- 29 Sep, 2015 40 commits
-
-
Arnd Bergmann authored
Most of the code in llog.c and llog_cat.c is completely unused on the client and can be removed, as a preparation for removing the dt_object code. Two tricky parts are: - In llog_cat_close(), we rely on the fact that llh_flags never contains LLOG_F_ZAP_WHEN_EMPTY, because nobody ever sets that flag. - In llog_read_header(), we check the return value of the lpi_cb callback, and again we know that it cannot be LLOG_PROC_BREAK or LLOG_DEL_RECORD and can remove the respective code path. Signed-off-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
Oleg Drokin <green@linuxhacker.ru> Signed-off-by:
Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-
Oleg Drokin authored
These functions only make sense on the server. Signed-off-by:
-
Oleg Drokin authored
This is only used on the servers to evict clients. Signed-off-by:
-
Arnd Bergmann authored
The CFS_TIME_T macro serves no real purpose as we stopped using time_t and changed over to time64_t, so we can remove the last remaining uses of this. Two uses of this macro are incorrect and refer to jiffies values rather than time_t, and one refers to an inode timespec that gets changed separately. Signed-off-by:
-
Oleg Drokin authored
This is only used on the server to keep track of alive clients and feeds into ping evictor (that was removed from the client code). Also remove struct obd's obd_exports_timed and struct obd_export's exp_obd_chain_timed used to keep track of that Signed-off-by:
-
Arnd Bergmann authored
More code that makes no sense on the client and that can be removed without replacement. Signed-off-by:
-
Arnd Bergmann authored
A bunch of API functions deal with time values but are now completely unused in lustre. This removes them in order to remove all references to time_t from the header files. Signed-off-by:
-
Arnd Bergmann authored
This is a simple cleanup that I did after noticing that the abstraction for the timer functions in completely pointless, and the one user (ptlrpc) can just as well call the native Linux functions. For good measure, this also removes the empty libcfs_arch_init() and libcfs_arch_cleanup() functions that are defined in the same file. Signed-off-by:
-
Arnd Bergmann authored
lov_stripe_md_cmp lov_lum_lsm_cmp lov_lum_swab_if_needed functions have not been in used for a long time, so lets just remove them. Signed-off-by:
-
Arnd Bergmann authored
This code is never used on the client and can simply be removed. Signed-off-by:
-
Arnd Bergmann authored
All request timestamps and deadlines in lustre are recorded in time_t and timeval units, which overflow in 2038 on 32-bit systems. In this patch, I'm converting them to time64_t and timespec64, respectively. Unfortunately, this makes a relatively large patch, but I could not find an obvious way to split it up some more without breaking atomicity of the change. Also unfortunately, this introduces two instances of div_u64_rem() in the request path, which can be slow on 32-bit architectures. This can probably be avoided by a larger restructuring of the code, but it is unlikely that lustre is used in performance critical setups on 32-bit architectures, so it seems better to optimize for correctness rather than speed here. Signed-off-by:
-
Arnd Bergmann authored
Here we use an unsigned long to store the timeout for gc, which is probably safe until 2106, but this patch converts it to use ktime_get_real_seconds() and time64_t for consistency. Signed-off-by:
-
Arnd Bergmann authored
The exp_flvr_expire and imp_sec_expire are defined as 'unsigned long', which doesn't overflow until 2106, but to be on the safe side, this changes the code to use time64_t like we do everywhere else. Signed-off-by:
-
Arnd Bergmann authored
These three are timestamps that are sent over the wire in mdc_lib and the obd logging 64-bit values, but are generated using the 32-bit get_seconds() function, which will eventually overflow. Changing them to use 64-bit ktime_get_real_seconds() solves the problem. Signed-off-by:
-
Arnd Bergmann authored
The ll_setattr_raw() function prints the new inode timestamps along with the current time using '%lu', which overflows in 2106. This changes the printing of the current time for now, the other two will change when we migrate the VFS code to use 64-bit timestamps. Signed-off-by:
-
Arnd Bergmann authored
This adapts the format string and get_seconds() call to not overflow in 2038 in the libcfs_debug_dumplog_internal() function. Signed-off-by:
-
Arnd Bergmann authored
ptlrpc_enc_page_pool computes time deltas using 'long' values from get_seconds(). This is probably safe beyond y2038, but it's better to go use monotonic times and 64-bit here for consistency. Signed-off-by:
-
Arnd Bergmann authored
The lustre selftest code has multiple time stamps that are kept as 'time_t' or 'unsigned long' and can therefore overflow on 32-bit systems. This changes the code to use time64_t instead. Signed-off-by:
-
Arnd Bergmann authored
The ni_last_alive member of lnet_ni uses a 'long' to store a timestamp, which breaks on 32-bit systems in 2038. This changes it to use time64_t and the respective functions for it. Signed-off-by:
-
Arnd Bergmann authored
The connection starting/failing time stamps will overflow in 2038 on 32-bit machines, so we need to use time64_t and ktime_get_real_seconds() instead. Signed-off-by:
-
Arnd Bergmann authored
The llite debugfs interface contains timestamps that are computed from timeval, which overflows in 2038 on 32-bit systems. This changes the output to use a timespec64 type to avoid the overflow. I also change the format to print the sub-second portion as 9 digits (nanoseconds) for clarity, rather than printing six digits without leading zeroes. Signed-off-by:
-
Arnd Bergmann authored
The adaptive timeout handling stores absolute times in 32-bit time_t quantities, which will overflow in 2038. This changes it to use time64_t. Signed-off-by:
-
Arnd Bergmann authored
This time is only printed in debugfs, and can be easily converted to 64-bit to avoid overflowing on 32-bit systems in 2038. Signed-off-by:
-
Arnd Bergmann authored
The ldlm pool calculates elapsed time by comparing the previous and current get_seconds() values, which is unsafe on 32-bit machines after 2038. This changes the code to use time64_t and ktime_get_real_seconds(), keeping the 'real' instead of 'monotonic' time because of the debug prints. Signed-off-by:
-
Arnd Bergmann authored
The wire protocol for the ping uses a 64-bit seconds/microseconds pair, but this won't work when one side uses a 32-bit timeval to look up the current time beyond 2038. This changes the code to use ktime_get_real_ts64() to create a timestamp that has the right format on all machines. Signed-off-by:
-
Arnd Bergmann authored
The s2dhms computes the day/hour/minute/second values from a time_t, which stops working in 2038. This changes the code to take a time64_t argument, and use div_u64_rem() to implement the first division. Signed-off-by:
-
Arnd Bergmann authored
Several functions in Lustre call cfs_srand with do_gettimeofday as the seed to get a pseudo-random number. There is no bug here, but changing it to use ktime_get_ts64() gets us closer to deprecating do_gettimeofday() and makes it slightly more random. Affected functions are: lnet_shuffle_seed, init_lustre_lite and class_handle_init Signed-off-by:
-
Arnd Bergmann authored
ibn_incarnation is a 64-bit value, but using timeval to compute it will cause an overflow in 2038. This changes it to use ktime_get_real_ts64() instead. Signed-off-by:
-
Arnd Bergmann authored
The cfs_duration_usec() function has a timeval as its output, which we want to avoid in general because of the y2038 problem. There are only two locations remaining in lustre, so we can for now replace one with jiffies_to_timeval(), which is a generic kernel function that does the same thing, the other can just use jiffies_to_usecs() and completely avoid the timeval. This is not a full solution yet, but it's a small step that lets us build a larger portion of lustre without this reference to timeval in a header file, and avoid triggering automated checking tools that wants to warn about timeval. Signed-off-by:
-
Arnd Bergmann authored
The cfs_duration_sec() converts a relative jiffies value into seconds, and returns that number as a time_t. We know that a 32-bit type is enough here, because the result is order of magnitudes smaller than the difference in jiffies that is also expressed as a 'long', so we can safely replace the time_t type with long as well. Signed-off-by:
-
Arnd Bergmann authored
This ioctl function passes a 64-bit time argument but then performs a computation with a 32-bit get_seconds() value. In order to avoid overflow here, this changes the code to use 64-bit math and ktime_get_real_seconds(). Signed-off-by:
-
Arnd Bergmann authored
The lnet_eq_wait_locked tries to wait for time to pass or an event to wake up the wait queue. The entire logic seems to be a very elaborate reimplementation of wait_event(). I'm not trying to clean up the entire logic here, but this at least gets rid of the multi-way conversion between miliseconds, timeval and jiffies. Signed-off-by:
-
Arnd Bergmann authored
The rq_at_index member of ptlrpc_request is incorrectly declared as time_t, when it is only used as an index into an array, and assigned from a __u32 variable. This changes the type to u32, so we can kill off another use of time_t. Signed-off-by:
-
Arnd Bergmann authored
process_param2_config() tries to print how much time has passed across a call_usermodehelper() function, and uses struct timeval for that. We want to remove this structure, so this is better expressed in terms of ktime_t and ktime_us_delta(). Signed-off-by:
-
Arnd Bergmann authored
The l_last_activity struct member is used to keep track lock hold times, and it is printed for debugging purposes. For the elapsed time, we can use 'long' here, but it's better to use time64_t for storing the real time to avoid an overflow in 2038. Signed-off-by:
-
Arnd Bergmann authored
Some lustre debugfs files contain the current time. Make sure we use a format here that does not overflow in 2038. Signed-off-by:
-
Arnd Bergmann authored
The lustre tracefile has a timestamp defined as __u32 ph_sec; __u64 ph_usec; which seems completely backwards, as the microsecond portion of a time stamp will always fit into a __u32 value, while the second portion will overflow in 2038 or 2106 (in case of unsigned seconds). Changing this would unfortunately change the format in an incompatible way, breaking all existing user space tools that access the data. This uses ktime_get_real_ts64() to replace the insufficient do_gettimeofday() and then truncates the seconds portion to an u32 type, along with comments to explain the result. A possible alternative would be the use of ktime_get_ts64() to read a monotonic timestamp that never overflows, but this would trigger a check in user space 'hdr->ph_sec < (1 << 30)' that attempts to ensure that the values are within a reasonable range. Signed-off-by: -
Arnd Bergmann authored
We want to get rid of all uses of time_t, and it turns out that obd.h contains a bunch of them that are completely unused. This removes those structures, along with a couple of other structures and functions in the same file that also turned out to be unused. Signed-off-by:
-
Arnd Bergmann authored
Lustre has 64-bit timestamps in its network data structures, but on 32 bit systems, it converts them directly into time_t, which is 32 bit wide. This changes the code to use 64-bit time stamps for files. The Linux VFS code still uses time_t though, and will be changed in a separate patch series. Signed-off-by:
-
Arnd Bergmann authored
The recently introduced lnet_peer_set_alive() function uses get_seconds() to read the current time into a shared variable, but all other uses of that variable compare it to jiffies values. This changes the current use to jiffies as well for consistency. Signed-off-by:
Oleg Drokin <oleg.drokin@intel.com> Signed-off-by:
-
