[apple/hfs.git] / core / hfs_cprotect.c

/*
 * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */
#if CONFIG_PROTECT

#include <sys/mount.h>
#include <sys/random.h>
#include <sys/xattr.h>
#include <sys/vnode_if.h>
#include <sys/fcntl.h>
#include <libkern/OSByteOrder.h>
#include <libkern/crypto/sha1.h>
#include <sys/proc.h>
#include <sys/kauth.h>
#include <sys/sysctl.h>
#include <sys/ubc.h>

#include "hfs.h"
#include "hfs_cnode.h"
#include "hfs_fsctl.h"
#include "hfs_cprotect.h"
#include "hfs_iokit.h"

#if HFS_CONFIG_KEY_ROLL
#include "hfs_key_roll.h"
#endif

#define PTR_ADD(type, base, offset)             (type)((uintptr_t)(base) + (offset))

extern int (**hfs_vnodeop_p) (void *);

/*
 * CP private functions
 */
static int cp_root_major_vers(mount_t mp);
static int cp_getxattr(cnode_t *, struct hfsmount *hfsmp, struct cprotect **);
static void cp_entry_dealloc(hfsmount_t *hfsmp, struct cprotect *entry);
static int cp_restore_keys(struct cprotect *, struct hfsmount *hfsmp, struct cnode *);
static int cp_lock_vnode_callback(vnode_t, void *);
static int cp_vnode_is_eligible (vnode_t);
static int cp_check_access (cnode_t *cp, struct hfsmount *hfsmp, int vnop);
static int cp_unwrap(struct hfsmount *, struct cprotect *, struct cnode *);
static void cp_init_access(aks_cred_t access, struct cnode *cp);

// -- cp_key_pair accessors --

void cpkp_init(cp_key_pair_t *cpkp, uint16_t max_pers_key_len,
                           uint16_t max_cached_key_len)
{
        cpkp->cpkp_max_pers_key_len = max_pers_key_len;
        cpkp->cpkp_pers_key_len = 0;
        cpx_init(cpkp_cpx(cpkp), max_cached_key_len);

        // Default to using offsets
        cpx_set_use_offset_for_iv(cpkp_cpx(cpkp), true);
}

uint16_t cpkp_max_pers_key_len(const cp_key_pair_t *cpkp)
{
        return cpkp->cpkp_max_pers_key_len;
}

uint16_t cpkp_pers_key_len(const cp_key_pair_t *cpkp)
{
        return cpkp->cpkp_pers_key_len;
}

static bool cpkp_has_pers_key(const cp_key_pair_t *cpkp)
{
        return cpkp->cpkp_pers_key_len > 0;
}

static void *cpkp_pers_key(const cp_key_pair_t *cpkp)
{
        return PTR_ADD(void *, &cpkp->cpkp_cpx, cpx_sizex(cpkp_cpx(cpkp)));
}

static void cpkp_set_pers_key_len(cp_key_pair_t *cpkp, uint16_t key_len)
{
        if (key_len > cpkp->cpkp_max_pers_key_len)
                panic("hfs_cprotect: key too big!");
        cpkp->cpkp_pers_key_len = key_len;
}

#pragma clang diagnostic push
#pragma clang diagnostic ignored "-Wcast-qual"
cpx_t cpkp_cpx(const cp_key_pair_t *cpkp)
{
        // Cast to remove const qualifier
        return (cpx_t)&cpkp->cpkp_cpx;
}
#pragma clang diagnostic pop

size_t cpkp_size(uint16_t pers_key_len, uint16_t cached_key_len)
{
        return sizeof(cp_key_pair_t) + pers_key_len + cpx_size(cached_key_len);
}

size_t cpkp_sizex(const cp_key_pair_t *cpkp)
{
        return cpkp_size(cpkp->cpkp_max_pers_key_len, cpx_max_key_len(cpkp_cpx(cpkp)));
}

void cpkp_flush(cp_key_pair_t *cpkp)
{
        cpx_flush(cpkp_cpx(cpkp));
        cpkp->cpkp_pers_key_len = 0;
        bzero(cpkp_pers_key(cpkp), cpkp->cpkp_max_pers_key_len);
}

bool cpkp_can_copy(const cp_key_pair_t *src, const cp_key_pair_t *dst)
{
        return (cpkp_pers_key_len(src) <= dst->cpkp_max_pers_key_len
                        && cpx_can_copy(cpkp_cpx(src), cpkp_cpx(dst)));
}

void cpkp_copy(const cp_key_pair_t *src, cp_key_pair_t *dst)
{
        const uint16_t key_len = cpkp_pers_key_len(src);
        cpkp_set_pers_key_len(dst, key_len);
        memcpy(cpkp_pers_key(dst), cpkp_pers_key(src), key_len);
        cpx_copy(cpkp_cpx(src), cpkp_cpx(dst));
}

// --

bool cp_is_supported_version(uint16_t vers)
{
        return vers == CP_VERS_4 || vers == CP_VERS_5;
}

/*
 * Return the appropriate key and, if requested, the physical offset and
 * maximum length for a particular I/O operation.
 */
void cp_io_params(__unused hfsmount_t *hfsmp, cprotect_t cpr,
                                  __unused off_rsrc_t off_rsrc,
                                  __unused int direction, cp_io_params_t *io_params)
{
#if HFS_CONFIG_KEY_ROLL
        hfs_cp_key_roll_ctx_t *ckr = cpr->cp_key_roll_ctx;

        if (ckr && off_rsrc < ckr->ckr_off_rsrc) {
                /*
                 * When we're in the process of rolling an extent, ckr_off_rsrc will
                 * indicate the end of the extent.
                 */
                const off_rsrc_t roll_loc = ckr->ckr_off_rsrc
                        - hfs_blk_to_bytes(ckr->ckr_roll_extent.blockCount,
                                                           hfsmp->blockSize);

                if (off_rsrc < roll_loc) {
                        io_params->max_len              = roll_loc - off_rsrc;
                        io_params->phys_offset  = -1;
                } else {
                        /*
                         * We should never get reads to the extent we're rolling
                         * because the pages should be locked in the UBC.  If we
                         * did get reads it's not obvious what the right thing to
                         * do is either: we could read from the old location, but
                         * we might have written later data to the new location,
                         * or we could read from the new location, but data might
                         * not have been written there yet.
                         *
                         * Note that whilst raw encrypted reads don't lock any
                         * pages, or take a cluster_read_direct lock, the call to
                         * hfs_key_roll_up_to in hfs_vnop_read will have ensured
                         * that the file has been rolled beyond the offset being
                         * read so this path should never be taken in that case.
                         */
                        hfs_assert(direction == VNODE_WRITE);

                        // For release builds, just in case...
                        if (direction == VNODE_READ) {
                                // Use the old key and offset
                                goto old_key;
                        }

                        io_params->max_len = ckr->ckr_off_rsrc - off_rsrc;
                        io_params->phys_offset = hfs_blk_to_bytes(ckr->ckr_roll_extent.startBlock,
                                                                                                          hfsmp->blockSize) + off_rsrc - roll_loc;
                }

                // Use new key
                io_params->cpx = cpkp_cpx(&ckr->ckr_keys);
                return;
        }
old_key:
        // Use old key...
#endif

        io_params->max_len = INT64_MAX;
        io_params->phys_offset = -1;
        io_params->cpx = cpkp_cpx(&cpr->cp_keys);
}

static void cp_flush_cached_keys(cprotect_t cpr)
{
        cpx_flush(cpkp_cpx(&cpr->cp_keys));
#if HFS_CONFIG_KEY_ROLL
        if (cpr->cp_key_roll_ctx)
                cpx_flush(cpkp_cpx(&cpr->cp_key_roll_ctx->ckr_keys));
#endif
}

static bool cp_needs_pers_key(cprotect_t cpr)
{
        if (CP_CLASS(cpr->cp_pclass) == PROTECTION_CLASS_F)
                return !cpx_has_key(cpkp_cpx(&cpr->cp_keys));
        else
                return !cpkp_has_pers_key(&cpr->cp_keys);
}

static cp_key_revision_t cp_initial_key_revision(__unused hfsmount_t *hfsmp)
{
        return 1;
}

cp_key_revision_t cp_next_key_revision(cp_key_revision_t rev)
{
        rev = (rev + 0x0100) ^ (mach_absolute_time() & 0xff);
        if (!rev)
                rev = 1;
        return rev;
}

/*
 * Allocate and initialize a cprotect blob for a new cnode.
 * Called from hfs_getnewvnode: cnode is locked exclusive.
 * 
 * Read xattr data off the cnode. Then, if conditions permit,
 * unwrap the file key and cache it in the cprotect blob.
 */
int
cp_entry_init(struct cnode *cp, struct mount *mp)
{
        struct cprotect *entry = NULL;
        int error = 0;
        struct hfsmount *hfsmp = VFSTOHFS(mp);

        /*
         * The cnode should be locked at this point, regardless of whether or not
         * we are creating a new item in the namespace or vending a vnode on behalf
         * of lookup.  The only time we tell getnewvnode to skip the lock is when 
         * constructing a resource fork vnode. But a resource fork vnode must come
         * after the regular data fork cnode has already been constructed.
         */
        if (!cp_fs_protected (mp)) {
                cp->c_cpentry = NULL;
                return 0;
        }

        if (!S_ISREG(cp->c_mode) && !S_ISDIR(cp->c_mode)) {
                cp->c_cpentry = NULL;
                return 0;
        }

        if (hfsmp->hfs_running_cp_major_vers == 0) {
                panic ("hfs cp: no running mount point version! ");             
        }

        hfs_assert(cp->c_cpentry == NULL);

        error = cp_getxattr(cp, hfsmp, &entry);
        if (error == ENOATTR) {
                /*
                 * Normally, we should always have a CP EA for a file or directory that
                 * we are initializing here. However, there are some extenuating circumstances,
                 * such as the root directory immediately following a newfs_hfs.
                 *
                 * As a result, we leave code here to deal with an ENOATTR which will always
                 * default to a 'D/NONE' key, though we don't expect to use it much.
                 */
                cp_key_class_t target_class = PROTECTION_CLASS_D;

                if (S_ISDIR(cp->c_mode)) {
                        target_class = PROTECTION_CLASS_DIR_NONE;
                }

                cp_key_revision_t key_revision = cp_initial_key_revision(hfsmp);

                /* allow keybag to override our class preferences */
                error = cp_new (&target_class, hfsmp, cp, cp->c_mode, CP_KEYWRAP_DIFFCLASS,
                                                key_revision, (cp_new_alloc_fn)cp_entry_alloc, (void **)&entry);
                if (error == 0) {
                        entry->cp_pclass = target_class;
                        entry->cp_key_os_version = cp_os_version();
                        entry->cp_key_revision = key_revision;
                        error = cp_setxattr (cp, entry, hfsmp, cp->c_fileid, XATTR_CREATE);
                }
        }

        /* 
         * Bail out if:
         * a) error was not ENOATTR (we got something bad from the getxattr call)
         * b) we encountered an error setting the xattr above.
         * c) we failed to generate a new cprotect data structure.
         */
        if (error) {
                goto out;
        }       

        cp->c_cpentry = entry;

out:
        if (error == 0) {
                entry->cp_backing_cnode = cp;
        }
        else {
                if (entry) {
                        cp_entry_destroy(hfsmp, entry);
                }
                cp->c_cpentry = NULL;
        }

        return error;
}

/*
 * cp_setup_newentry
 * 
 * Generate a keyless cprotect structure for use with the new AppleKeyStore kext.
 * Since the kext is now responsible for vending us both wrapped/unwrapped keys
 * we need to create a keyless xattr upon file / directory creation. When we have the inode value
 * and the file/directory is established, then we can ask it to generate keys.  Note that
 * this introduces a potential race;  If the device is locked and the wrapping
 * keys are purged between the time we call this function and the time we ask it to generate
 * keys for us, we could have to fail the open(2) call and back out the entry.
 */

int cp_setup_newentry (struct hfsmount *hfsmp, struct cnode *dcp,
                                           cp_key_class_t suppliedclass, mode_t cmode,
                                           struct cprotect **tmpentry)
{
        int isdir = 0;
        struct cprotect *entry = NULL;
        uint32_t target_class = hfsmp->default_cp_class;
        suppliedclass = CP_CLASS(suppliedclass);

        if (hfsmp->hfs_running_cp_major_vers == 0) {
                panic ("CP: major vers not set in mount!");
        }
        
        if (S_ISDIR (cmode))  {
                isdir = 1;
        }

        /* Decide the target class.  Input argument takes priority. */
        if (cp_is_valid_class (isdir, suppliedclass)) {
                /* caller supplies -1 if it was not specified so we will default to the mount point value */
                target_class = suppliedclass;
                /*
                 * One exception, F is never valid for a directory
                 * because its children may inherit and userland will be
                 * unable to read/write to the files.
                 */
                if (isdir) {
                        if (target_class == PROTECTION_CLASS_F) {
                                *tmpentry = NULL;
                                return EINVAL;
                        }
                }
        }
        else {
                /* 
                 * If no valid class was supplied, behave differently depending on whether or not
                 * the item being created is a file or directory.
                 * 
                 * for FILE:
                 *              If parent directory has a non-zero class, use that.
                 *              If parent directory has a zero class (not set), then attempt to
                 *              apply the mount point default.
                 * 
                 * for DIRECTORY:
                 *              Directories always inherit from the parent; if the parent
                 *              has a NONE class set, then we can continue to use that.
                 */
                if ((dcp) && (dcp->c_cpentry)) {
                        uint32_t parentclass = CP_CLASS(dcp->c_cpentry->cp_pclass);
                        /* If the parent class is not valid, default to the mount point value */
                        if (cp_is_valid_class(1, parentclass)) {
                                if (isdir) {
                                        target_class = parentclass;     
                                }
                                else if (parentclass != PROTECTION_CLASS_DIR_NONE) {
                                        /* files can inherit so long as it's not NONE */
                                        target_class = parentclass;
                                }
                        }
                        /* Otherwise, we already defaulted to the mount point's default */
                }
        }

        /* Generate the cprotect to vend out */
        entry = cp_entry_alloc(NULL, 0, 0, NULL);
        if (entry == NULL) {
                *tmpentry = NULL;
                return ENOMEM;
        }       

        /* 
         * We don't have keys yet, so fill in what we can.  At this point
         * this blob has no keys and it has no backing xattr.  We just know the
         * target class.
         */
        entry->cp_flags = CP_NO_XATTR;
        /* Note this is only the effective class */
        entry->cp_pclass = target_class;
        *tmpentry = entry;

        return 0;
}

/*
 * Set up an initial key/class pair for a disassociated cprotect entry.
 * This function is used to generate transient keys that will never be
 * written to disk.  We use class F for this since it provides the exact
 * semantics that are needed here.  Because we never attach this blob to
 * a cnode directly, we take a pointer to the cprotect struct.
 *
 * This function is primarily used in the HFS FS truncation codepath
 * where we may rely on AES symmetry to relocate encrypted data from
 * one spot in the disk to another.
 */
int cpx_gentempkeys(cpx_t *pcpx, __unused struct hfsmount *hfsmp)
{
        cpx_t cpx = cpx_alloc(CP_MAX_KEYSIZE);

        cpx_set_key_len(cpx, CP_MAX_KEYSIZE);
        read_random(cpx_key(cpx), CP_MAX_KEYSIZE);
        cpx_set_use_offset_for_iv(cpx, true);

        *pcpx = cpx;

        return 0;
}

/*
 * Tear down and clear a cprotect blob for a closing file.
 * Called at hfs_reclaim_cnode: cnode is locked exclusive.
 */
void
cp_entry_destroy(hfsmount_t *hfsmp, struct cprotect *entry_ptr)
{
        if (entry_ptr == NULL) {
                /* nothing to clean up */
                return;
        }
        cp_entry_dealloc(hfsmp, entry_ptr);
}


int
cp_fs_protected (mount_t mnt) 
{
        return (vfs_flags(mnt) & MNT_CPROTECT);
}


/*
 * Return a pointer to underlying cnode if there is one for this vnode.
 * Done without taking cnode lock, inspecting only vnode state.
 */
struct cnode *
cp_get_protected_cnode(struct vnode *vp)
{
        if (!cp_vnode_is_eligible(vp)) {
                return NULL;
        }

        if (!cp_fs_protected(VTOVFS(vp))) {
                /* mount point doesn't support it */
                return NULL;
        }

        return vnode_fsnode(vp);
}


/*
 * Sets *class to persistent class associated with vnode,
 * or returns error.
 */
int
cp_vnode_getclass(struct vnode *vp, cp_key_class_t *class)
{
        struct cprotect *entry;
        int error = 0;
        struct cnode *cp;
        int took_truncate_lock = 0;
        struct hfsmount *hfsmp = NULL;

        /* Is this an interesting vp? */
        if (!cp_vnode_is_eligible (vp)) {
                return EBADF;
        }

        /* Is the mount point formatted for content protection? */
        if (!cp_fs_protected(VTOVFS(vp))) {
                return ENOTSUP;
        }

        cp = VTOC(vp);
        hfsmp = VTOHFS(vp);

        /*
         * Take the truncate lock up-front in shared mode because we may need
         * to manipulate the CP blob. Pend lock events until we're done here.
         */
        hfs_lock_truncate (cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT);
        took_truncate_lock = 1;

        /*
         * We take only the shared cnode lock up-front.  If it turns out that
         * we need to manipulate the CP blob to write a key out, drop the
         * shared cnode lock and acquire an exclusive lock.
         */
        error = hfs_lock(cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT);
        if (error) {
                hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
                return error;
        }

        /* pull the class from the live entry */
        entry = cp->c_cpentry;

        if (entry == NULL) {
                panic("Content Protection: uninitialized cnode %p", cp);
        }
        
        /* Note that we may not have keys yet, but we know the target class. */

        if (error == 0) {
                *class = CP_CLASS(entry->cp_pclass);
        }

        if (took_truncate_lock) {
                hfs_unlock_truncate(cp, HFS_LOCK_DEFAULT);
        }

        hfs_unlock(cp);
        return error;
}

/*
 * Sets persistent class for this file or directory.
 * If vnode cannot be protected (system file, non-regular file, non-hfs), EBADF.
 * If the new class can't be accessed now, EPERM.
 * Otherwise, record class and re-wrap key if the mount point is content-protected.
 */
int
cp_vnode_setclass(struct vnode *vp, cp_key_class_t newclass)
{
        struct cnode *cp;
        struct cprotect *entry = 0;
        int error = 0;
        int took_truncate_lock = 0;
        struct hfsmount *hfsmp = NULL;
        int isdir = 0;

        if (vnode_isdir (vp)) {
                isdir = 1;
        }

        /* Ensure we only use the effective class here */
        newclass = CP_CLASS(newclass);

        if (!cp_is_valid_class(isdir, newclass)) {
                printf("hfs: CP: cp_setclass called with invalid class %d\n", newclass);
                return EINVAL;
        }

        /* Is this an interesting vp? */
        if (!cp_vnode_is_eligible(vp)) {
                return EBADF;
        }

        /* Is the mount point formatted for content protection? */
        if (!cp_fs_protected(VTOVFS(vp))) {
                return ENOTSUP;
        }

        hfsmp = VTOHFS(vp);
        if (hfsmp->hfs_flags & HFS_READ_ONLY) {
                return EROFS;
        }

        /*
         * Take the cnode truncate lock exclusive because we want to manipulate the
         * CP blob. The lock-event handling code is doing the same.  This also forces
         * all pending IOs to drain before we can re-write the persistent and cache keys.
         */
        cp = VTOC(vp);
        hfs_lock_truncate (cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
        took_truncate_lock = 1;

        /*
         * The truncate lock is not sufficient to guarantee the CP blob
         * isn't being used.  We must wait for existing writes to finish.
         */
        vnode_waitforwrites(vp, 0, 0, 0, "cp_vnode_setclass");

        if (hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)) {
                return EINVAL;
        }

        entry = cp->c_cpentry;
        if (entry == NULL) {
                error = EINVAL;
                goto out;
        }

        /* 
         * re-wrap per-file key with new class.  
         * Generate an entirely new key if switching to F. 
         */
        if (vnode_isreg(vp)) {
                /*
                 * The vnode is a file.  Before proceeding with the re-wrap, we need
                 * to unwrap the keys before proceeding.  This is to ensure that 
                 * the destination class's properties still work appropriately for the
                 * target class (since B allows I/O but an unwrap prior to the next unlock
                 * will not be allowed).
                 */
                if (!cpx_has_key(cpkp_cpx(&entry->cp_keys))) {
                        error = cp_restore_keys (entry, hfsmp, cp);
                        if (error) {
                                goto out;
                        }
                }

                if (newclass == PROTECTION_CLASS_F) {
                        /* Verify that file is blockless if switching to class F */
                        if (cp->c_datafork->ff_size > 0) {
                                error = EINVAL;
                                goto out;
                        }

                        cp_key_pair_t *cpkp;
                        cprotect_t new_entry = cp_entry_alloc(NULL, 0, CP_MAX_KEYSIZE, &cpkp);

                        if (!new_entry) {
                                error = ENOMEM;
                                goto out;
                        }

                        /* newclass is only the effective class */
                        new_entry->cp_pclass = newclass;
                        new_entry->cp_key_os_version = cp_os_version();
                        new_entry->cp_key_revision = cp_next_key_revision(entry->cp_key_revision);

                        cpx_t cpx = cpkp_cpx(cpkp);

                        /* Class F files are not wrapped, so they continue to use MAX_KEYSIZE */
                        cpx_set_key_len(cpx, CP_MAX_KEYSIZE);
                        read_random (cpx_key(cpx), CP_MAX_KEYSIZE);

                        cp_replace_entry(hfsmp, cp, new_entry);

                        error = 0;
                        goto out;
                }

                /* Deny the setclass if file is to be moved from F to something else */
                if (entry->cp_pclass == PROTECTION_CLASS_F) {
                        error = EPERM;
                        goto out;
                }

                if (!cpkp_has_pers_key(&entry->cp_keys)) {
                        struct cprotect *new_entry = NULL;
                        /*
                         * We want to fail if we can't wrap to the target class. By not setting
                         * CP_KEYWRAP_DIFFCLASS, we tell keygeneration that if it can't wrap
                         * to 'newclass' then error out.
                         */
                        uint32_t flags = 0;
                        error = cp_generate_keys (hfsmp, cp, newclass, flags,  &new_entry);
                        if (error == 0) {
                                cp_replace_entry (hfsmp, cp, new_entry);
                        }
                        /* Bypass the setxattr code below since generate_keys does it for us */
                        goto out;
                }

                cprotect_t new_entry;
                error = cp_rewrap(cp, hfsmp, &newclass, &entry->cp_keys, entry,
                                                  (cp_new_alloc_fn)cp_entry_alloc, (void **)&new_entry);
                if (error) {
                        /* we didn't have perms to set this class. leave file as-is and error out */
                        goto out;
                }

#if HFS_CONFIG_KEY_ROLL
                hfs_cp_key_roll_ctx_t *new_key_roll_ctx = NULL;
                if (entry->cp_key_roll_ctx) {
                        error = cp_rewrap(cp, hfsmp, &newclass, &entry->cp_key_roll_ctx->ckr_keys,
                                                          entry->cp_key_roll_ctx,
                                                          (cp_new_alloc_fn)hfs_key_roll_ctx_alloc,
                                                          (void **)&new_key_roll_ctx);

                        if (error) {
                                cp_entry_dealloc(hfsmp, new_entry);
                                goto out;
                        }

                        new_entry->cp_key_roll_ctx = new_key_roll_ctx;
                }
#endif

                new_entry->cp_pclass = newclass;

                cp_replace_entry(hfsmp, cp, new_entry);
                entry = new_entry;
        }
        else if (vnode_isdir(vp)) {
                /* For directories, just update the pclass.  newclass is only effective class */
                entry->cp_pclass = newclass;
                error = 0;      
        }
        else {
                /* anything else, just error out */
                error = EINVAL;
                goto out;       
        }
        
        /* 
         * We get here if the new class was F, or if we were re-wrapping a cprotect that already
         * existed. If the keys were never generated, then they'll skip the setxattr calls.
         */

        error = cp_setxattr(cp, cp->c_cpentry, VTOHFS(vp), 0, XATTR_REPLACE);
        if (error == ENOATTR) {
                error = cp_setxattr(cp, cp->c_cpentry, VTOHFS(vp), 0, XATTR_CREATE);
        }

out:

        if (took_truncate_lock) {
                hfs_unlock_truncate (cp, HFS_LOCK_DEFAULT);
        }
        hfs_unlock(cp);
        return error;
}


int cp_vnode_transcode(vnode_t vp, cp_key_t *k)
{
        struct cnode *cp;
        struct cprotect *entry = 0;
        int error = 0;
        int took_truncate_lock = 0;
        struct hfsmount *hfsmp = NULL;

        /* Structures passed between HFS and AKS */
        struct aks_cred_s access_in;
        struct aks_wrapped_key_s wrapped_key_in, wrapped_key_out;

        /* Is this an interesting vp? */
        if (!cp_vnode_is_eligible(vp)) {
                return EBADF;
        }

        /* Is the mount point formatted for content protection? */
        if (!cp_fs_protected(VTOVFS(vp))) {
                return ENOTSUP;
        }

        cp = VTOC(vp);
        hfsmp = VTOHFS(vp);

        /*
         * Take the cnode truncate lock exclusive because we want to manipulate the
         * CP blob. The lock-event handling code is doing the same.  This also forces
         * all pending IOs to drain before we can re-write the persistent and cache keys.
         */
        hfs_lock_truncate (cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
        took_truncate_lock = 1;

        if (hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT)) {
                return EINVAL;
        }

        entry = cp->c_cpentry;
        if (entry == NULL) {
                error = EINVAL;
                goto out;
        }

        /* Send the per-file key in wrapped form for re-wrap with the current class information
         * Send NULLs in the output parameters of the wrapper() and AKS will do the rest.
         * Don't need to process any outputs, so just clear the locks and pass along the error. */
        if (vnode_isreg(vp)) {

                /* Picked up the following from cp_wrap().
                 * If needed, more comments available there. */

                if (CP_CLASS(entry->cp_pclass) == PROTECTION_CLASS_F) {
                        error = EINVAL;
                        goto out;
                }

                cp_init_access(&access_in, cp);

                bzero(&wrapped_key_in, sizeof(wrapped_key_in));
                bzero(&wrapped_key_out, sizeof(wrapped_key_out));

                cp_key_pair_t *cpkp = &entry->cp_keys;

#if HFS_CONFIG_KEY_ROLL
                if (entry->cp_key_roll_ctx)
                        cpkp = &entry->cp_key_roll_ctx->ckr_keys;
#endif

                wrapped_key_in.key = cpkp_pers_key(cpkp);
                wrapped_key_in.key_len = cpkp_pers_key_len(cpkp);

                if (!wrapped_key_in.key_len) {
                        error = EINVAL;
                        goto out;
                }

                /* Use the actual persistent class when talking to AKS */
                wrapped_key_in.dp_class = entry->cp_pclass;
                wrapped_key_out.key = k->key;
                wrapped_key_out.key_len = k->len;

                error = hfs_backup_key(&access_in,
                                                           &wrapped_key_in,
                                                           &wrapped_key_out);

                if(error)
                        error = EPERM;
                else
                        k->len = wrapped_key_out.key_len;
        }

out:
        if (took_truncate_lock) {
                hfs_unlock_truncate (cp, HFS_LOCK_DEFAULT);
        }
        hfs_unlock(cp);
        return error;
}


/*
 * Check permission for the given operation (read, write) on this node.
 * Additionally, if the node needs work, do it:
 * - create a new key for the file if one hasn't been set before
 * - write out the xattr if it hasn't already been saved
 * - unwrap the key if needed
 *
 * Takes cnode lock, and upgrades to exclusive if modifying cprotect.
 *
 * Note that this function does *NOT* take the cnode truncate lock.  This is because
 * the thread calling us may already have the truncate lock.  It is not necessary
 * because either we successfully finish this function before the keys are tossed
 * and the IO will fail, or the keys are tossed and then this function will fail.
 * Either way, the cnode lock still ultimately guards the keys.  We only rely on the
 * truncate lock to protect us against tossing the keys as a cluster call is in-flight.
 */
int
cp_handle_vnop(struct vnode *vp, int vnop, int ioflag)
{
        struct cprotect *entry;
        int error = 0;
        struct hfsmount *hfsmp = NULL;
        struct cnode *cp = NULL;

        /*
         * First, do validation against the vnode before proceeding any further:
         * Is this vnode originating from a valid content-protected filesystem ?
         */
        if (cp_vnode_is_eligible(vp) == 0) {
                /*
                 * It is either not HFS or not a file/dir.  Just return success. This is a valid
                 * case if servicing i/o against another filesystem type from VFS
                 */
                return 0;
        }

        if (cp_fs_protected (VTOVFS(vp)) == 0) {
                /*
                 * The underlying filesystem does not support content protection.  This is also
                 * a valid case.  Simply return success.
                 */
                return 0;
        }

        /*
         * At this point, we know we have a HFS vnode that backs a file or directory on a
         * filesystem that supports content protection
         */
        cp = VTOC(vp);

        if ((error = hfs_lock(cp, HFS_SHARED_LOCK, HFS_LOCK_DEFAULT))) {
                return error;
        }

        entry = cp->c_cpentry;

        if (entry == NULL) {
                /*
                 * If this cnode is not content protected, simply return success.
                 * Note that this function is called by all I/O-based call sites
                 * when CONFIG_PROTECT is enabled during XNU building.
                 */

                /* 
                 * All files should have cprotect structs.  It's possible to encounter
                 * a directory from a V2.0 CP system but all files should have protection
                 * EAs
                 */
                if (vnode_isreg(vp)) {
                        error = EPERM;
                }

                goto out;
        }

        vp = CTOV(cp, 0);
        if (vp == NULL) {
                /* is it a rsrc */
                vp = CTOV(cp,1);
                if (vp == NULL) {
                        error = EINVAL;
                        goto out;
                }
        }
        hfsmp = VTOHFS(vp);

        if ((error = cp_check_access(cp, hfsmp, vnop))) {
                /* check for raw encrypted access before bailing out */
                if ((ioflag & IO_ENCRYPTED)
#if HFS_CONFIG_KEY_ROLL
                        // If we're rolling, we need the keys
                        && !hfs_is_key_rolling(cp)
#endif
                        && (vnop == CP_READ_ACCESS)) {
                        /*
                         * read access only + asking for the raw encrypted bytes
                         * is legitimate, so reset the error value to 0
                         */
                        error = 0;
                }
                else {
                        goto out;
                }
        }

        if (!ISSET(entry->cp_flags, CP_NO_XATTR)) {
                if (!S_ISREG(cp->c_mode))
                        goto out;

                // If we have a persistent key and the cached key, we're done
                if (!cp_needs_pers_key(entry)
                        && cpx_has_key(cpkp_cpx(&entry->cp_keys))) {
                        goto out;
                }
        }

        /* upgrade to exclusive lock */
        if (lck_rw_lock_shared_to_exclusive(&cp->c_rwlock) == FALSE) {
                if ((error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT))) {
                        return error;
                }
        } else {
                cp->c_lockowner = current_thread();
        }

        /* generate new keys if none have ever been saved */
        if (cp_needs_pers_key(entry)) {
                struct cprotect *newentry = NULL;
                /* 
                 * It's ok if this ends up being wrapped in a different class than 'pclass'.
                 * class modification is OK here. 
                 */             
                uint32_t flags = CP_KEYWRAP_DIFFCLASS;

                error = cp_generate_keys (hfsmp, cp, CP_CLASS(cp->c_cpentry->cp_pclass), flags, &newentry);     
                if (error == 0) {
                        cp_replace_entry (hfsmp, cp, newentry);
                        entry = newentry;
                }
                else {
                        goto out;
                }
        }

        /* unwrap keys if needed */
        if (!cpx_has_key(cpkp_cpx(&entry->cp_keys))) {
                if ((vnop == CP_READ_ACCESS) && (ioflag & IO_ENCRYPTED)) {
                        /* no need to try to restore keys; they are not going to be used */
                        error = 0;
                }
                else {
                        error = cp_restore_keys(entry, hfsmp, cp);
                        if (error) {
                                goto out;
                        }
                }
        }

        /* write out the xattr if it's new */
        if (entry->cp_flags & CP_NO_XATTR)
                error = cp_setxattr(cp, entry, VTOHFS(cp->c_vp), 0, XATTR_CREATE);

out:

        hfs_unlock(cp);
        return error;
}

#if HFS_TMPDBG
#if !SECURE_KERNEL
static void cp_log_eperm (struct vnode* vp, int pclass, boolean_t create) {
        char procname[256] = {};
        const char *fname = "unknown";
        const char *dbgop = "open";

        int ppid = proc_selfpid();
        /* selfname does a strlcpy so we're OK */
        proc_selfname(procname, sizeof(procname));
        if (vp && vp->v_name) {
                /* steal from the namecache */
                fname = vp->v_name;
        }

        if (create) {
                dbgop = "create";       
        }
        
        printf("proc %s (pid %d) class %d, op: %s failure @ file %s\n", procname, ppid, pclass, dbgop, fname);
}
#endif
#endif


int
cp_handle_open(struct vnode *vp, int mode)
{
        struct cnode *cp = NULL ;
        struct cprotect *entry = NULL;
        struct hfsmount *hfsmp;
        int error = 0;

        /* If vnode not eligible, just return success */
        if (!cp_vnode_is_eligible(vp)) {
                return 0;
        }

        /* If mount point not properly set up, then also return success */
        if (!cp_fs_protected(VTOVFS(vp))) {
                return 0;
        }

        cp = VTOC(vp);

        // Allow if raw encrypted mode requested
        if (ISSET(mode, FENCRYPTED)) {
#if HFS_CONFIG_KEY_ROLL
                // If we're rolling, we need the keys
                hfs_lock_always(cp, HFS_SHARED_LOCK);
                bool rolling = hfs_is_key_rolling(cp);
                hfs_unlock(cp);
                if (!rolling)
                        return 0;
#else
                return 0;
#endif
        }
        if (ISSET(mode, FUNENCRYPTED)) {
                return 0;
        }

        /* We know the vnode is in a valid state. Acquire cnode and validate */
        hfsmp = VTOHFS(vp);

        if ((error = hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT))) {
                return error;
        }

        entry = cp->c_cpentry;
        if (entry == NULL) {
                /* 
                 * If the mount is protected and we couldn't get a cprotect for this vnode,
                 * then it's not valid for opening.
                 */
                if (vnode_isreg(vp)) {
                        error = EPERM;
                }
                goto out;
        }

        if (!S_ISREG(cp->c_mode))
                goto out;

        /*
         * Does the cnode have keys yet?  If not, then generate them.
         */
        if (cp_needs_pers_key(entry)) {
                struct cprotect *newentry = NULL;
                /* Allow the keybag to override our class preferences */
                uint32_t flags = CP_KEYWRAP_DIFFCLASS;
                error = cp_generate_keys (hfsmp, cp, CP_CLASS(cp->c_cpentry->cp_pclass), flags, &newentry);
                if (error == 0) {
                        cp_replace_entry (hfsmp, cp, newentry);
                        entry = newentry;
                }       
                else {
                        goto out;
                }
        }       

        /*
         * We want to minimize the number of unwraps that we'll have to do since 
         * the cost can vary, depending on the platform we're running. 
         */
        switch (CP_CLASS(entry->cp_pclass)) {
                case PROTECTION_CLASS_B:
                        if (mode & O_CREAT) {
                                /* 
                                 * Class B always allows creation.  Since O_CREAT was passed through
                                 * we infer that this was a newly created vnode/cnode.  Even though a potential
                                 * race exists when multiple threads attempt to create/open a particular
                                 * file, only one can "win" and actually create it.  VFS will unset the
                                 * O_CREAT bit on the loser.     
                                 * 
                                 * Note that skipping the unwrap check here is not a security issue -- 
                                 * we have to unwrap the key permanently upon the first I/O.
                                 */
                                break;
                        }
                        
                        if (cpx_has_key(cpkp_cpx(&entry->cp_keys)) && !ISSET(mode, FENCRYPTED)) {
                                /*
                                 * For a class B file, attempt the unwrap if we have the key in
                                 * core already. 
                                 * The device could have just transitioned into the lock state, and 
                                 * this vnode may not yet have been purged from the vnode cache (which would
                                 * remove the keys). 
                                 */
                                struct aks_cred_s access_in;
                                struct aks_wrapped_key_s wrapped_key_in;

                                cp_init_access(&access_in, cp);
                                bzero(&wrapped_key_in, sizeof(wrapped_key_in));
                                wrapped_key_in.key = cpkp_pers_key(&entry->cp_keys);
                                wrapped_key_in.key_len = cpkp_pers_key_len(&entry->cp_keys);
                                /* Use the persistent class when talking to AKS */
                                wrapped_key_in.dp_class = entry->cp_pclass;
                                error = hfs_unwrap_key(&access_in, &wrapped_key_in, NULL);
                                if (error) {
                                        error = EPERM;
                                }
                                break;
                        }
                        /* otherwise, fall through to attempt the unwrap/restore */
                case PROTECTION_CLASS_A:
                case PROTECTION_CLASS_C:
                        /*
                         * At this point, we know that we need to attempt an unwrap if needed; we want
                         * to makes sure that open(2) fails properly if the device is either just-locked
                         * or never made it past first unlock.  Since the keybag serializes access to the
                         * unwrapping keys for us and only calls our VFS callback once they've been purged, 
                         * we will get here in two cases:
                         * 
                         * A) we're in a window before the wrapping keys are purged; this is OK since when they get 
                         * purged, the vnode will get flushed if needed.
                         * 
                         * B) The keys are already gone.  In this case, the restore_keys call below will fail. 
                         *
                         * Since this function is bypassed entirely if we're opening a raw encrypted file, 
                         * we can always attempt the restore.
                         */
                        if (!cpx_has_key(cpkp_cpx(&entry->cp_keys))) {
                                error = cp_restore_keys(entry, hfsmp, cp);
                        }
        
                        if (error) {
                                error = EPERM;
                        }
        
                        break;

                case PROTECTION_CLASS_D:
                default:
                        break;
        }

out:

#if HFS_TMPDBG
#if !SECURE_KERNEL
        if ((hfsmp->hfs_cp_verbose) && (error == EPERM)) {
                cp_log_eperm (vp, CP_CLASS(entry->cp_pclass), false);
        }
#endif
#endif

        hfs_unlock(cp);
        return error;
}


/*
 * cp_getrootxattr:
 * Gets the EA we set on the root folder (fileid 1) to get information about the
 * version of Content Protection that was used to write to this filesystem.
 * Note that all multi-byte fields are written to disk little endian so they must be
 * converted to native endian-ness as needed.
 */
int
cp_getrootxattr(struct hfsmount* hfsmp, struct cp_root_xattr *outxattr) 
{
        void    *buf;

        /*
         * We allow for an extra 64 bytes to cater for upgrades.  This wouldn't
         * be necessary if the xattr routines just returned what we asked for.
         */
        size_t bufsize = roundup(sizeof(struct cp_root_xattr) + 64, 64);

        int error = 0;

        hfs_assert(outxattr);

        buf = hfs_malloc(bufsize);

        uio_t uio = uio_create(1, 0, UIO_SYSSPACE, UIO_READ);

        uio_addiov(uio, CAST_USER_ADDR_T(buf), bufsize);

        size_t attrsize = bufsize;

        struct vnop_getxattr_args args = {
                .a_uio = uio,
                .a_name = CONTENT_PROTECTION_XATTR_NAME,
                .a_size = &attrsize
        };

        error = hfs_getxattr_internal(NULL, &args, hfsmp, 1);

        uio_free(uio);

        if (error != 0) {
                goto out;
        }

        if (attrsize < CP_ROOT_XATTR_MIN_LEN) {
                error = HFS_EINCONSISTENT;
                goto out;
        }

        const struct cp_root_xattr *xattr = buf;

        bzero(outxattr, sizeof(*outxattr));

        /* Now convert the multi-byte fields to native endianness */
        outxattr->major_version = OSSwapLittleToHostInt16(xattr->major_version);
        outxattr->minor_version = OSSwapLittleToHostInt16(xattr->minor_version);
        outxattr->flags = OSSwapLittleToHostInt64(xattr->flags);

        if (outxattr->major_version >= CP_VERS_5) {
                if (attrsize < sizeof(struct cp_root_xattr)) {
                        error = HFS_EINCONSISTENT;
                        goto out;
                }
#if HFS_CONFIG_KEY_ROLL
                outxattr->auto_roll_min_version = OSSwapLittleToHostInt32(xattr->auto_roll_min_version);
                outxattr->auto_roll_max_version = OSSwapLittleToHostInt32(xattr->auto_roll_max_version);
#endif
        }

out:
        hfs_free(buf, bufsize);
        return error;
}

/*
 * cp_setrootxattr:
 * Sets the EA we set on the root folder (fileid 1) to get information about the
 * version of Content Protection that was used to write to this filesystem.
 * Note that all multi-byte fields are written to disk little endian so they must be
 * converted to little endian as needed.
 *
 * This will be written to the disk when it detects the EA is not there, or when we need
 * to make a modification to the on-disk version that can be done in-place.
 */
int
cp_setrootxattr(struct hfsmount *hfsmp, struct cp_root_xattr *newxattr)
{
        int error = 0;
        struct vnop_setxattr_args args;

        args.a_desc = NULL;
        args.a_vp = NULL;
        args.a_name = CONTENT_PROTECTION_XATTR_NAME;
        args.a_uio = NULL; //pass data ptr instead
        args.a_options = 0;
        args.a_context = NULL; //no context needed, only done from mount.

        const uint64_t flags = newxattr->flags;

        /* Now convert the multi-byte fields to little endian before writing to disk. */
        newxattr->flags = OSSwapHostToLittleInt64(newxattr->flags);

        int xattr_size = sizeof(struct cp_root_xattr);

#if HFS_CONFIG_KEY_ROLL
        bool upgraded = false;

        if (newxattr->auto_roll_min_version || newxattr->auto_roll_max_version) {
                if (newxattr->major_version < CP_VERS_5) {
                        printf("hfs: upgrading to cp version %u\n", CP_CURRENT_VERS);

                        newxattr->major_version = CP_CURRENT_VERS;
                        newxattr->minor_version = CP_MINOR_VERS;

                        upgraded = true;
                }

                newxattr->auto_roll_min_version = OSSwapHostToLittleInt32(newxattr->auto_roll_min_version);
                newxattr->auto_roll_max_version = OSSwapHostToLittleInt32(newxattr->auto_roll_max_version);
        } else if (newxattr->major_version == CP_VERS_4)
                xattr_size = offsetof(struct cp_root_xattr, auto_roll_min_version);
#endif

        newxattr->major_version = OSSwapHostToLittleInt16(newxattr->major_version);
        newxattr->minor_version = OSSwapHostToLittleInt16(newxattr->minor_version);

        error = hfs_setxattr_internal(NULL, (caddr_t)newxattr,
                        xattr_size, &args, hfsmp, 1);

        if (!error) {
                hfsmp->cproot_flags = flags;
#if HFS_CONFIG_KEY_ROLL
                if (upgraded)
                        hfsmp->hfs_running_cp_major_vers = CP_CURRENT_VERS;
#endif
        }

        return error;
}


/*
 * Stores new xattr data on the cnode.
 * cnode lock held exclusive (if available).
 *
 * This function is also invoked during file creation.
 */
int cp_setxattr(struct cnode *cp, struct cprotect *entry, struct hfsmount *hfsmp,
                                uint32_t fileid, int options)
{
        int error = 0;
        cp_key_pair_t *cpkp = &entry->cp_keys;
#if HFS_CONFIG_KEY_ROLL
        bool rolling = entry->cp_key_roll_ctx != NULL;

        if (rolling && entry->cp_key_roll_ctx->ckr_off_rsrc == INT64_MAX) {
                // We've finished rolling, but we still have the context
                rolling = false;
                cpkp = &entry->cp_key_roll_ctx->ckr_keys;
        }
#endif

        if (hfsmp->hfs_flags & HFS_READ_ONLY) {
                return EROFS;
        }

        if (hfsmp->hfs_running_cp_major_vers < CP_CURRENT_VERS) {
                // Upgrade
                printf("hfs: upgrading to cp version %u\n", CP_CURRENT_VERS);

                struct cp_root_xattr root_xattr;

                error = cp_getrootxattr(hfsmp, &root_xattr);
                if (error)
                        return error;

                root_xattr.major_version = CP_CURRENT_VERS;
                root_xattr.minor_version = CP_MINOR_VERS;

                error = cp_setrootxattr(hfsmp, &root_xattr);
                if (error)
                        return error;

                hfsmp->hfs_running_cp_major_vers = CP_CURRENT_VERS;
        }

        struct cp_xattr_v5 *xattr;
        xattr = hfs_malloc(sizeof(*xattr));

        xattr->xattr_major_version      = OSSwapHostToLittleConstInt16(CP_VERS_5);
        xattr->xattr_minor_version      = OSSwapHostToLittleConstInt16(CP_MINOR_VERS);
        xattr->flags                            = 0;
#if HFS_CONFIG_KEY_ROLL
        if (rolling)
                xattr->flags |= CP_XAF_KEY_ROLLING;
#endif
        xattr->persistent_class         = OSSwapHostToLittleInt32(entry->cp_pclass);
        xattr->key_os_version           = OSSwapHostToLittleInt32(entry->cp_key_os_version);
        xattr->key_revision                     = OSSwapHostToLittleInt16(entry->cp_key_revision);

        uint16_t key_len = cpkp_pers_key_len(cpkp);
        xattr->key_len  = OSSwapHostToLittleInt16(key_len);
        memcpy(xattr->persistent_key, cpkp_pers_key(cpkp), key_len);

        size_t xattr_len = offsetof(struct cp_xattr_v5, persistent_key) + key_len;

#if HFS_CONFIG_KEY_ROLL
        if (rolling) {
                struct cp_roll_info *roll_info = PTR_ADD(struct cp_roll_info *, xattr, xattr_len);

                roll_info->off_rsrc = OSSwapHostToLittleInt64(entry->cp_key_roll_ctx->ckr_off_rsrc);

                key_len = cpkp_pers_key_len(&entry->cp_key_roll_ctx->ckr_keys);
                roll_info->key_len = OSSwapHostToLittleInt16(key_len);

                memcpy(roll_info->key, cpkp_pers_key(&entry->cp_key_roll_ctx->ckr_keys), key_len);

                xattr_len += offsetof(struct cp_roll_info, key) + key_len;
        }
#endif

        struct vnop_setxattr_args args = {
                .a_vp           = cp ? cp->c_vp : NULL,
                .a_name         = CONTENT_PROTECTION_XATTR_NAME,
                .a_options      = options,
                .a_context      = vfs_context_current(),
        };

        error = hfs_setxattr_internal(cp, xattr, xattr_len, &args, hfsmp, fileid);

        hfs_free(xattr, sizeof(*xattr));

        if (error == 0 ) {
                entry->cp_flags &= ~CP_NO_XATTR;
        }

        return error;
}

/*
 * Used by an fcntl to query the underlying FS for its content protection version #
 */

int
cp_get_root_major_vers(vnode_t vp, uint32_t *level) 
{
        int err = 0;
        struct hfsmount *hfsmp = NULL;
        struct mount *mp = NULL;

        mp = VTOVFS(vp);

        /* check if it supports content protection */
        if (cp_fs_protected(mp) == 0) {
                return ENOTSUP;
        }

        hfsmp = VFSTOHFS(mp);
        /* figure out the level */

        err = cp_root_major_vers(mp);

        if (err == 0) {
                *level = hfsmp->hfs_running_cp_major_vers;
        }
        /* in error case, cp_root_major_vers will just return EINVAL. Use that */

        return err;
}

/* Used by fcntl to query default protection level of FS */
int cp_get_default_level (struct vnode *vp, uint32_t *level) {
        int err = 0;
        struct hfsmount *hfsmp = NULL;
        struct mount *mp = NULL;

        mp = VTOVFS(vp);

        /* check if it supports content protection */
        if (cp_fs_protected(mp) == 0) {
                return ENOTSUP;
        }

        hfsmp = VFSTOHFS(mp);
        /* figure out the default */

        *level = hfsmp->default_cp_class;
        return err;
}

/********************
 * Private Functions
 *******************/

static int
cp_root_major_vers(mount_t mp)
{
        int err = 0;
        struct cp_root_xattr xattr;
        struct hfsmount *hfsmp = NULL;

        hfsmp = vfs_fsprivate(mp);
        err = cp_getrootxattr (hfsmp, &xattr);

        if (err == 0) {
                hfsmp->hfs_running_cp_major_vers = xattr.major_version;
        }
        else {
                return EINVAL;
        }

        return 0;
}

static int
cp_vnode_is_eligible(struct vnode *vp)
{
        return !vnode_issystem(vp) && (vnode_isreg(vp) || vnode_isdir(vp));
}

#if DEBUG
static const uint32_t cp_magic1 = 0x7b727063;   // cpr{
static const uint32_t cp_magic2 = 0x7270637d;   // }cpr
#endif

struct cprotect *
cp_entry_alloc(cprotect_t old, uint16_t pers_key_len,
                           uint16_t cached_key_len, cp_key_pair_t **pcpkp)
{
        struct cprotect *cp_entry;

        if (pers_key_len > CP_MAX_WRAPPEDKEYSIZE)
                return (NULL);

        size_t size = (sizeof(struct cprotect) - sizeof(cp_key_pair_t)
                                   + cpkp_size(pers_key_len, cached_key_len));

#if DEBUG
        size += 4;      // Extra for magic2
#endif

        cp_entry = hfs_malloc(size);

        if (old) {
                memcpy(cp_entry, old, offsetof(struct cprotect, cp_keys));

#if HFS_CONFIG_KEY_ROLL
                // We don't copy the key roll context
                cp_entry->cp_key_roll_ctx = NULL;
#endif
        } else {
                bzero(cp_entry, offsetof(struct cprotect, cp_keys));
        }

#if DEBUG
        cp_entry->cp_magic1 = cp_magic1;
        *PTR_ADD(uint32_t *, cp_entry, size - 4) = cp_magic2;
#endif

        cpkp_init(&cp_entry->cp_keys, pers_key_len, cached_key_len);

        /*
         * If we've been passed the old entry, then we are in the process of
         * rewrapping in which case we need to copy the cached key.  This is
         * important for class B files when the device is locked because we
         * won't be able to unwrap whilst in this state, yet we still need the
         * unwrapped key.
         */
        if (old)
                cpx_copy(cpkp_cpx(&old->cp_keys), cpkp_cpx(&cp_entry->cp_keys));

        if (pcpkp)
                *pcpkp = &cp_entry->cp_keys;

        return cp_entry;
}

static void
cp_entry_dealloc(__unused hfsmount_t *hfsmp, struct cprotect *entry)
{
#if HFS_CONFIG_KEY_ROLL
        hfs_release_key_roll_ctx(hfsmp, entry);
#endif

        cpkp_flush(&entry->cp_keys);

        size_t entry_size = (sizeof(struct cprotect) - sizeof(cp_key_pair_t)
                                                 + cpkp_sizex(&entry->cp_keys));

#if DEBUG
        hfs_assert(entry->cp_magic1 == cp_magic1);
        hfs_assert(*PTR_ADD(uint32_t *, entry, (sizeof(struct cprotect) - sizeof(cp_key_pair_t)
                                                                                + cpkp_sizex(&entry->cp_keys) == cp_magic2)));

        entry_size += 4;        // Extra for magic2
#endif

        hfs_free(entry, entry_size);
}

static int cp_read_xattr_v4(__unused hfsmount_t *hfsmp, struct cp_xattr_v4 *xattr,
                                                        size_t xattr_len, cprotect_t *pcpr, cp_getxattr_options_t options)
{
        /* Endian swap the multi-byte fields into host endianness from L.E. */
        xattr->xattr_major_version = OSSwapLittleToHostInt16(xattr->xattr_major_version);
        xattr->xattr_minor_version = OSSwapLittleToHostInt16(xattr->xattr_minor_version);
        xattr->key_size = OSSwapLittleToHostInt32(xattr->key_size);
        xattr->flags = OSSwapLittleToHostInt32(xattr->flags);
        xattr->persistent_class = OSSwapLittleToHostInt32(xattr->persistent_class);
        xattr->key_os_version = OSSwapLittleToHostInt32(xattr->key_os_version);

        /*
         * Prevent a buffer overflow, and validate the key length obtained from the
         * EA. If it's too big, then bail out, because the EA can't be trusted at this
         * point.
         */
        if (xattr->key_size > CP_MAX_WRAPPEDKEYSIZE)
                return HFS_EINCONSISTENT;

        size_t min_len = offsetof(struct cp_xattr_v4, persistent_key) + xattr->key_size;
        if (xattr_len < min_len)
                return HFS_EINCONSISTENT;

        /*
         * Class F files have no backing key; their keylength should be 0,
         * though they should have the proper flags set.
         *
         * A request to instantiate a CP for a class F file should result
         * in a bzero'd cp that just says class F, with key_flushed set.
         */
        if (CP_CLASS(xattr->persistent_class) == PROTECTION_CLASS_F
                || ISSET(xattr->flags, CP_XAF_NEEDS_KEYS)) {
                xattr->key_size = 0;
        }

        /* set up entry with information from xattr */
        cp_key_pair_t *cpkp;
        cprotect_t entry;
        
        if (ISSET(options, CP_GET_XATTR_BASIC_INFO)) {
                /* caller passed in a pre-allocated structure to get the basic info */
                entry = *pcpr;
                bzero(entry, offsetof(struct cprotect, cp_keys));
        }
        else {
                entry = cp_entry_alloc(NULL, xattr->key_size, CP_MAX_CACHEBUFLEN, &cpkp);
        }

        entry->cp_pclass = xattr->persistent_class;
        entry->cp_key_os_version = xattr->key_os_version;


        if (!ISSET(options, CP_GET_XATTR_BASIC_INFO)) {
                if (xattr->key_size) {
                        cpkp_set_pers_key_len(cpkp, xattr->key_size);
                        memcpy(cpkp_pers_key(cpkp), xattr->persistent_key, xattr->key_size);
                }

                *pcpr = entry;
        }
        else if (xattr->key_size) {
                SET(entry->cp_flags, CP_HAS_A_KEY);
        }

        return 0;
}

int cp_read_xattr_v5(hfsmount_t *hfsmp, struct cp_xattr_v5 *xattr,
                                         size_t xattr_len, cprotect_t *pcpr, cp_getxattr_options_t options)
{
        if (xattr->xattr_major_version == OSSwapHostToLittleConstInt16(CP_VERS_4)) {
                return cp_read_xattr_v4(hfsmp, (struct cp_xattr_v4 *)xattr, xattr_len, pcpr, options);
        }

        xattr->xattr_major_version      = OSSwapLittleToHostInt16(xattr->xattr_major_version);

        if (xattr->xattr_major_version != CP_VERS_5) {
                printf("hfs: cp_getxattr: unsupported xattr version %d\n",
                           xattr->xattr_major_version);
                return ENOTSUP;
        }

        size_t min_len = offsetof(struct cp_xattr_v5, persistent_key);

        if (xattr_len < min_len)
                return HFS_EINCONSISTENT;

        xattr->xattr_minor_version      = OSSwapLittleToHostInt16(xattr->xattr_minor_version);
        xattr->flags                            = OSSwapLittleToHostInt32(xattr->flags);
        xattr->persistent_class         = OSSwapLittleToHostInt32(xattr->persistent_class);
        xattr->key_os_version           = OSSwapLittleToHostInt32(xattr->key_os_version);
        xattr->key_revision                     = OSSwapLittleToHostInt16(xattr->key_revision);
        xattr->key_len                          = OSSwapLittleToHostInt16(xattr->key_len);

        uint16_t pers_key_len = xattr->key_len;

        min_len += pers_key_len;
        if (xattr_len < min_len)
                return HFS_EINCONSISTENT;

#if HFS_CONFIG_KEY_ROLL
        struct cp_roll_info *roll_info = NULL;

        if (ISSET(xattr->flags, CP_XAF_KEY_ROLLING)) {
                roll_info = PTR_ADD(struct cp_roll_info *, xattr, min_len);

                min_len += offsetof(struct cp_roll_info, key);

                if (xattr_len < min_len)
                        return HFS_EINCONSISTENT;

                roll_info->off_rsrc = OSSwapLittleToHostInt64(roll_info->off_rsrc);

        if (roll_info->off_rsrc % hfsmp->blockSize)
            return HFS_EINCONSISTENT;

                roll_info->key_len = OSSwapLittleToHostInt16(roll_info->key_len);

                min_len += roll_info->key_len;
                if (xattr_len < min_len)
                        return HFS_EINCONSISTENT;
        }
#endif

        cp_key_pair_t *cpkp;
        cprotect_t entry;
        
        /* 
         * If option CP_GET_XATTR_BASIC_INFO is set, we only return basic
         * information about the file's protection (and not the key) and
         * we store the result in the structure the caller passed to us.
         */
        if (ISSET(options, CP_GET_XATTR_BASIC_INFO)) {
                entry = *pcpr;
                bzero(entry, offsetof(struct cprotect, cp_keys));
#if HFS_CONFIG_KEY_ROLL
                if (ISSET(xattr->flags, CP_XAF_KEY_ROLLING)) {
                        SET(entry->cp_flags, CP_KEY_IS_ROLLING);
                }
#endif
        } else {
                entry = cp_entry_alloc(NULL, xattr->key_len, CP_MAX_CACHEBUFLEN, &cpkp);
        }

        entry->cp_pclass                        = xattr->persistent_class;
        entry->cp_key_os_version        = xattr->key_os_version;
        entry->cp_key_revision          = xattr->key_revision;

        if (!ISSET(options, CP_GET_XATTR_BASIC_INFO)) {
                if (xattr->key_len) {
                        cpkp_set_pers_key_len(cpkp, xattr->key_len);
                        memcpy(cpkp_pers_key(cpkp), xattr->persistent_key, xattr->key_len);
                }

#if HFS_CONFIG_KEY_ROLL
                if (roll_info) {
                        entry->cp_key_roll_ctx = hfs_key_roll_ctx_alloc(NULL, roll_info->key_len,
                                                                                                                        CP_MAX_CACHEBUFLEN, &cpkp);

                        entry->cp_key_roll_ctx->ckr_off_rsrc = roll_info->off_rsrc;

                        if (roll_info->key_len) {
                                cpkp_set_pers_key_len(cpkp, roll_info->key_len);
                                memcpy(cpkp_pers_key(cpkp), roll_info->key, roll_info->key_len);
                        }
                }
#endif

                *pcpr = entry;
        }
        else if (xattr->key_len) {
                SET(entry->cp_flags, CP_HAS_A_KEY);
        }

        return 0;
}

/*
 * Initializes a new cprotect entry with xattr data from the cnode.
 * cnode lock held shared
 */
static int
cp_getxattr(struct cnode *cp, struct hfsmount *hfsmp, cprotect_t *outentry)
{
        size_t xattr_len;
        struct cp_xattr_v5 *xattr;

        xattr = hfs_malloc(xattr_len = sizeof(*xattr));

        int error = hfs_xattr_read(cp->c_vp, CONTENT_PROTECTION_XATTR_NAME,
                                                           xattr, &xattr_len);

        if (!error) {
                if (xattr_len < CP_XATTR_MIN_LEN)
                        error = HFS_EINCONSISTENT;
                else
                        error = cp_read_xattr_v5(hfsmp, xattr, xattr_len, outentry, 0);
        }

#if DEBUG
        if (error && error != ENOATTR) {
                printf("cp_getxattr: bad cp xattr (%d):\n", error);
                for (size_t i = 0; i < xattr_len; ++i)
                        printf("%02x ", ((uint8_t *)xattr)[i]);
                printf("\n");
        }
#endif

        hfs_free(xattr, sizeof(*xattr));

        return error;
}

/*
 * If permitted, restore entry's unwrapped key from the persistent key.
 * If not, clear key and set CP_KEY_FLUSHED.
 * cnode lock held exclusive
 */
static int
cp_restore_keys(struct cprotect *entry, struct hfsmount *hfsmp, struct cnode *cp)
{
        int error = 0;

        error = cp_unwrap(hfsmp, entry, cp);
        if (error) {
                cp_flush_cached_keys(entry);
                error = EPERM;
        }
        return error;
}

void cp_device_locked_callback(mount_t mp, cp_lock_state_t state)
{
        struct hfsmount *hfsmp;

        /*
         * When iterating the various mount points that may
         * be present on a content-protected device, we need to skip
         * those that do not have it enabled.
         */
        if (!cp_fs_protected(mp)) {
                return;
        }

        hfsmp = VFSTOHFS(mp);

        hfsmp->hfs_cp_lock_state = state;

        if (state == CP_LOCKED_STATE) {
                /* 
                 * We respond only to lock events.  Since cprotect structs
                 * decrypt/restore keys lazily, the unlock events don't
                 * actually cause anything to happen.
                 */
                vnode_iterate(mp, 0, cp_lock_vnode_callback, (void *)(uintptr_t)state);
        }
}

/*
 * Deny access to protected files if keys have been locked.
 */
static int
cp_check_access(struct cnode *cp, struct hfsmount *hfsmp, int vnop __unused)
{
        int error = 0;

        /* 
         * For now it's OK to examine the state variable here without
         * holding the HFS lock.  This is only a short-circuit; if the state
         * transitions (or is in transition) after we examine this field, we'd
         * have to handle that anyway. 
         */
        if (hfsmp->hfs_cp_lock_state == CP_UNLOCKED_STATE) {
                return 0;
        }

        if (!cp->c_cpentry) {
                /* unprotected node */
                return 0;
        }

        if (!S_ISREG(cp->c_mode)) {
                return 0;
        }

        /* Deny all access for class A files */
        switch (CP_CLASS(cp->c_cpentry->cp_pclass)) {
                case PROTECTION_CLASS_A: {
                        error = EPERM;
                        break;
                }
                default:
                        error = 0;
                        break;
        }

        return error;
}

/*
 * Respond to a lock or unlock event.
 * On lock: clear out keys from memory, then flush file contents.
 * On unlock: nothing (function not called).
 */
static int
cp_lock_vnode_callback(struct vnode *vp, void *arg)
{
        cnode_t *cp = NULL;
        struct cprotect *entry = NULL;
        int error = 0;
        int locked = 1;
        unsigned long action = 0;
        int took_truncate_lock = 0;

        error = vnode_getwithref (vp);
        if (error) {
                return error;
        }

        cp = VTOC(vp);

        /*
         * When cleaning cnodes due to a lock event, we must
         * take the truncate lock AND the cnode lock.  By taking
         * the truncate lock here, we force (nearly) all pending IOs
         * to drain before we can acquire the truncate lock.  All HFS cluster
         * io calls except for swapfile IO need to acquire the truncate lock
         * prior to calling into the cluster layer.
         */
        hfs_lock_truncate (cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_DEFAULT);
        took_truncate_lock = 1;

        hfs_lock(cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);

        entry = cp->c_cpentry;
        if (!entry) {
                /* unprotected vnode: not a regular file */
                goto out;
        }

        action = (unsigned long) arg;
        switch (action) {
                case CP_LOCKED_STATE: {
                        vfs_context_t ctx;
                        if (CP_CLASS(entry->cp_pclass) != PROTECTION_CLASS_A ||
                                vnode_isdir(vp)) {
                                /*
                                 * There is no change at lock for other classes than A.
                                 * B is kept in memory for writing, and class F (for VM) does
                                 * not have a wrapped key, so there is no work needed for
                                 * wrapping/unwrapping.
                                 *
                                 * Note that 'class F' is relevant here because if
                                 * hfs_vnop_strategy does not take the cnode lock
                                 * to protect the cp blob across IO operations, we rely
                                 * implicitly on the truncate lock to be held when doing IO.
                                 * The only case where the truncate lock is not held is during
                                 * swapfile IO because HFS just funnels the VNOP_PAGEOUT
                                 * directly to cluster_pageout.
                                 */
                                goto out;
                        }

                        /* Before doing anything else, zero-fill sparse ranges as needed */
                        ctx = vfs_context_current();
                        (void) hfs_filedone (vp, ctx, 0);

                        /* first, sync back dirty pages */
                        hfs_unlock (cp);
                        ubc_msync (vp, 0, ubc_getsize(vp), NULL, UBC_PUSHALL | UBC_INVALIDATE | UBC_SYNC);
                        hfs_lock (cp, HFS_EXCLUSIVE_LOCK, HFS_LOCK_ALLOW_NOEXISTS);

                        /* flush keys:
                         * There was a concern here(9206856) about flushing keys before nand layer is done using them.
                         * But since we are using ubc_msync with UBC_SYNC, it blocks until all IO is completed.
                         * Once IOFS caches or is done with these keys, it calls the completion routine in IOSF.
                         * Which in turn calls buf_biodone() and eventually unblocks ubc_msync()
                         * Also verified that the cached data in IOFS is overwritten by other data, and there
                         * is no key leakage in that layer.
                         */

                        cp_flush_cached_keys(entry);

                        /* some write may have arrived in the mean time. dump those pages */
                        hfs_unlock(cp);
                        locked = 0;

                        ubc_msync (vp, 0, ubc_getsize(vp), NULL, UBC_INVALIDATE | UBC_SYNC);
                        break;
                }
                case CP_UNLOCKED_STATE: {
                        /* no-op */
                        break;
                }
                default:
                        panic("Content Protection: unknown lock action %lu\n", action);
        }

out:
        if (locked) {
                hfs_unlock(cp);
        }

        if (took_truncate_lock) {
                hfs_unlock_truncate (cp, HFS_LOCK_DEFAULT);
        }

        vnode_put (vp);
        return error;
}


/* 
 * cp_rewrap:
 *
 * Generate a new wrapped key based on the existing cache key.
 */

int
cp_rewrap(struct cnode *cp, __unused hfsmount_t *hfsmp,
                  cp_key_class_t *newclass, cp_key_pair_t *cpkp, const void *old_holder,
                  cp_new_alloc_fn alloc_fn, void **pholder)
{
        struct cprotect *entry = cp->c_cpentry;

        uint8_t new_persistent_key[CP_MAX_WRAPPEDKEYSIZE];
        unsigned keylen = CP_MAX_WRAPPEDKEYSIZE;
        int error = 0;
        const cp_key_class_t key_class = CP_CLASS(*newclass);

        /* Structures passed between HFS and AKS */
        struct aks_cred_s access_in;
        struct aks_wrapped_key_s wrapped_key_in;
        struct aks_wrapped_key_s wrapped_key_out;

        /*
         * PROTECTION_CLASS_F is in-use by VM swapfile; it represents a transient
         * key that is only good as long as the file is open.  There is no
         * wrapped key, so there isn't anything to wrap.
         */
        if (key_class == PROTECTION_CLASS_F) {
                return EINVAL;
        }

        cp_init_access(&access_in, cp);

        bzero(&wrapped_key_in, sizeof(wrapped_key_in));
        wrapped_key_in.key = cpkp_pers_key(cpkp);
        wrapped_key_in.key_len = cpkp_pers_key_len(cpkp);
        /* Use the persistent class when talking to AKS */
        wrapped_key_in.dp_class = entry->cp_pclass;

        bzero(&wrapped_key_out, sizeof(wrapped_key_out));
        wrapped_key_out.key = new_persistent_key;
        wrapped_key_out.key_len = keylen;

        /*
         * inode is passed here to find the backup bag wrapped blob
         * from userspace.  This lookup will occur shortly after creation
         * and only if the file still exists.  Beyond this lookup the
         * inode is not used.  Technically there is a race, we practically
         * don't lose.
         */
        error = hfs_rewrap_key(&access_in,
                                                   key_class, /* new class */
                                                   &wrapped_key_in,
                                                   &wrapped_key_out);

        keylen = wrapped_key_out.key_len;

        if (error == 0) {
                /*
                 * Verify that AKS returned to us a wrapped key of the 
                 * target class requested.   
                 */
                /* Get the effective class here */
                cp_key_class_t effective = CP_CLASS(wrapped_key_out.dp_class);
                if (effective != key_class) {
                        /* 
                         * Fail the operation if defaults or some other enforcement
                         * dictated that the class be wrapped differently. 
                         */

                        /* TODO: Invalidate the key when 12170074 unblocked */
                        return EPERM;
                }

                /* Allocate a new cpentry */
                cp_key_pair_t *new_cpkp;
                *pholder = alloc_fn(old_holder, keylen, CP_MAX_CACHEBUFLEN, &new_cpkp);

                /* copy the new key into the entry */
                cpkp_set_pers_key_len(new_cpkp, keylen);
                memcpy(cpkp_pers_key(new_cpkp), new_persistent_key, keylen);

                /* Actually record/store what AKS reported back, not the effective class stored in newclass */
                *newclass = wrapped_key_out.dp_class;
        }
        else {
                error = EPERM;
        }

        return error;
}

static int cpkp_unwrap(cnode_t *cp, cp_key_class_t key_class, cp_key_pair_t *cpkp)
{
        int error = 0;
        uint8_t iv_key[CP_IV_KEYSIZE];
        cpx_t cpx = cpkp_cpx(cpkp);

        /* Structures passed between HFS and AKS */
        struct aks_cred_s access_in;
        struct aks_wrapped_key_s wrapped_key_in;
        struct aks_raw_key_s key_out;

        cp_init_access(&access_in, cp);

        bzero(&wrapped_key_in, sizeof(wrapped_key_in));
        wrapped_key_in.key = cpkp_pers_key(cpkp);
        wrapped_key_in.key_len = cpkp_max_pers_key_len(cpkp);
        /* Use the persistent class when talking to AKS */
        wrapped_key_in.dp_class = key_class;

        bzero(&key_out, sizeof(key_out));
        key_out.iv_key = iv_key;
        key_out.key = cpx_key(cpx);
        /*
         * The unwrapper should validate/set the key length for
         * the IV key length and the cache key length, however we need
         * to supply the correct buffer length so that AKS knows how
         * many bytes it has to work with.
         */
        key_out.iv_key_len = CP_IV_KEYSIZE;
        key_out.key_len = cpx_max_key_len(cpx);

        error = hfs_unwrap_key(&access_in, &wrapped_key_in, &key_out);
        if (!error) {
                if (key_out.key_len == 0 || key_out.key_len > CP_MAX_CACHEBUFLEN) {
                        panic ("cp_unwrap: invalid key length! (%ul)\n", key_out.key_len);
                }

                if (key_out.iv_key_len != CP_IV_KEYSIZE)
                        panic ("cp_unwrap: invalid iv key length! (%ul)\n", key_out.iv_key_len);

                cpx_set_key_len(cpx, key_out.key_len);

                cpx_set_aes_iv_key(cpx, iv_key);
                cpx_set_is_sep_wrapped_key(cpx, ISSET(key_out.flags, AKS_RAW_KEY_WRAPPEDKEY));
        } else {
                error = EPERM;
        }

        return error;
}

static int
cp_unwrap(__unused struct hfsmount *hfsmp, struct cprotect *entry, struct cnode *cp)
{
        /*
         * PROTECTION_CLASS_F is in-use by VM swapfile; it represents a transient
         * key that is only good as long as the file is open.  There is no
         * wrapped key, so there isn't anything to unwrap.
         */
        if (CP_CLASS(entry->cp_pclass) == PROTECTION_CLASS_F) {
                return EPERM;
        }

        int error = cpkp_unwrap(cp, entry->cp_pclass, &entry->cp_keys);

#if HFS_CONFIG_KEY_ROLL
        if (!error && entry->cp_key_roll_ctx) {
                error = cpkp_unwrap(cp, entry->cp_pclass, &entry->cp_key_roll_ctx->ckr_keys);
                if (error)
                        cpx_flush(cpkp_cpx(&entry->cp_keys));
        }
#endif

        return error;
}

/*
 * cp_generate_keys
 *
 * Take a cnode that has already been initialized and establish persistent and
 * cache keys for it at this time. Note that at the time this is called, the
 * directory entry has already been created and we are holding the cnode lock
 * on 'cp'.
 * 
 */
int cp_generate_keys (struct hfsmount *hfsmp, struct cnode *cp, cp_key_class_t targetclass,
                uint32_t keyflags, struct cprotect **newentry) 
{

        int error = 0;
        struct cprotect *newcp = NULL;
        *newentry = NULL;

        /* Target class must be an effective class only */
        targetclass = CP_CLASS(targetclass);

        /* Validate that it has a cprotect already */
        if (cp->c_cpentry == NULL) {
                /* We can't do anything if it shouldn't be protected. */
                return 0;
        }       

        /* Asserts for the underlying cprotect */
        if (cp->c_cpentry->cp_flags & CP_NO_XATTR) {
                /* should already have an xattr by this point. */
                error = EINVAL;
                goto out;
        }

        if (S_ISREG(cp->c_mode)) {
                if (!cp_needs_pers_key(cp->c_cpentry)) {
                        error = EINVAL;
                        goto out;
                }
        }

        cp_key_revision_t key_revision = cp_initial_key_revision(hfsmp);

        error = cp_new (&targetclass, hfsmp, cp, cp->c_mode, keyflags, key_revision,
                                        (cp_new_alloc_fn)cp_entry_alloc, (void **)&newcp);
        if (error) {
                /* 
                 * Key generation failed. This is not necessarily fatal
                 * since the device could have transitioned into the lock 
                 * state before we called this.  
                 */     
                error = EPERM;
                goto out;
        }

        newcp->cp_pclass                        = targetclass;
        newcp->cp_key_os_version        = cp_os_version();
        newcp->cp_key_revision          = key_revision;

        /*
         * If we got here, then we have a new cprotect.
         * Attempt to write the new one out.
         */
        error = cp_setxattr (cp, newcp, hfsmp, cp->c_fileid, XATTR_REPLACE);

        if (error) {
                /* Tear down the new cprotect; Tell MKB that it's invalid. Bail out */
                /* TODO: rdar://12170074 needs to be fixed before we can tell MKB */
                if (newcp) {
                        cp_entry_destroy(hfsmp, newcp);
                }       
                goto out;
        }

        /* 
         * If we get here then we can assert that:
         * 1) generated wrapped/unwrapped keys.
         * 2) wrote the new keys to disk.
         * 3) cprotect is ready to go.
         */

        *newentry = newcp;

out:
        return error;

}

void cp_replace_entry (hfsmount_t *hfsmp, struct cnode *cp, struct cprotect *newentry)
{
        if (cp->c_cpentry) {
#if HFS_CONFIG_KEY_ROLL
                // Transfer the tentative reservation
                if (cp->c_cpentry->cp_key_roll_ctx && newentry->cp_key_roll_ctx) {
                        newentry->cp_key_roll_ctx->ckr_tentative_reservation
                                = cp->c_cpentry->cp_key_roll_ctx->ckr_tentative_reservation;

                        cp->c_cpentry->cp_key_roll_ctx->ckr_tentative_reservation = NULL;
                }
#endif

                cp_entry_destroy (hfsmp, cp->c_cpentry);
        }
        cp->c_cpentry = newentry;
        newentry->cp_backing_cnode = cp;

        return;
}


/*
 * cp_new
 *
 * Given a double-pointer to a cprotect, generate keys (either in-kernel or from keystore),
 * allocate a cprotect, and vend it back to the caller.
 * 
 * Additionally, decide if keys are even needed -- directories get cprotect data structures
 * but they do not have keys.
 *
 */

int
cp_new(cp_key_class_t *newclass_eff, __unused struct hfsmount *hfsmp, struct cnode *cp,
           mode_t cmode, int32_t keyflags, cp_key_revision_t key_revision,
           cp_new_alloc_fn alloc_fn, void **pholder)
{
        int error = 0;
        uint8_t new_key[CP_MAX_CACHEBUFLEN];
        unsigned new_key_len = CP_MAX_CACHEBUFLEN;  /* AKS tell us the proper key length, how much of this is used */
        uint8_t new_persistent_key[CP_MAX_WRAPPEDKEYSIZE];
        unsigned new_persistent_len = CP_MAX_WRAPPEDKEYSIZE;
        uint8_t iv_key[CP_IV_KEYSIZE];
        unsigned iv_key_len = CP_IV_KEYSIZE;
        int iswrapped = 0;
        cp_key_class_t key_class = CP_CLASS(*newclass_eff);

        /* Structures passed between HFS and AKS */
        struct aks_cred_s access_in;
        struct aks_wrapped_key_s wrapped_key_out;
        struct aks_raw_key_s key_out;

        /* Sanity check that it's a file or directory here */
        if (!(S_ISREG(cmode)) && !(S_ISDIR(cmode))) {
                return EPERM;
        }

        /*
         * Step 1: Generate Keys if needed.
         * 
         * For class F files, the kernel provides the key.
         * PROTECTION_CLASS_F is in-use by VM swapfile; it represents a transient
         * key that is only good as long as the file is open.  There is no
         * wrapped key, so there isn't anything to wrap.
         *
         * For class A->D files, the key store provides the key 
         * 
         * For Directories, we only give them a class ; no keys.
         */
        if (S_ISDIR (cmode)) {
                /* Directories */
                new_persistent_len = 0;
                new_key_len = 0;

                error = 0;
        }
        else {
                /* Must be a file */         
                if (key_class == PROTECTION_CLASS_F) {
                        /* class F files are not wrapped; they can still use the max key size */
                        new_key_len = CP_MAX_KEYSIZE;
                        read_random (&new_key[0], new_key_len);
                        new_persistent_len = 0;

                        error = 0;
                }
                else {
                        /* 
                         * The keystore is provided the file ID so that it can associate
                         * the wrapped backup blob with this key from userspace. This 
                         * lookup occurs after successful file creation.  Beyond this, the
                         * file ID is not used.  Note that there is a potential race here if
                         * the file ID is re-used.  
                         */
                        cp_init_access(&access_in, cp);
                
                        bzero(&key_out, sizeof(key_out));
                        key_out.key = new_key;
                        key_out.iv_key = iv_key;
                        /* 
                         * AKS will override our key length fields, but we need to supply
                         * the length of the buffer in those length fields so that 
                         * AKS knows hoa many bytes it has to work with.
                         */
                        key_out.key_len = new_key_len;
                        key_out.iv_key_len = iv_key_len;

                        bzero(&wrapped_key_out, sizeof(wrapped_key_out));
                        wrapped_key_out.key = new_persistent_key;
                        wrapped_key_out.key_len = new_persistent_len;

                        access_in.key_revision = key_revision;

                        error = hfs_new_key(&access_in,
                                                                key_class,
                                                                &key_out,
                                                                &wrapped_key_out);

                        if (error) {
                                /* keybag returned failure */
                                error = EPERM;
                                goto cpnew_fail;
                        }

                        /* Now sanity-check the output from new_key */
                        if (key_out.key_len == 0 || key_out.key_len > CP_MAX_CACHEBUFLEN) {
                                panic ("cp_new: invalid key length! (%ul) \n", key_out.key_len);
                        }

                        if (key_out.iv_key_len != CP_IV_KEYSIZE) {
                                panic ("cp_new: invalid iv key length! (%ul) \n", key_out.iv_key_len);
                        }       
                
                        /* 
                         * AKS is allowed to override our preferences and wrap with a 
                         * different class key for policy reasons. If we were told that 
                         * any class other than the one specified is unacceptable then error out 
                         * if that occurred.  Check that the effective class returned by 
                         * AKS is the same as our effective new class 
                         */
                        if (CP_CLASS(wrapped_key_out.dp_class) != key_class) {
                                if (!ISSET(keyflags, CP_KEYWRAP_DIFFCLASS)) {
                                        error = EPERM;
                                        /* TODO: When 12170074 fixed, release/invalidate the key! */
                                        goto cpnew_fail;
                                }
                        }

                        *newclass_eff = wrapped_key_out.dp_class;
                        new_key_len = key_out.key_len;
                        iv_key_len = key_out.iv_key_len;
                        new_persistent_len = wrapped_key_out.key_len;

                        /* Is the key a SEP wrapped key? */
                        if (key_out.flags & AKS_RAW_KEY_WRAPPEDKEY) {
                                iswrapped = 1;
                        }
                }
        }

        /*
         * Step 2: allocate cprotect and initialize it.
         */

        cp_key_pair_t *cpkp;
        *pholder = alloc_fn(NULL, new_persistent_len, new_key_len, &cpkp);
        if (*pholder == NULL) {
                return ENOMEM;
        }

        /* Copy the cache key & IV keys into place if needed. */
        if (new_key_len > 0) {
                cpx_t cpx = cpkp_cpx(cpkp);

                cpx_set_key_len(cpx, new_key_len);
                memcpy(cpx_key(cpx), new_key, new_key_len);

                /* Initialize the IV key */
                if (key_class != PROTECTION_CLASS_F)
                        cpx_set_aes_iv_key(cpx, iv_key);

                cpx_set_is_sep_wrapped_key(cpx, iswrapped);
        }
        if (new_persistent_len > 0) {
                cpkp_set_pers_key_len(cpkp, new_persistent_len);
                memcpy(cpkp_pers_key(cpkp), new_persistent_key, new_persistent_len);
        }

cpnew_fail:

#if HFS_TMPDBG
#if !SECURE_KERNEL
        if ((hfsmp->hfs_cp_verbose) && (error == EPERM)) {
                /* Only introspect the data fork */
                cp_log_eperm (cp->c_vp, *newclass_eff, true);
        }
#endif
#endif

        return error;
}

/* Initialize the aks_cred_t structure passed to AKS */
static void cp_init_access(aks_cred_t access, struct cnode *cp)
{
        vfs_context_t context = vfs_context_current();
        kauth_cred_t cred = vfs_context_ucred(context);
        proc_t proc = vfs_context_proc(context);

        bzero(access, sizeof(*access));

        /* Note: HFS uses 32-bit fileID, even though inode is a 64-bit value */
        access->inode = cp->c_fileid;
        access->pid = proc_pid(proc);
        access->uid = kauth_cred_getuid(cred);

        if (cp->c_cpentry)
                access->key_revision = cp->c_cpentry->cp_key_revision;

        return;
}

#if HFS_CONFIG_KEY_ROLL

errno_t cp_set_auto_roll(hfsmount_t *hfsmp,
                                                 const hfs_key_auto_roll_args_t *args)
{
        // 64 bytes should be OK on the stack
        _Static_assert(sizeof(struct cp_root_xattr) < 64, "cp_root_xattr too big!");

        struct cp_root_xattr xattr;
        errno_t ret;

        ret = cp_getrootxattr(hfsmp, &xattr);
        if (ret)
                return ret;

        ret = hfs_start_transaction(hfsmp);
        if (ret)
                return ret;

        xattr.auto_roll_min_version = args->min_key_os_version;
        xattr.auto_roll_max_version = args->max_key_os_version;

        bool roll_old_class_gen = ISSET(args->flags, HFS_KEY_AUTO_ROLL_OLD_CLASS_GENERATION);

        if (roll_old_class_gen)
                SET(xattr.flags, CP_ROOT_AUTO_ROLL_OLD_CLASS_GENERATION);
        else
                CLR(xattr.flags, CP_ROOT_AUTO_ROLL_OLD_CLASS_GENERATION);

        ret = cp_setrootxattr(hfsmp, &xattr);

        errno_t ret2 = hfs_end_transaction(hfsmp);

        if (!ret)
                ret = ret2;

        if (ret)
                return ret;

        hfs_lock_mount(hfsmp);
        hfsmp->hfs_auto_roll_min_key_os_version = args->min_key_os_version;
        hfsmp->hfs_auto_roll_max_key_os_version = args->max_key_os_version;
        hfs_unlock_mount(hfsmp);

        return ret;
}

bool cp_should_auto_roll(hfsmount_t *hfsmp, cprotect_t cpr)
{
        if (cpr->cp_key_roll_ctx) {
                // Already rolling
                return false;
        }

        // Only automatically roll class A, B & C
        if (CP_CLASS(cpr->cp_pclass) < PROTECTION_CLASS_A
                || CP_CLASS(cpr->cp_pclass) > PROTECTION_CLASS_C) {
                return false;
        }

        if (!cpkp_has_pers_key(&cpr->cp_keys))
                return false;

        /*
         * Remember, the class generation stored in HFS+ is updated at the *end*,
         * so it's old if it matches the generation we have stored.
         */
        if (ISSET(hfsmp->cproot_flags, CP_ROOT_AUTO_ROLL_OLD_CLASS_GENERATION)
                && cp_get_crypto_generation(cpr->cp_pclass) == hfsmp->cp_crypto_generation) {
                return true;
        }

        if (!hfsmp->hfs_auto_roll_min_key_os_version
                && !hfsmp->hfs_auto_roll_max_key_os_version) {
                // No minimum or maximum set
                return false;
        }

        if (hfsmp->hfs_auto_roll_min_key_os_version
                && cpr->cp_key_os_version < hfsmp->hfs_auto_roll_min_key_os_version) {
                // Before minimum
                return false;
        }

        if (hfsmp->hfs_auto_roll_max_key_os_version
                && cpr->cp_key_os_version >= hfsmp->hfs_auto_roll_max_key_os_version) {
                // Greater than maximum
                return false;
        }

        return true;
}

#endif // HFS_CONFIG_KEY_ROLL

errno_t cp_handle_strategy(buf_t bp)
{
        vnode_t vp = buf_vnode(bp);
        cnode_t *cp = NULL;

        if (bufattr_rawencrypted(buf_attr(bp))
                || !(cp = cp_get_protected_cnode(vp))
                || !cp->c_cpentry) {
                // Nothing to do
                return 0;
        }

        /*
         * For filesystem resize, we may not have access to the underlying
         * file's cache key for whatever reason (device may be locked).
         * However, we do not need it since we are going to use the
         * temporary HFS-wide resize key which is generated once we start
         * relocating file content.  If this file's I/O should be done
         * using the resize key, it will have been supplied already, so do
         * not attach the file's cp blob to the buffer.
         */
        if (ISSET(cp->c_cpentry->cp_flags, CP_RELOCATION_INFLIGHT))
                return 0;

#if HFS_CONFIG_KEY_ROLL
        /*
         * We don't require any locks here.  Pages will be locked so no
         * key rolling can take place until this I/O has completed.
         */
        if (!cp->c_cpentry->cp_key_roll_ctx)
#endif
        {
                // Fast path
                cpx_t cpx = cpkp_cpx(&cp->c_cpentry->cp_keys);

                if (cpx_has_key(cpx)) {
                        bufattr_setcpx(buf_attr(bp), cpx);
                        return 0;
                }
        }

        /*
         * We rely mostly (see note below) upon the truncate lock to
         * protect the CP cache key from getting tossed prior to our IO
         * finishing here.  Nearly all cluster io calls to manipulate file
         * payload from HFS take the truncate lock before calling into the
         * cluster layer to ensure the file size does not change, or that
         * they have exclusive right to change the EOF of the file.  That
         * same guarantee protects us here since the code that deals with
         * CP lock events must now take the truncate lock before doing
         * anything.
         *
         * If you want to change content protection structures, then the
         * truncate lock is not sufficient; you must take the truncate
         * lock and then wait for outstanding writes to complete.  This is
         * necessary because asynchronous I/O only holds the truncate lock
         * whilst I/O is being queued.
         *
         * One exception should be the VM swapfile IO, because HFS will
         * funnel the VNOP_PAGEOUT directly into a cluster_pageout call
         * for the swapfile code only without holding the truncate lock.
         * This is because individual swapfiles are maintained at
         * fixed-length sizes by the VM code.  In non-swapfile IO we use
         * PAGEOUT_V2 semantics which allow us to create our own UPL and
         * thus take the truncate lock before calling into the cluster
         * layer.  In that case, however, we are not concerned with the CP
         * blob being wiped out in the middle of the IO because there
         * isn't anything to toss; the VM swapfile key stays in-core as
         * long as the file is open.
         */

        off_rsrc_t off_rsrc = off_rsrc_make(buf_lblkno(bp) * GetLogicalBlockSize(vp),
                                                                                VNODE_IS_RSRC(vp));
        cp_io_params_t io_params;


        /*
         * We want to take the cnode lock here and because the vnode write
         * count is a pseudo-lock, we need to do something to preserve
         * lock ordering; the cnode lock comes before the write count.
         * Ideally, the write count would be incremented after the
         * strategy routine returns, but that becomes complicated if the
         * strategy routine where to call buf_iodone before returning.
         * For now, we drop the write count here and then pick it up again
         * later.
         */
        if (!ISSET(buf_flags(bp), B_READ) && !ISSET(buf_flags(bp), B_RAW))
                vnode_writedone(vp);

        hfs_lock_always(cp, HFS_SHARED_LOCK);
        cp_io_params(VTOHFS(vp), cp->c_cpentry, off_rsrc,
                                 ISSET(buf_flags(bp), B_READ) ? VNODE_READ : VNODE_WRITE,
                                 &io_params);
        hfs_unlock(cp);

        /*
         * Last chance: If this data protected I/O does not have unwrapped
         * keys present, then try to get them.  We already know that it
         * should, by this point.
         */
        if (!cpx_has_key(io_params.cpx)) {
                int io_op = ( (buf_flags(bp) & B_READ) ? CP_READ_ACCESS : CP_WRITE_ACCESS);
                errno_t error = cp_handle_vnop(vp, io_op, 0);
                if (error) {
                        /*
                         * We have to be careful here.  By this point in the I/O
                         * path, VM or the cluster engine has prepared a buf_t
                         * with the proper file offsets and all the rest, so
                         * simply erroring out will result in us leaking this
                         * particular buf_t.  We need to properly decorate the
                         * buf_t just as buf_strategy would so as to make it
                         * appear that the I/O errored out with the particular
                         * error code.
                         */
                        if (!ISSET(buf_flags(bp), B_READ) && !ISSET(buf_flags(bp), B_RAW))
                                vnode_startwrite(vp);
                        buf_seterror (bp, error);
                        buf_biodone(bp);
                        return error;
                }

                hfs_lock_always(cp, HFS_SHARED_LOCK);
                cp_io_params(VTOHFS(vp), cp->c_cpentry, off_rsrc,
                                         ISSET(buf_flags(bp), B_READ) ? VNODE_READ : VNODE_WRITE,
                                         &io_params);
                hfs_unlock(cp);
        }

        hfs_assert(buf_count(bp) <= io_params.max_len);
        bufattr_setcpx(buf_attr(bp), io_params.cpx);

        if (!ISSET(buf_flags(bp), B_READ) && !ISSET(buf_flags(bp), B_RAW))
                vnode_startwrite(vp);

        return 0;
}

#endif /* CONFIG_PROTECT */