+ * Description:
+ * Copy in data to a kernel buffer from space in the
+ * source map. The original space may be optionally
+ * deallocated.
+ *
+ * If successful, returns a new copy object.
+ */
+static kern_return_t
+vm_map_copyin_kernel_buffer(
+ vm_map_t src_map,
+ vm_map_offset_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ vm_map_copy_t *copy_result)
+{
+ kern_return_t kr;
+ vm_map_copy_t copy;
+ vm_size_t kalloc_size;
+
+ if (len > msg_ool_size_small)
+ return KERN_INVALID_ARGUMENT;
+
+ kalloc_size = (vm_size_t)(cpy_kdata_hdr_sz + len);
+
+ copy = (vm_map_copy_t)kalloc(kalloc_size);
+ if (copy == VM_MAP_COPY_NULL)
+ return KERN_RESOURCE_SHORTAGE;
+ copy->type = VM_MAP_COPY_KERNEL_BUFFER;
+ copy->size = len;
+ copy->offset = 0;
+
+ kr = copyinmap(src_map, src_addr, copy->cpy_kdata, (vm_size_t)len);
+ if (kr != KERN_SUCCESS) {
+ kfree(copy, kalloc_size);
+ return kr;
+ }
+ if (src_destroy) {
+ (void) vm_map_remove(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ vm_map_round_page(src_addr + len,
+ VM_MAP_PAGE_MASK(src_map)),
+ (VM_MAP_REMOVE_INTERRUPTIBLE |
+ VM_MAP_REMOVE_WAIT_FOR_KWIRE |
+ (src_map == kernel_map) ? VM_MAP_REMOVE_KUNWIRE : 0));
+ }
+ *copy_result = copy;
+ return KERN_SUCCESS;
+}
+
+/*
+ * Routine: vm_map_copyout_kernel_buffer [internal use only]
+ *
+ * Description:
+ * Copy out data from a kernel buffer into space in the
+ * destination map. The space may be otpionally dynamically
+ * allocated.
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ */
+static int vm_map_copyout_kernel_buffer_failures = 0;
+static kern_return_t
+vm_map_copyout_kernel_buffer(
+ vm_map_t map,
+ vm_map_address_t *addr, /* IN/OUT */
+ vm_map_copy_t copy,
+ boolean_t overwrite,
+ boolean_t consume_on_success)
+{
+ kern_return_t kr = KERN_SUCCESS;
+ thread_t thread = current_thread();
+
+ /*
+ * check for corrupted vm_map_copy structure
+ */
+ if (copy->size > msg_ool_size_small || copy->offset)
+ panic("Invalid vm_map_copy_t sz:%lld, ofst:%lld",
+ (long long)copy->size, (long long)copy->offset);
+
+ if (!overwrite) {
+
+ /*
+ * Allocate space in the target map for the data
+ */
+ *addr = 0;
+ kr = vm_map_enter(map,
+ addr,
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(map)),
+ (vm_map_offset_t) 0,
+ VM_FLAGS_ANYWHERE,
+ VM_OBJECT_NULL,
+ (vm_object_offset_t) 0,
+ FALSE,
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+ if (kr != KERN_SUCCESS)
+ return kr;
+ }
+
+ /*
+ * Copyout the data from the kernel buffer to the target map.
+ */
+ if (thread->map == map) {
+
+ /*
+ * If the target map is the current map, just do
+ * the copy.
+ */
+ assert((vm_size_t) copy->size == copy->size);
+ if (copyout(copy->cpy_kdata, *addr, (vm_size_t) copy->size)) {
+ kr = KERN_INVALID_ADDRESS;
+ }
+ }
+ else {
+ vm_map_t oldmap;
+
+ /*
+ * If the target map is another map, assume the
+ * target's address space identity for the duration
+ * of the copy.
+ */
+ vm_map_reference(map);
+ oldmap = vm_map_switch(map);
+
+ assert((vm_size_t) copy->size == copy->size);
+ if (copyout(copy->cpy_kdata, *addr, (vm_size_t) copy->size)) {
+ vm_map_copyout_kernel_buffer_failures++;
+ kr = KERN_INVALID_ADDRESS;
+ }
+
+ (void) vm_map_switch(oldmap);
+ vm_map_deallocate(map);
+ }
+
+ if (kr != KERN_SUCCESS) {
+ /* the copy failed, clean up */
+ if (!overwrite) {
+ /*
+ * Deallocate the space we allocated in the target map.
+ */
+ (void) vm_map_remove(
+ map,
+ vm_map_trunc_page(*addr,
+ VM_MAP_PAGE_MASK(map)),
+ vm_map_round_page((*addr +
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(map))),
+ VM_MAP_PAGE_MASK(map)),
+ VM_MAP_NO_FLAGS);
+ *addr = 0;
+ }
+ } else {
+ /* copy was successful, dicard the copy structure */
+ if (consume_on_success) {
+ kfree(copy, copy->size + cpy_kdata_hdr_sz);
+ }
+ }
+
+ return kr;
+}
+
+/*
+ * Macro: vm_map_copy_insert
+ *
+ * Description:
+ * Link a copy chain ("copy") into a map at the
+ * specified location (after "where").
+ * Side effects:
+ * The copy chain is destroyed.
+ * Warning:
+ * The arguments are evaluated multiple times.
+ */
+#define vm_map_copy_insert(map, where, copy) \
+MACRO_BEGIN \
+ vm_map_store_copy_insert(map, where, copy); \
+ zfree(vm_map_copy_zone, copy); \
+MACRO_END
+
+void
+vm_map_copy_remap(
+ vm_map_t map,
+ vm_map_entry_t where,
+ vm_map_copy_t copy,
+ vm_map_offset_t adjustment,
+ vm_prot_t cur_prot,
+ vm_prot_t max_prot,
+ vm_inherit_t inheritance)
+{
+ vm_map_entry_t copy_entry, new_entry;
+
+ for (copy_entry = vm_map_copy_first_entry(copy);
+ copy_entry != vm_map_copy_to_entry(copy);
+ copy_entry = copy_entry->vme_next) {
+ /* get a new VM map entry for the map */
+ new_entry = vm_map_entry_create(map,
+ !map->hdr.entries_pageable);
+ /* copy the "copy entry" to the new entry */
+ vm_map_entry_copy(new_entry, copy_entry);
+ /* adjust "start" and "end" */
+ new_entry->vme_start += adjustment;
+ new_entry->vme_end += adjustment;
+ /* clear some attributes */
+ new_entry->inheritance = inheritance;
+ new_entry->protection = cur_prot;
+ new_entry->max_protection = max_prot;
+ new_entry->behavior = VM_BEHAVIOR_DEFAULT;
+ /* take an extra reference on the entry's "object" */
+ if (new_entry->is_sub_map) {
+ assert(!new_entry->use_pmap); /* not nested */
+ vm_map_lock(VME_SUBMAP(new_entry));
+ vm_map_reference(VME_SUBMAP(new_entry));
+ vm_map_unlock(VME_SUBMAP(new_entry));
+ } else {
+ vm_object_reference(VME_OBJECT(new_entry));
+ }
+ /* insert the new entry in the map */
+ vm_map_store_entry_link(map, where, new_entry);
+ /* continue inserting the "copy entries" after the new entry */
+ where = new_entry;
+ }
+}
+
+
+boolean_t
+vm_map_copy_validate_size(
+ vm_map_t dst_map,
+ vm_map_copy_t copy,
+ vm_map_size_t size)
+{
+ if (copy == VM_MAP_COPY_NULL)
+ return FALSE;
+ switch (copy->type) {
+ case VM_MAP_COPY_OBJECT:
+ case VM_MAP_COPY_KERNEL_BUFFER:
+ if (size == copy->size)
+ return TRUE;
+ break;
+ case VM_MAP_COPY_ENTRY_LIST:
+ /*
+ * potential page-size rounding prevents us from exactly
+ * validating this flavor of vm_map_copy, but we can at least
+ * assert that it's within a range.
+ */
+ if (copy->size >= size &&
+ copy->size <= vm_map_round_page(size,
+ VM_MAP_PAGE_MASK(dst_map)))
+ return TRUE;
+ break;
+ default:
+ break;
+ }
+ return FALSE;
+}
+
+
+/*
+ * Routine: vm_map_copyout
+ *
+ * Description:
+ * Copy out a copy chain ("copy") into newly-allocated
+ * space in the destination map.
+ *
+ * If successful, consumes the copy object.
+ * Otherwise, the caller is responsible for it.
+ */
+
+kern_return_t
+vm_map_copyout(
+ vm_map_t dst_map,
+ vm_map_address_t *dst_addr, /* OUT */
+ vm_map_copy_t copy)
+{
+ return vm_map_copyout_internal(dst_map, dst_addr, copy,
+ TRUE, /* consume_on_success */
+ VM_PROT_DEFAULT,
+ VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+}
+
+kern_return_t
+vm_map_copyout_internal(
+ vm_map_t dst_map,
+ vm_map_address_t *dst_addr, /* OUT */
+ vm_map_copy_t copy,
+ boolean_t consume_on_success,
+ vm_prot_t cur_protection,
+ vm_prot_t max_protection,
+ vm_inherit_t inheritance)
+{
+ vm_map_size_t size;
+ vm_map_size_t adjustment;
+ vm_map_offset_t start;
+ vm_object_offset_t vm_copy_start;
+ vm_map_entry_t last;
+ vm_map_entry_t entry;
+ vm_map_entry_t hole_entry;
+
+ /*
+ * Check for null copy object.
+ */
+
+ if (copy == VM_MAP_COPY_NULL) {
+ *dst_addr = 0;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check for special copy object, created
+ * by vm_map_copyin_object.
+ */
+
+ if (copy->type == VM_MAP_COPY_OBJECT) {
+ vm_object_t object = copy->cpy_object;
+ kern_return_t kr;
+ vm_object_offset_t offset;
+
+ offset = vm_object_trunc_page(copy->offset);
+ size = vm_map_round_page((copy->size +
+ (vm_map_size_t)(copy->offset -
+ offset)),
+ VM_MAP_PAGE_MASK(dst_map));
+ *dst_addr = 0;
+ kr = vm_map_enter(dst_map, dst_addr, size,
+ (vm_map_offset_t) 0, VM_FLAGS_ANYWHERE,
+ object, offset, FALSE,
+ VM_PROT_DEFAULT, VM_PROT_ALL,
+ VM_INHERIT_DEFAULT);
+ if (kr != KERN_SUCCESS)
+ return(kr);
+ /* Account for non-pagealigned copy object */
+ *dst_addr += (vm_map_offset_t)(copy->offset - offset);
+ if (consume_on_success)
+ zfree(vm_map_copy_zone, copy);
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check for special kernel buffer allocated
+ * by new_ipc_kmsg_copyin.
+ */
+
+ if (copy->type == VM_MAP_COPY_KERNEL_BUFFER) {
+ return vm_map_copyout_kernel_buffer(dst_map, dst_addr,
+ copy, FALSE,
+ consume_on_success);
+ }
+
+
+ /*
+ * Find space for the data
+ */
+
+ vm_copy_start = vm_map_trunc_page((vm_map_size_t)copy->offset,
+ VM_MAP_COPY_PAGE_MASK(copy));
+ size = vm_map_round_page((vm_map_size_t)copy->offset + copy->size,
+ VM_MAP_COPY_PAGE_MASK(copy))
+ - vm_copy_start;
+
+
+StartAgain: ;
+
+ vm_map_lock(dst_map);
+ if( dst_map->disable_vmentry_reuse == TRUE) {
+ VM_MAP_HIGHEST_ENTRY(dst_map, entry, start);
+ last = entry;
+ } else {
+ if (dst_map->holelistenabled) {
+ hole_entry = (vm_map_entry_t)dst_map->holes_list;
+
+ if (hole_entry == NULL) {
+ /*
+ * No more space in the map?
+ */
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+
+ last = hole_entry;
+ start = last->vme_start;
+ } else {
+ assert(first_free_is_valid(dst_map));
+ start = ((last = dst_map->first_free) == vm_map_to_entry(dst_map)) ?
+ vm_map_min(dst_map) : last->vme_end;
+ }
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(dst_map));
+ }
+
+ while (TRUE) {
+ vm_map_entry_t next = last->vme_next;
+ vm_map_offset_t end = start + size;
+
+ if ((end > dst_map->max_offset) || (end < start)) {
+ if (dst_map->wait_for_space) {
+ if (size <= (dst_map->max_offset - dst_map->min_offset)) {
+ assert_wait((event_t) dst_map,
+ THREAD_INTERRUPTIBLE);
+ vm_map_unlock(dst_map);
+ thread_block(THREAD_CONTINUE_NULL);
+ goto StartAgain;
+ }
+ }
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+
+ if (dst_map->holelistenabled) {
+ if (last->vme_end >= end)
+ break;
+ } else {
+ /*
+ * If there are no more entries, we must win.
+ *
+ * OR
+ *
+ * If there is another entry, it must be
+ * after the end of the potential new region.
+ */
+
+ if (next == vm_map_to_entry(dst_map))
+ break;
+
+ if (next->vme_start >= end)
+ break;
+ }
+
+ last = next;
+
+ if (dst_map->holelistenabled) {
+ if (last == (vm_map_entry_t) dst_map->holes_list) {
+ /*
+ * Wrapped around
+ */
+ vm_map_unlock(dst_map);
+ return(KERN_NO_SPACE);
+ }
+ start = last->vme_start;
+ } else {
+ start = last->vme_end;
+ }
+ start = vm_map_round_page(start,
+ VM_MAP_PAGE_MASK(dst_map));
+ }
+
+ if (dst_map->holelistenabled) {
+ if (vm_map_lookup_entry(dst_map, last->vme_start, &last)) {
+ panic("Found an existing entry (%p) instead of potential hole at address: 0x%llx.\n", last, (unsigned long long)last->vme_start);
+ }
+ }
+
+
+ adjustment = start - vm_copy_start;
+ if (! consume_on_success) {
+ /*
+ * We're not allowed to consume "copy", so we'll have to
+ * copy its map entries into the destination map below.
+ * No need to re-allocate map entries from the correct
+ * (pageable or not) zone, since we'll get new map entries
+ * during the transfer.
+ * We'll also adjust the map entries's "start" and "end"
+ * during the transfer, to keep "copy"'s entries consistent
+ * with its "offset".
+ */
+ goto after_adjustments;
+ }
+
+ /*
+ * Since we're going to just drop the map
+ * entries from the copy into the destination
+ * map, they must come from the same pool.
+ */
+
+ if (copy->cpy_hdr.entries_pageable != dst_map->hdr.entries_pageable) {
+ /*
+ * Mismatches occur when dealing with the default
+ * pager.
+ */
+ zone_t old_zone;
+ vm_map_entry_t next, new;
+
+ /*
+ * Find the zone that the copies were allocated from
+ */
+
+ entry = vm_map_copy_first_entry(copy);
+
+ /*
+ * Reinitialize the copy so that vm_map_copy_entry_link
+ * will work.
+ */
+ vm_map_store_copy_reset(copy, entry);
+ copy->cpy_hdr.entries_pageable = dst_map->hdr.entries_pageable;
+
+ /*
+ * Copy each entry.
+ */
+ while (entry != vm_map_copy_to_entry(copy)) {
+ new = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+ vm_map_entry_copy_full(new, entry);
+ assert(!new->iokit_acct);
+ if (new->is_sub_map) {
+ /* clr address space specifics */
+ new->use_pmap = FALSE;
+ }
+ vm_map_copy_entry_link(copy,
+ vm_map_copy_last_entry(copy),
+ new);
+ next = entry->vme_next;
+ old_zone = entry->from_reserved_zone ? vm_map_entry_reserved_zone : vm_map_entry_zone;
+ zfree(old_zone, entry);
+ entry = next;
+ }
+ }
+
+ /*
+ * Adjust the addresses in the copy chain, and
+ * reset the region attributes.
+ */
+
+ for (entry = vm_map_copy_first_entry(copy);
+ entry != vm_map_copy_to_entry(copy);
+ entry = entry->vme_next) {
+ if (VM_MAP_PAGE_SHIFT(dst_map) == PAGE_SHIFT) {
+ /*
+ * We're injecting this copy entry into a map that
+ * has the standard page alignment, so clear
+ * "map_aligned" (which might have been inherited
+ * from the original map entry).
+ */
+ entry->map_aligned = FALSE;
+ }
+
+ entry->vme_start += adjustment;
+ entry->vme_end += adjustment;
+
+ if (entry->map_aligned) {
+ assert(VM_MAP_PAGE_ALIGNED(entry->vme_start,
+ VM_MAP_PAGE_MASK(dst_map)));
+ assert(VM_MAP_PAGE_ALIGNED(entry->vme_end,
+ VM_MAP_PAGE_MASK(dst_map)));
+ }
+
+ entry->inheritance = VM_INHERIT_DEFAULT;
+ entry->protection = VM_PROT_DEFAULT;
+ entry->max_protection = VM_PROT_ALL;
+ entry->behavior = VM_BEHAVIOR_DEFAULT;
+
+ /*
+ * If the entry is now wired,
+ * map the pages into the destination map.
+ */
+ if (entry->wired_count != 0) {
+ register vm_map_offset_t va;
+ vm_object_offset_t offset;
+ register vm_object_t object;
+ vm_prot_t prot;
+ int type_of_fault;
+
+ object = VME_OBJECT(entry);
+ offset = VME_OFFSET(entry);
+ va = entry->vme_start;
+
+ pmap_pageable(dst_map->pmap,
+ entry->vme_start,
+ entry->vme_end,
+ TRUE);
+
+ while (va < entry->vme_end) {
+ register vm_page_t m;
+
+ /*
+ * Look up the page in the object.
+ * Assert that the page will be found in the
+ * top object:
+ * either
+ * the object was newly created by
+ * vm_object_copy_slowly, and has
+ * copies of all of the pages from
+ * the source object
+ * or
+ * the object was moved from the old
+ * map entry; because the old map
+ * entry was wired, all of the pages
+ * were in the top-level object.
+ * (XXX not true if we wire pages for
+ * reading)
+ */
+ vm_object_lock(object);
+
+ m = vm_page_lookup(object, offset);
+ if (m == VM_PAGE_NULL || !VM_PAGE_WIRED(m) ||
+ m->absent)
+ panic("vm_map_copyout: wiring %p", m);
+
+ /*
+ * ENCRYPTED SWAP:
+ * The page is assumed to be wired here, so it
+ * shouldn't be encrypted. Otherwise, we
+ * couldn't enter it in the page table, since
+ * we don't want the user to see the encrypted
+ * data.
+ */
+ ASSERT_PAGE_DECRYPTED(m);
+
+ prot = entry->protection;
+
+ if (override_nx(dst_map, VME_ALIAS(entry)) &&
+ prot)
+ prot |= VM_PROT_EXECUTE;
+
+ type_of_fault = DBG_CACHE_HIT_FAULT;
+
+ vm_fault_enter(m, dst_map->pmap, va, prot, prot,
+ VM_PAGE_WIRED(m), FALSE, FALSE,
+ FALSE, VME_ALIAS(entry),
+ ((entry->iokit_acct ||
+ (!entry->is_sub_map &&
+ !entry->use_pmap))
+ ? PMAP_OPTIONS_ALT_ACCT
+ : 0),
+ NULL, &type_of_fault);
+
+ vm_object_unlock(object);
+
+ offset += PAGE_SIZE_64;
+ va += PAGE_SIZE;
+ }
+ }
+ }
+
+after_adjustments:
+
+ /*
+ * Correct the page alignment for the result
+ */
+
+ *dst_addr = start + (copy->offset - vm_copy_start);
+
+ /*
+ * Update the hints and the map size
+ */
+
+ if (consume_on_success) {
+ SAVE_HINT_MAP_WRITE(dst_map, vm_map_copy_last_entry(copy));
+ } else {
+ SAVE_HINT_MAP_WRITE(dst_map, last);
+ }
+
+ dst_map->size += size;
+
+ /*
+ * Link in the copy
+ */
+
+ if (consume_on_success) {
+ vm_map_copy_insert(dst_map, last, copy);
+ } else {
+ vm_map_copy_remap(dst_map, last, copy, adjustment,
+ cur_protection, max_protection,
+ inheritance);
+ }
+
+ vm_map_unlock(dst_map);
+
+ /*
+ * XXX If wiring_required, call vm_map_pageable
+ */
+
+ return(KERN_SUCCESS);
+}
+
+/*
+ * Routine: vm_map_copyin
+ *
+ * Description:
+ * see vm_map_copyin_common. Exported via Unsupported.exports.
+ *
+ */
+
+#undef vm_map_copyin
+
+kern_return_t
+vm_map_copyin(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ return(vm_map_copyin_common(src_map, src_addr, len, src_destroy,
+ FALSE, copy_result, FALSE));
+}
+
+/*
+ * Routine: vm_map_copyin_common
+ *
+ * Description:
+ * Copy the specified region (src_addr, len) from the
+ * source address space (src_map), possibly removing
+ * the region from the source address space (src_destroy).
+ *
+ * Returns:
+ * A vm_map_copy_t object (copy_result), suitable for
+ * insertion into another address space (using vm_map_copyout),
+ * copying over another address space region (using
+ * vm_map_copy_overwrite). If the copy is unused, it
+ * should be destroyed (using vm_map_copy_discard).
+ *
+ * In/out conditions:
+ * The source map should not be locked on entry.
+ */
+
+typedef struct submap_map {
+ vm_map_t parent_map;
+ vm_map_offset_t base_start;
+ vm_map_offset_t base_end;
+ vm_map_size_t base_len;
+ struct submap_map *next;
+} submap_map_t;
+
+kern_return_t
+vm_map_copyin_common(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ boolean_t src_destroy,
+ __unused boolean_t src_volatile,
+ vm_map_copy_t *copy_result, /* OUT */
+ boolean_t use_maxprot)
+{
+ int flags;
+
+ flags = 0;
+ if (src_destroy) {
+ flags |= VM_MAP_COPYIN_SRC_DESTROY;
+ }
+ if (use_maxprot) {
+ flags |= VM_MAP_COPYIN_USE_MAXPROT;
+ }
+ return vm_map_copyin_internal(src_map,
+ src_addr,
+ len,
+ flags,
+ copy_result);
+}
+kern_return_t
+vm_map_copyin_internal(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ int flags,
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ vm_map_entry_t tmp_entry; /* Result of last map lookup --
+ * in multi-level lookup, this
+ * entry contains the actual
+ * vm_object/offset.
+ */
+ vm_map_entry_t new_entry = VM_MAP_ENTRY_NULL; /* Map entry for copy */
+
+ vm_map_offset_t src_start; /* Start of current entry --
+ * where copy is taking place now
+ */
+ vm_map_offset_t src_end; /* End of entire region to be
+ * copied */
+ vm_map_offset_t src_base;
+ vm_map_t base_map = src_map;
+ boolean_t map_share=FALSE;
+ submap_map_t *parent_maps = NULL;
+
+ vm_map_copy_t copy; /* Resulting copy */
+ vm_map_address_t copy_addr;
+ vm_map_size_t copy_size;
+ boolean_t src_destroy;
+ boolean_t use_maxprot;
+
+ if (flags & ~VM_MAP_COPYIN_ALL_FLAGS) {
+ return KERN_INVALID_ARGUMENT;
+ }
+
+ src_destroy = (flags & VM_MAP_COPYIN_SRC_DESTROY) ? TRUE : FALSE;
+ use_maxprot = (flags & VM_MAP_COPYIN_USE_MAXPROT) ? TRUE : FALSE;
+
+ /*
+ * Check for copies of zero bytes.
+ */
+
+ if (len == 0) {
+ *copy_result = VM_MAP_COPY_NULL;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check that the end address doesn't overflow
+ */
+ src_end = src_addr + len;
+ if (src_end < src_addr)
+ return KERN_INVALID_ADDRESS;
+
+ /*
+ * If the copy is sufficiently small, use a kernel buffer instead
+ * of making a virtual copy. The theory being that the cost of
+ * setting up VM (and taking C-O-W faults) dominates the copy costs
+ * for small regions.
+ */
+ if ((len < msg_ool_size_small) &&
+ !use_maxprot &&
+ !(flags & VM_MAP_COPYIN_ENTRY_LIST))
+ return vm_map_copyin_kernel_buffer(src_map, src_addr, len,
+ src_destroy, copy_result);
+
+ /*
+ * Compute (page aligned) start and end of region
+ */
+ src_start = vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map));
+ src_end = vm_map_round_page(src_end,
+ VM_MAP_PAGE_MASK(src_map));
+
+ XPR(XPR_VM_MAP, "vm_map_copyin_common map 0x%x addr 0x%x len 0x%x dest %d\n", src_map, src_addr, len, src_destroy, 0);
+
+ /*
+ * Allocate a header element for the list.
+ *
+ * Use the start and end in the header to
+ * remember the endpoints prior to rounding.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(copy) =
+ vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy);
+ copy->type = VM_MAP_COPY_ENTRY_LIST;
+ copy->cpy_hdr.nentries = 0;
+ copy->cpy_hdr.entries_pageable = TRUE;
+#if 00
+ copy->cpy_hdr.page_shift = src_map->hdr.page_shift;
+#else
+ /*
+ * The copy entries can be broken down for a variety of reasons,
+ * so we can't guarantee that they will remain map-aligned...
+ * Will need to adjust the first copy_entry's "vme_start" and
+ * the last copy_entry's "vme_end" to be rounded to PAGE_MASK
+ * rather than the original map's alignment.
+ */
+ copy->cpy_hdr.page_shift = PAGE_SHIFT;
+#endif
+
+ vm_map_store_init( &(copy->cpy_hdr) );
+
+ copy->offset = src_addr;
+ copy->size = len;
+
+ new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+
+#define RETURN(x) \
+ MACRO_BEGIN \
+ vm_map_unlock(src_map); \
+ if(src_map != base_map) \
+ vm_map_deallocate(src_map); \
+ if (new_entry != VM_MAP_ENTRY_NULL) \
+ vm_map_copy_entry_dispose(copy,new_entry); \
+ vm_map_copy_discard(copy); \
+ { \
+ submap_map_t *_ptr; \
+ \
+ for(_ptr = parent_maps; _ptr != NULL; _ptr = parent_maps) { \
+ parent_maps=parent_maps->next; \
+ if (_ptr->parent_map != base_map) \
+ vm_map_deallocate(_ptr->parent_map); \
+ kfree(_ptr, sizeof(submap_map_t)); \
+ } \
+ } \
+ MACRO_RETURN(x); \
+ MACRO_END
+
+ /*
+ * Find the beginning of the region.
+ */
+
+ vm_map_lock(src_map);
+
+ /*
+ * Lookup the original "src_addr" rather than the truncated
+ * "src_start", in case "src_start" falls in a non-map-aligned
+ * map entry *before* the map entry that contains "src_addr"...
+ */
+ if (!vm_map_lookup_entry(src_map, src_addr, &tmp_entry))
+ RETURN(KERN_INVALID_ADDRESS);
+ if(!tmp_entry->is_sub_map) {
+ /*
+ * ... but clip to the map-rounded "src_start" rather than
+ * "src_addr" to preserve map-alignment. We'll adjust the
+ * first copy entry at the end, if needed.
+ */
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ }
+ if (src_start < tmp_entry->vme_start) {
+ /*
+ * Move "src_start" up to the start of the
+ * first map entry to copy.
+ */
+ src_start = tmp_entry->vme_start;
+ }
+ /* set for later submap fix-up */
+ copy_addr = src_start;
+
+ /*
+ * Go through entries until we get to the end.
+ */
+
+ while (TRUE) {
+ register
+ vm_map_entry_t src_entry = tmp_entry; /* Top-level entry */
+ vm_map_size_t src_size; /* Size of source
+ * map entry (in both
+ * maps)
+ */
+
+ register
+ vm_object_t src_object; /* Object to copy */
+ vm_object_offset_t src_offset;
+
+ boolean_t src_needs_copy; /* Should source map
+ * be made read-only
+ * for copy-on-write?
+ */
+
+ boolean_t new_entry_needs_copy; /* Will new entry be COW? */
+
+ boolean_t was_wired; /* Was source wired? */
+ vm_map_version_t version; /* Version before locks
+ * dropped to make copy
+ */
+ kern_return_t result; /* Return value from
+ * copy_strategically.
+ */
+ while(tmp_entry->is_sub_map) {
+ vm_map_size_t submap_len;
+ submap_map_t *ptr;
+
+ ptr = (submap_map_t *)kalloc(sizeof(submap_map_t));
+ ptr->next = parent_maps;
+ parent_maps = ptr;
+ ptr->parent_map = src_map;
+ ptr->base_start = src_start;
+ ptr->base_end = src_end;
+ submap_len = tmp_entry->vme_end - src_start;
+ if(submap_len > (src_end-src_start))
+ submap_len = src_end-src_start;
+ ptr->base_len = submap_len;
+
+ src_start -= tmp_entry->vme_start;
+ src_start += VME_OFFSET(tmp_entry);
+ src_end = src_start + submap_len;
+ src_map = VME_SUBMAP(tmp_entry);
+ vm_map_lock(src_map);
+ /* keep an outstanding reference for all maps in */
+ /* the parents tree except the base map */
+ vm_map_reference(src_map);
+ vm_map_unlock(ptr->parent_map);
+ if (!vm_map_lookup_entry(
+ src_map, src_start, &tmp_entry))
+ RETURN(KERN_INVALID_ADDRESS);
+ map_share = TRUE;
+ if(!tmp_entry->is_sub_map)
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ src_entry = tmp_entry;
+ }
+ /* we are now in the lowest level submap... */
+
+ if ((VME_OBJECT(tmp_entry) != VM_OBJECT_NULL) &&
+ (VME_OBJECT(tmp_entry)->phys_contiguous)) {
+ /* This is not, supported for now.In future */
+ /* we will need to detect the phys_contig */
+ /* condition and then upgrade copy_slowly */
+ /* to do physical copy from the device mem */
+ /* based object. We can piggy-back off of */
+ /* the was wired boolean to set-up the */
+ /* proper handling */
+ RETURN(KERN_PROTECTION_FAILURE);
+ }
+ /*
+ * Create a new address map entry to hold the result.
+ * Fill in the fields from the appropriate source entries.
+ * We must unlock the source map to do this if we need
+ * to allocate a map entry.
+ */
+ if (new_entry == VM_MAP_ENTRY_NULL) {
+ version.main_timestamp = src_map->timestamp;
+ vm_map_unlock(src_map);
+
+ new_entry = vm_map_copy_entry_create(copy, !copy->cpy_hdr.entries_pageable);
+
+ vm_map_lock(src_map);
+ if ((version.main_timestamp + 1) != src_map->timestamp) {
+ if (!vm_map_lookup_entry(src_map, src_start,
+ &tmp_entry)) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ if (!tmp_entry->is_sub_map)
+ vm_map_clip_start(src_map, tmp_entry, src_start);
+ continue; /* restart w/ new tmp_entry */
+ }
+ }
+
+ /*
+ * Verify that the region can be read.
+ */
+ if (((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE &&
+ !use_maxprot) ||
+ (src_entry->max_protection & VM_PROT_READ) == 0)
+ RETURN(KERN_PROTECTION_FAILURE);
+
+ /*
+ * Clip against the endpoints of the entire region.
+ */
+
+ vm_map_clip_end(src_map, src_entry, src_end);
+
+ src_size = src_entry->vme_end - src_start;
+ src_object = VME_OBJECT(src_entry);
+ src_offset = VME_OFFSET(src_entry);
+ was_wired = (src_entry->wired_count != 0);
+
+ vm_map_entry_copy(new_entry, src_entry);
+ if (new_entry->is_sub_map) {
+ /* clr address space specifics */
+ new_entry->use_pmap = FALSE;
+ }
+
+ /*
+ * Attempt non-blocking copy-on-write optimizations.
+ */
+
+ if (src_destroy &&
+ (src_object == VM_OBJECT_NULL ||
+ (src_object->internal && !src_object->true_share
+ && !map_share))) {
+ /*
+ * If we are destroying the source, and the object
+ * is internal, we can move the object reference
+ * from the source to the copy. The copy is
+ * copy-on-write only if the source is.
+ * We make another reference to the object, because
+ * destroying the source entry will deallocate it.
+ */
+ vm_object_reference(src_object);
+
+ /*
+ * Copy is always unwired. vm_map_copy_entry
+ * set its wired count to zero.
+ */
+
+ goto CopySuccessful;
+ }
+
+
+ RestartCopy:
+ XPR(XPR_VM_MAP, "vm_map_copyin_common src_obj 0x%x ent 0x%x obj 0x%x was_wired %d\n",
+ src_object, new_entry, VME_OBJECT(new_entry),
+ was_wired, 0);
+ if ((src_object == VM_OBJECT_NULL ||
+ (!was_wired && !map_share && !tmp_entry->is_shared)) &&
+ vm_object_copy_quickly(
+ &VME_OBJECT(new_entry),
+ src_offset,
+ src_size,
+ &src_needs_copy,
+ &new_entry_needs_copy)) {
+
+ new_entry->needs_copy = new_entry_needs_copy;
+
+ /*
+ * Handle copy-on-write obligations
+ */
+
+ if (src_needs_copy && !tmp_entry->needs_copy) {
+ vm_prot_t prot;
+
+ prot = src_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(src_map, VME_ALIAS(src_entry))
+ && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ vm_object_pmap_protect(
+ src_object,
+ src_offset,
+ src_size,
+ (src_entry->is_shared ?
+ PMAP_NULL
+ : src_map->pmap),
+ src_entry->vme_start,
+ prot);
+
+ assert(tmp_entry->wired_count == 0);
+ tmp_entry->needs_copy = TRUE;
+ }
+
+ /*
+ * The map has never been unlocked, so it's safe
+ * to move to the next entry rather than doing
+ * another lookup.
+ */
+
+ goto CopySuccessful;
+ }
+
+ /*
+ * Take an object reference, so that we may
+ * release the map lock(s).
+ */
+
+ assert(src_object != VM_OBJECT_NULL);
+ vm_object_reference(src_object);
+
+ /*
+ * Record the timestamp for later verification.
+ * Unlock the map.
+ */
+
+ version.main_timestamp = src_map->timestamp;
+ vm_map_unlock(src_map); /* Increments timestamp once! */
+
+ /*
+ * Perform the copy
+ */
+
+ if (was_wired) {
+ CopySlowly:
+ vm_object_lock(src_object);
+ result = vm_object_copy_slowly(
+ src_object,
+ src_offset,
+ src_size,
+ THREAD_UNINT,
+ &VME_OBJECT(new_entry));
+ VME_OFFSET_SET(new_entry, 0);
+ new_entry->needs_copy = FALSE;
+
+ }
+ else if (src_object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC &&
+ (tmp_entry->is_shared || map_share)) {
+ vm_object_t new_object;
+
+ vm_object_lock_shared(src_object);
+ new_object = vm_object_copy_delayed(
+ src_object,
+ src_offset,
+ src_size,
+ TRUE);
+ if (new_object == VM_OBJECT_NULL)
+ goto CopySlowly;
+
+ VME_OBJECT_SET(new_entry, new_object);
+ assert(new_entry->wired_count == 0);
+ new_entry->needs_copy = TRUE;
+ assert(!new_entry->iokit_acct);
+ assert(new_object->purgable == VM_PURGABLE_DENY);
+ new_entry->use_pmap = TRUE;
+ result = KERN_SUCCESS;
+
+ } else {
+ vm_object_offset_t new_offset;
+ new_offset = VME_OFFSET(new_entry);
+ result = vm_object_copy_strategically(src_object,
+ src_offset,
+ src_size,
+ &VME_OBJECT(new_entry),
+ &new_offset,
+ &new_entry_needs_copy);
+ if (new_offset != VME_OFFSET(new_entry)) {
+ VME_OFFSET_SET(new_entry, new_offset);
+ }
+
+ new_entry->needs_copy = new_entry_needs_copy;
+ }
+
+ if (result != KERN_SUCCESS &&
+ result != KERN_MEMORY_RESTART_COPY) {
+ vm_map_lock(src_map);
+ RETURN(result);
+ }
+
+ /*
+ * Throw away the extra reference
+ */
+
+ vm_object_deallocate(src_object);
+
+ /*
+ * Verify that the map has not substantially
+ * changed while the copy was being made.
+ */
+
+ vm_map_lock(src_map);
+
+ if ((version.main_timestamp + 1) == src_map->timestamp)
+ goto VerificationSuccessful;
+
+ /*
+ * Simple version comparison failed.
+ *
+ * Retry the lookup and verify that the
+ * same object/offset are still present.
+ *
+ * [Note: a memory manager that colludes with
+ * the calling task can detect that we have
+ * cheated. While the map was unlocked, the
+ * mapping could have been changed and restored.]
+ */
+
+ if (!vm_map_lookup_entry(src_map, src_start, &tmp_entry)) {
+ if (result != KERN_MEMORY_RESTART_COPY) {
+ vm_object_deallocate(VME_OBJECT(new_entry));
+ VME_OBJECT_SET(new_entry, VM_OBJECT_NULL);
+ assert(!new_entry->iokit_acct);
+ new_entry->use_pmap = TRUE;
+ }
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+
+ src_entry = tmp_entry;
+ vm_map_clip_start(src_map, src_entry, src_start);
+
+ if ((((src_entry->protection & VM_PROT_READ) == VM_PROT_NONE) &&
+ !use_maxprot) ||
+ ((src_entry->max_protection & VM_PROT_READ) == 0))
+ goto VerificationFailed;
+
+ if (src_entry->vme_end < new_entry->vme_end) {
+ assert(VM_MAP_PAGE_ALIGNED(src_entry->vme_end,
+ VM_MAP_COPY_PAGE_MASK(copy)));
+ new_entry->vme_end = src_entry->vme_end;
+ src_size = new_entry->vme_end - src_start;
+ }
+
+ if ((VME_OBJECT(src_entry) != src_object) ||
+ (VME_OFFSET(src_entry) != src_offset) ) {
+
+ /*
+ * Verification failed.
+ *
+ * Start over with this top-level entry.
+ */
+
+ VerificationFailed: ;
+
+ vm_object_deallocate(VME_OBJECT(new_entry));
+ tmp_entry = src_entry;
+ continue;
+ }
+
+ /*
+ * Verification succeeded.
+ */
+
+ VerificationSuccessful: ;
+
+ if (result == KERN_MEMORY_RESTART_COPY)
+ goto RestartCopy;
+
+ /*
+ * Copy succeeded.
+ */
+
+ CopySuccessful: ;
+
+ /*
+ * Link in the new copy entry.
+ */
+
+ vm_map_copy_entry_link(copy, vm_map_copy_last_entry(copy),
+ new_entry);
+
+ /*
+ * Determine whether the entire region
+ * has been copied.
+ */
+ src_base = src_start;
+ src_start = new_entry->vme_end;
+ new_entry = VM_MAP_ENTRY_NULL;
+ while ((src_start >= src_end) && (src_end != 0)) {
+ submap_map_t *ptr;
+
+ if (src_map == base_map) {
+ /* back to the top */
+ break;
+ }
+
+ ptr = parent_maps;
+ assert(ptr != NULL);
+ parent_maps = parent_maps->next;
+
+ /* fix up the damage we did in that submap */
+ vm_map_simplify_range(src_map,
+ src_base,
+ src_end);
+
+ vm_map_unlock(src_map);
+ vm_map_deallocate(src_map);
+ vm_map_lock(ptr->parent_map);
+ src_map = ptr->parent_map;
+ src_base = ptr->base_start;
+ src_start = ptr->base_start + ptr->base_len;
+ src_end = ptr->base_end;
+ if (!vm_map_lookup_entry(src_map,
+ src_start,
+ &tmp_entry) &&
+ (src_end > src_start)) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ kfree(ptr, sizeof(submap_map_t));
+ if (parent_maps == NULL)
+ map_share = FALSE;
+ src_entry = tmp_entry->vme_prev;
+ }
+
+ if ((VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT) &&
+ (src_start >= src_addr + len) &&
+ (src_addr + len != 0)) {
+ /*
+ * Stop copying now, even though we haven't reached
+ * "src_end". We'll adjust the end of the last copy
+ * entry at the end, if needed.
+ *
+ * If src_map's aligment is different from the
+ * system's page-alignment, there could be
+ * extra non-map-aligned map entries between
+ * the original (non-rounded) "src_addr + len"
+ * and the rounded "src_end".
+ * We do not want to copy those map entries since
+ * they're not part of the copied range.
+ */
+ break;
+ }
+
+ if ((src_start >= src_end) && (src_end != 0))
+ break;
+
+ /*
+ * Verify that there are no gaps in the region
+ */
+
+ tmp_entry = src_entry->vme_next;
+ if ((tmp_entry->vme_start != src_start) ||
+ (tmp_entry == vm_map_to_entry(src_map))) {
+ RETURN(KERN_INVALID_ADDRESS);
+ }
+ }
+
+ /*
+ * If the source should be destroyed, do it now, since the
+ * copy was successful.
+ */
+ if (src_destroy) {
+ (void) vm_map_delete(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ src_end,
+ ((src_map == kernel_map) ?
+ VM_MAP_REMOVE_KUNWIRE :
+ VM_MAP_NO_FLAGS),
+ VM_MAP_NULL);
+ } else {
+ /* fix up the damage we did in the base map */
+ vm_map_simplify_range(
+ src_map,
+ vm_map_trunc_page(src_addr,
+ VM_MAP_PAGE_MASK(src_map)),
+ vm_map_round_page(src_end,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+
+ vm_map_unlock(src_map);
+
+ if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT) {
+ vm_map_offset_t original_start, original_offset, original_end;
+
+ assert(VM_MAP_COPY_PAGE_MASK(copy) == PAGE_MASK);
+
+ /* adjust alignment of first copy_entry's "vme_start" */
+ tmp_entry = vm_map_copy_first_entry(copy);
+ if (tmp_entry != vm_map_copy_to_entry(copy)) {
+ vm_map_offset_t adjustment;
+
+ original_start = tmp_entry->vme_start;
+ original_offset = VME_OFFSET(tmp_entry);
+
+ /* map-align the start of the first copy entry... */
+ adjustment = (tmp_entry->vme_start -
+ vm_map_trunc_page(
+ tmp_entry->vme_start,
+ VM_MAP_PAGE_MASK(src_map)));
+ tmp_entry->vme_start -= adjustment;
+ VME_OFFSET_SET(tmp_entry,
+ VME_OFFSET(tmp_entry) - adjustment);
+ copy_addr -= adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ /* ... adjust for mis-aligned start of copy range */
+ adjustment =
+ (vm_map_trunc_page(copy->offset,
+ PAGE_MASK) -
+ vm_map_trunc_page(copy->offset,
+ VM_MAP_PAGE_MASK(src_map)));
+ if (adjustment) {
+ assert(page_aligned(adjustment));
+ assert(adjustment < VM_MAP_PAGE_SIZE(src_map));
+ tmp_entry->vme_start += adjustment;
+ VME_OFFSET_SET(tmp_entry,
+ (VME_OFFSET(tmp_entry) +
+ adjustment));
+ copy_addr += adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ }
+
+ /*
+ * Assert that the adjustments haven't exposed
+ * more than was originally copied...
+ */
+ assert(tmp_entry->vme_start >= original_start);
+ assert(VME_OFFSET(tmp_entry) >= original_offset);
+ /*
+ * ... and that it did not adjust outside of a
+ * a single 16K page.
+ */
+ assert(vm_map_trunc_page(tmp_entry->vme_start,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_trunc_page(original_start,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+
+ /* adjust alignment of last copy_entry's "vme_end" */
+ tmp_entry = vm_map_copy_last_entry(copy);
+ if (tmp_entry != vm_map_copy_to_entry(copy)) {
+ vm_map_offset_t adjustment;
+
+ original_end = tmp_entry->vme_end;
+
+ /* map-align the end of the last copy entry... */
+ tmp_entry->vme_end =
+ vm_map_round_page(tmp_entry->vme_end,
+ VM_MAP_PAGE_MASK(src_map));
+ /* ... adjust for mis-aligned end of copy range */
+ adjustment =
+ (vm_map_round_page((copy->offset +
+ copy->size),
+ VM_MAP_PAGE_MASK(src_map)) -
+ vm_map_round_page((copy->offset +
+ copy->size),
+ PAGE_MASK));
+ if (adjustment) {
+ assert(page_aligned(adjustment));
+ assert(adjustment < VM_MAP_PAGE_SIZE(src_map));
+ tmp_entry->vme_end -= adjustment;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ }
+
+ /*
+ * Assert that the adjustments haven't exposed
+ * more than was originally copied...
+ */
+ assert(tmp_entry->vme_end <= original_end);
+ /*
+ * ... and that it did not adjust outside of a
+ * a single 16K page.
+ */
+ assert(vm_map_round_page(tmp_entry->vme_end,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_round_page(original_end,
+ VM_MAP_PAGE_MASK(src_map)));
+ }
+ }
+
+ /* Fix-up start and end points in copy. This is necessary */
+ /* when the various entries in the copy object were picked */
+ /* up from different sub-maps */
+
+ tmp_entry = vm_map_copy_first_entry(copy);
+ copy_size = 0; /* compute actual size */
+ while (tmp_entry != vm_map_copy_to_entry(copy)) {
+ assert(VM_MAP_PAGE_ALIGNED(
+ copy_addr + (tmp_entry->vme_end -
+ tmp_entry->vme_start),
+ VM_MAP_COPY_PAGE_MASK(copy)));
+ assert(VM_MAP_PAGE_ALIGNED(
+ copy_addr,
+ VM_MAP_COPY_PAGE_MASK(copy)));
+
+ /*
+ * The copy_entries will be injected directly into the
+ * destination map and might not be "map aligned" there...
+ */
+ tmp_entry->map_aligned = FALSE;
+
+ tmp_entry->vme_end = copy_addr +
+ (tmp_entry->vme_end - tmp_entry->vme_start);
+ tmp_entry->vme_start = copy_addr;
+ assert(tmp_entry->vme_start < tmp_entry->vme_end);
+ copy_addr += tmp_entry->vme_end - tmp_entry->vme_start;
+ copy_size += tmp_entry->vme_end - tmp_entry->vme_start;
+ tmp_entry = (struct vm_map_entry *)tmp_entry->vme_next;
+ }
+
+ if (VM_MAP_PAGE_SHIFT(src_map) != PAGE_SHIFT &&
+ copy_size < copy->size) {
+ /*
+ * The actual size of the VM map copy is smaller than what
+ * was requested by the caller. This must be because some
+ * PAGE_SIZE-sized pages are missing at the end of the last
+ * VM_MAP_PAGE_SIZE(src_map)-sized chunk of the range.
+ * The caller might not have been aware of those missing
+ * pages and might not want to be aware of it, which is
+ * fine as long as they don't try to access (and crash on)
+ * those missing pages.
+ * Let's adjust the size of the "copy", to avoid failing
+ * in vm_map_copyout() or vm_map_copy_overwrite().
+ */
+ assert(vm_map_round_page(copy_size,
+ VM_MAP_PAGE_MASK(src_map)) ==
+ vm_map_round_page(copy->size,
+ VM_MAP_PAGE_MASK(src_map)));
+ copy->size = copy_size;
+ }
+
+ *copy_result = copy;
+ return(KERN_SUCCESS);
+
+#undef RETURN
+}
+
+kern_return_t
+vm_map_copy_extract(
+ vm_map_t src_map,
+ vm_map_address_t src_addr,
+ vm_map_size_t len,
+ vm_map_copy_t *copy_result, /* OUT */
+ vm_prot_t *cur_prot, /* OUT */
+ vm_prot_t *max_prot)
+{
+ vm_map_offset_t src_start, src_end;
+ vm_map_copy_t copy;
+ kern_return_t kr;
+
+ /*
+ * Check for copies of zero bytes.
+ */
+
+ if (len == 0) {
+ *copy_result = VM_MAP_COPY_NULL;
+ return(KERN_SUCCESS);
+ }
+
+ /*
+ * Check that the end address doesn't overflow
+ */
+ src_end = src_addr + len;
+ if (src_end < src_addr)
+ return KERN_INVALID_ADDRESS;
+
+ /*
+ * Compute (page aligned) start and end of region
+ */
+ src_start = vm_map_trunc_page(src_addr, PAGE_MASK);
+ src_end = vm_map_round_page(src_end, PAGE_MASK);
+
+ /*
+ * Allocate a header element for the list.
+ *
+ * Use the start and end in the header to
+ * remember the endpoints prior to rounding.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ vm_map_copy_first_entry(copy) =
+ vm_map_copy_last_entry(copy) = vm_map_copy_to_entry(copy);
+ copy->type = VM_MAP_COPY_ENTRY_LIST;
+ copy->cpy_hdr.nentries = 0;
+ copy->cpy_hdr.entries_pageable = TRUE;
+
+ vm_map_store_init(©->cpy_hdr);
+
+ copy->offset = 0;
+ copy->size = len;
+
+ kr = vm_map_remap_extract(src_map,
+ src_addr,
+ len,
+ FALSE, /* copy */
+ ©->cpy_hdr,
+ cur_prot,
+ max_prot,
+ VM_INHERIT_SHARE,
+ TRUE); /* pageable */
+ if (kr != KERN_SUCCESS) {
+ vm_map_copy_discard(copy);
+ return kr;
+ }
+
+ *copy_result = copy;
+ return KERN_SUCCESS;
+}
+
+/*
+ * vm_map_copyin_object:
+ *
+ * Create a copy object from an object.
+ * Our caller donates an object reference.
+ */
+
+kern_return_t
+vm_map_copyin_object(
+ vm_object_t object,
+ vm_object_offset_t offset, /* offset of region in object */
+ vm_object_size_t size, /* size of region in object */
+ vm_map_copy_t *copy_result) /* OUT */
+{
+ vm_map_copy_t copy; /* Resulting copy */
+
+ /*
+ * We drop the object into a special copy object
+ * that contains the object directly.
+ */
+
+ copy = (vm_map_copy_t) zalloc(vm_map_copy_zone);
+ copy->c_u.hdr.rb_head_store.rbh_root = (void*)(int)SKIP_RB_TREE;
+ copy->type = VM_MAP_COPY_OBJECT;
+ copy->cpy_object = object;
+ copy->offset = offset;
+ copy->size = size;
+
+ *copy_result = copy;
+ return(KERN_SUCCESS);
+}
+
+static void
+vm_map_fork_share(
+ vm_map_t old_map,
+ vm_map_entry_t old_entry,
+ vm_map_t new_map)
+{
+ vm_object_t object;
+ vm_map_entry_t new_entry;
+
+ /*
+ * New sharing code. New map entry
+ * references original object. Internal
+ * objects use asynchronous copy algorithm for
+ * future copies. First make sure we have
+ * the right object. If we need a shadow,
+ * or someone else already has one, then
+ * make a new shadow and share it.
+ */
+
+ object = VME_OBJECT(old_entry);
+ if (old_entry->is_sub_map) {
+ assert(old_entry->wired_count == 0);
+#ifndef NO_NESTED_PMAP
+ if(old_entry->use_pmap) {
+ kern_return_t result;
+
+ result = pmap_nest(new_map->pmap,
+ (VME_SUBMAP(old_entry))->pmap,
+ (addr64_t)old_entry->vme_start,
+ (addr64_t)old_entry->vme_start,
+ (uint64_t)(old_entry->vme_end - old_entry->vme_start));
+ if(result)
+ panic("vm_map_fork_share: pmap_nest failed!");
+ }
+#endif /* NO_NESTED_PMAP */
+ } else if (object == VM_OBJECT_NULL) {
+ object = vm_object_allocate((vm_map_size_t)(old_entry->vme_end -
+ old_entry->vme_start));
+ VME_OFFSET_SET(old_entry, 0);
+ VME_OBJECT_SET(old_entry, object);
+ old_entry->use_pmap = TRUE;
+ assert(!old_entry->needs_copy);
+ } else if (object->copy_strategy !=
+ MEMORY_OBJECT_COPY_SYMMETRIC) {
+
+ /*
+ * We are already using an asymmetric
+ * copy, and therefore we already have
+ * the right object.
+ */
+
+ assert(! old_entry->needs_copy);
+ }
+ else if (old_entry->needs_copy || /* case 1 */
+ object->shadowed || /* case 2 */
+ (!object->true_share && /* case 3 */
+ !old_entry->is_shared &&
+ (object->vo_size >
+ (vm_map_size_t)(old_entry->vme_end -
+ old_entry->vme_start)))) {
+
+ /*
+ * We need to create a shadow.
+ * There are three cases here.
+ * In the first case, we need to
+ * complete a deferred symmetrical
+ * copy that we participated in.
+ * In the second and third cases,
+ * we need to create the shadow so
+ * that changes that we make to the
+ * object do not interfere with
+ * any symmetrical copies which
+ * have occured (case 2) or which
+ * might occur (case 3).
+ *
+ * The first case is when we had
+ * deferred shadow object creation
+ * via the entry->needs_copy mechanism.
+ * This mechanism only works when
+ * only one entry points to the source
+ * object, and we are about to create
+ * a second entry pointing to the
+ * same object. The problem is that
+ * there is no way of mapping from
+ * an object to the entries pointing
+ * to it. (Deferred shadow creation
+ * works with one entry because occurs
+ * at fault time, and we walk from the
+ * entry to the object when handling
+ * the fault.)
+ *
+ * The second case is when the object
+ * to be shared has already been copied
+ * with a symmetric copy, but we point
+ * directly to the object without
+ * needs_copy set in our entry. (This
+ * can happen because different ranges
+ * of an object can be pointed to by
+ * different entries. In particular,
+ * a single entry pointing to an object
+ * can be split by a call to vm_inherit,
+ * which, combined with task_create, can
+ * result in the different entries
+ * having different needs_copy values.)
+ * The shadowed flag in the object allows
+ * us to detect this case. The problem
+ * with this case is that if this object
+ * has or will have shadows, then we
+ * must not perform an asymmetric copy
+ * of this object, since such a copy
+ * allows the object to be changed, which
+ * will break the previous symmetrical
+ * copies (which rely upon the object
+ * not changing). In a sense, the shadowed
+ * flag says "don't change this object".
+ * We fix this by creating a shadow
+ * object for this object, and sharing
+ * that. This works because we are free
+ * to change the shadow object (and thus
+ * to use an asymmetric copy strategy);
+ * this is also semantically correct,
+ * since this object is temporary, and
+ * therefore a copy of the object is
+ * as good as the object itself. (This
+ * is not true for permanent objects,
+ * since the pager needs to see changes,
+ * which won't happen if the changes
+ * are made to a copy.)
+ *
+ * The third case is when the object
+ * to be shared has parts sticking
+ * outside of the entry we're working
+ * with, and thus may in the future
+ * be subject to a symmetrical copy.
+ * (This is a preemptive version of
+ * case 2.)
+ */
+ VME_OBJECT_SHADOW(old_entry,
+ (vm_map_size_t) (old_entry->vme_end -
+ old_entry->vme_start));
+
+ /*
+ * If we're making a shadow for other than
+ * copy on write reasons, then we have
+ * to remove write permission.
+ */
+
+ if (!old_entry->needs_copy &&
+ (old_entry->protection & VM_PROT_WRITE)) {
+ vm_prot_t prot;
+
+ prot = old_entry->protection & ~VM_PROT_WRITE;
+
+ if (override_nx(old_map, VME_ALIAS(old_entry)) && prot)
+ prot |= VM_PROT_EXECUTE;
+
+ if (old_map->mapped_in_other_pmaps) {
+ vm_object_pmap_protect(
+ VME_OBJECT(old_entry),
+ VME_OFFSET(old_entry),
+ (old_entry->vme_end -
+ old_entry->vme_start),
+ PMAP_NULL,
+ old_entry->vme_start,
+ prot);
+ } else {
+ pmap_protect(old_map->pmap,
+ old_entry->vme_start,
+ old_entry->vme_end,
+ prot);
+ }
+ }
+
+ old_entry->needs_copy = FALSE;
+ object = VME_OBJECT(old_entry);
+ }
+
+
+ /*
+ * If object was using a symmetric copy strategy,
+ * change its copy strategy to the default
+ * asymmetric copy strategy, which is copy_delay
+ * in the non-norma case and copy_call in the
+ * norma case. Bump the reference count for the
+ * new entry.
+ */
+
+ if(old_entry->is_sub_map) {
+ vm_map_lock(VME_SUBMAP(old_entry));
+ vm_map_reference(VME_SUBMAP(old_entry));
+ vm_map_unlock(VME_SUBMAP(old_entry));
+ } else {
+ vm_object_lock(object);
+ vm_object_reference_locked(object);
+ if (object->copy_strategy == MEMORY_OBJECT_COPY_SYMMETRIC) {
+ object->copy_strategy = MEMORY_OBJECT_COPY_DELAY;
+ }
+ vm_object_unlock(object);
+ }
+
+ /*
+ * Clone the entry, using object ref from above.
+ * Mark both entries as shared.
+ */
+
+ new_entry = vm_map_entry_create(new_map, FALSE); /* Never the kernel
+ * map or descendants */
+ vm_map_entry_copy(new_entry, old_entry);
+ old_entry->is_shared = TRUE;
+ new_entry->is_shared = TRUE;
+
+ /*
+ * Insert the entry into the new map -- we
+ * know we're inserting at the end of the new
+ * map.
+ */
+
+ vm_map_store_entry_link(new_map, vm_map_last_entry(new_map), new_entry);
+
+ /*
+ * Update the physical map
+ */
+
+ if (old_entry->is_sub_map) {
+ /* Bill Angell pmap support goes here */
+ } else {
+ pmap_copy(new_map->pmap, old_map->pmap, new_entry->vme_start,
+ old_entry->vme_end - old_entry->vme_start,
+ old_entry->vme_start);
+ }
+}
+
+static boolean_t
+vm_map_fork_copy(
+ vm_map_t old_map,
+ vm_map_entry_t *old_entry_p,
+ vm_map_t new_map)
+{
+ vm_map_entry_t old_entry = *old_entry_p;
+ vm_map_size_t entry_size = old_entry->vme_end - old_entry->vme_start;
+ vm_map_offset_t start = old_entry->vme_start;
+ vm_map_copy_t copy;
+ vm_map_entry_t last = vm_map_last_entry(new_map);
+
+ vm_map_unlock(old_map);
+ /*
+ * Use maxprot version of copyin because we
+ * care about whether this memory can ever
+ * be accessed, not just whether it's accessible
+ * right now.
+ */
+ if (vm_map_copyin_maxprot(old_map, start, entry_size, FALSE, ©)
+ != KERN_SUCCESS) {
+ /*
+ * The map might have changed while it
+ * was unlocked, check it again. Skip
+ * any blank space or permanently
+ * unreadable region.
+ */
+ vm_map_lock(old_map);
+ if (!vm_map_lookup_entry(old_map, start, &last) ||
+ (last->max_protection & VM_PROT_READ) == VM_PROT_NONE) {
+ last = last->vme_next;
+ }
+ *old_entry_p = last;
+
+ /*
+ * XXX For some error returns, want to
+ * XXX skip to the next element. Note
+ * that INVALID_ADDRESS and
+ * PROTECTION_FAILURE are handled above.
+ */
+
+ return FALSE;
+ }
+
+ /*
+ * Insert the copy into the new map
+ */
+
+ vm_map_copy_insert(new_map, last, copy);
+
+ /*
+ * Pick up the traversal at the end of
+ * the copied region.
+ */
+
+ vm_map_lock(old_map);
+ start += entry_size;
+ if (! vm_map_lookup_entry(old_map, start, &last)) {
+ last = last->vme_next;
+ } else {
+ if (last->vme_start == start) {
+ /*
+ * No need to clip here and we don't
+ * want to cause any unnecessary
+ * unnesting...
+ */
+ } else {
+ vm_map_clip_start(old_map, last, start);
+ }
+ }
+ *old_entry_p = last;
+
+ return TRUE;
+}
+
+/*
+ * vm_map_fork:
+ *
+ * Create and return a new map based on the old
+ * map, according to the inheritance values on the
+ * regions in that map.
projects / apple / xnu.git / blobdiff