than an ino_t depending on the - * combination of largefile feature macros. Rather than try to guess, - * we just internally store them in the largest know type. Hopefully - * it's enough. */ -#define INO_T int64 +/* Starting from protocol version 26, we always use 64-bit + * ino_t and dev_t internally, even if this platform does not + * allow files to have 64-bit inums. That's because the + * receiver needs to find duplicate (dev,ino) tuples to detect + * hardlinks, and it might have files coming from a platform + * that has 64-bit inums. + * + * The only exception is if we're on a platform with no 64-bit type at + * all. + * + * Because we use read_longint() to get these off the wire, if you + * transfer devices or hardlinks with dev or inum > 2**32 to a machine + * with no 64-bit types then you will get an overflow error. Probably + * not many people have that combination of machines, and you can + * avoid it by not preserving hardlinks or not transferring device + * nodes. It's not clear that any other behaviour is better. + * + * Note that if you transfer devices from a 64-bit-devt machine (say, + * Solaris) to a 32-bit-devt machine (say, Linux-2.2/x86) then the + * device numbers will be truncated. But it's a kind of silly thing + * to do anyhow. + */ +#define INO64_T int64 +#define DEV64_T int64 #ifndef MIN #define MIN(a,b) ((a)<(b)?(a):(b)) @@ -308,9 +326,13 @@ struct file_struct { time_t modtime; OFF_T length; mode_t mode; - INO_T inode; - dev_t dev; - dev_t rdev; + + INO64_T inode; + /** Device this file lives upon */ + DEV64_T dev; + + /** If this is a device node, the device number. */ + DEV64_T rdev; uid_t uid; gid_t gid; char *basename; |