+ memory_object_size_t tsize;
+ kern_return_t kret;
+ int request_flags = 0;
+ int flush_flags = MEMORY_OBJECT_RETURN_NONE;
+
+ if ( !UBCINFOEXISTS(vp))
+ return (0);
+ if ((flags & (UBC_INVALIDATE | UBC_PUSHDIRTY | UBC_PUSHALL)) == 0)
+ return (0);
+ if (end_off <= beg_off)
+ return (1);
+
+ if (flags & UBC_INVALIDATE)
+ /*
+ * discard the resident pages
+ */
+ request_flags = (MEMORY_OBJECT_DATA_FLUSH | MEMORY_OBJECT_DATA_NO_CHANGE);
+
+ if (flags & UBC_SYNC)
+ /*
+ * wait for all the I/O to complete before returning
+ */
+ request_flags |= MEMORY_OBJECT_IO_SYNC;
+
+ if (flags & UBC_PUSHDIRTY)
+ /*
+ * we only return the dirty pages in the range
+ */
+ flush_flags = MEMORY_OBJECT_RETURN_DIRTY;
+
+ if (flags & UBC_PUSHALL)
+ /*
+ * then return all the interesting pages in the range (both
+ * dirty and precious) to the pager
+ */
+ flush_flags = MEMORY_OBJECT_RETURN_ALL;
+
+ beg_off = trunc_page_64(beg_off);
+ end_off = round_page_64(end_off);
+ tsize = (memory_object_size_t)end_off - beg_off;
+
+ /* flush and/or invalidate pages in the range requested */
+ kret = memory_object_lock_request(vp->v_ubcinfo->ui_control,
+ beg_off, tsize,
+ (memory_object_offset_t *)resid_off,
+ io_errno, flush_flags, request_flags,
+ VM_PROT_NO_CHANGE);
+
+ return ((kret == KERN_SUCCESS) ? 1 : 0);
+}
+
+
+/*
+ * ubc_map
+ *
+ * Explicitly map a vnode that has an associate ubc_info, and add a reference
+ * to it for the ubc system, if there isn't one already, so it will not be
+ * recycled while it's in use, and set flags on the ubc_info to indicate that
+ * we have done this
+ *
+ * Parameters: vp The vnode to map
+ * flags The mapping flags for the vnode; this
+ * will be a combination of one or more of
+ * PROT_READ, PROT_WRITE, and PROT_EXEC
+ *
+ * Returns: 0 Success
+ * EPERM Permission was denied
+ *
+ * Notes: An I/O reference on the vnode must already be held on entry
+ *
+ * If there is no ubc_info associated with the vnode, this function
+ * will return success.
+ *
+ * If a permission error occurs, this function will return
+ * failure; all other failures will cause this function to return
+ * success.
+ *
+ * IMPORTANT: This is an internal use function, and its symbols
+ * are not exported, hence its error checking is not very robust.
+ * It is primarily used by:
+ *
+ * o mmap(), when mapping a file
+ * o When mapping a shared file (a shared library in the
+ * shared segment region)
+ * o When loading a program image during the exec process
+ *
+ * ...all of these uses ignore the return code, and any fault that
+ * results later because of a failure is handled in the fix-up path
+ * of the fault handler. The interface exists primarily as a
+ * performance hint.
+ *
+ * Given that third party implementation of the type of interfaces
+ * that would use this function, such as alternative executable
+ * formats, etc., are unsupported, this function is not exported
+ * for general use.
+ *
+ * The extra reference is held until the VM system unmaps the
+ * vnode from its own context to maintain a vnode reference in
+ * cases like open()/mmap()/close(), which leave the backing
+ * object referenced by a mapped memory region in a process
+ * address space.
+ */
+__private_extern__ int
+ubc_map(vnode_t vp, int flags)
+{
+ struct ubc_info *uip;
+ int error = 0;
+ int need_ref = 0;
+ int need_wakeup = 0;
+
+ if (UBCINFOEXISTS(vp)) {
+
+ vnode_lock(vp);
+ uip = vp->v_ubcinfo;
+
+ while (ISSET(uip->ui_flags, UI_MAPBUSY)) {
+ SET(uip->ui_flags, UI_MAPWAITING);
+ (void) msleep(&uip->ui_flags, &vp->v_lock,
+ PRIBIO, "ubc_map", NULL);
+ }
+ SET(uip->ui_flags, UI_MAPBUSY);
+ vnode_unlock(vp);
+
+ error = VNOP_MMAP(vp, flags, vfs_context_current());
+
+ /*
+ * rdar://problem/22587101 required that we stop propagating
+ * EPERM up the stack. Otherwise, we would have to funnel up
+ * the error at all the call sites for memory_object_map().
+ * The risk is in having to undo the map/object/entry state at
+ * all these call sites. It would also affect more than just mmap()
+ * e.g. vm_remap().
+ *
+ * if (error != EPERM)
+ * error = 0;
+ */
+
+ error = 0;
+
+ vnode_lock_spin(vp);
+
+ if (error == 0) {
+ if ( !ISSET(uip->ui_flags, UI_ISMAPPED))
+ need_ref = 1;
+ SET(uip->ui_flags, (UI_WASMAPPED | UI_ISMAPPED));
+ if (flags & PROT_WRITE) {
+ SET(uip->ui_flags, UI_MAPPEDWRITE);
+ }
+ }
+ CLR(uip->ui_flags, UI_MAPBUSY);
+
+ if (ISSET(uip->ui_flags, UI_MAPWAITING)) {
+ CLR(uip->ui_flags, UI_MAPWAITING);
+ need_wakeup = 1;
+ }
+ vnode_unlock(vp);
+
+ if (need_wakeup)
+ wakeup(&uip->ui_flags);
+
+ if (need_ref) {
+ /*
+ * Make sure we get a ref as we can't unwind from here
+ */
+ if (vnode_ref_ext(vp, 0, VNODE_REF_FORCE))
+ panic("%s : VNODE_REF_FORCE failed\n", __FUNCTION__);
+ }
+ }
+ return (error);
+}
+
+
+/*
+ * ubc_destroy_named
+ *
+ * Destroy the named memory object associated with the ubc_info control object
+ * associated with the designated vnode, if there is a ubc_info associated
+ * with the vnode, and a control object is associated with it
+ *
+ * Parameters: vp The designated vnode
+ *
+ * Returns: (void)
+ *
+ * Notes: This function is called on vnode termination for all vnodes,
+ * and must therefore not assume that there is a ubc_info that is
+ * associated with the vnode, nor that there is a control object
+ * associated with the ubc_info.
+ *
+ * If all the conditions necessary are present, this function
+ * calls memory_object_destory(), which will in turn end up
+ * calling ubc_unmap() to release any vnode references that were
+ * established via ubc_map().
+ *
+ * IMPORTANT: This is an internal use function that is used
+ * exclusively by the internal use function vclean().
+ */
+__private_extern__ void
+ubc_destroy_named(vnode_t vp)
+{
+ memory_object_control_t control;
+ struct ubc_info *uip;
+ kern_return_t kret;
+
+ if (UBCINFOEXISTS(vp)) {
+ uip = vp->v_ubcinfo;
+
+ /* Terminate the memory object */
+ control = ubc_getobject(vp, UBC_HOLDOBJECT);
+ if (control != MEMORY_OBJECT_CONTROL_NULL) {
+ kret = memory_object_destroy(control, 0);
+ if (kret != KERN_SUCCESS)
+ panic("ubc_destroy_named: memory_object_destroy failed");
+ }
+ }
+}
+
+
+/*
+ * ubc_isinuse
+ *
+ * Determine whether or not a vnode is currently in use by ubc at a level in
+ * excess of the requested busycount
+ *
+ * Parameters: vp The vnode to check
+ * busycount The threshold busy count, used to bias
+ * the count usually already held by the
+ * caller to avoid races
+ *
+ * Returns: 1 The vnode is in use over the threshold
+ * 0 The vnode is not in use over the
+ * threshold
+ *
+ * Notes: Because the vnode is only held locked while actually asking
+ * the use count, this function only represents a snapshot of the
+ * current state of the vnode. If more accurate information is
+ * required, an additional busycount should be held by the caller
+ * and a non-zero busycount used.
+ *
+ * If there is no ubc_info associated with the vnode, this
+ * function will report that the vnode is not in use by ubc.
+ */
+int
+ubc_isinuse(struct vnode *vp, int busycount)
+{
+ if ( !UBCINFOEXISTS(vp))
+ return (0);
+ return(ubc_isinuse_locked(vp, busycount, 0));
+}
+
+
+/*
+ * ubc_isinuse_locked
+ *
+ * Determine whether or not a vnode is currently in use by ubc at a level in
+ * excess of the requested busycount
+ *
+ * Parameters: vp The vnode to check
+ * busycount The threshold busy count, used to bias
+ * the count usually already held by the
+ * caller to avoid races
+ * locked True if the vnode is already locked by
+ * the caller
+ *
+ * Returns: 1 The vnode is in use over the threshold
+ * 0 The vnode is not in use over the
+ * threshold
+ *
+ * Notes: If the vnode is not locked on entry, it is locked while
+ * actually asking the use count. If this is the case, this
+ * function only represents a snapshot of the current state of
+ * the vnode. If more accurate information is required, the
+ * vnode lock should be held by the caller, otherwise an
+ * additional busycount should be held by the caller and a
+ * non-zero busycount used.
+ *
+ * If there is no ubc_info associated with the vnode, this
+ * function will report that the vnode is not in use by ubc.
+ */
+int
+ubc_isinuse_locked(struct vnode *vp, int busycount, int locked)
+{
+ int retval = 0;
+
+
+ if (!locked)
+ vnode_lock_spin(vp);
+
+ if ((vp->v_usecount - vp->v_kusecount) > busycount)
+ retval = 1;
+
+ if (!locked)
+ vnode_unlock(vp);
+ return (retval);
+}
+
+
+/*
+ * ubc_unmap
+ *
+ * Reverse the effects of a ubc_map() call for a given vnode
+ *
+ * Parameters: vp vnode to unmap from ubc
+ *
+ * Returns: (void)
+ *
+ * Notes: This is an internal use function used by vnode_pager_unmap().
+ * It will attempt to obtain a reference on the supplied vnode,
+ * and if it can do so, and there is an associated ubc_info, and
+ * the flags indicate that it was mapped via ubc_map(), then the
+ * flag is cleared, the mapping removed, and the reference taken
+ * by ubc_map() is released.
+ *
+ * IMPORTANT: This MUST only be called by the VM
+ * to prevent race conditions.
+ */
+__private_extern__ void
+ubc_unmap(struct vnode *vp)
+{
+ struct ubc_info *uip;
+ int need_rele = 0;
+ int need_wakeup = 0;
+
+ if (vnode_getwithref(vp))
+ return;
+
+ if (UBCINFOEXISTS(vp)) {
+ bool want_fsevent = false;
+
+ vnode_lock(vp);
+ uip = vp->v_ubcinfo;
+
+ while (ISSET(uip->ui_flags, UI_MAPBUSY)) {
+ SET(uip->ui_flags, UI_MAPWAITING);
+ (void) msleep(&uip->ui_flags, &vp->v_lock,
+ PRIBIO, "ubc_unmap", NULL);
+ }
+ SET(uip->ui_flags, UI_MAPBUSY);
+
+ if (ISSET(uip->ui_flags, UI_ISMAPPED)) {
+ if (ISSET(uip->ui_flags, UI_MAPPEDWRITE))
+ want_fsevent = true;
+
+ need_rele = 1;
+
+ /*
+ * We want to clear the mapped flags after we've called
+ * VNOP_MNOMAP to avoid certain races and allow
+ * VNOP_MNOMAP to call ubc_is_mapped_writable.
+ */
+ }
+ vnode_unlock(vp);
+
+ if (need_rele) {
+ vfs_context_t ctx = vfs_context_current();
+
+ (void)VNOP_MNOMAP(vp, ctx);
+
+#if CONFIG_FSE
+ /*
+ * Why do we want an fsevent here? Normally the
+ * content modified fsevent is posted when a file is
+ * closed and only if it's written to via conventional
+ * means. It's perfectly legal to close a file and
+ * keep your mappings and we don't currently track
+ * whether it was written to via a mapping.
+ * Therefore, we need to post an fsevent here if the
+ * file was mapped writable. This may result in false
+ * events, i.e. we post a notification when nothing
+ * has really changed.
+ */
+ if (want_fsevent && need_fsevent(FSE_CONTENT_MODIFIED, vp)) {
+ add_fsevent(FSE_CONTENT_MODIFIED, ctx,
+ FSE_ARG_VNODE, vp,
+ FSE_ARG_DONE);
+ }
+#endif
+
+ vnode_rele(vp);
+ }
+
+ vnode_lock_spin(vp);
+
+ if (need_rele)
+ CLR(uip->ui_flags, UI_ISMAPPED | UI_MAPPEDWRITE);
+
+ CLR(uip->ui_flags, UI_MAPBUSY);
+
+ if (ISSET(uip->ui_flags, UI_MAPWAITING)) {
+ CLR(uip->ui_flags, UI_MAPWAITING);
+ need_wakeup = 1;
+ }
+ vnode_unlock(vp);
+
+ if (need_wakeup)
+ wakeup(&uip->ui_flags);
+
+ }
+ /*
+ * the drop of the vnode ref will cleanup
+ */
+ vnode_put(vp);
+}
+
+
+/*
+ * ubc_page_op
+ *
+ * Manipulate individual page state for a vnode with an associated ubc_info
+ * with an associated memory object control.
+ *
+ * Parameters: vp The vnode backing the page
+ * f_offset A file offset interior to the page
+ * ops The operations to perform, as a bitmap
+ * (see below for more information)
+ * phys_entryp The address of a ppnum_t; may be NULL
+ * to ignore
+ * flagsp A pointer to an int to contain flags;
+ * may be NULL to ignore
+ *
+ * Returns: KERN_SUCCESS Success
+ * KERN_INVALID_ARGUMENT If the memory object control has no VM
+ * object associated
+ * KERN_INVALID_OBJECT If UPL_POP_PHYSICAL and the object is
+ * not physically contiguous
+ * KERN_INVALID_OBJECT If !UPL_POP_PHYSICAL and the object is
+ * physically contiguous
+ * KERN_FAILURE If the page cannot be looked up
+ *
+ * Implicit Returns:
+ * *phys_entryp (modified) If phys_entryp is non-NULL and
+ * UPL_POP_PHYSICAL
+ * *flagsp (modified) If flagsp is non-NULL and there was
+ * !UPL_POP_PHYSICAL and a KERN_SUCCESS
+ *
+ * Notes: For object boundaries, it is considerably more efficient to
+ * ensure that f_offset is in fact on a page boundary, as this
+ * will avoid internal use of the hash table to identify the
+ * page, and would therefore skip a number of early optimizations.
+ * Since this is a page operation anyway, the caller should try
+ * to pass only a page aligned offset because of this.
+ *
+ * *flagsp may be modified even if this function fails. If it is
+ * modified, it will contain the condition of the page before the
+ * requested operation was attempted; these will only include the
+ * bitmap flags, and not the PL_POP_PHYSICAL, UPL_POP_DUMP,
+ * UPL_POP_SET, or UPL_POP_CLR bits.
+ *
+ * The flags field may contain a specific operation, such as
+ * UPL_POP_PHYSICAL or UPL_POP_DUMP:
+ *
+ * o UPL_POP_PHYSICAL Fail if not contiguous; if
+ * *phys_entryp and successful, set
+ * *phys_entryp
+ * o UPL_POP_DUMP Dump the specified page
+ *
+ * Otherwise, it is treated as a bitmap of one or more page
+ * operations to perform on the final memory object; allowable
+ * bit values are:
+ *
+ * o UPL_POP_DIRTY The page is dirty
+ * o UPL_POP_PAGEOUT The page is paged out
+ * o UPL_POP_PRECIOUS The page is precious
+ * o UPL_POP_ABSENT The page is absent
+ * o UPL_POP_BUSY The page is busy
+ *
+ * If the page status is only being queried and not modified, then
+ * not other bits should be specified. However, if it is being
+ * modified, exactly ONE of the following bits should be set:
+ *
+ * o UPL_POP_SET Set the current bitmap bits
+ * o UPL_POP_CLR Clear the current bitmap bits
+ *
+ * Thus to effect a combination of setting an clearing, it may be
+ * necessary to call this function twice. If this is done, the
+ * set should be used before the clear, since clearing may trigger
+ * a wakeup on the destination page, and if the page is backed by
+ * an encrypted swap file, setting will trigger the decryption
+ * needed before the wakeup occurs.
+ */
+kern_return_t
+ubc_page_op(
+ struct vnode *vp,
+ off_t f_offset,
+ int ops,
+ ppnum_t *phys_entryp,
+ int *flagsp)
+{
+ memory_object_control_t control;
+
+ control = ubc_getobject(vp, UBC_FLAGS_NONE);
+ if (control == MEMORY_OBJECT_CONTROL_NULL)
+ return KERN_INVALID_ARGUMENT;
+
+ return (memory_object_page_op(control,
+ (memory_object_offset_t)f_offset,
+ ops,
+ phys_entryp,
+ flagsp));
+}
+
+
+/*
+ * ubc_range_op
+ *
+ * Manipulate page state for a range of memory for a vnode with an associated
+ * ubc_info with an associated memory object control, when page level state is
+ * not required to be returned from the call (i.e. there are no phys_entryp or
+ * flagsp parameters to this call, and it takes a range which may contain
+ * multiple pages, rather than an offset interior to a single page).
+ *
+ * Parameters: vp The vnode backing the page
+ * f_offset_beg A file offset interior to the start page
+ * f_offset_end A file offset interior to the end page
+ * ops The operations to perform, as a bitmap
+ * (see below for more information)
+ * range The address of an int; may be NULL to
+ * ignore
+ *
+ * Returns: KERN_SUCCESS Success
+ * KERN_INVALID_ARGUMENT If the memory object control has no VM
+ * object associated
+ * KERN_INVALID_OBJECT If the object is physically contiguous
+ *
+ * Implicit Returns:
+ * *range (modified) If range is non-NULL, its contents will
+ * be modified to contain the number of
+ * bytes successfully operated upon.
+ *
+ * Notes: IMPORTANT: This function cannot be used on a range that
+ * consists of physically contiguous pages.
+ *
+ * For object boundaries, it is considerably more efficient to
+ * ensure that f_offset_beg and f_offset_end are in fact on page
+ * boundaries, as this will avoid internal use of the hash table
+ * to identify the page, and would therefore skip a number of
+ * early optimizations. Since this is an operation on a set of
+ * pages anyway, the caller should try to pass only a page aligned
+ * offsets because of this.
+ *
+ * *range will be modified only if this function succeeds.
+ *
+ * The flags field MUST contain a specific operation; allowable
+ * values are:
+ *
+ * o UPL_ROP_ABSENT Returns the extent of the range
+ * presented which is absent, starting
+ * with the start address presented
+ *
+ * o UPL_ROP_PRESENT Returns the extent of the range
+ * presented which is present (resident),
+ * starting with the start address
+ * presented
+ * o UPL_ROP_DUMP Dump the pages which are found in the
+ * target object for the target range.
+ *
+ * IMPORTANT: For UPL_ROP_ABSENT and UPL_ROP_PRESENT; if there are
+ * multiple regions in the range, only the first matching region
+ * is returned.
+ */
+kern_return_t
+ubc_range_op(
+ struct vnode *vp,
+ off_t f_offset_beg,
+ off_t f_offset_end,
+ int ops,
+ int *range)
+{
+ memory_object_control_t control;
+
+ control = ubc_getobject(vp, UBC_FLAGS_NONE);
+ if (control == MEMORY_OBJECT_CONTROL_NULL)
+ return KERN_INVALID_ARGUMENT;
+
+ return (memory_object_range_op(control,
+ (memory_object_offset_t)f_offset_beg,
+ (memory_object_offset_t)f_offset_end,
+ ops,
+ range));
+}
+
+
+/*
+ * ubc_create_upl
+ *
+ * Given a vnode, cause the population of a portion of the vm_object; based on
+ * the nature of the request, the pages returned may contain valid data, or
+ * they may be uninitialized.
+ *
+ * Parameters: vp The vnode from which to create the upl
+ * f_offset The start offset into the backing store
+ * represented by the vnode
+ * bufsize The size of the upl to create
+ * uplp Pointer to the upl_t to receive the
+ * created upl; MUST NOT be NULL
+ * plp Pointer to receive the internal page
+ * list for the created upl; MAY be NULL
+ * to ignore
+ *
+ * Returns: KERN_SUCCESS The requested upl has been created
+ * KERN_INVALID_ARGUMENT The bufsize argument is not an even
+ * multiple of the page size
+ * KERN_INVALID_ARGUMENT There is no ubc_info associated with
+ * the vnode, or there is no memory object
+ * control associated with the ubc_info
+ * memory_object_upl_request:KERN_INVALID_VALUE
+ * The supplied upl_flags argument is
+ * invalid
+ * Implicit Returns:
+ * *uplp (modified)
+ * *plp (modified) If non-NULL, the value of *plp will be
+ * modified to point to the internal page
+ * list; this modification may occur even
+ * if this function is unsuccessful, in
+ * which case the contents may be invalid
+ *
+ * Note: If successful, the returned *uplp MUST subsequently be freed
+ * via a call to ubc_upl_commit(), ubc_upl_commit_range(),
+ * ubc_upl_abort(), or ubc_upl_abort_range().
+ */
+kern_return_t
+ubc_create_upl_external(
+ struct vnode *vp,
+ off_t f_offset,
+ int bufsize,
+ upl_t *uplp,
+ upl_page_info_t **plp,
+ int uplflags)
+{
+ return (ubc_create_upl_kernel(vp, f_offset, bufsize, uplp, plp, uplflags, vm_tag_bt()));
+}
+
+kern_return_t
+ubc_create_upl_kernel(
+ struct vnode *vp,
+ off_t f_offset,
+ int bufsize,
+ upl_t *uplp,
+ upl_page_info_t **plp,
+ int uplflags,
+ vm_tag_t tag)
+{
+ memory_object_control_t control;
+ kern_return_t kr;
+
+ if (plp != NULL)
+ *plp = NULL;
+ *uplp = NULL;
+
+ if (bufsize & 0xfff)
+ return KERN_INVALID_ARGUMENT;
+
+ if (bufsize > MAX_UPL_SIZE_BYTES)
+ return KERN_INVALID_ARGUMENT;
+
+ if (uplflags & (UPL_UBC_MSYNC | UPL_UBC_PAGEOUT | UPL_UBC_PAGEIN)) {
+
+ if (uplflags & UPL_UBC_MSYNC) {
+ uplflags &= UPL_RET_ONLY_DIRTY;
+
+ uplflags |= UPL_COPYOUT_FROM | UPL_CLEAN_IN_PLACE |
+ UPL_SET_INTERNAL | UPL_SET_LITE;
+
+ } else if (uplflags & UPL_UBC_PAGEOUT) {
+ uplflags &= UPL_RET_ONLY_DIRTY;
+
+ if (uplflags & UPL_RET_ONLY_DIRTY)
+ uplflags |= UPL_NOBLOCK;
+
+ uplflags |= UPL_FOR_PAGEOUT | UPL_CLEAN_IN_PLACE |
+ UPL_COPYOUT_FROM | UPL_SET_INTERNAL | UPL_SET_LITE;
+ } else {
+ uplflags |= UPL_RET_ONLY_ABSENT |
+ UPL_NO_SYNC | UPL_CLEAN_IN_PLACE |
+ UPL_SET_INTERNAL | UPL_SET_LITE;
+
+ /*
+ * if the requested size == PAGE_SIZE, we don't want to set
+ * the UPL_NOBLOCK since we may be trying to recover from a
+ * previous partial pagein I/O that occurred because we were low
+ * on memory and bailed early in order to honor the UPL_NOBLOCK...
+ * since we're only asking for a single page, we can block w/o fear
+ * of tying up pages while waiting for more to become available
+ */
+ if (bufsize > PAGE_SIZE)
+ uplflags |= UPL_NOBLOCK;
+ }
+ } else {
+ uplflags &= ~UPL_FOR_PAGEOUT;
+
+ if (uplflags & UPL_WILL_BE_DUMPED) {
+ uplflags &= ~UPL_WILL_BE_DUMPED;
+ uplflags |= (UPL_NO_SYNC|UPL_SET_INTERNAL);
+ } else
+ uplflags |= (UPL_NO_SYNC|UPL_CLEAN_IN_PLACE|UPL_SET_INTERNAL);
+ }
+ control = ubc_getobject(vp, UBC_FLAGS_NONE);
+ if (control == MEMORY_OBJECT_CONTROL_NULL)
+ return KERN_INVALID_ARGUMENT;
+
+ kr = memory_object_upl_request(control, f_offset, bufsize, uplp, NULL, NULL, uplflags, tag);
+ if (kr == KERN_SUCCESS && plp != NULL)
+ *plp = UPL_GET_INTERNAL_PAGE_LIST(*uplp);
+ return kr;
+}
+
+
+/*
+ * ubc_upl_maxbufsize
+ *
+ * Return the maximum bufsize ubc_create_upl( ) will take.
+ *
+ * Parameters: none
+ *
+ * Returns: maximum size buffer (in bytes) ubc_create_upl( ) will take.
+ */
+upl_size_t
+ubc_upl_maxbufsize(
+ void)
+{
+ return(MAX_UPL_SIZE_BYTES);
+}
+
+/*
+ * ubc_upl_map
+ *
+ * Map the page list assocated with the supplied upl into the kernel virtual
+ * address space at the virtual address indicated by the dst_addr argument;
+ * the entire upl is mapped
+ *
+ * Parameters: upl The upl to map
+ * dst_addr The address at which to map the upl
+ *
+ * Returns: KERN_SUCCESS The upl has been mapped
+ * KERN_INVALID_ARGUMENT The upl is UPL_NULL
+ * KERN_FAILURE The upl is already mapped
+ * vm_map_enter:KERN_INVALID_ARGUMENT
+ * A failure code from vm_map_enter() due
+ * to an invalid argument
+ */
+kern_return_t
+ubc_upl_map(
+ upl_t upl,
+ vm_offset_t *dst_addr)
+{
+ return (vm_upl_map(kernel_map, upl, dst_addr));
+}
+
+
+/*
+ * ubc_upl_unmap
+ *
+ * Unmap the page list assocated with the supplied upl from the kernel virtual
+ * address space; the entire upl is unmapped.
+ *
+ * Parameters: upl The upl to unmap
+ *
+ * Returns: KERN_SUCCESS The upl has been unmapped
+ * KERN_FAILURE The upl is not currently mapped
+ * KERN_INVALID_ARGUMENT If the upl is UPL_NULL
+ */
+kern_return_t
+ubc_upl_unmap(
+ upl_t upl)
+{
+ return(vm_upl_unmap(kernel_map, upl));
+}
+
+
+/*
+ * ubc_upl_commit
+ *
+ * Commit the contents of the upl to the backing store
+ *
+ * Parameters: upl The upl to commit
+ *
+ * Returns: KERN_SUCCESS The upl has been committed
+ * KERN_INVALID_ARGUMENT The supplied upl was UPL_NULL
+ * KERN_FAILURE The supplied upl does not represent
+ * device memory, and the offset plus the
+ * size would exceed the actual size of
+ * the upl
+ *
+ * Notes: In practice, the only return value for this function should be
+ * KERN_SUCCESS, unless there has been data structure corruption;
+ * since the upl is deallocated regardless of success or failure,
+ * there's really nothing to do about this other than panic.
+ *
+ * IMPORTANT: Use of this function should not be mixed with use of
+ * ubc_upl_commit_range(), due to the unconditional deallocation
+ * by this function.
+ */
+kern_return_t
+ubc_upl_commit(
+ upl_t upl)
+{
+ upl_page_info_t *pl;
+ kern_return_t kr;
+
+ pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
+ kr = upl_commit(upl, pl, MAX_UPL_SIZE_BYTES >> PAGE_SHIFT);
+ upl_deallocate(upl);
+ return kr;
+}
+
+
+/*
+ * ubc_upl_commit
+ *
+ * Commit the contents of the specified range of the upl to the backing store
+ *
+ * Parameters: upl The upl to commit
+ * offset The offset into the upl
+ * size The size of the region to be committed,
+ * starting at the specified offset
+ * flags commit type (see below)
+ *
+ * Returns: KERN_SUCCESS The range has been committed
+ * KERN_INVALID_ARGUMENT The supplied upl was UPL_NULL
+ * KERN_FAILURE The supplied upl does not represent
+ * device memory, and the offset plus the
+ * size would exceed the actual size of
+ * the upl
+ *
+ * Notes: IMPORTANT: If the commit is successful, and the object is now
+ * empty, the upl will be deallocated. Since the caller cannot
+ * check that this is the case, the UPL_COMMIT_FREE_ON_EMPTY flag
+ * should generally only be used when the offset is 0 and the size
+ * is equal to the upl size.
+ *
+ * The flags argument is a bitmap of flags on the rage of pages in
+ * the upl to be committed; allowable flags are:
+ *
+ * o UPL_COMMIT_FREE_ON_EMPTY Free the upl when it is
+ * both empty and has been
+ * successfully committed
+ * o UPL_COMMIT_CLEAR_DIRTY Clear each pages dirty
+ * bit; will prevent a
+ * later pageout
+ * o UPL_COMMIT_SET_DIRTY Set each pages dirty
+ * bit; will cause a later
+ * pageout
+ * o UPL_COMMIT_INACTIVATE Clear each pages
+ * reference bit; the page
+ * will not be accessed
+ * o UPL_COMMIT_ALLOW_ACCESS Unbusy each page; pages
+ * become busy when an
+ * IOMemoryDescriptor is
+ * mapped or redirected,
+ * and we have to wait for
+ * an IOKit driver
+ *
+ * The flag UPL_COMMIT_NOTIFY_EMPTY is used internally, and should
+ * not be specified by the caller.
+ *
+ * The UPL_COMMIT_CLEAR_DIRTY and UPL_COMMIT_SET_DIRTY flags are
+ * mutually exclusive, and should not be combined.
+ */
+kern_return_t
+ubc_upl_commit_range(
+ upl_t upl,
+ upl_offset_t offset,
+ upl_size_t size,
+ int flags)
+{
+ upl_page_info_t *pl;
+ boolean_t empty;
+ kern_return_t kr;
+
+ if (flags & UPL_COMMIT_FREE_ON_EMPTY)
+ flags |= UPL_COMMIT_NOTIFY_EMPTY;
+
+ if (flags & UPL_COMMIT_KERNEL_ONLY_FLAGS) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ pl = UPL_GET_INTERNAL_PAGE_LIST(upl);
+
+ kr = upl_commit_range(upl, offset, size, flags,
+ pl, MAX_UPL_SIZE_BYTES >> PAGE_SHIFT, &empty);
+
+ if((flags & UPL_COMMIT_FREE_ON_EMPTY) && empty)
+ upl_deallocate(upl);
+
+ return kr;
+}
+
+
+/*
+ * ubc_upl_abort_range
+ *
+ * Abort the contents of the specified range of the specified upl
+ *
+ * Parameters: upl The upl to abort
+ * offset The offset into the upl
+ * size The size of the region to be aborted,
+ * starting at the specified offset
+ * abort_flags abort type (see below)
+ *
+ * Returns: KERN_SUCCESS The range has been aborted
+ * KERN_INVALID_ARGUMENT The supplied upl was UPL_NULL
+ * KERN_FAILURE The supplied upl does not represent
+ * device memory, and the offset plus the
+ * size would exceed the actual size of
+ * the upl
+ *
+ * Notes: IMPORTANT: If the abort is successful, and the object is now
+ * empty, the upl will be deallocated. Since the caller cannot
+ * check that this is the case, the UPL_ABORT_FREE_ON_EMPTY flag
+ * should generally only be used when the offset is 0 and the size
+ * is equal to the upl size.
+ *
+ * The abort_flags argument is a bitmap of flags on the range of
+ * pages in the upl to be aborted; allowable flags are:
+ *
+ * o UPL_ABORT_FREE_ON_EMPTY Free the upl when it is both
+ * empty and has been successfully
+ * aborted
+ * o UPL_ABORT_RESTART The operation must be restarted
+ * o UPL_ABORT_UNAVAILABLE The pages are unavailable
+ * o UPL_ABORT_ERROR An I/O error occurred
+ * o UPL_ABORT_DUMP_PAGES Just free the pages
+ * o UPL_ABORT_NOTIFY_EMPTY RESERVED
+ * o UPL_ABORT_ALLOW_ACCESS RESERVED
+ *
+ * The UPL_ABORT_NOTIFY_EMPTY is an internal use flag and should
+ * not be specified by the caller. It is intended to fulfill the
+ * same role as UPL_COMMIT_NOTIFY_EMPTY does in the function
+ * ubc_upl_commit_range(), but is never referenced internally.
+ *
+ * The UPL_ABORT_ALLOW_ACCESS is defined, but neither set nor
+ * referenced; do not use it.
+ */
+kern_return_t
+ubc_upl_abort_range(
+ upl_t upl,
+ upl_offset_t offset,
+ upl_size_t size,
+ int abort_flags)
+{
+ kern_return_t kr;
+ boolean_t empty = FALSE;
+
+ if (abort_flags & UPL_ABORT_FREE_ON_EMPTY)
+ abort_flags |= UPL_ABORT_NOTIFY_EMPTY;
+
+ kr = upl_abort_range(upl, offset, size, abort_flags, &empty);
+
+ if((abort_flags & UPL_ABORT_FREE_ON_EMPTY) && empty)
+ upl_deallocate(upl);
+
+ return kr;
+}
+
+
+/*
+ * ubc_upl_abort
+ *
+ * Abort the contents of the specified upl
+ *
+ * Parameters: upl The upl to abort
+ * abort_type abort type (see below)
+ *
+ * Returns: KERN_SUCCESS The range has been aborted
+ * KERN_INVALID_ARGUMENT The supplied upl was UPL_NULL
+ * KERN_FAILURE The supplied upl does not represent
+ * device memory, and the offset plus the
+ * size would exceed the actual size of
+ * the upl
+ *
+ * Notes: IMPORTANT: If the abort is successful, and the object is now
+ * empty, the upl will be deallocated. Since the caller cannot
+ * check that this is the case, the UPL_ABORT_FREE_ON_EMPTY flag
+ * should generally only be used when the offset is 0 and the size
+ * is equal to the upl size.
+ *
+ * The abort_type is a bitmap of flags on the range of
+ * pages in the upl to be aborted; allowable flags are:
+ *
+ * o UPL_ABORT_FREE_ON_EMPTY Free the upl when it is both
+ * empty and has been successfully
+ * aborted
+ * o UPL_ABORT_RESTART The operation must be restarted
+ * o UPL_ABORT_UNAVAILABLE The pages are unavailable
+ * o UPL_ABORT_ERROR An I/O error occurred
+ * o UPL_ABORT_DUMP_PAGES Just free the pages
+ * o UPL_ABORT_NOTIFY_EMPTY RESERVED
+ * o UPL_ABORT_ALLOW_ACCESS RESERVED
+ *
+ * The UPL_ABORT_NOTIFY_EMPTY is an internal use flag and should
+ * not be specified by the caller. It is intended to fulfill the
+ * same role as UPL_COMMIT_NOTIFY_EMPTY does in the function
+ * ubc_upl_commit_range(), but is never referenced internally.
+ *
+ * The UPL_ABORT_ALLOW_ACCESS is defined, but neither set nor
+ * referenced; do not use it.
+ */
+kern_return_t
+ubc_upl_abort(
+ upl_t upl,
+ int abort_type)
+{
+ kern_return_t kr;
+
+ kr = upl_abort(upl, abort_type);
+ upl_deallocate(upl);
+ return kr;
+}
+
+
+/*
+ * ubc_upl_pageinfo
+ *
+ * Retrieve the internal page list for the specified upl
+ *
+ * Parameters: upl The upl to obtain the page list from
+ *
+ * Returns: !NULL The (upl_page_info_t *) for the page
+ * list internal to the upl
+ * NULL Error/no page list associated
+ *
+ * Notes: IMPORTANT: The function is only valid on internal objects
+ * where the list request was made with the UPL_INTERNAL flag.
+ *
+ * This function is a utility helper function, since some callers
+ * may not have direct access to the header defining the macro,
+ * due to abstraction layering constraints.
+ */
+upl_page_info_t *
+ubc_upl_pageinfo(
+ upl_t upl)
+{
+ return (UPL_GET_INTERNAL_PAGE_LIST(upl));
+}
+
+
+int
+UBCINFOEXISTS(const struct vnode * vp)
+{
+ return((vp) && ((vp)->v_type == VREG) && ((vp)->v_ubcinfo != UBC_INFO_NULL));
+}
+
+
+void
+ubc_upl_range_needed(
+ upl_t upl,
+ int index,
+ int count)
+{
+ upl_range_needed(upl, index, count);
+}
+
+boolean_t ubc_is_mapped(const struct vnode *vp, boolean_t *writable)
+{
+ if (!UBCINFOEXISTS(vp) || !ISSET(vp->v_ubcinfo->ui_flags, UI_ISMAPPED))
+ return FALSE;
+ if (writable)
+ *writable = ISSET(vp->v_ubcinfo->ui_flags, UI_MAPPEDWRITE);
+ return TRUE;
+}
+
+boolean_t ubc_is_mapped_writable(const struct vnode *vp)
+{
+ boolean_t writable;
+ return ubc_is_mapped(vp, &writable) && writable;
+}
+
+
+/*
+ * CODE SIGNING
+ */
+static volatile SInt32 cs_blob_size = 0;
+static volatile SInt32 cs_blob_count = 0;
+static SInt32 cs_blob_size_peak = 0;
+static UInt32 cs_blob_size_max = 0;
+static SInt32 cs_blob_count_peak = 0;
+
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_count, CTLFLAG_RD | CTLFLAG_LOCKED, (int *)(uintptr_t)&cs_blob_count, 0, "Current number of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size, CTLFLAG_RD | CTLFLAG_LOCKED, (int *)(uintptr_t)&cs_blob_size, 0, "Current size of all code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_count_peak, CTLFLAG_RD | CTLFLAG_LOCKED, &cs_blob_count_peak, 0, "Peak number of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size_peak, CTLFLAG_RD | CTLFLAG_LOCKED, &cs_blob_size_peak, 0, "Peak size of code signature blobs");
+SYSCTL_INT(_vm, OID_AUTO, cs_blob_size_max, CTLFLAG_RD | CTLFLAG_LOCKED, &cs_blob_size_max, 0, "Size of biggest code signature blob");
+
+/*
+ * Function: csblob_parse_teamid
+ *
+ * Description: This function returns a pointer to the team id
+ stored within the codedirectory of the csblob.
+ If the codedirectory predates team-ids, it returns
+ NULL.
+ This does not copy the name but returns a pointer to
+ it within the CD. Subsequently, the CD must be
+ available when this is used.
+*/
+
+static const char *
+csblob_parse_teamid(struct cs_blob *csblob)
+{
+ const CS_CodeDirectory *cd;
+
+ cd = csblob->csb_cd;
+
+ if (ntohl(cd->version) < CS_SUPPORTSTEAMID)
+ return NULL;
+
+ if (cd->teamOffset == 0)
+ return NULL;
+
+ const char *name = ((const char *)cd) + ntohl(cd->teamOffset);
+ if (cs_debug > 1)
+ printf("found team-id %s in cdblob\n", name);
+
+ return name;
+}
+
+
+kern_return_t
+ubc_cs_blob_allocate(
+ vm_offset_t *blob_addr_p,
+ vm_size_t *blob_size_p)
+{
+ kern_return_t kr;
+
+ *blob_addr_p = (vm_offset_t) kalloc_tag(*blob_size_p, VM_KERN_MEMORY_SECURITY);
+ if (*blob_addr_p == 0) {
+ kr = KERN_NO_SPACE;
+ } else {
+ kr = KERN_SUCCESS;
+ }
+ return kr;
+}
+
+void
+ubc_cs_blob_deallocate(
+ vm_offset_t blob_addr,
+ vm_size_t blob_size)
+{
+ kfree((void *) blob_addr, blob_size);
+}
+
+/*
+ * Some codesigned files use a lowest common denominator page size of
+ * 4KiB, but can be used on systems that have a runtime page size of
+ * 16KiB. Since faults will only occur on 16KiB ranges in
+ * cs_validate_range(), we can convert the original Code Directory to
+ * a multi-level scheme where groups of 4 hashes are combined to form
+ * a new hash, which represents 16KiB in the on-disk file. This can
+ * reduce the wired memory requirement for the Code Directory by
+ * 75%. Care must be taken for binaries that use the "fourk" VM pager
+ * for unaligned access, which may still attempt to validate on
+ * non-16KiB multiples for compatibility with 3rd party binaries.
+ */
+static boolean_t
+ubc_cs_supports_multilevel_hash(struct cs_blob *blob)
+{
+ const CS_CodeDirectory *cd;
+
+ /*
+ * Only applies to binaries that ship as part of the OS,
+ * primarily the shared cache.
+ */
+ if (!blob->csb_platform_binary || blob->csb_teamid != NULL) {
+ return FALSE;
+ }
+
+ /*
+ * If the runtime page size matches the code signing page
+ * size, there is no work to do.
+ */
+ if (PAGE_SHIFT <= blob->csb_hash_pageshift) {
+ return FALSE;
+ }
+
+ cd = blob->csb_cd;
+
+ /*
+ * There must be a valid integral multiple of hashes
+ */
+ if (ntohl(cd->nCodeSlots) & (PAGE_MASK >> blob->csb_hash_pageshift)) {
+ return FALSE;
+ }
+
+ /*
+ * Scatter lists must also have ranges that have an integral number of hashes
+ */
+ if ((ntohl(cd->version) >= CS_SUPPORTSSCATTER) && (ntohl(cd->scatterOffset))) {
+
+ const SC_Scatter *scatter = (const SC_Scatter*)
+ ((const char*)cd + ntohl(cd->scatterOffset));
+ /* iterate all scatter structs to make sure they are all aligned */
+ do {
+ uint32_t sbase = ntohl(scatter->base);
+ uint32_t scount = ntohl(scatter->count);
+
+ /* last scatter? */
+ if (scount == 0) {
+ break;
+ }
+
+ if (sbase & (PAGE_MASK >> blob->csb_hash_pageshift)) {
+ return FALSE;
+ }
+
+ if (scount & (PAGE_MASK >> blob->csb_hash_pageshift)) {
+ return FALSE;
+ }
+
+ scatter++;
+ } while(1);
+ }
+
+ /* Covered range must be a multiple of the new page size */
+ if (ntohl(cd->codeLimit) & PAGE_MASK) {
+ return FALSE;
+ }
+
+ /* All checks pass */
+ return TRUE;
+}
+
+/*
+ * All state and preconditions were checked before, so this
+ * function cannot fail.
+ */
+static void
+ubc_cs_convert_to_multilevel_hash(struct cs_blob *blob)
+{
+ const CS_CodeDirectory *old_cd, *cd;
+ CS_CodeDirectory *new_cd;
+ const CS_GenericBlob *entitlements;
+ vm_offset_t new_blob_addr;
+ vm_size_t new_blob_size;
+ vm_size_t new_cdsize;
+ kern_return_t kr;
+ int error;
+ size_t length;
+
+ uint32_t hashes_per_new_hash_shift = (uint32_t)(PAGE_SHIFT - blob->csb_hash_pageshift);
+
+ if (cs_debug > 1) {
+ printf("CODE SIGNING: Attempting to convert Code Directory for %lu -> %lu page shift\n",
+ (unsigned long)blob->csb_hash_pageshift, (unsigned long)PAGE_SHIFT);
+ }
+
+ old_cd = blob->csb_cd;
+
+ /* Up to the hashes, we can copy all data */
+ new_cdsize = ntohl(old_cd->hashOffset);
+ new_cdsize += (ntohl(old_cd->nCodeSlots) >> hashes_per_new_hash_shift) * old_cd->hashSize;
+
+ new_blob_size = sizeof(CS_SuperBlob);
+ new_blob_size += sizeof(CS_BlobIndex);
+ new_blob_size += new_cdsize;
+
+ if (blob->csb_entitlements_blob) {
+ /* We need to add a slot for the entitlements */
+ new_blob_size += sizeof(CS_BlobIndex);
+ new_blob_size += ntohl(blob->csb_entitlements_blob->length);
+ }
projects / apple / xnu.git / blobdiff