Before this major redesign, the slabs interface likely
was our messiest 2nd generation attempt at opaqueness.
Beyond the standard 'new', 'ref' & 'unref', there were
a total of 12 exported functions. Now, there are four.
The 1st step was to remove several of those functions.
These were quick to go since they were not used (yet):
. procps_slabnode_count
. procps_slabnode_getname
. procps_slabnode_getstack
Then, the following were internalized so users needn't
be burdened with implementation details in the future:
. procps_slabinfo_read (renamed: read_slabinfo_failed)
. procps_slabnode_stacks_alloc (renamed: stacks_alloc)
Still others evolved into the minimal interface we had
strived for in the other upgraded 3rd generation APIs:
. procps_slabnode_get -----------> procps_slabinfo_get
. separate stack_alloc/fill --> procps_slabinfo_select
. separate stacks_alloc/fill ---> procps_slabinfo_reap
. procps_slabnode_stacks_sort --> procps_slabinfo_sort
------------------------------------------------------
Beyond those reductions, the major modifications were:
. This API tries to be as forgiving as possible and as
such won't throw errors when a caller request makes no
sense. For example, if a 'get' or 'select' requested a
SLABNODE item (with no current means to id that node),
results will be zero. By the same token, should 'reap'
include a global SLABS item (meaning those values will
be duplicated in *every* node stack) it'll be allowed.
. If the above behavior is undesired, a new #define of
ENFORCE_LOGICAL can be used to restrict certain items.
. Permission problems will now be caught at 'new' time
thanks to a priming 'read' call. That read also serves
to make DELTA values potentially useful at 1st access.
. Separate slab/slabnode enumerators were consolidated
into one, simplifying validation & the results struct.
. Several internal parameter checks were relaxed since
they were already checked by the caller. Besides if we
cannot trust our own code we might as well hang it up.
. That sort provision was made more efficient and will
offer the ascending choice, in addition to descending.
------------------------------------------------------
Lastly, some additional thoughts regarding the future:
. It would not be difficult to expand 'select' to also
accept a nodeid, or to clone it as 'select_node'. And,
should the same be extended to 'get', a results struct
could be returned instead of signed long accommodating
the extra data type(s) like a node name (string data).
. The 'get' function is not currently affected by that
define ENFORCE_LOGICAL. However, at some future point
perhaps -EINVAL would be more appropriate than a zero.
Signed-off-by: Jim Warner <james.warner@comcast.net>
Where possible, libprocps files convey the name of the
actual source pseudo file under the '/proc' directory.
This patch brings slab into line with such a standard.
Signed-off-by: Jim Warner <james.warner@comcast.net>
The original test did not NULL the unallocated info pointer. This
was incorrectly calling the procps_pids_new() function.
Signed-off-by: Craig Small <csmall@dropbear.xyz>
This interface represented a 2nd generation attempt at
the opaque newlib approach. In other words, it did not
involve the 1st generation 'chains'. Instead, 'stacks'
were employed. But the interface wasn't user friendly.
Users were required to create their own stacks, before
calling 'getstack' to retrieve multiple results with a
single call. Even worse, sometimes 'read' was required
before calling 'get' when working with single results.
So this commit represents the 3rd generation approach.
We eliminate the burden of 'read' and creating stacks.
Rather, beyond those standard 'new', 'ref' and 'unref'
functions, we'll offer just 'get' (single result) plus
a 'select' function (for multiple results in 1 stack).
And along the way, this commit vastly expands the data
extracted from /proc/vmstat. All values that currently
exist (and their delta equivalents) are now available.
Deltas were included for everything because there's no
real runtime costs beyond using a little extra memory.
The only problem is a lack of documentation for all of
those fields, as is reflected in the references below.
Oh well, maybe someday someone will dig through kernel
sources & finally plug that rather large document gap.
[ as an aside, rather than using a 'strcmp' approach ]
[ when parsing the /proc/vmstat file, as is found in ]
[ the <meminfo> module, we exploit those hash search ]
[ provisions that are found in the <search.h> header ]
Reference(s):
http://www.spinics.net/lists/linux-man/msg09096.htmlhttp://www.linuxinsight.com/proc_vmstat.html
Signed-off-by: Jim Warner <james.warner@comcast.net>
With an eye to the future, rather than managing a long
switch/case construct in the procps_stat_get function,
we'll adopt the approach used in the <meminfo> module.
By making 'get' processing table driven too, just like
'select' is already, that may ease future maintenance.
Along the way a rather large boo-boo was fixed dealing
with that /proc/stat 'procs_running' field which crept
in with the commit referenced below. It wasn't spelled
correctly and thus was never captured as PROC_RUNNING.
Reference(s):
commit abc71a46ad
Signed-off-by: Jim Warner <james.warner@comcast.net>
When this interface was normalized/standardized, under
the commit referenced below, the parameters were never
validated in the 'get' function. Let's plug that hole.
Reference(s):
commit 407f1b71de
Signed-off-by: Jim Warner <james.warner@comcast.net>
Update the pids tests for tweaked API. Unsure why the toomany
test doesn't work, may point to a problem in the change as it used
to trigger, or the library has flexibility.
In that reference below a specific systemd problem was
fixed in the commit shown. However lurking deep within
the <pids> interface was yet one final case where NULL
could be returned, involving 'strv' and the following:
. a user requested both a single string vector (always
returned as a normal string) & the vectorized version,
as with PROCPS_PIDS_CMDLINE and PROCPS_PIDS_CMDLINE_V.
. a user simply duplicated some vectorized enum items.
The root of that NULL problem is the fact those single
string vectors shared the same proc_t field with their
true vectorized version. So while multiple occurrences
for most strings could be satisfied with strdup versus
the normal ownership usurpation, those true vectorized
fields could not be quite so easily copied/duplicated.
Thus newlib chose to return a NULL result.strv pointer
under either of the above scenarios (which perhaps was
just a user boo-boo in the first place). In any event,
the NULL was a potential for true string vectors only.
Now, since newlib is the sole caller into the readproc
module, separate fields have been created for what are
just normal strings (never vectorized) and those which
remain the true vectorized versions. And, former flags
which only worked if combined, now act as stand alone.
Thus, both PROCPS_PIDS_CMDLINE & PROCPS_PIDS_CMDLINE_V
can be used simultaneously (as they should have been).
Also with this patch, items which a user duplicates in
the stack (beyond the first such item) will return the
the string "[ duplicate ENUM_ID ]". This practice will
apply to both single strings and true vectorized ones.
In addition to informing users of their error, it will
also mean potential NULLs need now never be a concern.
Reference(s);
http://www.freelists.org/post/procps/systemd-binary-vs-library
commit 0580a7b4c6
Signed-off-by: Jim Warner <james.warner@comcast.net>
Calls to free() have now been reintroduce in the new()
function to quiet coverity warnings. Those free's were
removed originally as that library 'new' was returning
with a fatal error and a caller should end abnormally.
Plus, it is virtually impossible to fail a malloc call
under linux. And lastly, they required braces with the
if statement making the code considerably less pretty.
[ commit also includes 2 unrelated whitespace tweaks ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
A potential conflict arises should should some program
attempt to access systemd data using our library built
without that support (--disable-systemd). And while we
could argue that all callers should check against NULL
string pointers, it's not expected of libprocps users.
So we'll guarantee 'em valid string consisting of "?".
Reference(s):
http://www.freelists.org/post/procps/systemd-binary-vs-library
Signed-off-by: Jim Warner <james.warner@comcast.net>
Rather than adopt an approach offered below, involving
changing all enumerator names to convey types, perhaps
this much more modest change is a worthwhile 1st step.
By just changing the comments to 'right justification'
it dramatically improves readability, at least it does
to my eye. Perhaps such an approach can postpone that
massive alternative effort for the foreseeable future.
Reference(s):
http://www.freelists.org/post/procps/newlib-drip-drip,4
Signed-off-by: Jim Warner <james.warner@comcast.net>
This is just beginning an effort to minimize/normalize
the sheer variety of results types in use for our API.
In taking these first baby steps, a few anomalies were
found. There'll no doubt be many more yet to discover.
. the _FLT_ (fault) fields were already signed long in
the proc_t (even though their sscanf format used %lu).
. although strtoul will alway return an unsigned long,
all of the _VM_ fields were made signed long just like
other memory fields (& signed sorts more efficiently).
Reference(s):
http://www.freelists.org/post/procps/newlib-drip-drip,4
Signed-off-by: Jim Warner <james.warner@comcast.net>
This represents a rather major interface redesign. The
following highlights most of the changes/enhancements.
. The 'read' interface (employed by pgrep & pidof) saw
the biggest change. The 'open', 'next' and 'shut' guys
all went bye-bye, replaced by a single 'get' function.
. The items specified at 'new' time no longer serve as
the maximum. In fact, items & numitems are now treated
as optional, should callers prefer to wait until later
when the 'reset' function would then become mandatory.
. Even at 'reset' time, the stacks are not tied to any
sort of maximum. They will grow dynamically as needed.
. The order of some parameters was changed to parallel
that found in our other APIs. Specifically, when items
& numitems are needed they're specified in that order.
. A user will no longer be prevented from concurrently
employing any accessor functions. In other words, that
'get' (old 'read') won't preclude 'reap' and 'select'.
. A duplicate enumerator was found dealing with locked
resident pages. So, the name VM_LOCK was eliminated in
favor of VM_RSS_LOCKED, which is way more descriptive.
. The struct address returned to callers following any
reap() or select() is now more sharable as pids_fetch.
. Some input parameter names were changed to make them
more descriptive of the intended purpose/requirements.
------------------------------------------------------
Internally, there were numerous implementation changes
made that did not directly impact any potential users.
. That #define FPRINT_STACKS was eliminated along with
the associated supporting function and its invocation.
. Addresses returned following 'reap' or 'select' will
now be NULL delimited, so one has the option of stacks
access via the total count or this new NULL fencepost.
. Input params were simplified and generalized in both
oldproc_open() & close() to enable more than 1 PROCTAB
to be open simultaneously, which was required for get.
. The PROCPS_PIDS_logical_end enum was relocated after
the Item_table making the need to keep it synchronized
more apparent (if the table expands it's right there).
. The 'Public function' section of the source file was
subdivided into 1) the three basic required functions;
and 2) functions that can sometimes vary between APIs.
Signed-off-by: Jim Warner <james.warner@comcast.net>
if (info->flags | PROC_UID)
Something OR a non-zero constant is always true.
Looks like it should be and'ed for the standard flag masking
pattern.
References:
Coverity #99118
Signed-off-by: Craig Small <csmall@dropbear.xyz>
A while loop used n >=0
A branch inside the loop had a test for n<0
That's never going to happen!
So the branch must go. There is similiar code outside
the branch. Yes, I miss pythons while else here.
Also cleaned up the ***-awful indentation this function had.
No wonder that branch lived for so long there.
References:
Coverity #99119
Signed-off-by: Craig Small <csmall@dropbear.xyz>
procps_slabnode_stacks_sort() would check if stacks was NULL
*after* the pointer was derefenced in the initialisation of
the variable p.
p is now assigned after the check.
References:
Coverity #99160
Because of the vast quantities of virtual memory which
may be allocated, it initially seemed like a good idea
to provide for a widest possible range through the use
of a 'ull_int' result type. However, on second thought
the implementation was a bit flawed for these reasons:
. that underlying meminfo_data variable 'VmallocTotal'
is 'unsigned long' not a required 'unsigned long long'
. there wasn't a convenient way to value it since each
variable was set with a strtoul() call, not strtoull()
So this patch will standardize on the 'ul_int' results
type (and reduce the associated delta to 's_int' too).
For now, we'll rely on protections under a 64-bit arch
where a 'ull_int' & 'ul_int' yield identical capacity.
Signed-off-by: Jim Warner <james.warner@comcast.net>
A recent buglet in the <meminfo> interface, where many
delta values could legitimately be negative, suggested
a review of this module where a minus was forced to 0.
As it turns out, there are two delta items that indeed
could be negative. They involve the current processes.
So henceforth, DELTA_PROC_BLOCKED & DELTA_PROC_RUNNING
will now be allowed to go negative. I believe that all
other items can only grow. But, if we find more later,
at least the adaptation approach has been established.
[ this whole business of checking for less than zero ]
[ harks back to an old kernel anomaly where negative ]
[ tics were sometimes experienced. top was affected. ]
Signed-off-by: Jim Warner <james.warner@comcast.net>
When the code for DELTA values (among other stuff) was
stolen from the recently revised <stat> interface, the
concept of ever growing values was propagated too. But
here we must manage both growing and shrinking values.
Thus former protections against a negative delta don't
have any place in this module and are hereby banished.
Signed-off-by: Jim Warner <james.warner@comcast.net>
Thus function returns a pointer, not an int, so if there is
an error return NULL and not -errno.
proc/meminfo.c: In function ‘procps_meminfo_select’:
proc/meminfo.c:994:20: warning: return makes pointer from integer
without a cast [-Wint-conversion]
return -ENOMEM;
References:
commit 407f1b71de
This represents the refinement of this interface after
the <stat> API was redesigned. We now follow a pattern
of 'get' for single item retrieval & 'select' for when
multiple items are desired, with just 1 function call.
And again following the <stat> lead this interface now
provides for delta values encompassing most items. The
reason I went cuckoo nuts with those deltas is because
they are essentially free. At the cost of a little RAM
and just one memcpy there's no other price to be paid.
Signed-off-by: Jim Warner <james.warner@comcast.net>
In the commit referenced below it was acknowledged the
procps_stat_get() function will intentionally restrict
returned values by excluding DELTA items among others.
Upon reflection that decision seemed rather arbitrary,
and so has now been lifted. And, while the initial use
of a DELTA will return 0, subsequent calls to that get
function will now return the difference (providing one
second, at least, has elapsed between any such calls).
Reference(s):
. reference to design decisions
commit abc71a46ad
Signed-off-by: Jim Warner <james.warner@comcast.net>
This commit is just my usual tweaking after an initial
submission, once the dust has settled & some more desk
checking was performed. Here are the changes included:
. get function need not check the 'stat_was_read' flag
( first time in save_sec == 0 so branch always taken )
. some table logic appeared outside the proper section
. cleanup_stack's loop made to work like assign_result
. eliminated an include of procps.h from stat.h header
. removed 'ext_numitems' from the stacks_extent struct
. changed the alignment of three prototypes for stat.h
. reorganized the pointers for 'stacks_extents' struct
. improved types indentation in header for readability
. lastly, some really gratuitous changes made to align
- comments within that private 'stacks_alloc' function
Signed-off-by: Jim Warner <james.warner@comcast.net
This commit represents a complete redesign of the stat
interface. Gone are the confusing 8 separate accessors
along with their 2 additional read functions. In their
place we have just 3 accessors, with no read required.
That old interface also suffered an inflexibility with
respect to structures. Now we deal with an unchanging
standard 'result' struct enabling future changes where
the binary interface will no longer need to be broken.
And gone is that former unnecessary typedef, used when
dealing with jiffies. Now the standard C type is used.
Our new API also adds some brand new functionality. If
a caller plans to employ successive 'select' or 'reap'
invocations, then delta values are available (which is
actually only what that top program is interested in).
At some future point a 'sort' function could be easily
introduced to complement the 'reap' function. However,
I saw no need for it at present and so it was omitted.
There were several design decisions which everyone may
not agree with. In support I'll offer these rationals:
. The 'get' function returns a signed long long result
which means a potential loss of some significance. But
I felt the ability to distinguish actual errors (minus
values) from true zero results were worth such a risk.
. The DELTA item enumerators were also made signed and
smaller than their parents. And they are intentionally
grouped as last so as to emphasize those distinctions.
. The SYS type items were excluded from the new 'reap'
function. It would not make sense to duplicate them in
each results stack. They're limited to 'get'/'select'.
. By the same token, some items (DELTA, etc.) will not
be allowed under that 'get' routine. That function was
already open to significant internal overhead (through
subsequent calls like in vmstat.c). That is why it has
been limited via 1 second between reads of /proc/stat.
Lastly, when we finally get around to documenting this
interface there's a real potential toe stubber when it
comes to the numa node portion. The libnuma.so doesn't
really provide any means to retrieve the active nodes.
Thus, any total reported by <stat> is just the highest
node number plus one, as reported by the numa library.
Any unused/inactive nodes are identified through these
. PROCPS_STAT_TIC_ID shows as PROCPS_STAT_NODE_INVALID
By the same token after the STAT_REAP_CPUS_ONLY 'reap'
. PROCPS_STAT_TIC_NUMA_NODE = PROCPS_STAT_NODE_INVALID
Reference(s):
http://www.freelists.org/post/procps/newlib-stat-interface
Signed-off-by: Jim Warner <james.warner@comcast.net>
During work on the new <stat> interface the typedef of
jiffs was eliminated along with the clang warning. The
exact same warning in sysinfo.h now goes bye-bye also.
Reference(s):
./proc/sysinfo.h:16:28: warning: redefinition of typedef 'jiff' is a C11 feature [-Wtypedef-redefinition]
typedef unsigned long long jiff;
Signed-off-by: Jim Warner <james.warner@comcast.net>
The previous commit did not take into account operator evaluation
order so always gave 0. Fixed the macro and added some checks for
the macro.
References:
commit 9abf7d879d
Linux release numbers are x.y.z we assumed but never protected
against x> 0x7fff and y,z > 0xff before.
Linux release in 1991, 1 major release per 6 years so we're fine
until 7452, unless they do way too many minor rels (max being 39)
procps_uptime, previously just plain uptime() used to put the
uptime (as a double) in the first argument and return uptime
(as an int).
It meant if you ran
myuptime2 = procps_uptime(&myuptime1, NULL);
You might get different results for myuptime1 and myuptime2 because
they are different types.
Most library calls use the return value to return the status,
procps_uptime was in the middle.
Until now.
This function will return 0 on success. If you want (for whatever
reason) uptime as an int then cast it.
All of the procps binaries didn't use the return value for uptime
except ps which set a variable to it but never used it anywhere.
While procps_diskstat_ref and procps_diskstat_unref were declared
they actually weren't implemented in the library. Linking some
binary that used these functions gave the following errors:
CCLD vmstat
vmstat.o: In function `disksum_format':
vmstat.c:719: undefined reference to `procps_diskstat_unref'
vmstat.o: In function `diskformat':
vmstat.c:591: undefined reference to `procps_diskstat_unref'
In an attempt to normalize the 'stat' interface, we'll
first shed any reminders of the old readproc interface
by changing file names to be more descriptive & brief.
Signed-off-by: Jim Warner <james.warner@comcast.net>
After the commit referenced below the potential exists
for a SEGV (resulting from an out-of-bounds Item_table
reference if PROCPS_PIDS_physical_end is encountered).
So this patch eliminates that PROCPS_PIDS_physical_end
as no longer necessary and completes the task of using
PROCPS_PIDS_logical_end as a sole necessary fencepost.
Reference(s):
commit e7585992d9
Without breaking either ABI or API just rearrange some
stuff to provide a little more consistent source file.
[ these were prompted by some work on the <stat> API ]
[ in preparation for him borrowing some of our code. ]
Reference(s):
Signed-off-by: Jim Warner <james.warner@comcast.net>
Generalised the library API tests and created a common
test-runner. Instead of copy and pasting the same code in
each time.
First cut of pids API test, for _new so far.
Remove a remnant of this new API's evolution. Here we
no longer fill stacks, rather we reap and select them.
Signed-off-by: Jim Warner <james.warner@comcast.net>
By co-mingling both external/internal identifiers with
actual implementation code, potential future additions
to our API would have been considerably more difficult.
So, this patch will now rely solely on internal/hidden
identifiers serving as fenceposts in validation logic.
And if the following convention is used for new fields
we will maintain that enumerator alphabetic ordering a
a little longer (even though 2 user fields now don't):
. PROCPS_PIDS_XTRA_FOO_A, PROCPS_PIDS_XTRA_FOO_B, etc.
Reference(s):
http://www.freelists.org/post/procps/me-too-newlib,7
Signed-off-by: Jim Warner <james.warner@comcast.net>
Unfortunately, reading that '/proc/sys/kernel/pid_max'
file returns a newline, which we will now account for.
[ also, we should use the existing dedicated buffer ]
Reference(s):
. original 'procps_pid_length' introduction
commit ccb6ae8de1
Signed-off-by: Jim Warner <james.warner@comcast.net>
escaped_copy(): only appears in ps, moved to ps/output.c
escape_strlist() only used in escape.c made static
escape_command() used in library, made internal
procps.h no longer includes escape.h
escape_str() used by library and ps so needs to be exported
definition put into procps.h including the odd define required.
Far from ideal to have it this way, will look at it another time
to have it all in, all out or split nicer so its not in the API;
perhaps a lib/ file?
readproc still had some of the old API hanging around that
was good while we were trying it out, but its time to say
goodbye:
readproc.h removed from public procps.h header file
enum ns_type - not used - removed
get_ns_name() - not defined - removed
get_ns_id() - not defined - removed
PROCTAB proc_t only used internal to library
readproctab() and 2,3 friends not used - removed
proc_data_t was used by readproctab23() - removed
readtask() - not used - removed
read_cmdline() - copy,renamed and made generic - remove original
freeproc() - not used - removed
get_proc_status - not used - removed, however there should be
a new function created that does this. Given a PID return data
about it instead of scanning the entire procfs. Maybe it already does.
Left as internal-to-library only functions:
readproc(), readeither(), look_up_our_self(), openproc(), closeproc()
Updated libprocps.sym to export only what we use.
Now that the conditional OOMEM_ENABLE has been removed
and, after reviewing current library support, it turns
out we've been using a deprecated /proc/<pid>/oom_adj.
What we should use instead is that more recent tunable
/proc/<pid>/oom_score_adj. This new field will provide
a range of -1000 thru +1000 (former was -17 thru +15).
Reference(s):
. removal of misguided OOMEM_ENABLE
commit 64238730fa
. linux, partial oom_adj revert (Nov, 2012)
commit fa0cbbf145aabbf29c6f28f8a11935c0b0fd86fc
. linux, removal oom_adj (Oct, 2012)
commit 01dc52ebdf472f77cca623ca693ca24cfc0f1bbe
Signed-off-by: Jim Warner <james.warner@comcast.net>
The includes used to define a lot of things a library include
should not. It was also a bit messy what was exposed in the library
and what was not.
get_pid_digits -> procps_pid_length and exported correctly
MALLOC attribute move into relevant .c files
NORETURN attribute moved to relevant .c, not used in library
PURE attribute removed, it wasn't used
KLONG/KLF/STRTOUKL were fixed for long, so now just use long
HIDDEN attribute removed. It was for 3 functions. The PROCPS_EXPORT
seems to do the same (opposite) thing.
likely/unlikely removed from most places, its highly debateable
this does anything useful as CPUs have gotten smarter about branches.
Re-arranged the includes, ALL external programs should just #include
<proc/procps.h> then proc/procps.h includes headers for files that
have exported functions. procps.h and the headers it includes should
not use items that are not exportable (e.g. hidden functions or
macros) they go in procps-private.h
Multiple scanf()s use the GNU-permitted %Lu. This is not supported in
other libraries and isn't to the POSIX specification. The L modifier
is only used for floats in POSIX.
Replacing %Lu with %llu is the same for GNU libc (scanf(3) says as much)
but means other libraries will work fine.
From master commit da715e3
References:
http://pubs.opengroup.org/onlinepubs/009695399/functions/fscanf.html
Beginning with linux-4.5, the following new fields are
being added under that /proc/<pid>/status pseudo file:
. RssAnon - size of resident anonymous memory
. RssFile - size of resident file mappings
. RssShmem - size of resident shared memory
This patch just represents the initial library and top
support, sharing a commit message with 2 more patches.
p.s. locked resident memory support was also added but
isn't directly related to the kernel 4.5 enhancements.
Reference(s):
commit 1f8e41d019
Signed-off-by: Jim Warner <james.warner@comcast.net>
Summarized below, miscellany addressed in this commit:
. deleted extraneous newline(s) for consistent spacing
Signed-off-by: Jim Warner <james.warner@comcast.net>
When reference counts were added to some string fields
the 3 true string vector fields were not duplicated as
were those other fields. Instead they were supposed to
disallow a duplicate stack reference beyond the first.
However, the actual implementation gave NULL for every
true vector field whenever such items were duplicated.
More importantly, such true string vector fields never
considered references to the shared proc_t source root
which would have forced the conversion of such vectors
into a single string form via the '_CVT' library flag.
So this commit restores the intended outcome with true
string vectors. There's only 1 valid reference allowed
and duplicates and converted fields will yield a NULL.
Signed-off-by: Jim Warner <james.warner@comcast.net>
