smb: client: compress: LZ77 code improvements cleanup

- Check data compressibility with some heuristics (copied from
  btrfs):
  - should_compress() final decision is is_compressible(data)

- Cleanup compress/lz77.h leaving only lz77_compress() exposed:
  - Move parts to compress/lz77.c, while removing the rest of it
    because they were either unused, used only once, were
    implemented wrong (thanks to David Howells for the help)

- Updated the compression parameters (still compatible with
  Windows implementation) trading off ~20% compression ratio
  for ~40% performance:
  - min match len: 3 -> 4
  - max distance: 8KiB -> 1KiB
  - hash table type: u32 * -> u64 *

Known bugs:
This implementation currently works fine in general, but breaks with
some payloads used during testing.  Investigation ongoing, to be
fixed in a next commit.

Signed-off-by: Enzo Matsumiya <[email protected]>
Co-developed-by: David Howells <[email protected]>
Signed-off-by: David Howells <[email protected]>
Signed-off-by: Steve French <[email protected]>

Author: ematsumiya

fs/smb/client/compress.c

@@ -15,6 +15,7 @@
 #include <linux/slab.h>
 #include <linux/kernel.h>
 #include <linux/uio.h>
+#include <linux/sort.h>
 
 #include "cifsglob.h"
 #include "../common/smb2pdu.h"
@@ -24,27 +25,361 @@
 #include "compress/lz77.h"
 #include "compress.h"
 
-int smb_compress(void *buf, const void *data, size_t *len)
+/*
+ * The heuristic_*() functions below try to determine data compressibility.
+ *
+ * Derived from fs/btrfs/compression.c, changing coding style, some parameters, and removing
+ * unused parts.
+ *
+ * Read that file for better and more detailed explanation of the calculations.
+ *
+ * The algorithms are ran in a collected sample of the input (uncompressed) data.
+ * The sample is formed of 2K reads in PAGE_SIZE intervals, with a maximum size of 4M.
+ *
+ * Parsing the sample goes from "low-hanging fruits" (fastest algorithms, likely compressible)
+ * to "need more analysis" (likely uncompressible).
+ */
+
+struct bucket {
+	unsigned int count;
+};
+
+/**
+ * calc_shannon_entropy() - Compute Shannon entropy of the sampled data.
+ * @bkt:	Bytes counts of the sample.
+ * @slen:	Size of the sample.
+ *
+ * Return: true if the level (percentage of number of bits that would be required to
+ *	   compress the data) is below the minimum threshold.
+ *
+ * Note:
+ * There _is_ an entropy level here that's > 65 (minimum threshold) that would indicate a
+ * possibility of compression, but compressing, or even further analysing, it would waste so much
+ * resources that it's simply not worth it.
+ *
+ * Also Shannon entropy is the last computed heuristic; if we got this far and ended up
+ * with uncertainty, just stay on the safe side and call it uncompressible.
+ */
+static bool calc_shannon_entropy(struct bucket *bkt, size_t slen)
+{
+	const size_t threshold = 65, max_entropy = 8 * ilog2(16);
+	size_t i, p, p2, len, sum = 0;
+
+#define pow4(n) (n * n * n * n)
+	len = ilog2(pow4(slen));
+
+	for (i = 0; i < 256 && bkt[i].count > 0; i++) {
+		p = bkt[i].count;
+		p2 = ilog2(pow4(p));
+		sum += p * (len - p2);
+	}
+
+	sum /= slen;
+
+	return ((sum * 100 / max_entropy) <= threshold);
+}
+
+/**
+ * calc_byte_distribution() - Compute byte distribution on the sampled data.
+ * @bkt:	Byte counts of the sample.
+ * @slen:	Size of the sample.
+ *
+ * Return:
+ * 1:	High probability (normal (Gaussian) distribution) of the data being compressible.
+ * 0:	A "hard no" for compression -- either a computed uniform distribution of the bytes (e.g.
+ *	random or encrypted data), or calc_shannon_entropy() returned false (see above).
+ * 2:	When computed byte distribution resulted in "low > n < high" grounds.
+ *	calc_shannon_entropy() should be used for a final decision.
+ */
+static int calc_byte_distribution(struct bucket *bkt, size_t slen)
 {
-	struct smb2_compression_hdr *hdr;
-	size_t buf_len, data_len;
+	const size_t low = 64, high = 200, threshold = slen * 90 / 100;
+	size_t sum = 0;
+	int i;
+
+	for (i = 0; i < low; i++)
+		sum += bkt[i].count;
+
+	if (sum > threshold)
+		return i;
+
+	for (; i < high && bkt[i].count > 0; i++) {
+		sum += bkt[i].count;
+		if (sum > threshold)
+			break;
+	}
+
+	if (i <= low)
+		return 1;
+
+	if (i >= high)
+		return 0;
+
+	return 2;
+}
+
+static bool check_ascii_bytes(const struct bucket *bkt)
+{
+	const size_t threshold = 64;
+	size_t count = 0;
+	int i;
+
+	for (i = 0; i < threshold; i++)
+		if (bkt[i].count > 0)
+			count++;
+
+	for (; i < 256; i++) {
+		if (bkt[i].count > 0) {
+			count++;
+			if (count > threshold)
+				break;
+		}
+	}
+
+	return (count < threshold);
+}
+
+static bool check_repeated_data(const u8 *sample, size_t len)
+{
+	size_t s = len / 2;
+
+	return (!memcmp(&sample[0], &sample[s], s));
+}
+
+static int cmp_bkt(const void *_a, const void *_b)
+{
+	const struct bucket *a = _a, *b = _b;
+
+	/* Reverse sort. */
+	if (a->count > b->count)
+		return -1;
+
+	return 1;
+}
+
+/*
+ * TODO:
+ * Support other iter types, if required.
+ * Only ITER_XARRAY is supported for now.
+ */
+static int collect_sample(const struct iov_iter *iter, ssize_t max, u8 *sample)
+{
+	struct folio *folios[16], *folio;
+	unsigned int nr, i, j, npages;
+	loff_t start = iter->xarray_start + iter->iov_offset;
+	pgoff_t last, index = start / PAGE_SIZE;
+	size_t len, off, foff;
+	ssize_t ret = 0;
+	void *p;
+	int s = 0;
+
+	last = (start + max - 1) / PAGE_SIZE;
+	do {
+		nr = xa_extract(iter->xarray, (void **)folios, index, last, ARRAY_SIZE(folios),
+				XA_PRESENT);
+		if (nr == 0)
+			return -EIO;
+
+		for (i = 0; i < nr; i++) {
+			folio = folios[i];
+			npages = folio_nr_pages(folio);
+			foff = start - folio_pos(folio);
+			off = foff % PAGE_SIZE;
+
+			for (j = foff / PAGE_SIZE; j < npages; j++) {
+				size_t len2;
+
+				len = min_t(size_t, max, PAGE_SIZE - off);
+				len2 = min_t(size_t, len, SZ_2K);
+
+				p = kmap_local_page(folio_page(folio, j));
+				memcpy(&sample[s], p, len2);
+				kunmap_local(p);
+
+				if (ret < 0)
+					return ret;
+
+				s += len2;
+
+				if (len2 < SZ_2K || s >= max - SZ_2K)
+					return s;
+
+				max -= len;
+				if (max <= 0)
+					return s;
+
+				start += len;
+				off = 0;
+				index++;
+			}
+		}
+	} while (nr == ARRAY_SIZE(folios));
+
+	return s;
+}
+
+/**
+ * is_compressible() - Determines if a chunk of data is compressible.
+ * @data: Iterator containing uncompressed data.
+ *
+ * Return:
+ * 0:		@data is not compressible
+ * 1:		@data is compressible
+ * -ENOMEM:	failed to allocate memory for sample buffer
+ *
+ * Tests shows that this function is quite reliable in predicting data compressibility,
+ * matching close to 1:1 with the behaviour of LZ77 compression success and failures.
+ */
+static int is_compressible(const struct iov_iter *data)
+{
+	const size_t read_size = SZ_2K, bkt_size = 256, max = SZ_4M;
+	struct bucket *bkt;
+	int i = 0, ret = 0;
+	size_t len;
+	u8 *sample;
+
+	len = iov_iter_count(data);
+	if (len < read_size)
+		return 0;
+
+	if (len - read_size > max)
+		len = max;
+
+	sample = kvzalloc(len, GFP_KERNEL);
+	if (!sample)
+		return -ENOMEM;
+
+	/* Sample 2K bytes per page of the uncompressed data. */
+	ret = collect_sample(data, len, sample);
+	if (ret < 0)
+		goto out;
+
+	len = ret;
+	ret = 1;
+
+	if (check_repeated_data(sample, len))
+		goto out;
+
+	bkt = kcalloc(bkt_size, sizeof(*bkt), GFP_KERNEL);
+	if (!bkt) {
+		kvfree(sample);
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < len; i++)
+		bkt[sample[i]].count++;
+
+	if (check_ascii_bytes(bkt))
+		goto out;
+
+	/* Sort in descending order */
+	sort(bkt, bkt_size, sizeof(*bkt), cmp_bkt, NULL);
+
+	ret = calc_byte_distribution(bkt, len);
+	if (ret != 2)
+		goto out;
+
+	ret = calc_shannon_entropy(bkt, len);
+out:
+	kvfree(sample);
+	kfree(bkt);
+
+	WARN(ret < 0, "%s: ret=%d\n", __func__, ret);
+
+	return !!ret;
+}
+
+bool should_compress(const struct cifs_tcon *tcon, const struct smb_rqst *rq)
+{
+	const struct smb2_hdr *shdr = rq->rq_iov->iov_base;
+
+	if (unlikely(!tcon || !tcon->ses || !tcon->ses->server))
+		return false;
+
+	if (!tcon->ses->server->compression.enabled)
+		return false;
+
+	if (!(tcon->share_flags & SMB2_SHAREFLAG_COMPRESS_DATA))
+		return false;
+
+	if (shdr->Command == SMB2_WRITE) {
+		const struct smb2_write_req *wreq = rq->rq_iov->iov_base;
+
+		if (wreq->Length < SMB_COMPRESS_MIN_LEN)
+			return false;
+
+		return is_compressible(&rq->rq_iter);
+	}
+
+	return (shdr->Command == SMB2_READ);
+}
+
+int smb_compress(struct TCP_Server_Info *server, struct smb_rqst *rq, compress_send_fn send_fn)
+{
+	struct iov_iter iter;
+	u32 slen, dlen;
+	void *src, *dst;
 	int ret;
 
-	buf_len = sizeof(struct smb2_write_req);
-	data_len = *len;
-	*len = 0;
-
-	hdr = buf;
-	hdr->ProtocolId = SMB2_COMPRESSION_TRANSFORM_ID;
-	hdr->OriginalCompressedSegmentSize = cpu_to_le32(data_len);
-	hdr->Offset = cpu_to_le32(buf_len);
-	hdr->Flags = SMB2_COMPRESSION_FLAG_NONE;
-	hdr->CompressionAlgorithm = SMB3_COMPRESS_LZ77;
-
-	/* XXX: add other algs here as they're implemented */
-	ret = lz77_compress(data, data_len, buf + SMB_COMPRESS_HDR_LEN + buf_len, &data_len);
-	if (!ret)
-		*len = SMB_COMPRESS_HDR_LEN + buf_len + data_len;
+	if (!server || !rq || !rq->rq_iov || !rq->rq_iov->iov_base)
+		return -EINVAL;
+
+	if (rq->rq_iov->iov_len != sizeof(struct smb2_write_req))
+		return -EINVAL;
+
+	slen = iov_iter_count(&rq->rq_iter);
+	src = kvzalloc(slen, GFP_KERNEL);
+	if (!src) {
+		ret = -ENOMEM;
+		goto err_free;
+	}
+
+	/* Keep the original iter intact. */
+	iter = rq->rq_iter;
+
+	if (!copy_from_iter_full(src, slen, &iter)) {
+		ret = -EIO;
+		goto err_free;
+	}
+
+	/*
+	 * This is just overprovisioning, as the algorithm will error out if @dst reaches 7/8
+	 * of @slen.
+	 */
+	dlen = slen;
+	dst = kvzalloc(dlen, GFP_KERNEL);
+	if (!dst) {
+		ret = -ENOMEM;
+		goto err_free;
+	}
+
+	ret = lz77_compress(src, slen, dst, &dlen);
+	if (!ret) {
+		struct smb2_compression_hdr hdr = { 0 };
+		struct smb_rqst comp_rq = { .rq_nvec = 3, };
+		struct kvec iov[3];
+
+		hdr.ProtocolId = SMB2_COMPRESSION_TRANSFORM_ID;
+		hdr.OriginalCompressedSegmentSize = cpu_to_le32(slen);
+		hdr.CompressionAlgorithm = SMB3_COMPRESS_LZ77;
+		hdr.Flags = SMB2_COMPRESSION_FLAG_NONE;
+		hdr.Offset = cpu_to_le32(rq->rq_iov[0].iov_len);
+
+		iov[0].iov_base = &hdr;
+		iov[0].iov_len = sizeof(hdr);
+		iov[1] = rq->rq_iov[0];
+		iov[2].iov_base = dst;
+		iov[2].iov_len = dlen;
+
+		comp_rq.rq_iov = iov;
+
+		ret = send_fn(server, 1, &comp_rq);
+	} else if (ret == -EMSGSIZE || dlen >= slen) {
+		ret = send_fn(server, 1, rq);
+	}
+err_free:
+	kvfree(dst);
+	kvfree(src);
 
 	return ret;
 }