static int SelectProcTimeCounterRangeMatch(char *name1, char *name2, time_t min, time_t max, char **names, char **line)
{
if ((min == CF_NOINT) || (max == CF_NOINT))
{
return false;
}
int i = GetProcColumnIndex(name1, name2, names);
if (i != -1)
{
time_t value = (time_t) TimeCounter2Int(line[i]);
if (value == CF_NOINT)
{
Log(LOG_LEVEL_INFO,
"Failed to extract a valid integer from %c => '%s' in process list",
name1[i], line[i]);
return false;
}
if ((min <= value) && (value <= max))
{
Log(LOG_LEVEL_VERBOSE, "Selection filter matched counter range"
" '%s/%s' = '%s' in [%jd,%jd] (= %jd secs)",
name1, name2, line[i], (intmax_t)min, (intmax_t)max, (intmax_t)value);
return true;
}
else
{
LogDebug(LOG_MOD_PS, "Selection filter REJECTED counter range"
" '%s/%s' = '%s' in [%jd,%jd] (= %jd secs)",
name1, name2, line[i],
(intmax_t) min, (intmax_t) max, (intmax_t) value);
return false;
}
}
return false;
}
static int SelectProcTimeCounterRangeMatch(char *name1, char *name2, time_t min, time_t max, char **names, char **line)
{
int i;
time_t value;
if (min == CF_NOINT || max == CF_NOINT)
{
return false;
}
if ((i = GetProcColumnIndex(name1, name2, names)) != -1)
{
value = (time_t) TimeCounter2Int(line[i]);
if (value == CF_NOINT)
{
CfOut(cf_inform, "", "Failed to extract a valid integer from %c => \"%s\" in process list\n", name1[i],
line[i]);
return false;
}
if (min <= value && value <= max)
{
CfOut(cf_verbose, "", "Selection filter matched counter range %s/%s = %s in [%ld,%ld] (= %ld secs)\n",
name1, name2, line[i], min, max, value);
return true;
}
else
{
CfDebug("Selection filter REJECTED counter range %s/%s = %s in [%ld,%ld] (= %ld secs)\n", name1, name2,
line[i], min, max, value);
return false;
}
}
return false;
}
static void MaybeFixStartTime(const char *line,
time_t pstime,
char **names,
char **fields)
{
/* Since start times can be very imprecise (e.g. just a past day's
* date, or a past year), calculate a better value from elapsed
* time, if available: */
int k = GetProcColumnIndex("ELAPSED", "ELAPSED", names);
if (k != -1)
{
const long elapsed = TimeCounter2Int(fields[k]);
if (elapsed != CF_NOINT) /* Only use if parsed successfully ! */
{
int j = GetProcColumnIndex("STIME", "START", names), ns[3];
/* Trust the reported value if it matches hh:mm[:ss], though: */
if (sscanf(fields[j], "%d:%d:%d", ns, ns + 1, ns + 2) < 2)
{
time_t value = pstime - (time_t) elapsed;
Log(LOG_LEVEL_DEBUG,
"processes: Replacing parsed start time %s with %s",
fields[j], ctime(&value));
free(fields[j]);
xasprintf(fields + j, "%ld", value);
}
}
else if (fields[k])
{
Log(LOG_LEVEL_VERBOSE,
"Parsing problem was in ELAPSED field of '%s'",
line);
}
}
}
static int SplitProcLine(const char *proc, time_t pstime,
char **names, int *start, int *end,
char **line)
{
if (proc == NULL || proc[0] == '\0')
{
return false;
}
memset(line, 0, sizeof(char *) * CF_PROCCOLS);
int prior = -1; /* End of last header-selected field. */
const size_t linelen = strlen(proc);
const char *sp = proc; /* Just after last space-separated field. */
/* Scan in parallel for two heuristics: space-delimited fields
* found using sp, and ones inferred from the column headers. */
/* TODO: we test with isspace() mostly, which makes sense as TAB
* might be used to enable field alignment; but we make no attempt
* to account for the alignment offsets this may introduce, that
* would mess up the heuristic based on column headers. Hard to
* do robustly, as we'd need to assume a tab-width and apply the
* same accounting to the headers (where start and end are indices
* of bytes, we'd need a third array for tab-induced offsets
* between headers). */
for (int i = 0; i < CF_PROCCOLS && names[i] != NULL; i++)
{
/* Space-delimited heuristic, from sp to just before ep: */
while (isspace((unsigned char) sp[0]))
{
sp++;
}
const char *ep = sp;
/* Header-driven heuristic, field from proc[s] to proc[e].
* Start with the column header's position and maybe grow
* outwards. */
int s = start[i], e;
/* If the previous field over-spilled into this one, our start
* may be under our header, not directly below the header's
* start; but only believe this if the earlier field is
* followed by one space and the subsequent field does start
* under our header. Otherwise, allow that the prior field
* may be abutting this field, or this field may have
* overflowed left into the prior field causing the prior to
* (mistakenly) think it includes the present. */
if (s <= prior &&
prior + 2 <= end[i] &&
proc[prior + 1] == ' ' &&
proc[prior + 2] != '\0' &&
!isspace((unsigned char) proc[prior + 2]))
{
s = prior + 2;
}
if (i + 1 == CF_PROCCOLS || names[i + 1] == NULL)
{
e = linelen - 1;
/* Extend space-delimited field to line end: */
while (ep[0] && ep[0] != '\n')
{
ep++;
}
/* But trim trailing hspace: */
while (isspace((unsigned char)ep[-1]))
{
ep--;
}
}
else
{
e = end[i];
/* It's possible the present field has been shunted left
* to make way for an over-sized number or date to the
* right. This can confuse the numeric check below, on
* which left-overspill is conditioned - it fails to
* recognise this field as numeric, so doesn't check for
* it overspilling to the left. Look to see whether we
* should move e left a bit: this is a conservative check,
* that'll be revisited below. */
int back = start[i + 1];
while (back > s && !isspace((unsigned char) proc[back]))
{
back--;
}
if (back > s && back <= e &&
/* back > s implies isspace(proc[back]), with no
* further space before the next field; if this is a
* single space, it's credibly the separator between
* our field, shunted left, and this over-spilled
* field: */
proc[back] == ' ' &&
!isspace((unsigned char) proc[back - 1]))
{
/* So we have a non-empty field that ends under our
* header but before e; adjust e. */
e = back - 1;
}
/* Extend space-delimited to next space: */
while (ep[0] && !isspace((unsigned char)ep[0]))
{
ep++;
}
}
/* ep points at the space (or '\0') *following* the word or
* final field. */
/* Some ps stimes may contain spaces, e.g. "Jan 25" */
if (strcmp(names[i], "STIME") == 0 &&
ep - sp == 3)
{
const char *np = ep;
while (isspace((unsigned char) np[0]))
{
np++;
}
if (isdigit((unsigned char) np[0]))
{
do
{
np++;
} while (isdigit((unsigned char) np[0]));
ep = np;
}
}
/* Numeric columns, including times, are apt to grow leftwards
* and be right-aligned; identify candidates for this by
* proc[e] being a digit. Text columns are typically
* left-aligned and may grow rightwards; identify candidates
* for this by proc[s] being alphabetic. Some columns shall
* match both (e.g. STIME). Some numeric columns may grow
* left even as far as under the heading of the next column
* (seen with ps -fel's SZ spilling left into ADDR's space on
* Linux). While widening, it's OK to include a stray space;
* we'll trim that afterwards. Overspill from neighbouring
* columns can muck up alignment, so "digit" means any
* character that can appear in a numeric field (we may need
* to tweak "alphabetic" likewise for text fields; the user
* field can, for example, have a '+' appended). */
bool overspilt = false;
/* Numeric fields may include [-.:] and perhaps other
* punctuators: */
#define IsNumberish(c) (isdigit((unsigned char)(c)) || ispunct((unsigned char)(c)))
/* Right-aligned numeric: move s left until we run outside the
* field or find space. */
if (IsNumberish(proc[e]))
{
bool number = i > 0; /* Should we check for under-spill ? */
int outer = number ? end[i - 1] + 1 : 0;
while (s >= outer && !isspace((unsigned char) proc[s]))
{
if (number && !IsNumberish(proc[s]))
{
number = false;
}
s--;
}
/* Numeric field might overspill under previous header: */
if (s < outer)
{
int spill = s;
s = outer; /* By default, don't overlap previous column. */
if (number && IsNumberish(proc[spill]))
{
outer = start[i - 1];
/* Explore all the way to the start-column of the
* previous field's header. If we get there, in
* digits-and-punctuation, we've got two numeric
* fields that abut; we can't do better than assume
* the boundary is under the right end of the previous
* column label (which is what our parsing of the
* previous column assumed). So leave s where it is,
* just after the previous field's header's
* end-column. */
while (spill > outer)
{
spill--;
if (!IsNumberish(proc[spill]))
{
s = spill + 1; /* Confirmed overlap. */
break;
}
}
}
}
overspilt = IsNumberish(proc[e + 1]);
}
#undef IsNumberish
bool abut = false;
/* Left-aligned text or numeric misaligned by overspill; move
* e right (if there's any right to move into; last column
* already reaches end of proc): */
if (proc[e + 1] &&
(overspilt || isalpha((unsigned char) proc[s])))
{
assert(i + 1 < CF_PROCCOLS && names[i + 1]);
int outer = start[i + 1]; /* Start of next field's header. */
int beyond = end[i + 1]; /* End of next field's header. */
if (beyond > linelen)
{
/* Command is shorter than its header. */
assert(i + 2 >= CF_PROCCOLS || NULL == names[i + 2]);
beyond = linelen;
}
assert(beyond >= outer);
int out = e;
do
{
out++;
} while (out < beyond && !isspace((unsigned char) proc[out]));
if (out < outer)
{
/* Simple extension to the right, no overlap: we're on
* a space just before the next field's header. */
e = out - 1;
}
else if (out == beyond)
{
/* This looks like we're actually looking at the next
* field over-spilling left so far that it reaches
* under our header; but we did previously check for
* this and amend e left-wards if it's credible; so
* we're now looking at a case where it isn't;
* probably our columns abut, with no space in
* between. The best we can do is include the text
* between columns as part of both columns :-( */
abut = true;
prior = e; /* Before adjusting it: */
e = outer - 1;
}
else
{
/* Our word appears to over-spill under the next
* header. See if that's credible. If the next is a
* left-aligned field, it'll follow ours after exactly
* a simple space. If it's right-aligned (numeric),
* it'll still follow after a single space unless it
* has enough space to fit under its header after more
* than one space. */
assert(out < beyond && isspace((unsigned char) proc[out]));
if ((proc[out] == ' ' && proc[out + 1] &&
!isspace((unsigned char) proc[out + 1])) ||
(isdigit((unsigned char) proc[beyond]) &&
isspace((unsigned char) proc[beyond + 1])))
{
e = out - 1;
}
else
{
e = outer - 1;
}
}
}
if (!abut)
{
prior = e;
}
/* Strip off any leading and trailing spaces: */
while (isspace((unsigned char) proc[s]))
{
s++;
}
/* ... but stop if the field is empty ! */
while (s <= e && isspace((unsigned char) proc[e]))
{
e--;
}
/* Grumble if the two heuristics don't agree: */
size_t wordlen = ep - sp;
if (e < s ? ep > sp : (sp != proc + s || ep != proc + e + 1))
{
char word[CF_SMALLBUF];
if (wordlen >= CF_SMALLBUF)
{
wordlen = CF_SMALLBUF - 1;
}
memcpy(word, sp, wordlen);
word[wordlen] = '\0';
char column[CF_SMALLBUF];
if (s <= e)
{
/* Copy proc[s through e] inclusive: */
const size_t len = MIN(1 + e - s, CF_SMALLBUF - 1);
memcpy(column, proc + s, len);
column[len] = '\0';
}
else
{
column[0] = '\0';
}
Log(LOG_LEVEL_VERBOSE,
"Unreliable fuzzy parsing of ps output (%s) %s: '%s' != '%s'",
proc, names[i], word, column);
}
/* Fall back on word if column got an empty answer: */
line[i] = e < s ? xstrndup(sp, ep - sp) : xstrndup(proc + s, 1 + e - s);
sp = ep;
}
/* Since start times can be very imprecise (e.g. just a past day's
* date, or a past year), calculate a better value from elapsed
* time, if available: */
int k = GetProcColumnIndex("ELAPSED", "ELAPSED", names);
if (k != -1)
{
const long elapsed = TimeCounter2Int(line[k]);
if (elapsed != CF_NOINT) /* Only use if parsed successfully ! */
{
int j = GetProcColumnIndex("STIME", "START", names), ns[3];
/* Trust the reported value if it matches hh:mm[:ss], though: */
if (sscanf(line[j], "%d:%d:%d", ns, ns + 1, ns + 2) < 2)
{
time_t value = pstime - (time_t) elapsed;
Log(LOG_LEVEL_DEBUG,
"SplitProcLine: Replacing parsed start time %s with %s",
line[j], ctime(&value));
free(line[j]);
xasprintf(line + j, "%ld", value);
}
}
}
return true;
}
static int SplitProcLine(const char *proc,
char **names, int *start, int *end,
char **line)
{
if (proc == NULL || proc[0] == '\0')
{
return false;
}
memset(line, 0, sizeof(char *) * CF_PROCCOLS);
const char *sp = proc;
/* Scan in parallel for two heuristics: space-delimited fields
* found using sp, and ones inferred from the column headers. */
for (int i = 0; i < CF_PROCCOLS && names[i] != NULL; i++)
{
/* Space-delimited heuristic, from sp to just before ep: */
while (isspace((unsigned char) sp[0]))
{
sp++;
}
const char *ep = sp;
/* Header-driven heuristic, field from proc[s] to proc[e].
* Start with the column header's position and maybe grow
* outwards. */
int s = start[i], e;
if (i + 1 == CF_PROCCOLS || names[i + 1] == NULL)
{
e = strlen(proc) - 1;
/* Extend space-delimited field to line end: */
while (ep[0] && ep[0] != '\n')
{
ep++;
}
/* But trim trailing hspace: */
while (isspace((unsigned char)ep[-1]))
{
ep--;
}
}
else
{
e = end[i];
/* Extend space-delimited to next space: */
while (ep[0] && !isspace((unsigned char)ep[0]))
{
ep++;
}
}
/* ep points at the space (or '\0') *following* the word or
* final field. */
/* Some ps stimes may contain spaces, e.g. "Jan 25" */
if (strcmp(names[i], "STIME") == 0 &&
ep - sp == 3)
{
const char *np = ep;
while (isspace((unsigned char) np[0]))
{
np++;
}
if (isdigit((unsigned char) np[0]))
{
do
{
np++;
}
while (isdigit((unsigned char) np[0]));
ep = np;
}
}
/* Numeric columns, including times, are apt to grow leftwards
* and be right-aligned; identify candidates for this by
* proc[e] being a digit. Text columns are typically
* left-aligned and may grow rightwards; identify candidates
* for this by proc[s] being alphabetic. Some columns shall
* match both (e.g. STIME). Some numeric columns may grow
* left even as far as under the heading of the next column
* (seen with ps -fel's SZ spilling left into ADDR's space on
* Linux). While widening, it's OK to include a stray space;
* we'll trim that afterwards. Overspill from neighbouring
* columns can muck up alignment, so "digit" means any
* character that can appear in a numeric field (we may need
* to tweak "alphabetic" likewise for text fields; the user
* field can, for example, have a '+' appended). */
bool overspilt = false;
/* Numeric fields may include [-.:] and perhaps other
* punctuators: */
#define IsNumberish(c) (isdigit((unsigned char)(c)) || ispunct((unsigned char)(c)))
/* Right-aligned numeric: move s left until we run outside the
* field or find space. */
if (IsNumberish(proc[e]))
{
bool number = i > 0; /* Should we check for under-spill ? */
int outer = number ? end[i - 1] + 1 : 0;
while (s >= outer && !isspace((unsigned char) proc[s]))
{
if (number && !IsNumberish(proc[s]))
{
number = false;
}
s--;
}
/* Numeric field might overspill under previous header: */
if (s < outer)
{
int spill = s;
s = outer; /* By default, don't overlap previous column. */
if (number && IsNumberish(proc[spill]))
{
outer = start[i - 1];
/* Explore all the way to the start-column of the
* previous field's header. If we get there, in
* digits-and-punctuation, we've got two numeric
* fields that abut; we can't do better than assume
* the boundary is under the right end of the previous
* column label (which is what our parsing of the
* previous column assumed). So leave s where it is,
* just after the previous field's header's
* end-column. */
while (spill > outer)
{
spill--;
if (!IsNumberish(proc[spill]))
{
s = spill + 1; /* Confirmed overlap. */
break;
}
}
}
}
overspilt = IsNumberish(proc[e + 1]);
}
#undef IsNumberish
/* Left-aligned text or numeric misaligned by overspill;
* move e right (but never under next heading): */
if (overspilt || isalpha((unsigned char) proc[s]))
{
int outer;
if (i + 1 < CF_PROCCOLS && names[i + 1])
{
outer = start[i + 1];
}
else
{
outer = strlen(proc);
}
/* Simplify loop condition; we want e to end *before* next
* field, not on its first byte (or the terminator): */
outer--;
while (e < outer && !isspace((unsigned char) proc[e]))
{
e++;
}
}
/* Strip off any leading and trailing spaces: */
while (isspace((unsigned char) proc[s]))
{
s++;
}
/* ... but stop if the field is empty ! */
while (s <= e && isspace((unsigned char) proc[e]))
{
e--;
}
/* Grumble if the two heuristics don't agree: */
size_t wordlen = ep - sp;
if (e < s ? ep > sp : (sp != proc + s || ep != proc + e + 1))
{
char word[CF_SMALLBUF];
if (wordlen >= CF_SMALLBUF)
{
wordlen = CF_SMALLBUF - 1;
}
memcpy(word, sp, wordlen);
word[wordlen] = '\0';
char column[CF_SMALLBUF];
if (s <= e)
{
/* Copy proc[s through e] inclusive: */
const size_t len = MIN(1 + e - s, CF_SMALLBUF - 1);
memcpy(column, proc + s, len);
column[len] = '\0';
}
else
{
column[0] = '\0';
}
Log(LOG_LEVEL_INFO,
"Unacceptable model uncertainty examining process(%s): '%s' != '%s'",
proc, word, column);
}
/* Fall back on word if column got an empty answer: */
line[i] = e < s ? xstrndup(sp, ep - sp) : xstrndup(proc + s, 1 + e - s);
sp = ep;
}
/* Since start times can be very imprecise (e.g. just a past day's
* date, or a past year), calculate a better value from elapsed
* time, if available: */
int k = GetProcColumnIndex("ELAPSED", "ELAPSED", names);
if (k != -1)
{
const long elapsed = TimeCounter2Int(line[k]);
if (elapsed != CF_NOINT) /* Only use if parsed successfully ! */
{
int j = GetProcColumnIndex("STIME", "START", names), ns[3];
/* Trust the reported value if it matches hh:mm[:ss], though: */
if (sscanf(line[j], "%d:%d:%d", ns, ns + 1, ns + 2) < 2)
{
/* TODO: use time of ps-run, not time(NULL), which may be later. */
time_t value = time(NULL) - (time_t) elapsed;
Log(LOG_LEVEL_DEBUG,
"SplitProcLine: Replacing parsed start time %s with %s",
line[j], ctime(&value));
free(line[j]);
xasprintf(line + j, "%ld", value);
}
}
}
return true;
}
