error, nexterror, poperror, waserror – error handling functions


void error(char*)

void nexterror(void)

unhandled troff command .sp

#define poperror() (up->nerrlab--)

#define waserror() (setlabel(&up->errlab[up->nerrlab++]))


The kernel handles error conditions using non-local gotos, similar to setjmp(2), but using a stack of error labels to implement nested exception handling. This simplifies many of the internal interfaces by eliminating the need for returning and checking error codes at every level of the call stack, at the cost of requiring kernel routines to adhere to a strict discipline.

Each process has in its defining kernel Proc structure a stack of labels, NERR (currently 64) elements deep. A kernel function that must perform a clean up or recovery action on an error makes a stylised call to waserror, nexterror and poperror:


unhandled troff command .DT

if(waserror()){ /* recovery action */ nexterror(); } /* normal action */ poperror();

When called in the normal course of events, waserror registers an error handling block by pushing its label onto the stack, and returns zero. The return value of waserror should be tested as shown above. If non-zero (true), the calling function should perform the needed error recovery, ended by a call to nexterror to transfer control to the next location on the error stack. Typical recovery actions include deallocating memory, unlocking resources, and resetting state variables.

Within the recovery block, after handling an error condition, there must normally be a call to nexterror to transfer control to any error recovery lower down in the stack. The main exception is in the outermost function in a process, which must not call nexterror (there being nothing further on the stack), but calls pexit (see kproc(9)) instead, to terminate the process.

When the need to recover a particular resource has passed, a function that has called waserror must remove the corresponding label from the stack by calling poperror. This must be done before returning from the function; otherwise, a subsequent call to error will return to an obsolete activation record, with unpredictable but unpleasant consequences.

Error copies the given error message, which is limited to ERRMAX bytes, into the Proc.errstr of the current process, enables interrupts by calling spllo (native only), and finally calls nexterror to start invoking the recovery procedures currently stacked by waserror. The file /sys/src/9/port/error.h offers a wide selection of predefined error messages, suitable for almost any occasion. The message set by the most recent call to error can be obtained within the kernel by examining up->error and in an application, by using the %r directive of print(2).

A complex function can have nested error handlers. A waserror block will follow the acquisition of a resource, releasing it on error before calling nexterror, and a poperror will precede its release in the normal case. For example:


unhandled troff command .DT

void outer(Thing *t) { qlock(t); if(waserror()){ /* A */ qunlock(t); nexterror(); } m = mallocz(READSTR, 0); if(m == nil) error(Enomem); /* returns to A */ if(waserror()){ /* B */ free(m); nexterror(); /* invokes A */ } inner(t); poperror(); /* pops B */ free(m); poperror(); /* pops A */ qunlock(t); }

unhandled troff command .sp

void inner(Thing *t) { if(t->bad) error(Egreg); /* returns to B */ t->valid++; }




The description above has many instances of should, will, must and must not .


panic(9), kproc(9), splhi(9)