On Sunday 06 December 2015 22:04:05 Deepa Dinamani wrote:
The current representation of inode times in struct inode, struct iattr, struct kstat: struct timespec are not y2038 safe.
Add provision to convert them to use 64 bit times in the future.
struct inode uses struct inode time to maintain same size for times across 32 bit and 64 bit architectures. structs iattr and kstat have no such requirement since these data types are only used to pass values in and out of vfs layer.
Ok, ignore my comment on patch 1, I should have read all the way until here ;-)
I think you can safely merge patch 1 into patch 3, it will actually be clearer that way.
In addition, inode_time is defined as packed and aligned to a 4 byte boundary to make the structure use 12 bytes each instead of 16 bytes. This will help save RAM space as inode structure is cached in memory. The other structures are transient and such a change would not be beneficial.
Signed-off-by: Deepa Dinamani deepa.kernel@gmail.com
Ah, this also explains the use of inode_time: it's much nicer if we don't need it in the file systems but only in struct inode.
@@ -616,9 +622,15 @@ struct inode { }; dev_t i_rdev; loff_t i_size; +#ifdef CONFIG_FS_USES_64BIT_TIME
- struct inode_time i_atime;
- struct inode_time i_mtime;
- struct inode_time i_ctime;
+#else struct timespec i_atime; struct timespec i_mtime; struct timespec i_ctime; +#endif spinlock_t i_lock; /* i_blocks, i_bytes, maybe i_size */ unsigned short i_bytes; unsigned int i_blkbits; @@ -679,10 +691,17 @@ struct inode { void *i_private; /* fs or device private pointer */ }; +#ifdef CONFIG_FS_USES_64BIT_TIME +#define FS_INODE_SET_XTIME(xtime, inode, ts64) \
- ((inode)->i_##xtime = timespec64_to_inode_time(ts64))
+#define FS_INODE_GET_XTIME(xtime, inode) \
- (inode_time_to_timespec64((inode)->i_##xtime))
+#else #define FS_INODE_SET_XTIME(xtime, inode, ts) \ ((inode)->i_##xtime = (ts)) #define FS_INODE_GET_XTIME(xtime, inode) \ ((inode)->i_##xtime) +#endif
So this makes a lot of sense now. If it's the only use of timespec64_to_inode_time/inode_time_to_timespec64, you could also open-code that in the macro and avoid introducing the macro to start with.
We have had some discussion about limiting the range of the times in the inode to the range allowed by the file system at some point, so leaving them as inline functions will make it easier to extend them later that way.
static inline int inode_unhashed(struct inode *inode) { diff --git a/include/linux/stat.h b/include/linux/stat.h index 075cb0c..e3443a9 100644 --- a/include/linux/stat.h +++ b/include/linux/stat.h @@ -27,9 +27,15 @@ struct kstat { kgid_t gid; dev_t rdev; loff_t size;
- struct timespec atime;
- struct timespec mtime;
- struct timespec ctime;
+#ifdef CONFIG_FS_USES_64BIT_TIME
- struct timespec64 atime;
- struct timespec64 mtime;
- struct timespec64 ctime;
+#else
- struct timespec atime;
- struct timespec mtime;
- struct timespec ctime;
+#endif unsigned long blksize; unsigned long long blocks; };
Arnd