sys/systm.h Kernel.framework

Generic null operation; always returns 0 (success). Installed in device switch and operation tables for entry points that require no action.

Null device routine; always returns 0. Placed in insignificant entries of the bdevsw and cdevsw device switch tables.

Returns ENOTTY. Stub installed for device entry points that do not support ioctl.

Returns ENOSYS. Stub used for an otherwise-reasonable operation that is not supported by the current system binary.

Returns ENXIO. Stub installed in device switch tables for devices that are not configured.

Returns ENOTSUP, for an operation not supported on a specific object or file type. The xnu source notes the routine's name is historical and does not match the errno it returns.

Allocates and zero-fills a hash table of list heads whose length is the largest power of two not greater than count, and stores the bucket mask (length - 1) through hashmask for indexing the table. The type argument, a legacy BSD malloc type, is ignored. Panics if count is not positive; returns NULL only on allocation failure. Free the table with hashdestroy, passing the same hashmask.

Frees a hash table allocated by hashinit. hashmask must be the mask hashinit returned (asserted to be one less than a power of two); the type argument is ignored.

Copies len bytes from from to to, like bcopy, but recognizes overlapping source and destination ranges and copies in the direction that handles them correctly (memmove semantics).

Fetches one byte from the given user-space address of the current process via copyin. Returns the byte value (0-255), or -1 if the address cannot be read.

Identical to fubyte: fetches one byte from the given user-space address via copyin, returning the byte value or -1 on fault. Historically the variant that was safe from interrupt context; in xnu the two are the same.

Stores the low-order byte of byte at the given user-space address of the current process via copyout. Returns 0 on success, -1 on fault.

Identical to subyte: stores the low-order byte of byte at the given user-space address via copyout, returning 0 on success and -1 on fault. Historically the interrupt-safe variant; in xnu the two are the same.

Fetches a 32-bit word (sizeof(int), despite the long return type) from the given user-space address via copyin. Returns the value, or -1 on fault; note -1 is also a legitimate stored value, so the failure return is ambiguous.

Identical to fuword: fetches a 32-bit word from the given user-space address, returning the value or -1 on fault. Historically the interrupt-safe variant; in xnu the two are the same.

Stores the low-order 32 bits (sizeof(int), despite the long parameter type) of word at the given user-space address via copyout. Returns 0 on success, -1 on fault.

Identical to suword: stores the low-order 32 bits of word at the given user-space address, returning 0 on success and -1 on fault. Historically the interrupt-safe variant; in xnu the two are the same.

Returns nonzero if the current task's address map permits access to the user address range addr through addr + len, checking VM_PROT_READ when prot is B_READ and VM_PROT_WRITE otherwise (vm_map_check_protection). Checks map protections only; it does not fault pages in or guard against concurrent unmapping, so a subsequent copyin or copyout can still fail.

Schedules fcn to be called with arg after the interval given by ts, via a delayed thread call (thread_call_func_delayed); the callout runs in thread-call context. A NULL or zero ts schedules the function to run immediately. Cancel with bsd_untimeout.

Cancels a pending bsd_timeout callout identified by the function and argument pair (thread_call_func_cancel, without waiting for a running callout). The xnu source marks this mechanism deprecated because cancellation by function/argument is very inefficient; new code should use an allocated thread call instead.

For a block-device vnode (VBLK), re-derives the device's block size by calling the device's bdevsw d_psize entry, defaulting to DEV_BSIZE if the routine is missing or fails, and stores it in the vnode (v_specsize) under the vnode lock. Vnodes of other types, and majors beyond the bdevsw table, are ignored. Called by file systems after changing a device's block size so that subsequent buffer I/O against the device vnode uses the new size.

Converts a struct timeval interval to Mach absolute-time units, converting the seconds and microseconds components separately with clock_interval_to_absolutetime_interval and summing them.

Converts a struct timespec interval to Mach absolute-time units, converting the seconds and nanoseconds components separately with clock_interval_to_absolutetime_interval and summing them.

Allocates a throttle info structure used for low-priority I/O throttling and returns an opaque pointer for the file system to pass to throttle_info_update. A reference is taken on the structure; every create must be balanced by a throttle_info_release.

Attaches the throttle info structure to the mount point, taking a reference on it; the association is used by the cluster I/O layer. File systems that call throttle_info_create must call this from their mount routine and balance it with throttle_info_mount_rel in their unmount routine. Any throttle info previously attached to the mount is released first. A NULL mount or throttle info is ignored.

Releases the throttle info reference attached to the mount point by throttle_info_mount_ref and clears the association. File systems that called throttle_info_mount_ref must call this in their unmount routine.

Drops the reference taken by throttle_info_create, freeing the throttle info structure when the last reference is gone. Must be called once for every throttle_info_create. NULL is ignored.

Records an I/O against the given throttle info for throttled-I/O bookkeeping; usually called before every I/O. If the calling thread runs at a throttled I/O tier, opens or refreshes a lowpri I/O window that throttle_lowpri_io will later honor; I/Os flagged B_PASSIVE in flags do not open a throttle window. Low overhead and never sleeps. A NULL throttle_info is ignored.

Called by a thread at the end of an I/O to pay the low-priority I/O penalty. If the current uthread has an open throttle window, blocks until woken by the throttle timer or a rethrottle, waiting at most sleep_amount throttle I/O periods; sleep_amount 0 performs only bookkeeping and does not sleep. Closes the thread's window and drops its reference on the throttle info. Returns the number of times the thread blocked, 0 if it was not throttled. Kernel-fault paths are never throttled.

Sets the current thread's I/O policy; policy may be IOPOL_NORMAL, IOPOL_THROTTLE, IOPOL_PASSIVE, IOPOL_UTILITY, or IOPOL_STANDARD, with the semantics described in setiopolicy_np(2). Wrapper around proc_set_thread_policy(TASK_POLICY_IOPOL).

Returns the current thread's effective I/O policy (IOPOL_* value), as computed by proc_get_effective_thread_policy with TASK_POLICY_IO.

Returns nonzero if the current thread's effective I/O tier is above the metadata tier (numerically less than IOSCHED_METADATA_TIER), i.e. its I/O takes priority over file system metadata I/O.

Returns, through throttle_info_handle, a referenced handle for the throttle info of the device identified by the lowest set bit of throttle_mask (a mount's mnt_throttle_mask). Returns 0 on success, EINVAL if throttle_info_handle is NULL or throttle_mask is 0. The handle must be released with throttle_info_rel_by_mask.

Releases the throttle info reference held by a handle obtained from throttle_info_ref_by_mask.

Records an I/O against the throttle info identified by a handle from throttle_info_ref_by_mask; equivalent to throttle_info_update on the underlying throttle info, with the same flags (B_PASSIVE I/Os do not open a throttle window). Usually called before every I/O; low overhead, never sleeps.

Marks the throttle info for the given device number as disabled; used for devices whose driver performs its own I/O scheduling. If isfusion is true (the device is part of a Fusion pair), software throttling remains enabled with priority handling and the throttle periods are reset to hard-disk values. Panics if devno is not in [0, LOWPRI_MAX_NUM_DEV).