/* * This program may be freely redistributed, * but this entire comment MUST remain intact. * * Copyright (c) 2018, Eitan Adler * Copyright (c) 1984, 1989, William LeFebvre, Rice University * Copyright (c) 1989, 1990, 1992, William LeFebvre, Northwestern University * * $FreeBSD$ */ /* * This file contains various handy utilities used by top. */ #include "top.h" #include "utils.h" #include #include #include #include #include #include #include #include #include #include int atoiwi(const char *str) { size_t len; len = strlen(str); if (len != 0) { if (strncmp(str, "infinity", len) == 0 || strncmp(str, "all", len) == 0 || strncmp(str, "maximum", len) == 0) { return(Infinity); } else if (str[0] == '-') { return(Invalid); } else { return((int)strtol(str, NULL, 10)); } } return(0); } /* * itoa - convert integer (decimal) to ascii string for positive numbers * only (we don't bother with negative numbers since we know we * don't use them). */ /* * How do we know that 16 will suffice? * Because the biggest number that we will * ever convert will be 2^32-1, which is 10 * digits. */ _Static_assert(sizeof(int) <= 4, "buffer too small for this sized int"); char * itoa(unsigned int val) { static char buffer[16]; /* result is built here */ /* 16 is sufficient since the largest number we will ever convert will be 2^32-1, which is 10 digits. */ sprintf(buffer, "%u", val); return (buffer); } /* * itoa7(val) - like itoa, except the number is right justified in a 7 * character field. This code is a duplication of itoa instead of * a front end to a more general routine for efficiency. */ char * itoa7(int val) { static char buffer[16]; /* result is built here */ /* 16 is sufficient since the largest number we will ever convert will be 2^32-1, which is 10 digits. */ sprintf(buffer, "%6u", val); return (buffer); } /* * digits(val) - return number of decimal digits in val. Only works for * non-negative numbers. */ int __pure2 digits(int val) { int cnt = 0; if (val == 0) { return 1; } while (val > 0) { cnt++; val /= 10; } return(cnt); } /* * string_index(string, array) - find string in array and return index */ int string_index(const char *string, const char * const *array) { size_t i = 0; while (*array != NULL) { if (strcmp(string, *array) == 0) { return(i); } array++; i++; } return(-1); } /* * argparse(line, cntp) - parse arguments in string "line", separating them * out into an argv-like array, and setting *cntp to the number of * arguments encountered. This is a simple parser that doesn't understand * squat about quotes. */ const char ** argparse(char *line, int *cntp) { const char **ap; static const char *argv[1024] = {0}; *cntp = 1; ap = &argv[1]; while ((*ap = strsep(&line, " ")) != NULL) { if (**ap != '\0') { (*cntp)++; if (*cntp >= (int)nitems(argv)) { break; } ap++; } } return (argv); } /* * percentages(cnt, out, new, old, diffs) - calculate percentage change * between array "old" and "new", putting the percentages i "out". * "cnt" is size of each array and "diffs" is used for scratch space. * The array "old" is updated on each call. * The routine assumes modulo arithmetic. This function is especially * useful on for calculating cpu state percentages. */ long percentages(int cnt, int *out, long *new, long *old, long *diffs) { int i; long change; long total_change; long *dp; long half_total; /* initialization */ total_change = 0; dp = diffs; /* calculate changes for each state and the overall change */ for (i = 0; i < cnt; i++) { if ((change = *new - *old) < 0) { /* this only happens when the counter wraps */ change = (int) ((unsigned long)*new-(unsigned long)*old); } total_change += (*dp++ = change); *old++ = *new++; } /* avoid divide by zero potential */ if (total_change == 0) { total_change = 1; } /* calculate percentages based on overall change, rounding up */ half_total = total_change / 2l; for (i = 0; i < cnt; i++) { *out++ = (int)((*diffs++ * 1000 + half_total) / total_change); } /* return the total in case the caller wants to use it */ return(total_change); } /* format_time(seconds) - format number of seconds into a suitable * display that will fit within 6 characters. Note that this * routine builds its string in a static area. If it needs * to be called more than once without overwriting previous data, * then we will need to adopt a technique similar to the * one used for format_k. */ /* Explanation: We want to keep the output within 6 characters. For low values we use the format mm:ss. For values that exceed 999:59, we switch to a format that displays hours and fractions: hhh.tH. For values that exceed 999.9, we use hhhh.t and drop the "H" designator. For values that exceed 9999.9, we use "???". */ const char * format_time(long seconds) { static char result[10]; /* sanity protection */ if (seconds < 0 || seconds > (99999l * 360l)) { strcpy(result, " ???"); } else if (seconds >= (1000l * 60l)) { /* alternate (slow) method displaying hours and tenths */ sprintf(result, "%5.1fH", (double)seconds / (double)(60l * 60l)); /* It is possible that the sprintf took more than 6 characters. If so, then the "H" appears as result[6]. If not, then there is a \0 in result[6]. Either way, it is safe to step on. */ result[6] = '\0'; } else { /* standard method produces MMM:SS */ sprintf(result, "%3ld:%02ld", seconds / 60l, seconds % 60l); } return(result); } /* * format_k(amt) - format a kilobyte memory value, returning a string * suitable for display. Returns a pointer to a static * area that changes each call. "amt" is converted to a fixed * size humanize_number call */ /* * Compromise time. We need to return a string, but we don't want the * caller to have to worry about freeing a dynamically allocated string. * Unfortunately, we can't just return a pointer to a static area as one * of the common uses of this function is in a large call to sprintf where * it might get invoked several times. Our compromise is to maintain an * array of strings and cycle thru them with each invocation. We make the * array large enough to handle the above mentioned case. The constant * NUM_STRINGS defines the number of strings in this array: we can tolerate * up to NUM_STRINGS calls before we start overwriting old information. * Keeping NUM_STRINGS a power of two will allow an intelligent optimizer * to convert the modulo operation into something quicker. What a hack! */ #define NUM_STRINGS 8 char * format_k(int64_t amt) { static char retarray[NUM_STRINGS][16]; static int index_ = 0; char *ret; ret = retarray[index_]; index_ = (index_ + 1) % NUM_STRINGS; humanize_number(ret, 6, amt * 1024, "", HN_AUTOSCALE, HN_NOSPACE | HN_B); return (ret); } int find_pid(pid_t pid) { kvm_t *kd = NULL; struct kinfo_proc *pbase = NULL; int nproc; int ret = 0; kd = kvm_open(NULL, _PATH_DEVNULL, NULL, O_RDONLY, NULL); if (kd == NULL) { fprintf(stderr, "top: kvm_open() failed.\n"); quit(TOP_EX_SYS_ERROR); } pbase = kvm_getprocs(kd, KERN_PROC_PID, pid, &nproc); if (pbase == NULL) { goto done; } if ((nproc == 1) && (pbase->ki_pid == pid)) { ret = 1; } done: kvm_close(kd); return ret; }