dev/dtrace/dtrace.c

788  * These functions are called from probe context.  Because probe context is
791  * As a result, functions called from probe context may only call other DTrace
795  * loads are to be performed from probe context, they _must_ be in terms of
798  * Some functions in this block are not actually called from probe context;
800  * "Note:  not called from probe context."
816  * Atomically increment a specified error counter from probe context.
988 	 * If we hold the privilege to read from kernel memory, then
1001 	 * We're allowed to read from our own string table.
1025 	 * If we hold the privilege to read from kernel memory, then
1050 	 * If we hold the privilege to read from kernel memory, then
1194  * Copy src to dst, deriving the size and type from the specified (BYREF)
1519  * Note:  not called from probe context.  This function is called
1520  * asynchronously (and at a regular interval) from outside of probe context to
1545 		 * a dtrace_dynvar() allocation on this CPU (or from this CPU)
1856 		 * the hash bucket to prevent themselves from racing with
1923 				 * we'll try to allocate from a different
2055 	 * this hash chain, or another CPU is deleting an element from this
2633 			 * different CPU from the speculate(), and we must
2772  * Note:  not called from probe context.  This function is called
2773  * asynchronously from cross call context to clean any speculations that are
2818  * Note:  not called from probe context.  This function is called
2971 	 * This is the tougher case: we copy the string in question from
3035 			 * from tail-optimizing the call to dtrace_getarg().
3807 				 * notion that index() can start from a
3915 			 * strtok pointer from the mstate.  Note that this
3927 			 * (when we fetch addr from mstate->dtms_strtok)
4202 		 * Start from the back of the string, moving back toward the
4215 		 * Starting from the last character in the basename, move
4954  * Darwin 32-bit kernel may fetch from 64-bit user.
5490 			* Darwin kmem_zalloc() called from
5595 	 * thread calls panic() from dtrace_probe(), and that panic() is
5604 	 * allowing a return from panic().  Revisit someday.
5827 		 * allow this condition to be warned about.  (If this is from
5842  * is the function called by the provider to fire a probe -- from which all
5922 	 * Detect that here and cons up a viable state (from the probe_id).
5988 			 * from seeing one another's metaprobes.)
6034 			 * change out from under us so it's only legit to
6097 			 * the KILLED state, from which no further processing
6179 				 * value from the previous iteration of the
6499 				 * time to prevent it from being accumulated
6536 	/* FIXME: On Darwin the time spent leaving DTrace from this point to the rti is attributed
6546 			/* Return from error recursion. No accumulation, just clear the sign bit on t_dtrace_tracing. */
6587  * sections) are not _called_ from probe context.  (Any exceptions to this are
6588  * marked with a "Note:".)  Rather, they are called from elsewhere in the
6589  * DTrace framework to look-up probes in, add probes to and remove probes from
6759 	 * Find the bucket that we're removing this probe from.
6805  * These are random utility functions that are _not_ called from probe context.
7265  * Build a probe comparison key for use with dtrace_match_probe() from the
7449  * Unregister the specified provider from the DTrace framework.  This should
7517 	 * remove all of them from their hash chains and from the probe array.
7543 	 * The provider's probes have been removed from the hash chains and
7544 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
7545 	 * everyone has cleared out from any probe array processing.
7896 		 * mod_lock to prevent the list from being modified.  Note
7897 		 * that this also prevents the mod_busy bits from changing.
10270 	 * We disable the ECB by removing it from its probe.
10305 	 * The ECB has been disconnected from the probe; now sync to assure
10394 		 * predicate bits to prevent the ecb from activating at
10525  * Note:  called from cross call context.  This function switches the two
10556  * Note:  called from cross call context.  This function activates a buffer
10687  * Note:  called from probe context.  This function just increments the drop
10700  * Note:  called from probe context.  This function is called to reserve space
10802 			 * there.  We need to clear the buffer from the current
10874 				 * from one that hasn't wrapped; we need only
10961 	 *   we need to zero the data from the buffer offset to the wrapped
10969 	 *   space from the offset to the end of the buffer _and_ from the
11374 	 * are retained enablings.  We can safely load from dtrace_retained,
11416 		 * (The ECBs that result from priming an enabling aren't
12729 		 * hold on the credential to prevent it from disappearing on
12731 		 * credential from disappearing.  This means that we can
12732 		 * examine the credential and the zone from probe context.
12784 		 * the user has the dtrace_user privilege from a visibility
13045 		 * copy that option value from the grabbing state into the
13109 		 * it from failing to allocate a buffer due to failure to meet
13173 	 * probe (the data from this CPU will be processed first at user
13185 	 * We may have had an exit action from a BEGIN probe; only change our
13199 	 * atomically transition from processing none of a state's ECBs to
13307 		 * Prevent consumers from enabling destructive actions if DTrace
13533 				 * failure mode from dtrace_state_create() is a
13686 	 * from machine state; this is okay, too.
14647 	 * We cannot assert against DTRACE_DOF_MODE_NEVER here, because we are called from
14676 	 * We cannot assert against DTRACE_DOF_MODE_NEVER here, because we are called from
14833 	 * We're now going to lose the help from this process.
14873 			 * Remove the helper from the deferred list.
14909 dtrace_helpers_duplicate(proc_t *from, proc_t *to)
14919 	ASSERT(from->p_dtrace_helpers != NULL);
14922 	help = from->p_dtrace_helpers;
15066  * This routine is called from dtrace_module_unloaded().
15068  * from the kext shadow list.
15207 	 * A module can force symbols from kernel space using the plist key,
15212 	 * from kernel space (see dtrace_impl.h).  If this system state is set
15409 	 * We've removed all of the module's probes from the hash chains and
15410 	 * from the probe array.  Now issue a dtrace_sync() to be sure that
15411 	 * everyone has cleared out from any probe array processing.
15820 	 * eligible for symbols from userspace. The same dtrace that is currently
15889 		 * If we are the last dtrace client, switch back to lazy (from userspace) symbols
17479 		 * Finally, divide by three to prevent DTrace from eating too
17610 		 * The default is to wait for symbols from userspace (lazy symbols).
17626 	 * Called from bsd_init after all provider's *_init() routines have been
17653  * from the system.