get_seconds() returns an unsigned long can overflow on some architectures
and is deprecated because of that. In cachefs, we cast that number to
a a 32-bit integer, which will overflow in year 2106 on all architectures.
As confirmed by David Howells, the overflow probably isn't harmful
in the end, since the timestamps are only used to make the file names
unique, but they don't strictly have to be in monotonically increasing
order since the files only exist in order to be deleted as quickly
as possible.
Moving to ktime_get_real_seconds() avoids the deprecated interface.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
v2: only change to ktime_get_real_seconds() without printing the
extra leading digits
---
fs/cachefiles/namei.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c
index ab0bbe93b398..0a38978b8cd3 100644
--- a/fs/cachefiles/namei.c
+++ b/fs/cachefiles/namei.c
@@ -337,7 +337,7 @@ static int cachefiles_bury_object(struct cachefiles_cache *cache,
try_again:
/* first step is to make up a grave dentry in the graveyard */
sprintf(nbuffer, "%08x%08x",
- (uint32_t) get_seconds(),
+ (uint32_t) ktime_get_real_seconds(),
(uint32_t) atomic_inc_return(&cache->gravecounter));
/* do the multiway lock magic */
--
2.9.0
get_seconds() returns an unsigned long can overflow on some architectures
and is deprecated because of that. In cachefs, we cast that number to
a a 32-bit integer, which will overflow in year 2106 on all architectures.
The overflow probably isn't harmful in the end, since the timestamps are
only used to make the file names unique, but they don't strictly have to
be in monotonically increasing order. Moving to ktime_get_real_seconds()
avoids the deprecated interface, the question is whether we should still
truncate to 32 bits.
In this patch, I decided to not overflow, but instead to extend the
file names using the 64-bit timestamp, so they will be 17 characters
after 2106 rather than 16.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/cachefiles/namei.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c
index ab0bbe93b398..5705e29d4506 100644
--- a/fs/cachefiles/namei.c
+++ b/fs/cachefiles/namei.c
@@ -297,7 +297,7 @@ static int cachefiles_bury_object(struct cachefiles_cache *cache,
{
struct dentry *grave, *trap;
struct path path, path_to_graveyard;
- char nbuffer[8 + 8 + 1];
+ char nbuffer[16 + 8 + 1];
int ret;
_enter(",'%pd','%pd'", dir, rep);
@@ -336,8 +336,8 @@ static int cachefiles_bury_object(struct cachefiles_cache *cache,
try_again:
/* first step is to make up a grave dentry in the graveyard */
- sprintf(nbuffer, "%08x%08x",
- (uint32_t) get_seconds(),
+ sprintf(nbuffer, "%08llx%08x",
+ (uint64_t) ktime_get_real_seconds(),
(uint32_t) atomic_inc_return(&cache->gravecounter));
/* do the multiway lock magic */
--
2.9.0
afs uses 32-bit timestamps everywhere, but mixes signed and unsigned
usage, which is a bit inconsistent. In particular on 32-bit machines,
it currently uses unsigned timestamps (ranging from 1970 to 2106) for
locally modified files, but signed timestamps (rand 1902 to 2038) when
reading from a remote end. On 64-bit machines, we always interpret
timestamps as unsigned here.
This replaces the deprecated time_t and get_seconds() interfaces with the
modern time64_t and current_time() to locally store 64-bit timestamps,
taking care to use unsigned interpretation of the raw values everywhere,
which avoids the y2038 overflow and is consistent with the previous
usage on 64-bit machines.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/afs/afs.h | 6 +++---
fs/afs/inode.c | 4 +---
2 files changed, 4 insertions(+), 6 deletions(-)
diff --git a/fs/afs/afs.h b/fs/afs/afs.h
index b4ff1f7ae4ab..6ca50c293553 100644
--- a/fs/afs/afs.h
+++ b/fs/afs/afs.h
@@ -129,8 +129,8 @@ typedef u32 afs_access_t;
struct afs_file_status {
u64 size; /* file size */
afs_dataversion_t data_version; /* current data version */
- time_t mtime_client; /* last time client changed data */
- time_t mtime_server; /* last time server changed data */
+ time64_t mtime_client; /* last time client changed data */
+ time64_t mtime_server; /* last time server changed data */
unsigned abort_code; /* Abort if bulk-fetching this failed */
afs_file_type_t type; /* file type */
@@ -158,7 +158,7 @@ struct afs_file_status {
* AFS volume synchronisation information
*/
struct afs_volsync {
- time_t creation; /* volume creation time */
+ time64_t creation; /* volume creation time */
};
/*
diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index 479b7fdda124..0507e52e3330 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -213,9 +213,7 @@ struct inode *afs_iget_pseudo_dir(struct super_block *sb, bool root)
set_nlink(inode, 2);
inode->i_uid = GLOBAL_ROOT_UID;
inode->i_gid = GLOBAL_ROOT_GID;
- inode->i_ctime.tv_sec = get_seconds();
- inode->i_ctime.tv_nsec = 0;
- inode->i_atime = inode->i_mtime = inode->i_ctime;
+ inode->i_ctime = inode->i_atime = inode->i_mtime = current_time(inode);
inode->i_blocks = 0;
inode_set_iversion_raw(inode, 0);
inode->i_generation = 0;
--
2.9.0
get_seconds() is deprecated in favor of ktime_get_real_seconds(),
which returns a 64-bit timestamp.
In the SYSV file system, the superblock timestamp is only 32 bits
wide, and it is used to check whether a file system is clean, so
the best solution seems to be to force a wraparound and explicitly
convert it to an unsigned 32-bit value.
This is independent of the inode timestamps that are also 32-bit
wide on disk and that come from current_time().
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/sysv/inode.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/fs/sysv/inode.c b/fs/sysv/inode.c
index bec9f79adb25..499a20a5a010 100644
--- a/fs/sysv/inode.c
+++ b/fs/sysv/inode.c
@@ -35,7 +35,7 @@
static int sysv_sync_fs(struct super_block *sb, int wait)
{
struct sysv_sb_info *sbi = SYSV_SB(sb);
- unsigned long time = get_seconds(), old_time;
+ u32 time = (u32)ktime_get_real_seconds(), old_time;
mutex_lock(&sbi->s_lock);
@@ -46,8 +46,8 @@ static int sysv_sync_fs(struct super_block *sb, int wait)
*/
old_time = fs32_to_cpu(sbi, *sbi->s_sb_time);
if (sbi->s_type == FSTYPE_SYSV4) {
- if (*sbi->s_sb_state == cpu_to_fs32(sbi, 0x7c269d38 - old_time))
- *sbi->s_sb_state = cpu_to_fs32(sbi, 0x7c269d38 - time);
+ if (*sbi->s_sb_state == cpu_to_fs32(sbi, 0x7c269d38u - old_time))
+ *sbi->s_sb_state = cpu_to_fs32(sbi, 0x7c269d38u - time);
*sbi->s_sb_time = cpu_to_fs32(sbi, time);
mark_buffer_dirty(sbi->s_bh2);
}
--
2.9.0
get_seconds() is deprecated because of the 32-bit overflow and will
be removed. All callers in ufs also truncate to a 32-bit number, so
nothing changes during the conversion, but this should be harmless as the
superblock and cylinder group timestamps are not visible to user space,
except for checking the fs-dirty state, wich works fine across the
overflow.
This moves the call to get_seconds() into a new inline function, with
a comment explaining the constraints, while converting it to
ktime_get_real_seconds().
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/ufs/balloc.c | 4 ++--
fs/ufs/ialloc.c | 2 +-
fs/ufs/super.c | 4 ++--
fs/ufs/util.h | 14 ++++++++++++++
4 files changed, 19 insertions(+), 5 deletions(-)
diff --git a/fs/ufs/balloc.c b/fs/ufs/balloc.c
index e727ee07dbe4..075d3d9114c8 100644
--- a/fs/ufs/balloc.c
+++ b/fs/ufs/balloc.c
@@ -547,7 +547,7 @@ static u64 ufs_add_fragments(struct inode *inode, u64 fragment,
/*
* Block can be extended
*/
- ucg->cg_time = cpu_to_fs32(sb, get_seconds());
+ ucg->cg_time = ufs_get_seconds(sb);
for (i = newcount; i < (uspi->s_fpb - fragoff); i++)
if (ubh_isclr (UCPI_UBH(ucpi), ucpi->c_freeoff, fragno + i))
break;
@@ -639,7 +639,7 @@ static u64 ufs_alloc_fragments(struct inode *inode, unsigned cgno,
if (!ufs_cg_chkmagic(sb, ucg))
ufs_panic (sb, "ufs_alloc_fragments",
"internal error, bad magic number on cg %u", cgno);
- ucg->cg_time = cpu_to_fs32(sb, get_seconds());
+ ucg->cg_time = ufs_get_seconds(sb);
if (count == uspi->s_fpb) {
result = ufs_alloccg_block (inode, ucpi, goal, err);
diff --git a/fs/ufs/ialloc.c b/fs/ufs/ialloc.c
index e1ef0f0a1353..c678fff2a04d 100644
--- a/fs/ufs/ialloc.c
+++ b/fs/ufs/ialloc.c
@@ -89,7 +89,7 @@ void ufs_free_inode (struct inode * inode)
if (!ufs_cg_chkmagic(sb, ucg))
ufs_panic (sb, "ufs_free_fragments", "internal error, bad cg magic number");
- ucg->cg_time = cpu_to_fs32(sb, get_seconds());
+ ucg->cg_time = ufs_get_seconds(sb);
is_directory = S_ISDIR(inode->i_mode);
diff --git a/fs/ufs/super.c b/fs/ufs/super.c
index 488088141451..a4e07e910f1b 100644
--- a/fs/ufs/super.c
+++ b/fs/ufs/super.c
@@ -698,7 +698,7 @@ static int ufs_sync_fs(struct super_block *sb, int wait)
usb1 = ubh_get_usb_first(uspi);
usb3 = ubh_get_usb_third(uspi);
- usb1->fs_time = cpu_to_fs32(sb, get_seconds());
+ usb1->fs_time = ufs_get_seconds(sb);
if ((flags & UFS_ST_MASK) == UFS_ST_SUN ||
(flags & UFS_ST_MASK) == UFS_ST_SUNOS ||
(flags & UFS_ST_MASK) == UFS_ST_SUNx86)
@@ -1342,7 +1342,7 @@ static int ufs_remount (struct super_block *sb, int *mount_flags, char *data)
*/
if (*mount_flags & SB_RDONLY) {
ufs_put_super_internal(sb);
- usb1->fs_time = cpu_to_fs32(sb, get_seconds());
+ usb1->fs_time = ufs_get_seconds(sb);
if ((flags & UFS_ST_MASK) == UFS_ST_SUN
|| (flags & UFS_ST_MASK) == UFS_ST_SUNOS
|| (flags & UFS_ST_MASK) == UFS_ST_SUNx86)
diff --git a/fs/ufs/util.h b/fs/ufs/util.h
index 1907be6d5808..1fd3011ea623 100644
--- a/fs/ufs/util.h
+++ b/fs/ufs/util.h
@@ -590,3 +590,17 @@ static inline int ufs_is_data_ptr_zero(struct ufs_sb_private_info *uspi,
else
return *(__fs32 *)p == 0;
}
+
+static inline __fs32 ufs_get_seconds(struct super_block *sbp)
+{
+ time64_t now = ktime_get_real_seconds();
+
+ /* Signed 32-bit interpretation wraps around in 2038, which
+ * happens in ufs1 inode stamps but not ufs2 using 64-bits
+ * stamps. For superblock and blockgroup, let's assume
+ * unsigned 32-bit stamps, which are good until y2106.
+ * Wrap around rather than clamp here to make the dirty
+ * file system detection work in the superblock stamp.
+ */
+ return cpu_to_fs32(sbp, lower_32_bits(now));
+}
--
2.9.0
get_seconds() is deprecated because it will lead to a 32-bit overflow
in 2038 or 2106. We don't need the i_generation to be strictly
monotonic anyway, and other file systems like ext4 and xfs just use
prandom_u32(), so let's use the same one here.
If this is considered too slow, we could also use ktime_get_seconds()
or ktime_get_real_seconds() to keep the previous behavior.
Both of these return a time64_t and are not deprecated, but only
return a unique value once per second, and are predictable.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
mm/shmem.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/shmem.c b/mm/shmem.c
index 2cab84403055..387ae5323f56 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -29,6 +29,7 @@
#include <linux/pagemap.h>
#include <linux/file.h>
#include <linux/mm.h>
+#include <linux/random.h>
#include <linux/sched/signal.h>
#include <linux/export.h>
#include <linux/swap.h>
@@ -2187,7 +2188,7 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
inode_init_owner(inode, dir, mode);
inode->i_blocks = 0;
inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
- inode->i_generation = get_seconds();
+ inode->i_generation = prandom_u32();
info = SHMEM_I(inode);
memset(info, 0, (char *)inode - (char *)info);
spin_lock_init(&info->lock);
--
2.9.0
The handling of timestamps outside of the 1970..2038 range in the dlm
glue is rather inconsistent: on 32-bit architectures, this has always
wrapped around to negative timestamps in the 1902..1969 range, while on
64-bit kernels all timestamps are interpreted as positive 34 bit numbers
in the 1970..2514 year range.
Now that the VFS code handles 64-bit timestamps on all architectures,
we can make the behavior more consistent here, and return the same result
that we had on 64-bit already, making the file system y2038 safe in the
process. Outside of dlmglue, it already uses 64-bit on-disk timestamps
anway, so that part is fine.
For consistency, I'm changing ocfs2_pack_timespec() to clamp
anything outside of the supported range to the minimum and maximum
values. This avoids a possible ambiguity of values before 1970
in particular, which used to be interpreted as times at the end of the
2514 range previously.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/ocfs2/dlmglue.c | 26 +++++++++-----------------
1 file changed, 9 insertions(+), 17 deletions(-)
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
index 0ff424c6d17c..50610a9ed9f4 100644
--- a/fs/ocfs2/dlmglue.c
+++ b/fs/ocfs2/dlmglue.c
@@ -2121,10 +2121,10 @@ static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
/* LVB only has room for 64 bits of time here so we pack it for
* now. */
-static u64 ocfs2_pack_timespec(struct timespec *spec)
+static u64 ocfs2_pack_timespec(struct timespec64 *spec)
{
u64 res;
- u64 sec = spec->tv_sec;
+ u64 sec = clamp_t(time64_t, spec->tv_sec, 0, 0x3ffffffffull);
u32 nsec = spec->tv_nsec;
res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
@@ -2140,7 +2140,6 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode)
struct ocfs2_inode_info *oi = OCFS2_I(inode);
struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
struct ocfs2_meta_lvb *lvb;
- struct timespec ts;
lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
@@ -2161,15 +2160,12 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode)
lvb->lvb_igid = cpu_to_be32(i_gid_read(inode));
lvb->lvb_imode = cpu_to_be16(inode->i_mode);
lvb->lvb_inlink = cpu_to_be16(inode->i_nlink);
- ts = timespec64_to_timespec(inode->i_atime);
lvb->lvb_iatime_packed =
- cpu_to_be64(ocfs2_pack_timespec(&ts));
- ts = timespec64_to_timespec(inode->i_ctime);
+ cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
lvb->lvb_ictime_packed =
- cpu_to_be64(ocfs2_pack_timespec(&ts));
- ts = timespec64_to_timespec(inode->i_mtime);
+ cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
lvb->lvb_imtime_packed =
- cpu_to_be64(ocfs2_pack_timespec(&ts));
+ cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
lvb->lvb_iattr = cpu_to_be32(oi->ip_attr);
lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
@@ -2178,7 +2174,7 @@ static void __ocfs2_stuff_meta_lvb(struct inode *inode)
mlog_meta_lvb(0, lockres);
}
-static void ocfs2_unpack_timespec(struct timespec *spec,
+static void ocfs2_unpack_timespec(struct timespec64 *spec,
u64 packed_time)
{
spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
@@ -2187,7 +2183,6 @@ static void ocfs2_unpack_timespec(struct timespec *spec,
static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
{
- struct timespec ts;
struct ocfs2_inode_info *oi = OCFS2_I(inode);
struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
struct ocfs2_meta_lvb *lvb;
@@ -2215,15 +2210,12 @@ static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
inode->i_mode = be16_to_cpu(lvb->lvb_imode);
set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
- ocfs2_unpack_timespec(&ts,
+ ocfs2_unpack_timespec(&inode->i_atime,
be64_to_cpu(lvb->lvb_iatime_packed));
- inode->i_atime = timespec_to_timespec64(ts);
- ocfs2_unpack_timespec(&ts,
+ ocfs2_unpack_timespec(&inode->i_mtime,
be64_to_cpu(lvb->lvb_imtime_packed));
- inode->i_mtime = timespec_to_timespec64(ts);
- ocfs2_unpack_timespec(&ts,
+ ocfs2_unpack_timespec(&inode->i_ctime,
be64_to_cpu(lvb->lvb_ictime_packed));
- inode->i_ctime = timespec_to_timespec64(ts);
spin_unlock(&oi->ip_lock);
}
--
2.9.0
Using get_seconds() for timestamps is deprecated since it can lead
to overflows on 32-bit systems. While the interface generally doesn't
overflow until year 2106, the specific implementation of the TCP PAWS
algorithm breaks in 2038 when the intermediate signed 32-bit timestamps
overflow.
A related problem is that the local timestamps in CLOCK_REALTIME form
lead to unexpected behavior when settimeofday is called to set the system
clock backwards or forwards by more than 24 days.
While the first problem could be solved by using an overflow-safe method
of comparing the timestamps, a nicer solution is to use a monotonic
clocksource with ktime_get_seconds() that simply doesn't overflow (at
least not until 136 years after boot) and that doesn't change during
settimeofday().
To make 32-bit and 64-bit architectures behave the same way here, and
also save a few bytes in the tcp_options_received structure, I'm changing
the type to a 32-bit integer, which is now safe on all architectures.
Finally, the ts_recent_stamp field also (confusingly) gets used to store
a jiffies value in tcp_synq_overflow()/tcp_synq_no_recent_overflow().
This is currently safe, but changing the type to 32-bit requires
some small changes there to keep it working.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
v2: use time_before32()/time_after32() everywhere as suggested
Eric Dumazet
---
drivers/crypto/chelsio/chtls/chtls_cm.c | 2 +-
include/linux/tcp.h | 4 ++--
include/net/tcp.h | 17 ++++++++++-------
net/ipv4/tcp_input.c | 2 +-
net/ipv4/tcp_ipv4.c | 3 ++-
net/ipv4/tcp_minisocks.c | 8 ++++----
6 files changed, 20 insertions(+), 16 deletions(-)
diff --git a/drivers/crypto/chelsio/chtls/chtls_cm.c b/drivers/crypto/chelsio/chtls/chtls_cm.c
index 2bb6f0380758..0997e166ea57 100644
--- a/drivers/crypto/chelsio/chtls/chtls_cm.c
+++ b/drivers/crypto/chelsio/chtls/chtls_cm.c
@@ -1673,7 +1673,7 @@ static void chtls_timewait(struct sock *sk)
struct tcp_sock *tp = tcp_sk(sk);
tp->rcv_nxt++;
- tp->rx_opt.ts_recent_stamp = get_seconds();
+ tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
tp->srtt_us = 0;
tcp_time_wait(sk, TCP_TIME_WAIT, 0);
}
diff --git a/include/linux/tcp.h b/include/linux/tcp.h
index 72705eaf4b84..f911b9b09b16 100644
--- a/include/linux/tcp.h
+++ b/include/linux/tcp.h
@@ -89,7 +89,7 @@ struct tcp_sack_block {
struct tcp_options_received {
/* PAWS/RTTM data */
- long ts_recent_stamp;/* Time we stored ts_recent (for aging) */
+ int ts_recent_stamp;/* Time we stored ts_recent (for aging) */
u32 ts_recent; /* Time stamp to echo next */
u32 rcv_tsval; /* Time stamp value */
u32 rcv_tsecr; /* Time stamp echo reply */
@@ -425,7 +425,7 @@ struct tcp_timewait_sock {
/* The time we sent the last out-of-window ACK: */
u32 tw_last_oow_ack_time;
- long tw_ts_recent_stamp;
+ int tw_ts_recent_stamp;
#ifdef CONFIG_TCP_MD5SIG
struct tcp_md5sig_key *tw_md5_key;
#endif
diff --git a/include/net/tcp.h b/include/net/tcp.h
index 0448e7c5d2b4..535dfb7f2ab4 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -471,19 +471,20 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
*/
static inline void tcp_synq_overflow(const struct sock *sk)
{
- unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
- unsigned long now = jiffies;
+ unsigned int last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
+ unsigned int now = jiffies;
- if (time_after(now, last_overflow + HZ))
+ if (time_after32(now, last_overflow + HZ))
tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
}
/* syncookies: no recent synqueue overflow on this listening socket? */
static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
{
- unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
+ unsigned int last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
+ unsigned int now = jiffies;
- return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
+ return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
}
static inline u32 tcp_cookie_time(void)
@@ -1361,7 +1362,8 @@ static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
{
if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
return true;
- if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
+ if (unlikely(!time_before32(ktime_get_seconds(),
+ rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
return true;
/*
* Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
@@ -1391,7 +1393,8 @@ static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
However, we can relax time bounds for RST segments to MSL.
*/
- if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
+ if (rst && !time_before32(ktime_get_seconds(),
+ rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
return false;
return true;
}
diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 355d3dffd021..0eb314774aec 100644
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -3449,7 +3449,7 @@ static void tcp_send_challenge_ack(struct sock *sk, const struct sk_buff *skb)
static void tcp_store_ts_recent(struct tcp_sock *tp)
{
tp->rx_opt.ts_recent = tp->rx_opt.rcv_tsval;
- tp->rx_opt.ts_recent_stamp = get_seconds();
+ tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
}
static void tcp_replace_ts_recent(struct tcp_sock *tp, u32 seq)
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index bea17f1e8302..dc415c66a33a 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -155,7 +155,8 @@ int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp)
and use initial timestamp retrieved from peer table.
*/
if (tcptw->tw_ts_recent_stamp &&
- (!twp || (reuse && get_seconds() - tcptw->tw_ts_recent_stamp > 1))) {
+ (!twp || (reuse && time_after32(ktime_get_seconds(),
+ tcptw->tw_ts_recent_stamp)))) {
tp->write_seq = tcptw->tw_snd_nxt + 65535 + 2;
if (tp->write_seq == 0)
tp->write_seq = 1;
diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c
index 1dda1341a223..1f652beb79ca 100644
--- a/net/ipv4/tcp_minisocks.c
+++ b/net/ipv4/tcp_minisocks.c
@@ -144,7 +144,7 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
tw->tw_substate = TCP_TIME_WAIT;
tcptw->tw_rcv_nxt = TCP_SKB_CB(skb)->end_seq;
if (tmp_opt.saw_tstamp) {
- tcptw->tw_ts_recent_stamp = get_seconds();
+ tcptw->tw_ts_recent_stamp = ktime_get_seconds();
tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
}
@@ -189,7 +189,7 @@ tcp_timewait_state_process(struct inet_timewait_sock *tw, struct sk_buff *skb,
if (tmp_opt.saw_tstamp) {
tcptw->tw_ts_recent = tmp_opt.rcv_tsval;
- tcptw->tw_ts_recent_stamp = get_seconds();
+ tcptw->tw_ts_recent_stamp = ktime_get_seconds();
}
inet_twsk_put(tw);
@@ -534,7 +534,7 @@ struct sock *tcp_create_openreq_child(const struct sock *sk,
if (newtp->rx_opt.tstamp_ok) {
newtp->rx_opt.ts_recent = req->ts_recent;
- newtp->rx_opt.ts_recent_stamp = get_seconds();
+ newtp->rx_opt.ts_recent_stamp = ktime_get_seconds();
newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
} else {
newtp->rx_opt.ts_recent_stamp = 0;
@@ -600,7 +600,7 @@ struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
* it can be estimated (approximately)
* from another data.
*/
- tmp_opt.ts_recent_stamp = get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
+ tmp_opt.ts_recent_stamp = ktime_get_seconds() - ((TCP_TIMEOUT_INIT/HZ)<<req->num_timeout);
paws_reject = tcp_paws_reject(&tmp_opt, th->rst);
}
}
--
2.9.0
We exchange file timestamps with user space using psdev device
read/write operations with a fixed but architecture specific binary
layout.
On 32-bit systems, this uses a 'timespec' structure that is defined by
the C library to contain two 32-bit values for seconds and nanoseconds.
As we get ready for the year 2038 overflow of the 32-bit signed seconds,
the kernel now uses 64-bit timestamps internally, and user space will
do the same change by changing the 'timespec' definition in the future.
Unfortunately, this breaks the layout of the coda_vattr structure, so
we need to redefine that in terms of something that does not change.
I'm introducing a new 'struct vtimespec' structure here that keeps
the existing layout, and the same change has to be done in the coda
user space copy of linux/coda.h before anyone can use that on a 32-bit
architecture with 64-bit time_t.
An open question is what should happen to actual times past y2038,
as they are now truncated to the last valid date when sent to user
space, and interpreted as pre-1970 times when a timestamp with the
MSB set is read back into the kernel. Alternatively, we could
change the new timespec64_to_coda()/coda_to_timespec64() functions
to use a different interpretation and extend the available range
further to the future by disallowing past timestamps. This would
require more changes in the user space side though.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
Documentation/filesystems/coda.txt | 11 ++++++---
fs/coda/coda_linux.c | 50 +++++++++++++++++++++++++++++---------
include/uapi/linux/coda.h | 20 ++++++++++++---
3 files changed, 62 insertions(+), 19 deletions(-)
diff --git a/Documentation/filesystems/coda.txt b/Documentation/filesystems/coda.txt
index 61311356025d..ea5969068895 100644
--- a/Documentation/filesystems/coda.txt
+++ b/Documentation/filesystems/coda.txt
@@ -481,7 +481,10 @@ kernel support.
-
+ struct vtimespec {
+ long tv_sec; /* seconds */
+ long tv_nsec; /* nanoseconds */
+ };
struct coda_vattr {
enum coda_vtype va_type; /* vnode type (for create) */
@@ -493,9 +496,9 @@ kernel support.
long va_fileid; /* file id */
u_quad_t va_size; /* file size in bytes */
long va_blocksize; /* blocksize preferred for i/o */
- struct timespec va_atime; /* time of last access */
- struct timespec va_mtime; /* time of last modification */
- struct timespec va_ctime; /* time file changed */
+ struct vtimespec va_atime; /* time of last access */
+ struct vtimespec va_mtime; /* time of last modification */
+ struct vtimespec va_ctime; /* time file changed */
u_long va_gen; /* generation number of file */
u_long va_flags; /* flags defined for file */
dev_t va_rdev; /* device special file represents */
diff --git a/fs/coda/coda_linux.c b/fs/coda/coda_linux.c
index f3d543dd9a98..8addcd166908 100644
--- a/fs/coda/coda_linux.c
+++ b/fs/coda/coda_linux.c
@@ -66,6 +66,32 @@ unsigned short coda_flags_to_cflags(unsigned short flags)
return coda_flags;
}
+static struct timespec64 coda_to_timespec64(struct vtimespec ts)
+{
+ /*
+ * We interpret incoming timestamps as 'signed' to match traditional
+ * usage and support pre-1970 timestamps, but this breaks in y2038
+ * on 32-bit machines.
+ */
+ struct timespec64 ts64 = {
+ .tv_sec = ts.tv_sec,
+ .tv_nsec = ts.tv_nsec,
+ };
+
+ return ts64;
+}
+
+static struct vtimespec timespec64_to_coda(struct timespec64 ts64)
+{
+ /* clamp the timestamps to the maximum range rather than wrapping */
+ struct vtimespec ts = {
+ .tv_sec = lower_32_bits(clamp_t(time64_t, ts64.tv_sec,
+ LONG_MIN, LONG_MAX)),
+ .tv_nsec = ts64.tv_nsec,
+ };
+
+ return ts;
+}
/* utility functions below */
void coda_vattr_to_iattr(struct inode *inode, struct coda_vattr *attr)
@@ -105,11 +131,11 @@ void coda_vattr_to_iattr(struct inode *inode, struct coda_vattr *attr)
if (attr->va_size != -1)
inode->i_blocks = (attr->va_size + 511) >> 9;
if (attr->va_atime.tv_sec != -1)
- inode->i_atime = timespec_to_timespec64(attr->va_atime);
+ inode->i_atime = coda_to_timespec64(attr->va_atime);
if (attr->va_mtime.tv_sec != -1)
- inode->i_mtime = timespec_to_timespec64(attr->va_mtime);
+ inode->i_mtime = coda_to_timespec64(attr->va_mtime);
if (attr->va_ctime.tv_sec != -1)
- inode->i_ctime = timespec_to_timespec64(attr->va_ctime);
+ inode->i_ctime = coda_to_timespec64(attr->va_ctime);
}
@@ -130,12 +156,12 @@ void coda_iattr_to_vattr(struct iattr *iattr, struct coda_vattr *vattr)
vattr->va_uid = (vuid_t) -1;
vattr->va_gid = (vgid_t) -1;
vattr->va_size = (off_t) -1;
- vattr->va_atime.tv_sec = (time_t) -1;
- vattr->va_atime.tv_nsec = (time_t) -1;
- vattr->va_mtime.tv_sec = (time_t) -1;
- vattr->va_mtime.tv_nsec = (time_t) -1;
- vattr->va_ctime.tv_sec = (time_t) -1;
- vattr->va_ctime.tv_nsec = (time_t) -1;
+ vattr->va_atime.tv_sec = (long) -1;
+ vattr->va_atime.tv_nsec = (long) -1;
+ vattr->va_mtime.tv_sec = (long) -1;
+ vattr->va_mtime.tv_nsec = (long) -1;
+ vattr->va_ctime.tv_sec = (long) -1;
+ vattr->va_ctime.tv_nsec = (long) -1;
vattr->va_type = C_VNON;
vattr->va_fileid = -1;
vattr->va_gen = -1;
@@ -175,13 +201,13 @@ void coda_iattr_to_vattr(struct iattr *iattr, struct coda_vattr *vattr)
vattr->va_size = iattr->ia_size;
}
if ( valid & ATTR_ATIME ) {
- vattr->va_atime = timespec64_to_timespec(iattr->ia_atime);
+ vattr->va_atime = timespec64_to_coda(iattr->ia_atime);
}
if ( valid & ATTR_MTIME ) {
- vattr->va_mtime = timespec64_to_timespec(iattr->ia_mtime);
+ vattr->va_mtime = timespec64_to_coda(iattr->ia_mtime);
}
if ( valid & ATTR_CTIME ) {
- vattr->va_ctime = timespec64_to_timespec(iattr->ia_ctime);
+ vattr->va_ctime = timespec64_to_coda(iattr->ia_ctime);
}
}
diff --git a/include/uapi/linux/coda.h b/include/uapi/linux/coda.h
index 695fade33c64..027a8eb04423 100644
--- a/include/uapi/linux/coda.h
+++ b/include/uapi/linux/coda.h
@@ -211,6 +211,20 @@ struct CodaFid {
*/
enum coda_vtype { C_VNON, C_VREG, C_VDIR, C_VBLK, C_VCHR, C_VLNK, C_VSOCK, C_VFIFO, C_VBAD };
+#ifdef __linux__
+/*
+ * This matches the traditional Linux 'timespec' structure binary layout,
+ * before using 64-bit time_t everywhere. Overflows in y2038 on 32-bit
+ * architectures.
+ */
+struct vtimespec {
+ long tv_sec; /* seconds */
+ long tv_nsec; /* nanoseconds */
+};
+#else
+#define vtimespec timespec
+#endif
+
struct coda_vattr {
long va_type; /* vnode type (for create) */
u_short va_mode; /* files access mode and type */
@@ -220,9 +234,9 @@ struct coda_vattr {
long va_fileid; /* file id */
u_quad_t va_size; /* file size in bytes */
long va_blocksize; /* blocksize preferred for i/o */
- struct timespec va_atime; /* time of last access */
- struct timespec va_mtime; /* time of last modification */
- struct timespec va_ctime; /* time file changed */
+ struct vtimespec va_atime; /* time of last access */
+ struct vtimespec va_mtime; /* time of last modification */
+ struct vtimespec va_ctime; /* time file changed */
u_long va_gen; /* generation number of file */
u_long va_flags; /* flags defined for file */
cdev_t va_rdev; /* device special file represents */
--
2.9.0
The file creation time in the inode uses time_t which is defined differently
on 32-bit and 64-bit architectures and deprecated. The representation in
the inode uses an unsigned 32-bit number, but this gets wrapped around
after year 2038 when assigned to a time_t.
This changes the type to time64_t, so we can support the full range of
timestamps between 1970 and 2106 on 32-bit systems like we do on 64-bit
systems already, and matching what we do for the atime/ctime/mtime stamps
since the introduction of 64-bit timestamps in VFS.
Note: the otime stamp is not actually used anywhere at the moment in
the kernel, it is just set when writing a file, so none of this really
makes a difference unless we implement setting the btime field in the
getattr() callback.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/jfs/jfs_incore.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fs/jfs/jfs_incore.h b/fs/jfs/jfs_incore.h
index 1f26d1910409..d5c46f86b2ef 100644
--- a/fs/jfs/jfs_incore.h
+++ b/fs/jfs/jfs_incore.h
@@ -43,7 +43,7 @@ struct jfs_inode_info {
pxd_t ixpxd; /* inode extent descriptor */
dxd_t acl; /* dxd describing acl */
dxd_t ea; /* dxd describing ea */
- time_t otime; /* time created */
+ time64_t otime; /* time created */
uint next_index; /* next available directory entry index */
int acltype; /* Type of ACL */
short btorder; /* access order */
--
2.9.0
Now that the VFS has been converted from timespec to timespec64
timestamps, only the conversion to/from ntfs timestamps uses 32-bit
seconds.
This changes that last missing piece to get the ntfs implementation
y2038 safe on 32-bit architectures.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
fs/ntfs/inode.c | 12 ++++++------
fs/ntfs/time.h | 27 +++++++++++++++------------
2 files changed, 21 insertions(+), 18 deletions(-)
diff --git a/fs/ntfs/inode.c b/fs/ntfs/inode.c
index decaf75d1cd5..bd3221cbdd95 100644
--- a/fs/ntfs/inode.c
+++ b/fs/ntfs/inode.c
@@ -667,18 +667,18 @@ static int ntfs_read_locked_inode(struct inode *vi)
* mtime is the last change of the data within the file. Not changed
* when only metadata is changed, e.g. a rename doesn't affect mtime.
*/
- vi->i_mtime = timespec_to_timespec64(ntfs2utc(si->last_data_change_time));
+ vi->i_mtime = ntfs2utc(si->last_data_change_time);
/*
* ctime is the last change of the metadata of the file. This obviously
* always changes, when mtime is changed. ctime can be changed on its
* own, mtime is then not changed, e.g. when a file is renamed.
*/
- vi->i_ctime = timespec_to_timespec64(ntfs2utc(si->last_mft_change_time));
+ vi->i_ctime = ntfs2utc(si->last_mft_change_time);
/*
* Last access to the data within the file. Not changed during a rename
* for example but changed whenever the file is written to.
*/
- vi->i_atime = timespec_to_timespec64(ntfs2utc(si->last_access_time));
+ vi->i_atime = ntfs2utc(si->last_access_time);
/* Find the attribute list attribute if present. */
ntfs_attr_reinit_search_ctx(ctx);
@@ -2997,7 +2997,7 @@ int __ntfs_write_inode(struct inode *vi, int sync)
si = (STANDARD_INFORMATION*)((u8*)ctx->attr +
le16_to_cpu(ctx->attr->data.resident.value_offset));
/* Update the access times if they have changed. */
- nt = utc2ntfs(timespec64_to_timespec(vi->i_mtime));
+ nt = utc2ntfs(vi->i_mtime);
if (si->last_data_change_time != nt) {
ntfs_debug("Updating mtime for inode 0x%lx: old = 0x%llx, "
"new = 0x%llx", vi->i_ino, (long long)
@@ -3006,7 +3006,7 @@ int __ntfs_write_inode(struct inode *vi, int sync)
si->last_data_change_time = nt;
modified = true;
}
- nt = utc2ntfs(timespec64_to_timespec(vi->i_ctime));
+ nt = utc2ntfs(vi->i_ctime);
if (si->last_mft_change_time != nt) {
ntfs_debug("Updating ctime for inode 0x%lx: old = 0x%llx, "
"new = 0x%llx", vi->i_ino, (long long)
@@ -3015,7 +3015,7 @@ int __ntfs_write_inode(struct inode *vi, int sync)
si->last_mft_change_time = nt;
modified = true;
}
- nt = utc2ntfs(timespec64_to_timespec(vi->i_atime));
+ nt = utc2ntfs(vi->i_atime);
if (si->last_access_time != nt) {
ntfs_debug("Updating atime for inode 0x%lx: old = 0x%llx, "
"new = 0x%llx", vi->i_ino,
diff --git a/fs/ntfs/time.h b/fs/ntfs/time.h
index 01233989d5d1..24cd719f1fd2 100644
--- a/fs/ntfs/time.h
+++ b/fs/ntfs/time.h
@@ -36,16 +36,16 @@
* Convert the Linux UTC time @ts to its corresponding NTFS time and return
* that in little endian format.
*
- * Linux stores time in a struct timespec consisting of a time_t (long at
- * present) tv_sec and a long tv_nsec where tv_sec is the number of 1-second
- * intervals since 1st January 1970, 00:00:00 UTC and tv_nsec is the number of
- * 1-nano-second intervals since the value of tv_sec.
+ * Linux stores time in a struct timespec64 consisting of a time64_t tv_sec
+ * and a long tv_nsec where tv_sec is the number of 1-second intervals since
+ * 1st January 1970, 00:00:00 UTC and tv_nsec is the number of 1-nano-second
+ * intervals since the value of tv_sec.
*
* NTFS uses Microsoft's standard time format which is stored in a s64 and is
* measured as the number of 100-nano-second intervals since 1st January 1601,
* 00:00:00 UTC.
*/
-static inline sle64 utc2ntfs(const struct timespec ts)
+static inline sle64 utc2ntfs(const struct timespec64 ts)
{
/*
* Convert the seconds to 100ns intervals, add the nano-seconds
@@ -63,7 +63,10 @@ static inline sle64 utc2ntfs(const struct timespec ts)
*/
static inline sle64 get_current_ntfs_time(void)
{
- return utc2ntfs(current_kernel_time());
+ struct timespec64 ts;
+
+ ktime_get_coarse_real_ts64(&ts);
+ return utc2ntfs(ts);
}
/**
@@ -73,18 +76,18 @@ static inline sle64 get_current_ntfs_time(void)
* Convert the little endian NTFS time @time to its corresponding Linux UTC
* time and return that in cpu format.
*
- * Linux stores time in a struct timespec consisting of a time_t (long at
- * present) tv_sec and a long tv_nsec where tv_sec is the number of 1-second
- * intervals since 1st January 1970, 00:00:00 UTC and tv_nsec is the number of
- * 1-nano-second intervals since the value of tv_sec.
+ * Linux stores time in a struct timespec64 consisting of a time64_t tv_sec
+ * and a long tv_nsec where tv_sec is the number of 1-second intervals since
+ * 1st January 1970, 00:00:00 UTC and tv_nsec is the number of 1-nano-second
+ * intervals since the value of tv_sec.
*
* NTFS uses Microsoft's standard time format which is stored in a s64 and is
* measured as the number of 100 nano-second intervals since 1st January 1601,
* 00:00:00 UTC.
*/
-static inline struct timespec ntfs2utc(const sle64 time)
+static inline struct timespec64 ntfs2utc(const sle64 time)
{
- struct timespec ts;
+ struct timespec64 ts;
/* Subtract the NTFS time offset. */
u64 t = (u64)(sle64_to_cpu(time) - NTFS_TIME_OFFSET);
--
2.9.0
As Mathieu pointed out, my conversion to time64_t was incorrect and resulted
in negative times to be read from the RTC. The problem is that during the
conversion from a byte array to a time64_t, the 'unsigned char' variable
holding the top byte gets turned into a negative signed 32-bit integer
before being assigned to the 64-bit variable for any times after 1972.
This changes the logic to cast to an unsigned 32-bit number first for
the Macintosh time and then convert that to the Unix time, which then gives
us a time in the documented 1904..2040 year range. I decided not to use
the longer 1970..2106 range that other drivers use, for consistency with
the literal interpretation of the register, but that could be easily
changed if we decide we want to support any Mac after 2040.
Just to be on the safe side, I'm also adding a WARN_ON that will trigger
if either the year 2040 has come and is observed by this driver, or we
run into an RTC that got set back to a pre-1970 date for some reason
(the two are indistinguishable).
The same code exists in arch/m68k/ and is patched in an identical way now
in a separate patch.
Fixes: 5bfd643583b2 ("powerpc: use time64_t in read_persistent_clock")
Reported-by: Mathieu Malaterre <malat(a)debian.org>
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
arch/powerpc/platforms/powermac/time.c | 21 ++++++++++++++++-----
1 file changed, 16 insertions(+), 5 deletions(-)
diff --git a/arch/powerpc/platforms/powermac/time.c b/arch/powerpc/platforms/powermac/time.c
index 7c968e46736f..173a80630169 100644
--- a/arch/powerpc/platforms/powermac/time.c
+++ b/arch/powerpc/platforms/powermac/time.c
@@ -42,7 +42,11 @@
#define DBG(x...)
#endif
-/* Apparently the RTC stores seconds since 1 Jan 1904 */
+/*
+ * Offset between Unix time (1970-based) and Mac time (1904-based). Cuda and PMU
+ * times wrap in 2040. If we need to handle later times, the read_time functions
+ * need to be changed to interpret wrapped times as post-2040.
+ */
#define RTC_OFFSET 2082844800
/*
@@ -97,8 +101,11 @@ static time64_t cuda_get_time(void)
if (req.reply_len != 7)
printk(KERN_ERR "cuda_get_time: got %d byte reply\n",
req.reply_len);
- now = (req.reply[3] << 24) + (req.reply[4] << 16)
- + (req.reply[5] << 8) + req.reply[6];
+ now = (u32)((req.reply[3] << 24) + (req.reply[4] << 16) +
+ (req.reply[5] << 8) + req.reply[6]);
+ /* it's either after year 2040, or the RTC has gone backwards */
+ WARN_ON(now < RTC_OFFSET);
+
return now - RTC_OFFSET;
}
@@ -140,8 +147,12 @@ static time64_t pmu_get_time(void)
if (req.reply_len != 4)
printk(KERN_ERR "pmu_get_time: got %d byte reply from PMU\n",
req.reply_len);
- now = (req.reply[0] << 24) + (req.reply[1] << 16)
- + (req.reply[2] << 8) + req.reply[3];
+ now = (u32)((req.reply[0] << 24) + (req.reply[1] << 16) +
+ (req.reply[2] << 8) + req.reply[3]);
+
+ /* it's either after year 2040, or the RTC has gone backwards */
+ WARN_ON(now < RTC_OFFSET);
+
return now - RTC_OFFSET;
}
--
2.9.0
The ohci driver uses the get_seconds() function to implement the 32-bit
CSR_BUS_TIME register. This was added in 2010 commit a48777e03ad5
("firewire: add CSR BUS_TIME support").
As get_seconds() returns a 32-bit value (on 32-bit architectures), it
seems like a good fit for that register, but it is also deprecated because
of the y2038/y2106 overflow problem, and should be replaced throughout
the kernel with either ktime_get_real_seconds() or ktime_get_seconds().
I'm using the latter here, which uses monotonic time. This has the
advantage of behaving better during concurrent settimeofday() updates
or leap second adjustments and won't overflow a 32-bit integer, but
the downside of using CLOCK_MONOTONIC instead of CLOCK_REALTIME is
that the observed values are not related to external clocks.
If we instead need UTC but can live with clock jumps or overflows,
then we should use ktime_get_real_seconds() instead, retaining the
existing behavior.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/firewire/ohci.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/firewire/ohci.c b/drivers/firewire/ohci.c
index 45c048751f3b..5125841ea338 100644
--- a/drivers/firewire/ohci.c
+++ b/drivers/firewire/ohci.c
@@ -1765,7 +1765,7 @@ static u32 update_bus_time(struct fw_ohci *ohci)
if (unlikely(!ohci->bus_time_running)) {
reg_write(ohci, OHCI1394_IntMaskSet, OHCI1394_cycle64Seconds);
- ohci->bus_time = (lower_32_bits(get_seconds()) & ~0x7f) |
+ ohci->bus_time = (lower_32_bits(ktime_get_seconds()) & ~0x7f) |
(cycle_time_seconds & 0x40);
ohci->bus_time_running = true;
}
--
2.9.0
The patch titled
Subject: kernel/sys.c: remove get_monotonic_boottime()
has been added to the -mm tree. Its filename is
sysinfo-remove-get_monotonic_boottime.patch
This patch should soon appear at
http://ozlabs.org/~akpm/mmots/broken-out/sysinfo-remove-get_monotonic_boott…
and later at
http://ozlabs.org/~akpm/mmotm/broken-out/sysinfo-remove-get_monotonic_boott…
Before you just go and hit "reply", please:
a) Consider who else should be cc'ed
b) Prefer to cc a suitable mailing list as well
c) Ideally: find the original patch on the mailing list and do a
reply-to-all to that, adding suitable additional cc's
*** Remember to use Documentation/process/submit-checklist.rst when testing your code ***
The -mm tree is included into linux-next and is updated
there every 3-4 working days
------------------------------------------------------
From: Arnd Bergmann <arnd(a)arndb.de>
Subject: kernel/sys.c: remove get_monotonic_boottime()
get_monotonic_boottime() is deprecated because it uses the old 'timespec'
structure. This replaces one of the last callers with a call to
ktime_get_boottime.
Link: http://lkml.kernel.org/r/20180618150114.849216-1-arnd@arndb.de
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
Reviewed-by: Cyrill Gorcunov <gorcunov(a)gmail.com>
Reviewed-by: Andrew Morton <akpm(a)linux-foundation.org>
Cc: Thomas Gleixner <tglx(a)linutronix.de>
Cc: <y2038(a)lists.linaro.org>
Cc: Dominik Brodowski <linux(a)dominikbrodowski.net>
Signed-off-by: Andrew Morton <akpm(a)linux-foundation.org>
---
kernel/sys.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff -puN kernel/sys.c~sysinfo-remove-get_monotonic_boottime kernel/sys.c
--- a/kernel/sys.c~sysinfo-remove-get_monotonic_boottime
+++ a/kernel/sys.c
@@ -2523,11 +2523,11 @@ static int do_sysinfo(struct sysinfo *in
{
unsigned long mem_total, sav_total;
unsigned int mem_unit, bitcount;
- struct timespec tp;
+ struct timespec64 tp;
memset(info, 0, sizeof(struct sysinfo));
- get_monotonic_boottime(&tp);
+ ktime_get_boottime_ts64(&tp);
info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
_
Patches currently in -mm which might be from arnd(a)arndb.de are
crash-print-timestamp-using-time64_t.patch
sysinfo-remove-get_monotonic_boottime.patch
getnstimeofday64() is deprecated because of the inconsistent naming,
it is only a wrapper around ktime_get_real_ts64() now, which could be
used as a direct replacement.
However, it is generally better to use CLOCK_MONOTONIC timestamps
where possible, to avoid glitches with a concurrent settimeofday()
or leap second.
The uses in ipmi are either for debugging prints or for comparing against
a prior timestamp, so using a monotonic ktime_get_ts64() is probably
best here.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/char/ipmi/ipmi_si_intf.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/drivers/char/ipmi/ipmi_si_intf.c b/drivers/char/ipmi/ipmi_si_intf.c
index ad353be871bf..fb19c796f0fa 100644
--- a/drivers/char/ipmi/ipmi_si_intf.c
+++ b/drivers/char/ipmi/ipmi_si_intf.c
@@ -269,7 +269,7 @@ void debug_timestamp(char *msg)
{
struct timespec64 t;
- getnstimeofday64(&t);
+ ktime_get_ts64(&t);
pr_debug("**%s: %lld.%9.9ld\n", msg, (long long) t.tv_sec, t.tv_nsec);
}
#else
@@ -961,12 +961,12 @@ static inline int ipmi_thread_busy_wait(enum si_sm_result smi_result,
if (max_busy_us == 0 || smi_result != SI_SM_CALL_WITH_DELAY)
ipmi_si_set_not_busy(busy_until);
else if (!ipmi_si_is_busy(busy_until)) {
- getnstimeofday64(busy_until);
+ ktime_get_ts64(busy_until);
timespec64_add_ns(busy_until, max_busy_us*NSEC_PER_USEC);
} else {
struct timespec64 now;
- getnstimeofday64(&now);
+ ktime_get_ts64(&now);
if (unlikely(timespec64_compare(&now, busy_until) > 0)) {
ipmi_si_set_not_busy(busy_until);
return 0;
--
2.9.0
This function has been unused since commit 5ed0bdf21a85 ("drm: i915:
Use nsec based interfaces"). Let's remove the definition as well now
to help get rid of all uses of 'timespec'.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/gpu/drm/i915/i915_drv.h | 8 --------
1 file changed, 8 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 34c125e2d90c..5f61676e8fb8 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3673,14 +3673,6 @@ static inline unsigned long nsecs_to_jiffies_timeout(const u64 n)
return min_t(u64, MAX_JIFFY_OFFSET, nsecs_to_jiffies64(n) + 1);
}
-static inline unsigned long
-timespec_to_jiffies_timeout(const struct timespec *value)
-{
- unsigned long j = timespec_to_jiffies(value);
-
- return min_t(unsigned long, MAX_JIFFY_OFFSET, j + 1);
-}
-
/*
* If you need to wait X milliseconds between events A and B, but event B
* doesn't happen exactly after event A, you record the timestamp (jiffies) of
--
2.9.0
get_seconds() is deprecated because of the 32-bit time overflow
in y2038/y2106 on 32-bit architectures. The way it is used in
cper_next_record_id() causes an overflow in 2106 when unsigned UTC
seconds overflow, even on 64-bit architectures.
This starts using ktime_get_real_seconds() to give us more than 32 bits
of timestamp on all architectures, and then changes the algorithm to use
39 bits for the timestamp after the y2038 wrap date, plus an always-1
bit at the top. This gives us another 127 epochs of 136 years, with
strictly monotonically increasing sequence numbers across boots.
This is almost certainly overkill, but seems better than just extending
the deadline from 2038 to 2106.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/firmware/efi/cper.c | 17 +++++++++++++++--
1 file changed, 15 insertions(+), 2 deletions(-)
diff --git a/drivers/firmware/efi/cper.c b/drivers/firmware/efi/cper.c
index 3bf0dca378a6..b73fc4cab083 100644
--- a/drivers/firmware/efi/cper.c
+++ b/drivers/firmware/efi/cper.c
@@ -48,8 +48,21 @@ u64 cper_next_record_id(void)
{
static atomic64_t seq;
- if (!atomic64_read(&seq))
- atomic64_set(&seq, ((u64)get_seconds()) << 32);
+ if (!atomic64_read(&seq)) {
+ time64_t time = ktime_get_real_seconds();
+
+ /*
+ * This code is unlikely to still be needed in year 2106,
+ * but just in case, let's use a few more bits for timestamps
+ * after y2038 to be sure they keep increasing monotonically
+ * for the next few hundred years...
+ */
+ if (time < 0x80000000)
+ atomic64_set(&seq, (ktime_get_real_seconds()) << 32);
+ else
+ atomic64_set(&seq, 0x8000000000000000ull |
+ ktime_get_real_seconds() << 24);
+ }
return atomic64_inc_return(&seq);
}
--
2.9.0
The API got renamed for consistency and the old name will
be removed once the last users are gone.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
include/sound/pcm.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/sound/pcm.h b/include/sound/pcm.h
index 5d5daa190b08..f566611f0fef 100644
--- a/include/sound/pcm.h
+++ b/include/sound/pcm.h
@@ -1187,7 +1187,7 @@ static inline void snd_pcm_gettime(struct snd_pcm_runtime *runtime,
ktime_get_ts64(tv);
break;
case SNDRV_PCM_TSTAMP_TYPE_MONOTONIC_RAW:
- getrawmonotonic64(tv);
+ ktime_get_raw_ts64(tv);
break;
default:
#ifdef CONFIG_SND_TSTAMP_REALTIME
--
2.9.0
The two do the same thing, but we want to remove getnstimeofday64()
to have a more consistent interface.
It would be nice to use a monotonic clocksource here rather than
'real' time, but that would break the user interface.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/usb/mon/mon_bin.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/drivers/usb/mon/mon_bin.c b/drivers/usb/mon/mon_bin.c
index ad2c082bd0fb..ac2b4fcc265f 100644
--- a/drivers/usb/mon/mon_bin.c
+++ b/drivers/usb/mon/mon_bin.c
@@ -95,8 +95,8 @@ struct mon_bin_hdr {
unsigned short busnum; /* Bus number */
char flag_setup;
char flag_data;
- s64 ts_sec; /* getnstimeofday64 */
- s32 ts_usec; /* getnstimeofday64 */
+ s64 ts_sec; /* ktime_get_real_ts64 */
+ s32 ts_usec; /* ktime_get_real_ts64 */
int status;
unsigned int len_urb; /* Length of data (submitted or actual) */
unsigned int len_cap; /* Delivered length */
@@ -497,7 +497,7 @@ static void mon_bin_event(struct mon_reader_bin *rp, struct urb *urb,
struct mon_bin_hdr *ep;
char data_tag = 0;
- getnstimeofday64(&ts);
+ ktime_get_real_ts64(&ts);
spin_lock_irqsave(&rp->b_lock, flags);
@@ -637,7 +637,7 @@ static void mon_bin_error(void *data, struct urb *urb, int error)
unsigned int offset;
struct mon_bin_hdr *ep;
- getnstimeofday64(&ts);
+ ktime_get_real_ts64(&ts);
spin_lock_irqsave(&rp->b_lock, flags);
--
2.9.0
I noticed that the mic driver passes a 'struct timespec64' as part of
a message into an attached device, where it is used to set the current
system time.
This won't actually work if one of the two sides runs a 32-bit kernel and
the other runs a 64-bit kernel, since the structure layout is different
between the two.
I found this while replacing calls to the deprecated do_settimeofday64()
interface with the modern ktime_get_real_ts() variant, but it seems
appropriate to address both at the same time here.
To make sure we have a sane structure, let's define our own structure
using the layout of the 64-bit kernel.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/misc/mic/cosm/cosm_main.h | 5 ++++-
drivers/misc/mic/cosm/cosm_scif_server.c | 6 +++++-
drivers/misc/mic/cosm_client/cosm_scif_client.c | 6 +++++-
3 files changed, 14 insertions(+), 3 deletions(-)
diff --git a/drivers/misc/mic/cosm/cosm_main.h b/drivers/misc/mic/cosm/cosm_main.h
index f01156fca881..aa78cdf25e40 100644
--- a/drivers/misc/mic/cosm/cosm_main.h
+++ b/drivers/misc/mic/cosm/cosm_main.h
@@ -45,7 +45,10 @@ struct cosm_msg {
u64 id;
union {
u64 shutdown_status;
- struct timespec64 timespec;
+ struct {
+ u64 tv_sec;
+ u64 tv_nsec;
+ } timespec;
};
};
diff --git a/drivers/misc/mic/cosm/cosm_scif_server.c b/drivers/misc/mic/cosm/cosm_scif_server.c
index 05a63286741c..e94b7eac4a06 100644
--- a/drivers/misc/mic/cosm/cosm_scif_server.c
+++ b/drivers/misc/mic/cosm/cosm_scif_server.c
@@ -179,9 +179,13 @@ static void cosm_set_crashed(struct cosm_device *cdev)
static void cosm_send_time(struct cosm_device *cdev)
{
struct cosm_msg msg = { .id = COSM_MSG_SYNC_TIME };
+ struct timespec64 ts;
int rc;
- getnstimeofday64(&msg.timespec);
+ ktime_get_real_ts64(&ts);
+ msg.timespec.tv_sec = ts.tv_sec;
+ msg.timespec.tv_nsec = ts.tv_nsec;
+
rc = scif_send(cdev->epd, &msg, sizeof(msg), SCIF_SEND_BLOCK);
if (rc < 0)
dev_err(&cdev->dev, "%s %d scif_send failed rc %d\n",
diff --git a/drivers/misc/mic/cosm_client/cosm_scif_client.c b/drivers/misc/mic/cosm_client/cosm_scif_client.c
index beafc0da4027..225078cb51fd 100644
--- a/drivers/misc/mic/cosm_client/cosm_scif_client.c
+++ b/drivers/misc/mic/cosm_client/cosm_scif_client.c
@@ -63,7 +63,11 @@ static struct notifier_block cosm_reboot = {
/* Set system time from timespec value received from the host */
static void cosm_set_time(struct cosm_msg *msg)
{
- int rc = do_settimeofday64(&msg->timespec);
+ struct timespec64 ts = {
+ .tv_sec = msg->timespec.tv_sec,
+ .tv_nsec = msg->timespec.tv_nsec,
+ };
+ int rc = do_settimeofday64(&ts);
if (rc)
dev_err(&client_spdev->dev, "%s: %d settimeofday rc %d\n",
--
2.9.0
getnstimeofday64() is just a wrapper around the ktime accessor, so
we should use that directly.
I considered using ktime_get_boottime_ts64() (to avoid leap second
problems) or ktime_get_real_seconds() (to simplify the calculation,
but in the end concluded that the existing interface is probably
the most appropriate in this case.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
drivers/rtc/class.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/rtc/class.c b/drivers/rtc/class.c
index d37588f08055..7fa32c922617 100644
--- a/drivers/rtc/class.c
+++ b/drivers/rtc/class.c
@@ -68,7 +68,7 @@ static int rtc_suspend(struct device *dev)
return 0;
}
- getnstimeofday64(&old_system);
+ ktime_get_real_ts64(&old_system);
old_rtc.tv_sec = rtc_tm_to_time64(&tm);
@@ -110,7 +110,7 @@ static int rtc_resume(struct device *dev)
return 0;
/* snapshot the current rtc and system time at resume */
- getnstimeofday64(&new_system);
+ ktime_get_real_ts64(&new_system);
err = rtc_read_time(rtc, &tm);
if (err < 0) {
pr_debug("%s: fail to read rtc time\n", dev_name(&rtc->dev));
--
2.9.0
Commit b5793b0d92c9 ("posix-timers: Make compat syscalls depend on
CONFIG_COMPAT_32BIT_TIME") added support for building the nanosleep
compat system call on 32-bit architectures, but missed one change
in nanosleep_copyout(), which would trigger a BUG() as soon as we
switch any architecture over to use it.
This makes sure the TT_COMPAT handler is available when we need it.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
kernel/time/hrtimer.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 055a4a728c00..3e93c54bd3a1 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1659,7 +1659,7 @@ EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
int nanosleep_copyout(struct restart_block *restart, struct timespec64 *ts)
{
switch(restart->nanosleep.type) {
-#ifdef CONFIG_COMPAT
+#ifdef CONFIG_COMPAT_32BIT_TIME
case TT_COMPAT:
if (compat_put_timespec64(ts, restart->nanosleep.compat_rmtp))
return -EFAULT;
--
2.9.0
The machine check timestamp uses get_seconds(), which returns an 'unsigned long'
number that might overflow on 32-bit architectures (in the distant future)
and is therefore deprecated.
The normal replacement would be ktime_get_real_seconds(), but that needs to
use a sequence lock that might cause a deadlock if the mce happens at just
the wrong moment. The __ktime_get_real_seconds() skips that lock and is
safer here, but has a miniscule risk of returning the wrong time when we read
it on a 32-bit architecture at the same time as updating the epoch, i.e.
from before y2106 overflow time to after, or vice versa.
This seems to be an acceptable risk in this particular case, and is the
same thing we do in kdb.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
arch/x86/kernel/cpu/mcheck/mce.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index e4cf6ff1c2e1..b887415652ed 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -123,8 +123,8 @@ void mce_setup(struct mce *m)
{
memset(m, 0, sizeof(struct mce));
m->cpu = m->extcpu = smp_processor_id();
- /* We hope get_seconds stays lockless */
- m->time = get_seconds();
+ /* need the internal __ version to avoid deadlocks */
+ m->time = __ktime_get_real_seconds();
m->cpuvendor = boot_cpu_data.x86_vendor;
m->cpuid = cpuid_eax(1);
m->socketid = cpu_data(m->extcpu).phys_proc_id;
--
2.9.0
time_t and get_seconds() are deprecated because they will overflow on
32-bit architectures in the future. This is not a problem on 64-bit s390,
but we should use proper interfaces anyway.
Besides moving to the time64_t based interface, the CLOCK_MONOTONIC
based ktime_get_seconds() is preferred for kernel internal timekeeping
because it does not behave in unexpected ways during leap second changes
or settimeofday() calls.
Signed-off-by: Arnd Bergmann <arnd(a)arndb.de>
---
arch/s390/hypfs/inode.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/s390/hypfs/inode.c b/arch/s390/hypfs/inode.c
index 06b513d192b9..c681329fdeec 100644
--- a/arch/s390/hypfs/inode.c
+++ b/arch/s390/hypfs/inode.c
@@ -36,7 +36,7 @@ struct hypfs_sb_info {
kuid_t uid; /* uid used for files and dirs */
kgid_t gid; /* gid used for files and dirs */
struct dentry *update_file; /* file to trigger update */
- time_t last_update; /* last update time in secs since 1970 */
+ time64_t last_update; /* last update, CLOCK_MONOTONIC time */
struct mutex lock; /* lock to protect update process */
};
@@ -52,7 +52,7 @@ static void hypfs_update_update(struct super_block *sb)
struct hypfs_sb_info *sb_info = sb->s_fs_info;
struct inode *inode = d_inode(sb_info->update_file);
- sb_info->last_update = get_seconds();
+ sb_info->last_update = ktime_get_seconds();
inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
}
@@ -179,7 +179,7 @@ static ssize_t hypfs_write_iter(struct kiocb *iocb, struct iov_iter *from)
* to restart data collection in this case.
*/
mutex_lock(&fs_info->lock);
- if (fs_info->last_update == get_seconds()) {
+ if (fs_info->last_update == ktime_get_seconds()) {
rc = -EBUSY;
goto out;
}
--
2.9.0
The series aims at adding a new y2038-safe struct __kernel_itimerspec.
This is intended to replace the struct itimerspec at ABI level.
The series is a continuation of efforts to convert all syscalls with
time_t or time_t-derived structures to be y2038-safe.
Arnd, maybe this series can go along with the rest of syscalls that you
have in your y2038 tree?
Deepa Dinamani (3):
time: Introduce struct __kernel_itimerspec
time: Enable get/put_compat_itimerspec64 always
time: Change types to new y2038 safe __kernel_itimerspec
fs/timerfd.c | 8 ++++----
include/linux/compat.h | 9 ---------
include/linux/compat_time.h | 9 +++++++++
include/linux/syscalls.h | 10 +++++-----
include/linux/time.h | 4 ++--
include/linux/time64.h | 1 +
include/uapi/linux/time.h | 7 +++++++
kernel/compat.c | 29 -----------------------------
kernel/time/posix-timers.c | 12 +++++++-----
kernel/time/time.c | 25 +++++++++++++++++++++++--
10 files changed, 58 insertions(+), 56 deletions(-)
base-commit: 4b373f94fee5acf2ff4c1efbb3f702060379df1f
--
2.17.1