erofs-freebsd-out-tree/inode.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Copyright (C) 2017-2018 HUAWEI, Inc.
 *             https://www.huawei.com/
 * Copyright (C) 2021, Alibaba Cloud
 */

#include <sys/param.h>
#include <sys/endian.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/libkern.h>
#include <sys/limits.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/fnv_hash.h>
#include <sys/stat.h>
#include <sys/vnode.h>

#include "internal.h"

static bool
erofs_is_48bit(const struct erofs_mount *em)
{
	return ((em->feature_incompat & EROFS_FEATURE_INCOMPAT_48BIT) != 0);
}

static uint64_t
erofs_addrmask(const struct erofs_mount *em)
{
	if (erofs_is_48bit(em))
		return ((1ULL << 48) - 1);
	return (UINT32_MAX);
}

static dev_t
erofs_decode_dev(uint32_t dev)
{
	unsigned int major, minor;

	major = (dev & 0xfff00) >> 8;
	minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
	return (makedev(major, minor));
}

static uint32_t
erofs_inode_generation(const struct erofs_mount *em, uint64_t nid,
    const void *inode, size_t inode_size)
{
	uint8_t encoded_nid[sizeof(nid)];
	uint32_t generation;

	le64enc(encoded_nid, nid);
	generation = fnv_32_buf(encoded_nid, sizeof(encoded_nid),
	    em->generation_seed);
	generation = fnv_32_buf(inode, inode_size, generation);
	return (generation != 0 ? generation : 1);
}

static int
erofs_set_timestamp(struct erofs_node *en, uint64_t seconds,
    uint32_t nanoseconds)
{
	if (nanoseconds >= 1000000000 || seconds > (uint64_t)INT64_MAX)
		return (EINTEGRITY);
	en->mtime = seconds;
	en->mtime_nsec = nanoseconds;
	return (0);
}

static int
erofs_set_data_blocks(const struct erofs_mount *em, struct erofs_node *en,
    uint64_t compressed_blocks)
{
	if (en->datalayout == EROFS_INODE_COMPRESSED_FULL ||
	    en->datalayout == EROFS_INODE_COMPRESSED_COMPACT) {
		en->data_blocks = compressed_blocks;
		return (0);
	}
	if (en->size == 0) {
		en->data_blocks = 0;
		return (0);
	}
	if (en->size > UINT64_MAX - (em->block_size - 1))
		return (EINTEGRITY);
	en->data_blocks = roundup2(en->size, (uint64_t)em->block_size) >>
	    em->block_bits;
	return (0);
}

static int
erofs_validate_inline_data(const struct erofs_mount *em,
    const struct erofs_node *en)
{
	uint64_t image_size, inline_end, inline_off, inline_size, tail_start;

	if (en->datalayout != EROFS_INODE_FLAT_INLINE || en->size == 0)
		return (0);
	tail_start = roundup2(en->size, (uint64_t)em->block_size) -
	    em->block_size;
	inline_size = en->size - tail_start;
	if (__builtin_add_overflow(en->inode_off, en->inode_isize, &inline_off) ||
	    __builtin_add_overflow(inline_off, en->xattr_isize, &inline_off) ||
	    __builtin_add_overflow(inline_off, inline_size, &inline_end))
		return (EINTEGRITY);
	if ((inline_off & (em->block_size - 1)) + inline_size > em->block_size)
		return (EINTEGRITY);
	if (erofs_nid_in_metabox(en->nid)) {
		if (em->metabox_en == NULL || inline_end > em->metabox_en->size)
			return (EINTEGRITY);
		return (0);
	}
	if (em->blocks > (UINT64_MAX >> em->block_bits))
		return (EINTEGRITY);
	image_size = em->blocks << em->block_bits;
	if (inline_end > image_size || inline_end > em->dif0.mediasize)
		return (EINTEGRITY);
	return (0);
}

/*
 * Convert a logical nid to its inode-table byte offset.  Normal NIDs are
 * relative to the primary metadata area.  For metabox NIDs, bit 63 selects
 * the metabox backing inode and the remaining bits are relative to its data.
 * EROFS_NULL_ADDR is returned when the address cannot be represented.
 */
uint64_t
erofs_iloc(struct erofs_mount *em, uint64_t nid)
{
	uint64_t meta_offset, nid_lo, result;
	bool in_metabox;

	in_metabox = erofs_nid_in_metabox(nid);
	if (in_metabox && !erofs_sb_has_metabox(em))
		return (EROFS_NULL_ADDR);
	nid_lo = nid & EROFS_DIRENT_NID_MASK;
	if (nid_lo > (UINT64_MAX >> 5))
		return (EROFS_NULL_ADDR);
	result = nid_lo << 5;
	if (in_metabox)
		return (result);

	if (em->block_bits > 58)
		return (EROFS_NULL_ADDR);
	meta_offset = (uint64_t)em->meta_blkaddr << em->block_bits;
	if (result > UINT64_MAX - meta_offset)
		return (EROFS_NULL_ADDR);

	return (meta_offset + result);
}

/*
 * Check that a NID can address at least one compact inode slot without
 * crossing the declared primary image or metabox backing-file boundary.
 */
bool
erofs_nid_is_valid(struct erofs_mount *em, uint64_t nid)
{
	uint64_t image_size, off;

	off = erofs_iloc(em, nid);
	if (off == EROFS_NULL_ADDR)
		return (false);
	if (erofs_nid_in_metabox(nid)) {
		if (em->metabox_en == NULL || off > em->metabox_en->size)
			return (false);
		return (sizeof(struct erofs_inode_compact) <=
		    em->metabox_en->size - off);
	}
	if (em->blocks > (UINT64_MAX >> em->block_bits))
		return (false);
	image_size = em->blocks << em->block_bits;
	if (off > image_size || sizeof(struct erofs_inode_compact) >
	    image_size - off)
		return (false);
	if (off > em->dif0.mediasize || sizeof(struct erofs_inode_compact) >
	    em->dif0.mediasize - off)
		return (false);
	return (true);
}

/*
 * Read and decode a disk inode.
 *
 * Currently supports:
 * - compact / extended inode;
 * - plain / inline uncompressed layouts;
 * - basic 48-bit address parsing;
 * - compact inode epoch/fixed_nsec timestamp semantics;
 * - dot_omitted / nlink==1 i_format details.
 */
int
erofs_read_inode(struct erofs_mount *em, uint64_t nid, struct erofs_node *en)
{
	struct erofs_inode_compact *dic;
	struct erofs_inode_extended *die;
	struct erofs_inode_chunk_info chunk_info;
	union erofs_inode_i_nb inode_nb;
	void *buf;
	uint64_t addrmask, mtime, off, startblk;
	uint64_t compressed_blocks;
	uint32_t raw_rdev, startblk_lo;
	uint16_t ifmt, startblk_hi;
	int error;

	if (!erofs_nid_is_valid(em, nid))
		return (EINTEGRITY);
	off = erofs_iloc(em, nid);
	error = erofs_read_metadata(em, nid, off,
	    sizeof(struct erofs_inode_compact), &buf);
	if (error != 0)
		return (error);

	bzero(&en->size, sizeof(*en) - offsetof(struct erofs_node, size));
	en->nid = nid;
	en->inode_off = off;
	ifmt = le16toh(*(uint16_t *)buf);
	if ((ifmt & ~EROFS_I_ALL) != 0) {
		erofs_brelse(buf);
		return (EOPNOTSUPP);
	}
	en->datalayout = erofs_inode_datalayout(ifmt);
	if (en->datalayout >= EROFS_INODE_DATALAYOUT_MAX) {
		erofs_brelse(buf);
		return (EOPNOTSUPP);
	}
	en->compact_inode = (erofs_inode_version(ifmt) == 0);
	if (!en->compact_inode) {
		erofs_brelse(buf);
		error = erofs_read_metadata(em, nid, off,
		    sizeof(struct erofs_inode_extended), &buf);
		if (error != 0)
			return (error);
	}
	addrmask = erofs_addrmask(em);
	startblk = EROFS_NULL_ADDR;
	startblk_lo = 0;
	startblk_hi = 0;
	compressed_blocks = 0;
	raw_rdev = 0;
	bzero(&inode_nb, sizeof(inode_nb));
	dic = buf;
	if (en->compact_inode) {
		en->inode_isize = sizeof(struct erofs_inode_compact);
		en->generation = erofs_inode_generation(em, nid, buf,
		    en->inode_isize);
		en->mode = le16toh(dic->i_mode);
		en->size = le32toh(dic->i_size);
		en->ino = le32toh(dic->i_ino);
		en->uid = le16toh(dic->i_uid);
		en->gid = le16toh(dic->i_gid);
		en->xattr_isize = erofs_xattr_ibody_size(dic->i_xattr_icount);
		if (__builtin_add_overflow(em->epoch,
		    (uint64_t)le32toh(dic->i_mtime), &mtime)) {
			erofs_brelse(buf);
			return (EINTEGRITY);
		}
		error = erofs_set_timestamp(en, mtime, em->fixed_nsec);
		if (error != 0) {
			erofs_brelse(buf);
			return (error);
		}
		startblk_lo = le32toh(dic->i_u.startblk_lo);
		compressed_blocks = le32toh(dic->i_u.blocks_lo);
		raw_rdev = le32toh(dic->i_u.rdev);
		if (!S_ISDIR(en->mode) &&
		    ((ifmt >> EROFS_I_NLINK_1_BIT) & 0x1) != 0) {
			en->nlink = 1;
			inode_nb = dic->i_nb;
		} else {
			en->nlink = le16toh(dic->i_nb.nlink);
			addrmask = UINT32_MAX;
		}
	} else {
		die = buf;
		en->inode_isize = sizeof(struct erofs_inode_extended);
		en->generation = erofs_inode_generation(em, nid, buf,
		    en->inode_isize);
		en->mode = le16toh(die->i_mode);
		en->size = le64toh(die->i_size);
		en->ino = le32toh(die->i_ino);
		en->uid = le32toh(die->i_uid);
		en->gid = le32toh(die->i_gid);
		en->nlink = le32toh(die->i_nlink);
		inode_nb = die->i_nb;
		en->xattr_isize = erofs_xattr_ibody_size(die->i_xattr_icount);
		error = erofs_set_timestamp(en, le64toh(die->i_mtime),
		    le32toh(die->i_mtime_nsec));
		if (error != 0) {
			erofs_brelse(buf);
			return (error);
		}
		startblk_lo = le32toh(die->i_u.startblk_lo);
		compressed_blocks = le32toh(die->i_u.blocks_lo);
		raw_rdev = le32toh(die->i_u.rdev);
	}
	startblk_hi = le16toh(inode_nb.startblk_hi);
	compressed_blocks |= (uint64_t)le16toh(inode_nb.blocks_hi) << 32;
	if (en->size > (uint64_t)OFF_MAX) {
		erofs_brelse(buf);
		return (EINTEGRITY);
	}

	en->vtype = IFTOVT(en->mode);
	if (en->mode != 0 && en->vtype == VNON) {
		erofs_brelse(buf);
		return (EINTEGRITY);
	}
	en->inline_data = (en->datalayout == EROFS_INODE_FLAT_INLINE);
	en->dot_omitted = (en->vtype == VDIR) &&
	    (((ifmt >> EROFS_I_DOT_OMITTED_BIT) & 0x1) != 0);

	if (en->datalayout == EROFS_INODE_COMPRESSED_FULL ||
	    en->datalayout == EROFS_INODE_COMPRESSED_COMPACT) {
		error = z_erofs_fill_inode(em, en);
		if (error != 0) {
			erofs_brelse(buf);
			return (error);
		}
	} else if (en->datalayout == EROFS_INODE_CHUNK_BASED) {
		if (!erofs_sb_has_chunked_file(em) || en->vtype != VREG) {
			erofs_brelse(buf);
			return (EINTEGRITY);
		}
		if (en->compact_inode)
			chunk_info = dic->i_u.c;
		else
			chunk_info = die->i_u.c;
		if (le16toh(chunk_info.reserved) != 0) {
			erofs_brelse(buf);
			return (EINTEGRITY);
		}
		en->chunkformat = le16toh(chunk_info.format);
		if (en->chunkformat & ~EROFS_CHUNK_FORMAT_ALL) {
			erofs_brelse(buf);
			return (EOPNOTSUPP);
		}
		if ((en->chunkformat & EROFS_CHUNK_FORMAT_48BIT) != 0 &&
		    (en->chunkformat & EROFS_CHUNK_FORMAT_INDEXES) == 0) {
			erofs_brelse(buf);
			return (EINTEGRITY);
		}
		en->chunkbits = em->block_bits +
		    (en->chunkformat & EROFS_CHUNK_FORMAT_BLKBITS_MASK);
		if (en->chunkbits >= 64) {
			erofs_brelse(buf);
			return (EINTEGRITY);
		}
	} else if (en->datalayout != EROFS_INODE_FLAT_PLAIN &&
	    en->datalayout != EROFS_INODE_FLAT_INLINE) {
		erofs_brelse(buf);
		return (EOPNOTSUPP);
	}

	switch (en->vtype) {
	case VREG:
	case VDIR:
	case VLNK:
		if (en->datalayout == EROFS_INODE_CHUNK_BASED) {
			en->startblk = EROFS_NULL_ADDR;
			en->rdev = NODEV;
			break;
		}
		startblk = startblk_lo | ((uint64_t)startblk_hi << 32);
		if (en->datalayout == EROFS_INODE_FLAT_PLAIN &&
		    ((startblk ^ EROFS_NULL_ADDR) & addrmask) == 0)
			startblk = EROFS_NULL_ADDR;
		en->startblk = startblk;
		en->rdev = NODEV;
		break;
	case VCHR:
	case VBLK:
		en->startblk = EROFS_NULL_ADDR;
		en->rdev = erofs_decode_dev(raw_rdev);
		break;
	case VFIFO:
	case VSOCK:
		en->startblk = EROFS_NULL_ADDR;
		en->rdev = NODEV;
		break;
	default:
		erofs_brelse(buf);
		return (EINTEGRITY);
	}
	error = erofs_set_data_blocks(em, en, compressed_blocks);
	if (error == 0)
		error = erofs_validate_inline_data(em, en);
	if (error != 0) {
		erofs_brelse(buf);
		return (error);
	}

	erofs_brelse(buf);
	return (0);
}

static u_int
erofs_vfs_hash(uint64_t nid)
{

	return (fnv_32_buf(&nid, sizeof(nid), FNV1_32_INIT));
}

static int
erofs_vfs_hash_cmp(struct vnode *vp, void *pnid)
{
	struct erofs_node *en;

	en = VTOE(vp);
	return (en == NULL || en->nid != *(uint64_t *)pnid);
}

/*
 * Get vnode by raw on-disk nid.  The raw nid is also the FreeBSD fileid and
 * hash identity, so the metabox selector bit remains collision-free.
 * Uses the standard FreeBSD vfs_hash API.
 * (Linux equivalent: erofs_iget in Linux's inode.c)
 */
int
erofs_vget(struct mount *mp, ino_t ino, int flags, struct vnode **vpp)
{
	struct erofs_mount *em;
	struct erofs_node *en;
	struct thread *td;
	struct vnode *vp;
	uint64_t nid;
	u_int hash;
	bool shared;
	int error;

	td = curthread;
	nid = (uint64_t)ino;
	shared = (flags & LK_TYPE_MASK) == LK_SHARED;
	hash = erofs_vfs_hash(nid);
	error = vfs_hash_get(mp, hash, flags, td, vpp, erofs_vfs_hash_cmp,
	    &nid);
	if (error != 0 || *vpp != NULL)
		return (error);

	em = MTOE(mp);
	en = malloc(sizeof(*en), M_EROFS, M_WAITOK | M_ZERO);
	error = getnewvnode("erofs", mp, &erofs_vnodeops, &vp);
	if (error != 0) {
		free(en, M_EROFS);
		*vpp = NULL;
		return (error);
	}
	vp->v_data = en;
	en->vnode = vp;
	en->nid = nid;
	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
	error = insmntque(vp, mp);
	if (error != 0) {
		free(en, M_EROFS);
		vp->v_data = NULL;
		*vpp = NULL;
		return (error);
	}
	error = vfs_hash_insert(vp, hash, flags, td, vpp, erofs_vfs_hash_cmp,
	    &nid);
	if (error != 0 || *vpp != NULL)
		return (error);

	error = erofs_read_inode(em, nid, en);
	if (error != 0) {
		*vpp = NULL;
		vgone(vp);
		vput(vp);
		return (error);
	}
	vp->v_type = en->vtype;
	if (vp->v_type == VFIFO)
		vp->v_op = &erofs_fifoops;
	if ((uint64_t)ino == em->root_nid)
		vp->v_vflag |= VV_ROOT;
	vn_set_state(vp, VSTATE_CONSTRUCTED);
	if (shared)
		VOP_LOCK(vp, LK_DOWNGRADE);
	*vpp = vp;
	return (0);
}