The syscall provider makes available a probe at the entry to and return from every system call in the system. Because system calls are the primary interface between user-level applications and the operating system kernel, the syscall provider can offer tremendous insight into application behavior with respect to the system.
syscall provides a pair of probes for each system call: an entry probe that fires before the system call is entered, and a return probe that fires after the system call has completed but before control has transferred back to user-level. For all syscall probes, the function name is set to be the name of the instrumented system call and the module name is undefined.
The names of the system calls as provided by the syscall provider may be found in the /etc/name_to_sysnum file. Often, the system call names provided by syscall correspond to names in Section 2 of the man pages. However, some probes provided by the syscall provider do not directly correspond to any documented system call. The common reasons for this discrepancy are described in this section.
In some cases, the name of the system call as provided by the syscall provider is actually a reflection of an ancient implementation detail. For example, for reasons dating back to UNIX antiquity, the name of exit in /etc/name_to_sysnum is rexit. Similarly, the name of timeis gtime, and the name of execve is exece.
Some system calls as presented in man page section 2 are implemented as suboperations of an undocumented system call. For example, the system calls related to System V semaphores, such as semctl, semget, semids, semop, and semtimedop are implemented as suboperations of a single system call, semsys. The semsys system call takes as its first argument an implementation-specific subcode denoting the specific system call required: SEMCTL, SEMGET, SEMIDS, SEMOP, or SEMTIMEDOP, respectively. As a result of overloading a single system call to implement multiple system calls, there is only a single pair of syscall probes for System V semaphores:
syscall::semsys:entry and syscall::semsys:return
Unlike Oracle Solaris 10, Oracle Solaris 11 implements the following system interfaces as individual system calls:
These system calls implement a superset of the functionality of their old non-at-suffixed counterparts. They take an additional first argument that is either an open directory file descriptor, in which case the operation on a relative pathname is taken relative to the specified directory, or is the reserved value AT_FDCWD, in which case the operation takes place relative to the current working directory.
In Oracle Solaris 11, the following old system calls have been removed from the system. The libc interfaces remain, but they are reimplemented not as system calls in their own right, but as calls to the new system calls as indicated:
|
A 32-bit program that supports large files that exceed two gigabytes in size must be able to process 64-bit file offsets. Because large files require use of large offsets, these files are manipulated through a parallel set of system interfaces. For more information, see the lf64(5) man page. These interfaces are documented in lf64, but they do not have individual man pages. Each of these large file system call interfaces appears as its own syscall probe as shown in the following table.
|
Some system calls are private implementation details of Oracle Solaris subsystems that span the user-kernel boundary. As such, these system calls do not have man pages in man page section 2. Examples of system calls in this category include the signotify system call, which is used as part of the implementation of POSIX.4 message queues, and the utssys system call, which is used to implement fuser. For more information, see the fuser(1M) man page.
For entry probes, the arguments arg0 .. argn are the arguments to the system call. For return probes, both arg0 and arg1 contain the return value. A non-zero value in the D variable errno indicates system call failure.
The syscall provider uses DTrace's stability mechanism to describe its stabilities as shown in the following table. For more information about the stability mechanism, refer to DTrace Stability Mechanisms.
|