// SPDX-License-Identifier: GPL-2.0 /* * trace_events_filter - generic event filtering * * Copyright (C) 2009 Tom Zanussi <[email protected]> */ #include <linux/uaccess.h> #include <linux/module.h> #include <linux/ctype.h> #include <linux/mutex.h> #include <linux/perf_event.h> #include <linux/slab.h> #include "trace.h" #include "trace_output.h" #define DEFAULT_SYS_FILTER_MESSAGE … /* Due to token parsing '<=' must be before '<' and '>=' must be before '>' */ #define OPS … #undef C #define C(a, b) … enum filter_op_ids { … }; #undef C #define C(a, b) … static const char * ops[] = …; enum filter_pred_fn { … }; struct filter_pred { … }; /* * pred functions are OP_LE, OP_LT, OP_GE, OP_GT, and OP_BAND * pred_funcs_##type below must match the order of them above. */ #define PRED_FUNC_START … #define PRED_FUNC_MAX … #define ERRORS … #undef C #define C(a, b) … enum { … }; #undef C #define C(a, b) … static const char *err_text[] = …; /* Called after a '!' character but "!=" and "!~" are not "not"s */ static bool is_not(const char *str) { … } /** * struct prog_entry - a singe entry in the filter program * @target: Index to jump to on a branch (actually one minus the index) * @when_to_branch: The value of the result of the predicate to do a branch * @pred: The predicate to execute. */ struct prog_entry { … }; /** * update_preds - assign a program entry a label target * @prog: The program array * @N: The index of the current entry in @prog * @invert: What to assign a program entry for its branch condition * * The program entry at @N has a target that points to the index of a program * entry that can have its target and when_to_branch fields updated. * Update the current program entry denoted by index @N target field to be * that of the updated entry. This will denote the entry to update if * we are processing an "||" after an "&&". */ static void update_preds(struct prog_entry *prog, int N, int invert) { … } struct filter_parse_error { … }; static void parse_error(struct filter_parse_error *pe, int err, int pos) { … } parse_pred_fn; enum { … }; static void free_predicate(struct filter_pred *pred) { … } /* * Without going into a formal proof, this explains the method that is used in * parsing the logical expressions. * * For example, if we have: "a && !(!b || (c && g)) || d || e && !f" * The first pass will convert it into the following program: * * n1: r=a; l1: if (!r) goto l4; * n2: r=b; l2: if (!r) goto l4; * n3: r=c; r=!r; l3: if (r) goto l4; * n4: r=g; r=!r; l4: if (r) goto l5; * n5: r=d; l5: if (r) goto T * n6: r=e; l6: if (!r) goto l7; * n7: r=f; r=!r; l7: if (!r) goto F * T: return TRUE * F: return FALSE * * To do this, we use a data structure to represent each of the above * predicate and conditions that has: * * predicate, when_to_branch, invert, target * * The "predicate" will hold the function to determine the result "r". * The "when_to_branch" denotes what "r" should be if a branch is to be taken * "&&" would contain "!r" or (0) and "||" would contain "r" or (1). * The "invert" holds whether the value should be reversed before testing. * The "target" contains the label "l#" to jump to. * * A stack is created to hold values when parentheses are used. * * To simplify the logic, the labels will start at 0 and not 1. * * The possible invert values are 1 and 0. The number of "!"s that are in scope * before the predicate determines the invert value, if the number is odd then * the invert value is 1 and 0 otherwise. This means the invert value only * needs to be toggled when a new "!" is introduced compared to what is stored * on the stack, where parentheses were used. * * The top of the stack and "invert" are initialized to zero. * * ** FIRST PASS ** * * #1 A loop through all the tokens is done: * * #2 If the token is an "(", the stack is push, and the current stack value * gets the current invert value, and the loop continues to the next token. * The top of the stack saves the "invert" value to keep track of what * the current inversion is. As "!(a && !b || c)" would require all * predicates being affected separately by the "!" before the parentheses. * And that would end up being equivalent to "(!a || b) && !c" * * #3 If the token is an "!", the current "invert" value gets inverted, and * the loop continues. Note, if the next token is a predicate, then * this "invert" value is only valid for the current program entry, * and does not affect other predicates later on. * * The only other acceptable token is the predicate string. * * #4 A new entry into the program is added saving: the predicate and the * current value of "invert". The target is currently assigned to the * previous program index (this will not be its final value). * * #5 We now enter another loop and look at the next token. The only valid * tokens are ")", "&&", "||" or end of the input string "\0". * * #6 The invert variable is reset to the current value saved on the top of * the stack. * * #7 The top of the stack holds not only the current invert value, but also * if a "&&" or "||" needs to be processed. Note, the "&&" takes higher * precedence than "||". That is "a && b || c && d" is equivalent to * "(a && b) || (c && d)". Thus the first thing to do is to see if "&&" needs * to be processed. This is the case if an "&&" was the last token. If it was * then we call update_preds(). This takes the program, the current index in * the program, and the current value of "invert". More will be described * below about this function. * * #8 If the next token is "&&" then we set a flag in the top of the stack * that denotes that "&&" needs to be processed, break out of this loop * and continue with the outer loop. * * #9 Otherwise, if a "||" needs to be processed then update_preds() is called. * This is called with the program, the current index in the program, but * this time with an inverted value of "invert" (that is !invert). This is * because the value taken will become the "when_to_branch" value of the * program. * Note, this is called when the next token is not an "&&". As stated before, * "&&" takes higher precedence, and "||" should not be processed yet if the * next logical operation is "&&". * * #10 If the next token is "||" then we set a flag in the top of the stack * that denotes that "||" needs to be processed, break out of this loop * and continue with the outer loop. * * #11 If this is the end of the input string "\0" then we break out of both * loops. * * #12 Otherwise, the next token is ")", where we pop the stack and continue * this inner loop. * * Now to discuss the update_pred() function, as that is key to the setting up * of the program. Remember the "target" of the program is initialized to the * previous index and not the "l" label. The target holds the index into the * program that gets affected by the operand. Thus if we have something like * "a || b && c", when we process "a" the target will be "-1" (undefined). * When we process "b", its target is "0", which is the index of "a", as that's * the predicate that is affected by "||". But because the next token after "b" * is "&&" we don't call update_preds(). Instead continue to "c". As the * next token after "c" is not "&&" but the end of input, we first process the * "&&" by calling update_preds() for the "&&" then we process the "||" by * calling updates_preds() with the values for processing "||". * * What does that mean? What update_preds() does is to first save the "target" * of the program entry indexed by the current program entry's "target" * (remember the "target" is initialized to previous program entry), and then * sets that "target" to the current index which represents the label "l#". * That entry's "when_to_branch" is set to the value passed in (the "invert" * or "!invert"). Then it sets the current program entry's target to the saved * "target" value (the old value of the program that had its "target" updated * to the label). * * Looking back at "a || b && c", we have the following steps: * "a" - prog[0] = { "a", X, -1 } // pred, when_to_branch, target * "||" - flag that we need to process "||"; continue outer loop * "b" - prog[1] = { "b", X, 0 } * "&&" - flag that we need to process "&&"; continue outer loop * (Notice we did not process "||") * "c" - prog[2] = { "c", X, 1 } * update_preds(prog, 2, 0); // invert = 0 as we are processing "&&" * t = prog[2].target; // t = 1 * s = prog[t].target; // s = 0 * prog[t].target = 2; // Set target to "l2" * prog[t].when_to_branch = 0; * prog[2].target = s; * update_preds(prog, 2, 1); // invert = 1 as we are now processing "||" * t = prog[2].target; // t = 0 * s = prog[t].target; // s = -1 * prog[t].target = 2; // Set target to "l2" * prog[t].when_to_branch = 1; * prog[2].target = s; * * #13 Which brings us to the final step of the first pass, which is to set * the last program entry's when_to_branch and target, which will be * when_to_branch = 0; target = N; ( the label after the program entry after * the last program entry processed above). * * If we denote "TRUE" to be the entry after the last program entry processed, * and "FALSE" the program entry after that, we are now done with the first * pass. * * Making the above "a || b && c" have a program of: * prog[0] = { "a", 1, 2 } * prog[1] = { "b", 0, 2 } * prog[2] = { "c", 0, 3 } * * Which translates into: * n0: r = a; l0: if (r) goto l2; * n1: r = b; l1: if (!r) goto l2; * n2: r = c; l2: if (!r) goto l3; // Which is the same as "goto F;" * T: return TRUE; l3: * F: return FALSE * * Although, after the first pass, the program is correct, it is * inefficient. The simple sample of "a || b && c" could be easily been * converted into: * n0: r = a; if (r) goto T * n1: r = b; if (!r) goto F * n2: r = c; if (!r) goto F * T: return TRUE; * F: return FALSE; * * The First Pass is over the input string. The next too passes are over * the program itself. * * ** SECOND PASS ** * * Which brings us to the second pass. If a jump to a label has the * same condition as that label, it can instead jump to its target. * The original example of "a && !(!b || (c && g)) || d || e && !f" * where the first pass gives us: * * n1: r=a; l1: if (!r) goto l4; * n2: r=b; l2: if (!r) goto l4; * n3: r=c; r=!r; l3: if (r) goto l4; * n4: r=g; r=!r; l4: if (r) goto l5; * n5: r=d; l5: if (r) goto T * n6: r=e; l6: if (!r) goto l7; * n7: r=f; r=!r; l7: if (!r) goto F: * T: return TRUE; * F: return FALSE * * We can see that "l3: if (r) goto l4;" and at l4, we have "if (r) goto l5;". * And "l5: if (r) goto T", we could optimize this by converting l3 and l4 * to go directly to T. To accomplish this, we start from the last * entry in the program and work our way back. If the target of the entry * has the same "when_to_branch" then we could use that entry's target. * Doing this, the above would end up as: * * n1: r=a; l1: if (!r) goto l4; * n2: r=b; l2: if (!r) goto l4; * n3: r=c; r=!r; l3: if (r) goto T; * n4: r=g; r=!r; l4: if (r) goto T; * n5: r=d; l5: if (r) goto T; * n6: r=e; l6: if (!r) goto F; * n7: r=f; r=!r; l7: if (!r) goto F; * T: return TRUE * F: return FALSE * * In that same pass, if the "when_to_branch" doesn't match, we can simply * go to the program entry after the label. That is, "l2: if (!r) goto l4;" * where "l4: if (r) goto T;", then we can convert l2 to be: * "l2: if (!r) goto n5;". * * This will have the second pass give us: * n1: r=a; l1: if (!r) goto n5; * n2: r=b; l2: if (!r) goto n5; * n3: r=c; r=!r; l3: if (r) goto T; * n4: r=g; r=!r; l4: if (r) goto T; * n5: r=d; l5: if (r) goto T * n6: r=e; l6: if (!r) goto F; * n7: r=f; r=!r; l7: if (!r) goto F * T: return TRUE * F: return FALSE * * Notice, all the "l#" labels are no longer used, and they can now * be discarded. * * ** THIRD PASS ** * * For the third pass we deal with the inverts. As they simply just * make the "when_to_branch" get inverted, a simple loop over the * program to that does: "when_to_branch ^= invert;" will do the * job, leaving us with: * n1: r=a; if (!r) goto n5; * n2: r=b; if (!r) goto n5; * n3: r=c: if (!r) goto T; * n4: r=g; if (!r) goto T; * n5: r=d; if (r) goto T * n6: r=e; if (!r) goto F; * n7: r=f; if (r) goto F * T: return TRUE * F: return FALSE * * As "r = a; if (!r) goto n5;" is obviously the same as * "if (!a) goto n5;" without doing anything we can interpret the * program as: * n1: if (!a) goto n5; * n2: if (!b) goto n5; * n3: if (!c) goto T; * n4: if (!g) goto T; * n5: if (d) goto T * n6: if (!e) goto F; * n7: if (f) goto F * T: return TRUE * F: return FALSE * * Since the inverts are discarded at the end, there's no reason to store * them in the program array (and waste memory). A separate array to hold * the inverts is used and freed at the end. */ static struct prog_entry * predicate_parse(const char *str, int nr_parens, int nr_preds, parse_pred_fn parse_pred, void *data, struct filter_parse_error *pe) { … } static inline int do_filter_cpumask(int op, const struct cpumask *mask, const struct cpumask *cmp) { … } /* Optimisation of do_filter_cpumask() for scalar fields */ static inline int do_filter_scalar_cpumask(int op, unsigned int cpu, const struct cpumask *mask) { … } static inline int do_filter_cpumask_scalar(int op, const struct cpumask *mask, unsigned int cpu) { … } enum pred_cmp_types { … }; #define DEFINE_COMPARISON_PRED(type) … #define DEFINE_CPUMASK_COMPARISON_PRED(size) … #define DEFINE_EQUALITY_PRED(size) … DEFINE_COMPARISON_PRED(s64); DEFINE_COMPARISON_PRED(u64); DEFINE_COMPARISON_PRED(s32); DEFINE_COMPARISON_PRED(u32); DEFINE_COMPARISON_PRED(s16); DEFINE_COMPARISON_PRED(u16); DEFINE_COMPARISON_PRED(s8); DEFINE_COMPARISON_PRED(u8); DEFINE_CPUMASK_COMPARISON_PRED(64); DEFINE_CPUMASK_COMPARISON_PRED(32); DEFINE_CPUMASK_COMPARISON_PRED(16); DEFINE_CPUMASK_COMPARISON_PRED(8); DEFINE_EQUALITY_PRED(64); DEFINE_EQUALITY_PRED(32); DEFINE_EQUALITY_PRED(16); DEFINE_EQUALITY_PRED(8); /* user space strings temp buffer */ #define USTRING_BUF_SIZE … struct ustring_buffer { … }; static __percpu struct ustring_buffer *ustring_per_cpu; static __always_inline char *test_string(char *str) { … } static __always_inline char *test_ustring(char *str) { … } /* Filter predicate for fixed sized arrays of characters */ static int filter_pred_string(struct filter_pred *pred, void *event) { … } static __always_inline int filter_pchar(struct filter_pred *pred, char *str) { … } /* Filter predicate for char * pointers */ static int filter_pred_pchar(struct filter_pred *pred, void *event) { … } /* Filter predicate for char * pointers in user space*/ static int filter_pred_pchar_user(struct filter_pred *pred, void *event) { … } /* * Filter predicate for dynamic sized arrays of characters. * These are implemented through a list of strings at the end * of the entry. * Also each of these strings have a field in the entry which * contains its offset from the beginning of the entry. * We have then first to get this field, dereference it * and add it to the address of the entry, and at last we have * the address of the string. */ static int filter_pred_strloc(struct filter_pred *pred, void *event) { … } /* * Filter predicate for relative dynamic sized arrays of characters. * These are implemented through a list of strings at the end * of the entry as same as dynamic string. * The difference is that the relative one records the location offset * from the field itself, not the event entry. */ static int filter_pred_strrelloc(struct filter_pred *pred, void *event) { … } /* Filter predicate for CPUs. */ static int filter_pred_cpu(struct filter_pred *pred, void *event) { … } /* Filter predicate for current CPU vs user-provided cpumask */ static int filter_pred_cpu_cpumask(struct filter_pred *pred, void *event) { … } /* Filter predicate for cpumask field vs user-provided cpumask */ static int filter_pred_cpumask(struct filter_pred *pred, void *event) { … } /* Filter predicate for cpumask field vs user-provided scalar */ static int filter_pred_cpumask_cpu(struct filter_pred *pred, void *event) { … } /* Filter predicate for COMM. */ static int filter_pred_comm(struct filter_pred *pred, void *event) { … } /* Filter predicate for functions. */ static int filter_pred_function(struct filter_pred *pred, void *event) { … } /* * regex_match_foo - Basic regex callbacks * * @str: the string to be searched * @r: the regex structure containing the pattern string * @len: the length of the string to be searched (including '\0') * * Note: * - @str might not be NULL-terminated if it's of type DYN_STRING * RDYN_STRING, or STATIC_STRING, unless @len is zero. */ static int regex_match_full(char *str, struct regex *r, int len) { … } static int regex_match_front(char *str, struct regex *r, int len) { … } static int regex_match_middle(char *str, struct regex *r, int len) { … } static int regex_match_end(char *str, struct regex *r, int len) { … } static int regex_match_glob(char *str, struct regex *r, int len __maybe_unused) { … } /** * filter_parse_regex - parse a basic regex * @buff: the raw regex * @len: length of the regex * @search: will point to the beginning of the string to compare * @not: tell whether the match will have to be inverted * * This passes in a buffer containing a regex and this function will * set search to point to the search part of the buffer and * return the type of search it is (see enum above). * This does modify buff. * * Returns enum type. * search returns the pointer to use for comparison. * not returns 1 if buff started with a '!' * 0 otherwise. */ enum regex_type filter_parse_regex(char *buff, int len, char **search, int *not) { … } static void filter_build_regex(struct filter_pred *pred) { … } #ifdef CONFIG_FTRACE_STARTUP_TEST static int test_pred_visited_fn(struct filter_pred *pred, void *event); #else static int test_pred_visited_fn(struct filter_pred *pred, void *event) { return 0; } #endif static int filter_pred_fn_call(struct filter_pred *pred, void *event); /* return 1 if event matches, 0 otherwise (discard) */ int filter_match_preds(struct event_filter *filter, void *rec) { … } EXPORT_SYMBOL_GPL(…); static void remove_filter_string(struct event_filter *filter) { … } static void append_filter_err(struct trace_array *tr, struct filter_parse_error *pe, struct event_filter *filter) { … } static inline struct event_filter *event_filter(struct trace_event_file *file) { … } /* caller must hold event_mutex */ void print_event_filter(struct trace_event_file *file, struct trace_seq *s) { … } void print_subsystem_event_filter(struct event_subsystem *system, struct trace_seq *s) { … } static void free_prog(struct event_filter *filter) { … } static void filter_disable(struct trace_event_file *file) { … } static void __free_filter(struct event_filter *filter) { … } void free_event_filter(struct event_filter *filter) { … } static inline void __remove_filter(struct trace_event_file *file) { … } static void filter_free_subsystem_preds(struct trace_subsystem_dir *dir, struct trace_array *tr) { … } static inline void __free_subsystem_filter(struct trace_event_file *file) { … } static void filter_free_subsystem_filters(struct trace_subsystem_dir *dir, struct trace_array *tr) { … } int filter_assign_type(const char *type) { … } static enum filter_pred_fn select_comparison_fn(enum filter_op_ids op, int field_size, int field_is_signed) { … } static int filter_pred_fn_call(struct filter_pred *pred, void *event) { … } /* Called when a predicate is encountered by predicate_parse() */ static int parse_pred(const char *str, void *data, int pos, struct filter_parse_error *pe, struct filter_pred **pred_ptr) { … } enum { … }; /* * Read the filter string once to calculate the number of predicates * as well as how deep the parentheses go. * * Returns: * 0 - everything is fine (err is undefined) * -1 - too many ')' * -2 - too many '(' * -3 - No matching quote */ static int calc_stack(const char *str, int *parens, int *preds, int *err) { … } static int process_preds(struct trace_event_call *call, const char *filter_string, struct event_filter *filter, struct filter_parse_error *pe) { … } static inline void event_set_filtered_flag(struct trace_event_file *file) { … } static inline void event_set_filter(struct trace_event_file *file, struct event_filter *filter) { … } static inline void event_clear_filter(struct trace_event_file *file) { … } struct filter_list { … }; static int process_system_preds(struct trace_subsystem_dir *dir, struct trace_array *tr, struct filter_parse_error *pe, char *filter_string) { … } static int create_filter_start(char *filter_string, bool set_str, struct filter_parse_error **pse, struct event_filter **filterp) { … } static void create_filter_finish(struct filter_parse_error *pe) { … } /** * create_filter - create a filter for a trace_event_call * @tr: the trace array associated with these events * @call: trace_event_call to create a filter for * @filter_string: filter string * @set_str: remember @filter_str and enable detailed error in filter * @filterp: out param for created filter (always updated on return) * Must be a pointer that references a NULL pointer. * * Creates a filter for @call with @filter_str. If @set_str is %true, * @filter_str is copied and recorded in the new filter. * * On success, returns 0 and *@filterp points to the new filter. On * failure, returns -errno and *@filterp may point to %NULL or to a new * filter. In the latter case, the returned filter contains error * information if @set_str is %true and the caller is responsible for * freeing it. */ static int create_filter(struct trace_array *tr, struct trace_event_call *call, char *filter_string, bool set_str, struct event_filter **filterp) { … } int create_event_filter(struct trace_array *tr, struct trace_event_call *call, char *filter_str, bool set_str, struct event_filter **filterp) { … } /** * create_system_filter - create a filter for an event subsystem * @dir: the descriptor for the subsystem directory * @filter_str: filter string * @filterp: out param for created filter (always updated on return) * * Identical to create_filter() except that it creates a subsystem filter * and always remembers @filter_str. */ static int create_system_filter(struct trace_subsystem_dir *dir, char *filter_str, struct event_filter **filterp) { … } /* caller must hold event_mutex */ int apply_event_filter(struct trace_event_file *file, char *filter_string) { … } int apply_subsystem_event_filter(struct trace_subsystem_dir *dir, char *filter_string) { … } #ifdef CONFIG_PERF_EVENTS void ftrace_profile_free_filter(struct perf_event *event) { … } struct function_filter_data { … }; #ifdef CONFIG_FUNCTION_TRACER static char ** ftrace_function_filter_re(char *buf, int len, int *count) { … } static int ftrace_function_set_regexp(struct ftrace_ops *ops, int filter, int reset, char *re, int len) { … } static int __ftrace_function_set_filter(int filter, char *buf, int len, struct function_filter_data *data) { … } static int ftrace_function_check_pred(struct filter_pred *pred) { … } static int ftrace_function_set_filter_pred(struct filter_pred *pred, struct function_filter_data *data) { … } static bool is_or(struct prog_entry *prog, int i) { … } static int ftrace_function_set_filter(struct perf_event *event, struct event_filter *filter) { … } #else static int ftrace_function_set_filter(struct perf_event *event, struct event_filter *filter) { return -ENODEV; } #endif /* CONFIG_FUNCTION_TRACER */ int ftrace_profile_set_filter(struct perf_event *event, int event_id, char *filter_str) { … } #endif /* CONFIG_PERF_EVENTS */ #ifdef CONFIG_FTRACE_STARTUP_TEST #include <linux/types.h> #include <linux/tracepoint.h> #define CREATE_TRACE_POINTS #include "trace_events_filter_test.h" #define DATA_REC … #define YES … #define NO … static struct test_filter_data_t { … } test_filter_data[] = …; #undef DATA_REC #undef FILTER #undef YES #undef NO #define DATA_CNT … static int test_pred_visited; static int test_pred_visited_fn(struct filter_pred *pred, void *event) { … } static void update_pred_fn(struct event_filter *filter, char *fields) { … } static __init int ftrace_test_event_filter(void) { … } late_initcall(ftrace_test_event_filter); #endif /* CONFIG_FTRACE_STARTUP_TEST */
Codebase Browser
linux