ooni-probe-cli/internal/measurex/measurer.go

1062 lines
33 KiB
Go
Raw Normal View History

package measurex
//
// Measurer
//
// High-level API for running measurements. The code in here
// has been designed to easily implement the new websteps
// network experiment, which is quite complex. It should be
// possible to write most other experiments using a Measurer.
//
import (
"context"
"crypto/tls"
"errors"
"net"
"net/http"
"net/url"
"strings"
"time"
"github.com/apex/log"
"github.com/lucas-clemente/quic-go"
"github.com/ooni/probe-cli/v3/internal/model"
"github.com/ooni/probe-cli/v3/internal/netxlite"
)
// Measurer performs measurements. If you don't use a factory
// for creating this type, make sure you set all the MANDATORY fields.
type Measurer struct {
// Begin is when we started measuring (this field is MANDATORY).
Begin time.Time
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DNSLookupTimeout is the OPTIONAL timeout for performing
// a DNS lookup. If not set, we use a default value.
//
// Note that the underlying network implementation MAY use a
// shorter-than-you-selected watchdog timeout. In such a case,
// the shorter watchdog timeout will prevail.
DNSLookupTimeout time.Duration
// HTTPClient is the MANDATORY HTTP client for the WCTH.
HTTPClient model.HTTPClient
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// HTTPMaxBodySnapshotSize is the OPTIONAL maximum size,
// in bytes, of the response body snapshot we save. If this field
// is zero or negative, we'll use a small default value.
HTTPMaxBodySnapshotSize int64
// HTTPRoundTripTimeout is the OPTIONAL timeout for performing
// an HTTP round trip. If not set, we use a default value.
//
// Note that the underlying network implementation MAY use a
// shorter-than-you-selected watchdog timeout. In such a case,
// the shorter watchdog timeout will prevail.
HTTPRoundTripTimeout time.Duration
// Logger is the MANDATORY logger to use.
Logger model.Logger
// MeasureURLHelper is the OPTIONAL test helper to use when
// we're measuring using the MeasureURL function. If this field
// is not set, we'll not be using any helper.
MeasureURLHelper MeasureURLHelper
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// QUICHandshakeTimeout is the OPTIONAL timeout for performing
// a QUIC handshake. If not set, we use a default value.
//
// Note that the underlying network implementation MAY use a
// shorter-than-you-selected watchdog timeout. In such a case,
// the shorter watchdog timeout will prevail.
QUICHandshakeTimeout time.Duration
// Resolvers is the MANDATORY list of resolvers.
Resolvers []*ResolverInfo
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// TCPConnectTimeout is the OPTIONAL timeout for performing
// a tcp connect. If not set, we use a default value.
//
// Note that the underlying network implementation MAY use a
// shorter-than-you-selected watchdog timeout. In such a case,
// the shorter watchdog timeout will prevail.
TCPconnectTimeout time.Duration
// TLSHandshakeTimeout is the OPTIONAL timeout for performing
// a tls handshake. If not set, we use a default value.
//
// Note that the underlying network implementation MAY use a
// shorter-than-you-selected watchdog timeout. In such a case,
// the shorter watchdog timeout will prevail.
TLSHandshakeTimeout time.Duration
// TLSHandshaker is the MANDATORY TLS handshaker.
TLSHandshaker model.TLSHandshaker
}
// NewMeasurerWithDefaultSettings creates a new Measurer
// instance using the most default settings.
func NewMeasurerWithDefaultSettings() *Measurer {
return &Measurer{
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
Begin: time.Now(),
DNSLookupTimeout: 0,
HTTPClient: &http.Client{},
HTTPMaxBodySnapshotSize: 0,
HTTPRoundTripTimeout: 0,
Logger: log.Log,
MeasureURLHelper: nil,
QUICHandshakeTimeout: 0,
Resolvers: []*ResolverInfo{{
Network: "system",
Address: "",
}, {
Network: "udp",
Address: "8.8.4.4:53",
}},
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
TCPconnectTimeout: 0,
TLSHandshakeTimeout: 0,
TLSHandshaker: netxlite.NewTLSHandshakerStdlib(log.Log),
}
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DefaultDNSLookupTimeout is the default DNS lookup timeout.
const DefaultDNSLookupTimeout = 4 * time.Second
// dnsLookupTimeout selects the correct DNS lookup timeout.
func (mx *Measurer) dnsLookupTimeout() time.Duration {
if mx.DNSLookupTimeout > 0 {
return mx.DNSLookupTimeout
}
return DefaultDNSLookupTimeout
}
// LookupHostSystem performs a LookupHost using the system resolver.
func (mx *Measurer) LookupHostSystem(ctx context.Context, domain string) *DNSMeasurement {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.dnsLookupTimeout()
ol := NewOperationLogger(mx.Logger, "LookupHost %s with getaddrinfo", domain)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
db := &MeasurementDB{}
r := mx.NewResolverSystem(db, mx.Logger)
defer r.CloseIdleConnections()
_, err := r.LookupHost(ctx, domain)
ol.Stop(err)
return &DNSMeasurement{
Domain: domain,
Measurement: db.AsMeasurement(),
}
}
// lookupHostForeign performs a LookupHost using a "foreign" resolver.
func (mx *Measurer) lookupHostForeign(
ctx context.Context, domain string, r model.Resolver) *DNSMeasurement {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.dnsLookupTimeout()
ol := NewOperationLogger(mx.Logger, "LookupHost %s with %s", domain, r.Network())
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
db := &MeasurementDB{}
_, err := mx.WrapResolver(db, r).LookupHost(ctx, domain)
ol.Stop(err)
return &DNSMeasurement{
Domain: domain,
Measurement: db.AsMeasurement(),
}
}
// LookupHostUDP is like LookupHostSystem but uses an UDP resolver.
//
// Arguments:
//
// - ctx is the context allowing to timeout the operation;
//
// - domain is the domain to resolve (e.g., "x.org");
//
// - address is the UDP resolver address (e.g., "dns.google:53").
//
// Returns a DNSMeasurement.
func (mx *Measurer) LookupHostUDP(
ctx context.Context, domain, address string) *DNSMeasurement {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.dnsLookupTimeout()
ol := NewOperationLogger(mx.Logger, "LookupHost %s with %s/udp", domain, address)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
db := &MeasurementDB{}
r := mx.NewResolverUDP(db, mx.Logger, address)
defer r.CloseIdleConnections()
_, err := r.LookupHost(ctx, domain)
ol.Stop(err)
return &DNSMeasurement{
Domain: domain,
Measurement: db.AsMeasurement(),
}
}
// LookupHTTPSSvcUDP issues an HTTPSSvc query for the given domain.
//
// Arguments:
//
// - ctx is the context allowing to timeout the operation;
//
// - domain is the domain to resolve (e.g., "x.org");
//
// - address is the UDP resolver address (e.g., "dns.google:53").
//
// Returns a DNSMeasurement.
func (mx *Measurer) LookupHTTPSSvcUDP(
ctx context.Context, domain, address string) *DNSMeasurement {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.dnsLookupTimeout()
ol := NewOperationLogger(mx.Logger, "LookupHTTPSvc %s with %s/udp", domain, address)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
db := &MeasurementDB{}
r := mx.NewResolverUDP(db, mx.Logger, address)
defer r.CloseIdleConnections()
_, err := r.LookupHTTPS(ctx, domain)
ol.Stop(err)
return &DNSMeasurement{
Domain: domain,
Measurement: db.AsMeasurement(),
}
}
// lookupHTTPSSvcUDPForeign is like LookupHTTPSSvcUDP
// except that it uses a "foreign" resolver.
func (mx *Measurer) lookupHTTPSSvcUDPForeign(
ctx context.Context, domain string, r model.Resolver) *DNSMeasurement {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.dnsLookupTimeout()
ol := NewOperationLogger(mx.Logger, "LookupHTTPSvc %s with %s", domain, r.Address())
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
db := &MeasurementDB{}
_, err := mx.WrapResolver(db, r).LookupHTTPS(ctx, domain)
ol.Stop(err)
return &DNSMeasurement{
Domain: domain,
Measurement: db.AsMeasurement(),
}
}
// TCPConnect establishes a connection with a TCP endpoint.
//
// Arguments:
//
// - ctx is the context allowing to timeout the connect;
//
// - address is the TCP endpoint address (e.g., "8.8.4.4:443").
//
// Returns an EndpointMeasurement.
func (mx *Measurer) TCPConnect(ctx context.Context, address string) *EndpointMeasurement {
db := &MeasurementDB{}
conn, _ := mx.TCPConnectWithDB(ctx, db, address)
measurement := db.AsMeasurement()
if conn != nil {
conn.Close()
}
return &EndpointMeasurement{
Network: NetworkTCP,
Address: address,
Measurement: measurement,
}
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DefaultTCPConnectTimeout is the default TCP connect timeout.
const DefaultTCPConnectTimeout = 15 * time.Second
// tcpConnectTimeout selects the correct TCP connect timeout.
func (mx *Measurer) tcpConnectTimeout() time.Duration {
if mx.TCPconnectTimeout > 0 {
return mx.TCPconnectTimeout
}
return DefaultTCPConnectTimeout
}
// TCPConnectWithDB is like TCPConnect but does not create a new measurement,
// rather it just stores the events inside of the given DB.
func (mx *Measurer) TCPConnectWithDB(ctx context.Context, db WritableDB, address string) (Conn, error) {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.tcpConnectTimeout()
ol := NewOperationLogger(mx.Logger, "TCPConnect %s", address)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
d := mx.NewDialerWithoutResolver(db, mx.Logger)
defer d.CloseIdleConnections()
conn, err := d.DialContext(ctx, "tcp", address)
ol.Stop(err)
return conn, err
}
// TLSConnectAndHandshake connects and TLS handshakes with a TCP endpoint.
//
// Arguments:
//
// - ctx is the context allowing to timeout the whole operation;
//
// - address is the endpoint address (e.g., "1.1.1.1:443");
//
// - config contains the TLS config (see below).
//
// You MUST set the following config fields:
//
// - ServerName to the desired SNI or InsecureSkipVerify to
// skip the certificate name verification;
//
// - RootCAs to nextlite.NewDefaultCertPool() output;
//
// - NextProtos to the desired ALPN ([]string{"h2", "http/1.1"} for
// HTTPS and []string{"dot"} for DNS-over-TLS).
//
// Caveats:
//
// The mx.TLSHandshaker field could point to a TLS handshaker using
// the Go stdlib or one using gitlab.com/yawning/utls.git.
//
// In the latter case, the content of the ClientHello message
// will not only depend on the config field but also on the
// utls.ClientHelloID thay you're using.
//
// Returns an EndpointMeasurement.
func (mx *Measurer) TLSConnectAndHandshake(ctx context.Context,
address string, config *tls.Config) *EndpointMeasurement {
db := &MeasurementDB{}
conn, _ := mx.TLSConnectAndHandshakeWithDB(ctx, db, address, config)
measurement := db.AsMeasurement()
if conn != nil {
conn.Close()
}
return &EndpointMeasurement{
Network: NetworkTCP,
Address: address,
Measurement: measurement,
}
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DefaultTLSHandshakeTimeout is the default TLS handshake timeout.
const DefaultTLSHandshakeTimeout = 10 * time.Second
// tlsHandshakeTimeout selects the correct TLS handshake timeout.
func (mx *Measurer) tlsHandshakeTimeout() time.Duration {
if mx.TLSHandshakeTimeout > 0 {
return mx.TLSHandshakeTimeout
}
return DefaultTLSHandshakeTimeout
}
// TLSConnectAndHandshakeWithDB is like TLSConnectAndHandshake but
// uses the given DB instead of creating a new Measurement.
func (mx *Measurer) TLSConnectAndHandshakeWithDB(ctx context.Context,
db WritableDB, address string, config *tls.Config) (netxlite.TLSConn, error) {
conn, err := mx.TCPConnectWithDB(ctx, db, address)
if err != nil {
return nil, err
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.tlsHandshakeTimeout()
ol := NewOperationLogger(mx.Logger,
"TLSHandshake %s with sni=%s", address, config.ServerName)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
th := mx.WrapTLSHandshaker(db, mx.TLSHandshaker)
tlsConn, _, err := th.Handshake(ctx, conn, config)
ol.Stop(err)
if err != nil {
return nil, err
}
// cast safe according to the docs of netxlite's handshaker
return tlsConn.(netxlite.TLSConn), nil
}
// QUICHandshake connects and TLS handshakes with a QUIC endpoint.
//
// Arguments:
//
// - ctx is the context allowing to timeout the whole operation;
//
// - address is the endpoint address (e.g., "1.1.1.1:443");
//
// - config contains the TLS config (see below).
//
// You MUST set the following config fields:
//
// - ServerName to the desired SNI or InsecureSkipVerify to
// skip the certificate name verification;
//
// - RootCAs to nextlite.NewDefaultCertPool() output;
//
// - NextProtos to the desired ALPN ([]string{"h2", "http/1.1"} for
// HTTPS and []string{"dot"} for DNS-over-TLS).
//
// Returns an EndpointMeasurement.
func (mx *Measurer) QUICHandshake(ctx context.Context, address string,
config *tls.Config) *EndpointMeasurement {
db := &MeasurementDB{}
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
qconn, _ := mx.QUICHandshakeWithDB(ctx, db, address, config)
measurement := db.AsMeasurement()
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
if qconn != nil {
// TODO(bassosimone): close connection with correct message
qconn.CloseWithError(0, "")
}
return &EndpointMeasurement{
Network: NetworkUDP,
Address: address,
Measurement: measurement,
}
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DefaultQUICHandshakeTimeout is the default QUIC handshake timeout.
const DefaultQUICHandshakeTimeout = 10 * time.Second
// quicHandshakeTimeout selects the correct QUIC handshake timeout.
func (mx *Measurer) quicHandshakeTimeout() time.Duration {
if mx.QUICHandshakeTimeout > 0 {
return mx.QUICHandshakeTimeout
}
return DefaultQUICHandshakeTimeout
}
// QUICHandshakeWithDB is like QUICHandshake but uses the given
// db to store events rather than creating a temporary one and
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
// use it to generate a new Measurement.
func (mx *Measurer) QUICHandshakeWithDB(ctx context.Context, db WritableDB,
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
address string, config *tls.Config) (quic.EarlyConnection, error) {
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.quicHandshakeTimeout()
ol := NewOperationLogger(mx.Logger,
"QUICHandshake %s with sni=%s", address, config.ServerName)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
qd := mx.NewQUICDialerWithoutResolver(db, mx.Logger)
defer qd.CloseIdleConnections()
qconn, err := qd.DialContext(ctx, address, config, &quic.Config{})
ol.Stop(err)
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
return qconn, err
}
// HTTPEndpointGet performs a GET request for an HTTP endpoint.
//
// This function WILL NOT follow redirects. If there is a redirect
// you will see it inside the specific database table.
//
// Arguments:
//
// - ctx is the context allowing to timeout the operation;
//
// - epnt is the HTTP endpoint;
//
// - jar is the cookie jar to use.
//
// Returns a measurement. The returned measurement is empty if
// the endpoint is misconfigured or the URL has an unknown scheme.
func (mx *Measurer) HTTPEndpointGet(
ctx context.Context, epnt *HTTPEndpoint, jar http.CookieJar) *HTTPEndpointMeasurement {
resp, m, _ := mx.httpEndpointGet(ctx, epnt, jar)
if resp != nil {
resp.Body.Close()
}
return m
}
// HTTPEndpointGetWithoutCookies is like HTTPEndpointGet
// but does not require you to provide a CookieJar.
func (mx *Measurer) HTTPEndpointGetWithoutCookies(
ctx context.Context, epnt *HTTPEndpoint) *HTTPEndpointMeasurement {
return mx.HTTPEndpointGet(ctx, epnt, NewCookieJar())
}
var (
errUnknownHTTPEndpointURLScheme = errors.New("unknown HTTPEndpoint.URL.Scheme")
// ErrUnknownHTTPEndpointNetwork means that the given endpoint's
// network is of a type that we don't know how to handle.
ErrUnknownHTTPEndpointNetwork = errors.New("unknown HTTPEndpoint.Network")
)
// httpEndpointGet implements HTTPEndpointGet.
func (mx *Measurer) httpEndpointGet(ctx context.Context, epnt *HTTPEndpoint,
jar http.CookieJar) (*http.Response, *HTTPEndpointMeasurement, error) {
resp, m, err := mx.httpEndpointGetMeasurement(ctx, epnt, jar)
out := &HTTPEndpointMeasurement{
URL: epnt.URL.String(),
Network: epnt.Network,
Address: epnt.Address,
Measurement: m,
}
return resp, out, err
}
// httpEndpointGetMeasurement implements httpEndpointGet.
//
// This function returns a triple where:
//
// - the first element is a valid response on success a nil response on failure
//
// - the second element is always a valid Measurement
//
// - the third element is a nil error on success and an error on failure
func (mx *Measurer) httpEndpointGetMeasurement(ctx context.Context, epnt *HTTPEndpoint,
jar http.CookieJar) (resp *http.Response, m *Measurement, err error) {
db := &MeasurementDB{}
resp, err = mx.httpEndpointGetWithDB(ctx, epnt, db, jar)
m = db.AsMeasurement()
return
}
// HTTPEndpointGetWithDB is an HTTPEndpointGet that stores the
// events into the given WritableDB.
func (mx *Measurer) HTTPEndpointGetWithDB(ctx context.Context, epnt *HTTPEndpoint,
db WritableDB, jar http.CookieJar) (err error) {
switch epnt.Network {
case NetworkUDP:
_, err = mx.httpEndpointGetQUIC(ctx, db, epnt, jar)
case NetworkTCP:
_, err = mx.httpEndpointGetTCP(ctx, db, epnt, jar)
default:
err = ErrUnknownHTTPEndpointNetwork
}
return
}
// httpEndpointGetWithDB is an HTTPEndpointGet that stores the
// events into the given WritableDB.
func (mx *Measurer) httpEndpointGetWithDB(ctx context.Context, epnt *HTTPEndpoint,
db WritableDB, jar http.CookieJar) (resp *http.Response, err error) {
switch epnt.Network {
case NetworkUDP:
resp, err = mx.httpEndpointGetQUIC(ctx, db, epnt, jar)
case NetworkTCP:
resp, err = mx.httpEndpointGetTCP(ctx, db, epnt, jar)
default:
err = ErrUnknownHTTPEndpointNetwork
}
return
}
// httpEndpointGetTCP specializes HTTPSEndpointGet for HTTP and HTTPS.
func (mx *Measurer) httpEndpointGetTCP(ctx context.Context,
db WritableDB, epnt *HTTPEndpoint, jar http.CookieJar) (*http.Response, error) {
switch epnt.URL.Scheme {
case "http":
return mx.httpEndpointGetHTTP(ctx, db, epnt, jar)
case "https":
return mx.httpEndpointGetHTTPS(ctx, db, epnt, jar)
default:
return nil, errUnknownHTTPEndpointURLScheme
}
}
// httpEndpointGetHTTP specializes httpEndpointGetTCP for HTTP.
func (mx *Measurer) httpEndpointGetHTTP(ctx context.Context,
db WritableDB, epnt *HTTPEndpoint, jar http.CookieJar) (*http.Response, error) {
conn, err := mx.TCPConnectWithDB(ctx, db, epnt.Address)
if err != nil {
return nil, err
}
defer conn.Close() // we own it
clnt := NewHTTPClientWithoutRedirects(db, jar,
mx.NewHTTPTransportWithConn(mx.Logger, db, conn))
defer clnt.CloseIdleConnections()
return mx.httpClientDo(ctx, clnt, epnt)
}
// httpEndpointGetHTTPS specializes httpEndpointGetTCP for HTTPS.
func (mx *Measurer) httpEndpointGetHTTPS(ctx context.Context,
db WritableDB, epnt *HTTPEndpoint, jar http.CookieJar) (*http.Response, error) {
conn, err := mx.TLSConnectAndHandshakeWithDB(ctx, db, epnt.Address, &tls.Config{
ServerName: epnt.SNI,
NextProtos: epnt.ALPN,
RootCAs: netxlite.NewDefaultCertPool(),
})
if err != nil {
return nil, err
}
defer conn.Close() // we own it
clnt := NewHTTPClientWithoutRedirects(db, jar,
mx.NewHTTPTransportWithTLSConn(mx.Logger, db, conn))
defer clnt.CloseIdleConnections()
return mx.httpClientDo(ctx, clnt, epnt)
}
// httpEndpointGetQUIC specializes httpEndpointGetTCP for QUIC.
func (mx *Measurer) httpEndpointGetQUIC(ctx context.Context,
db WritableDB, epnt *HTTPEndpoint, jar http.CookieJar) (*http.Response, error) {
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
qconn, err := mx.QUICHandshakeWithDB(ctx, db, epnt.Address, &tls.Config{
ServerName: epnt.SNI,
NextProtos: epnt.ALPN,
RootCAs: netxlite.NewDefaultCertPool(),
})
if err != nil {
return nil, err
}
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
// TODO(bassosimone): close connection with correct message
defer qconn.CloseWithError(0, "") // we own it
clnt := NewHTTPClientWithoutRedirects(db, jar,
cli: upgrade to lucas-clemente/quic-go@v0.27.0 (#715) * quic-go upgrade: replaced Session/EarlySession with Connection/EarlyConnection * quic-go upgrade: added context to RoundTripper.Dial * quic-go upgrade: made corresponding changes to tutorial * quic-go upgrade: changed sess variable instances to qconn * quic-go upgrade: made corresponding changes to tutorial * cleanup: remove unnecessary comments Those comments made sense in terms of illustrating the changes but they're going to be less useful once we merge. * fix(go.mod): apparently we needed `go1.18.1 mod tidy` VSCode just warned me about this. It seems fine to apply this change as part of the pull request at hand. * cleanup(netxlite): http3dialer can be removed We used to use http3dialer to glue a QUIC dialer, which had a context as its first argument, to the Dial function used by the HTTP3 transport, which did not have a context as its first argument. Now that HTTP3 transport has a Dial function taking a context as its first argument, we don't need http3dialer anymore, since we can use the QUIC dialer directly. Cc: @DecFox * Revert "cleanup(netxlite): http3dialer can be removed" This reverts commit c62244c620cee5fadcc2ca89d8228c8db0b96add to investigate the build failure mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * chore(netx): show that test was already broken We didn't see the breakage before because we were not using the created transport, but the issue of using a nil dialer was already present before, we just didn't see it. Now we understand why removing the http3transport in c62244c620cee5fadcc2ca89d8228c8db0b96add did cause the breakage mentioned at https://github.com/ooni/probe-cli/pull/715#issuecomment-1119450484 * fix(netx): convert broken integration test to working unit test There's no point in using the network here. Add a fake dialer that breaks and ensure we're getting the expected error. We've now improved upon the original test because the original test was not doing anything while now we're testing whether we get back a QUIC dialer that _can be used_. After this commit, I can then readd the cleanup commit c62244c620cee5fadcc2ca89d8228c8db0b96add and it won't be broken anymore (at least, this is what I expected to happen). * Revert "Revert "cleanup(netxlite): http3dialer can be removed"" This reverts commit 0e254bfc6ba3bfd65365ce3d8de2c8ec51b925ff because now we should have fixed the broken test. Co-authored-by: decfox <decfox> Co-authored-by: Simone Basso <bassosimone@gmail.com>
2022-05-06 12:24:03 +02:00
mx.NewHTTPTransportWithQUICConn(mx.Logger, db, qconn))
defer clnt.CloseIdleConnections()
return mx.httpClientDo(ctx, clnt, epnt)
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// HTTPClientGET performs a GET operation of the given URL
// using the given HTTP client instance.
func (mx *Measurer) HTTPClientGET(
ctx context.Context, clnt model.HTTPClient, URL *url.URL) (*http.Response, error) {
return mx.httpClientDo(ctx, clnt, &HTTPEndpoint{
Domain: URL.Hostname(),
Network: "tcp",
Address: URL.Hostname(),
SNI: "", // not needed
ALPN: []string{}, // not needed
URL: URL,
Header: NewHTTPRequestHeaderForMeasuring(),
})
}
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
// DefaultHTTPRoundTripTimeout is the default HTTP round-trip timeout.
const DefaultHTTPRoundTripTimeout = 15 * time.Second
// httpRoundTripTimeout selects the correct HTTP round-trip timeout.
func (mx *Measurer) httpRoundTripTimeout() time.Duration {
if mx.HTTPRoundTripTimeout > 0 {
return mx.HTTPRoundTripTimeout
}
return DefaultHTTPRoundTripTimeout
}
func (mx *Measurer) httpClientDo(ctx context.Context,
clnt model.HTTPClient, epnt *HTTPEndpoint) (*http.Response, error) {
req, err := NewHTTPGetRequest(ctx, epnt.URL.String())
if err != nil {
return nil, err
}
req.Header = epnt.Header.Clone() // must clone because of parallel usage
refactor(measurex): allow to configure timeouts and max-snapshot-size (#645) This diff lightly refactors the code in measurex to allow a user to configure all possible timeouts and the max-snapshot-size. There is currently a little bit of tension between setting timeouts inside of measurex and the watchdog timeouts inside of netxlite. This tension has been documented. Let us repeat the issue also in this commit message. If you are using a masurex.Measurer configured with very large timeouts and the underlying netxlite implementation uses shorter whatchdog timeouts, then you are going to see shorter than expected timeouts. Ideally, we would like to have just a single timeout but there is no way to ask the context "hey, can you tell me if you already have a configured timeout?". It may be that the right solution is to modify netxlite to have some sort of root/library object with this configuration. If that's the case, then a Measurer could be refactored as follows: - create the underlying netxlite "library" - initialize the timeouts desired by the Measurer - create a Dialer, of whatever is needed - use it Now this is not possible because netxlite timeouts are internal static settings rather than attributes of a structure. Anyway, for now I'm happy with this just being documented. (I suspect this issue will need to be addresses when we'll write unit tests for measurex; at that time a proper solution should come out naturally due to the unit tests constraints.) I'm working on this refactoring, BTW, to facilitate rewriting `tor` using measurex (see https://github.com/ooni/probe/issues/1688).
2022-01-04 13:20:48 +01:00
timeout := mx.httpRoundTripTimeout()
ol := NewOperationLogger(mx.Logger,
"%s %s with %s/%s", req.Method, req.URL.String(), epnt.Address, epnt.Network)
ctx, cancel := context.WithTimeout(ctx, timeout)
defer cancel()
resp, err := clnt.Do(req.WithContext(ctx))
ol.Stop(err)
return resp, err
}
// HTTPEndpointGetParallel performs an HTTPEndpointGet for each
// input endpoint using a pool of background goroutines.
//
// You can choose the parallelism with the parallelism argument. If this
// argument is zero, or negative, we use a small default value.
//
// This function returns to the caller a channel where to read
// measurements from. The channel is closed when done.
func (mx *Measurer) HTTPEndpointGetParallel(ctx context.Context, parallelism int,
jar http.CookieJar, epnts ...*HTTPEndpoint) <-chan *HTTPEndpointMeasurement {
var (
done = make(chan interface{})
input = make(chan *HTTPEndpoint)
output = make(chan *HTTPEndpointMeasurement)
)
go func() {
defer close(input)
for _, epnt := range epnts {
input <- epnt
}
}()
if parallelism <= 0 {
parallelism = 3
}
for i := 0; i < parallelism; i++ {
go func() {
for epnt := range input {
output <- mx.HTTPEndpointGet(ctx, epnt, jar)
}
done <- true
}()
}
go func() {
for i := 0; i < parallelism; i++ {
<-done
}
close(output)
}()
return output
}
// ResolverNetwork identifies the network of a resolver.
type ResolverNetwork string
var (
// ResolverSystem is the system resolver (i.e., getaddrinfo)
ResolverSystem = ResolverNetwork("system")
// ResolverUDP is a resolver using DNS-over-UDP
ResolverUDP = ResolverNetwork("udp")
// ResolverForeign is a resolver that is not managed by
// this package. We can wrap it, but we don't be able to
// observe any event but Lookup{Host,HTTPSvc}
ResolverForeign = ResolverNetwork("foreign")
)
// ResolverInfo contains info about a DNS resolver.
type ResolverInfo struct {
// Network is the resolver's network (e.g., "doh", "udp")
Network ResolverNetwork
// Address is the address (e.g., "1.1.1.1:53", "https://1.1.1.1/dns-query")
Address string
// ForeignResolver is only used when Network's
// value equals the ResolverForeign constant.
ForeignResolver model.Resolver
}
// LookupURLHostParallel performs an LookupHost-like operation for each
// resolver that you provide as argument using a pool of goroutines.
//
// You can choose the parallelism with the parallelism argument. If this
// argument is zero, or negative, we use a small default value.
func (mx *Measurer) LookupURLHostParallel(ctx context.Context, parallelism int,
URL *url.URL, resos ...*ResolverInfo) <-chan *DNSMeasurement {
var (
done = make(chan interface{})
resolvers = make(chan *ResolverInfo)
output = make(chan *DNSMeasurement)
)
go func() {
defer close(resolvers)
for _, reso := range resos {
resolvers <- reso
}
}()
if parallelism <= 0 {
parallelism = 3
}
for i := 0; i < parallelism; i++ {
go func() {
for reso := range resolvers {
mx.lookupHostWithResolverInfo(ctx, reso, URL, output)
}
done <- true
}()
}
go func() {
for i := 0; i < parallelism; i++ {
<-done
}
close(output)
}()
return output
}
// lookupHostWithResolverInfo performs a LookupHost-like
// operation using the given ResolverInfo.
func (mx *Measurer) lookupHostWithResolverInfo(
ctx context.Context, reso *ResolverInfo, URL *url.URL,
output chan<- *DNSMeasurement) {
switch reso.Network {
case ResolverSystem:
output <- mx.LookupHostSystem(ctx, URL.Hostname())
case ResolverUDP:
output <- mx.LookupHostUDP(ctx, URL.Hostname(), reso.Address)
case ResolverForeign:
output <- mx.lookupHostForeign(ctx, URL.Hostname(), reso.ForeignResolver)
default:
return
}
switch URL.Scheme {
case "https":
default:
return
}
switch reso.Network {
case ResolverUDP:
output <- mx.LookupHTTPSSvcUDP(ctx, URL.Hostname(), reso.Address)
case ResolverForeign:
output <- mx.lookupHTTPSSvcUDPForeign(ctx, URL.Hostname(), reso.ForeignResolver)
}
}
// LookupHostParallel is like LookupURLHostParallel but we only
// have in input an hostname rather than a URL. As such, we cannot
// determine whether to perform HTTPSSvc lookups and so we aren't
// going to perform this kind of lookups in this case.
//
// You can choose the parallelism with the parallelism argument. If this
// argument is zero, or negative, we use a small default value.
func (mx *Measurer) LookupHostParallel(ctx context.Context,
parallelism int, hostname, port string) <-chan *DNSMeasurement {
out := make(chan *DNSMeasurement)
go func() {
defer close(out)
URL := &url.URL{
Scheme: "", // so we don't see https and we don't try HTTPSSvc
Host: net.JoinHostPort(hostname, port),
}
for m := range mx.LookupURLHostParallel(ctx, parallelism, URL) {
out <- &DNSMeasurement{Domain: hostname, Measurement: m.Measurement}
}
}()
return out
}
// MeasureURLHelper is a Test Helper that discovers additional
// endpoints after MeasureURL has finished discovering endpoints
// via the usual DNS mechanism. The MeasureURLHelper:
//
// - is used by experiments to call a real test helper, i.e.,
// a remote service providing extra endpoints
//
// - is used by test helpers to augment the set of endpoints
// discovered so far with the ones provided by a client.
type MeasureURLHelper interface {
// LookupExtraHTTPEndpoints searches for extra HTTP endpoints
// suitable for the given URL we're measuring.
//
// Arguments:
//
// - ctx is the context for timeout/cancellation/deadline
//
// - URL is the URL we're currently measuring
//
// - headers contains the HTTP headers we wish to use
//
// - epnts is the current list of endpoints
//
// This function SHOULD return a NEW list of extra endpoints
// it discovered and SHOULD NOT merge the epnts endpoints with
// extra endpoints it discovered. Therefore:
//
// - on any kind of error it MUST return nil, err
//
// - on success it MUST return the NEW endpoints it discovered
// as well as the TH measurement to be added to the measurement
// that the URL measurer is constructing.
//
// It is the caller's responsibility to merge the NEW list of
// endpoints with the ones it passed as argument.
//
// It is also the caller's responsibility to ENSURE that the
// newly returned endpoints only use the few headers that our
// test helper protocol allows one to set.
LookupExtraHTTPEndpoints(ctx context.Context, URL *url.URL,
headers http.Header, epnts ...*HTTPEndpoint) (
newEpnts []*HTTPEndpoint, thMeasurement *THMeasurement, err error)
}
// MeasureURL measures an HTTP or HTTPS URL. The DNS resolvers
// and the Test Helpers we use in this measurement are the ones
// configured into the database. The default is to use the system
// resolver and to use not Test Helper. Use RegisterWCTH and
// RegisterUDPResolvers (and other similar functions that have
// not been written at the moment of writing this note) to
// augment the set of resolvers and Test Helpers we use here.
//
// Arguments:
//
// - ctx is the context for timeout/cancellation.
//
// - parallelism is the number of parallel background goroutines
// to use to perform parallelizable operations (i.e., operations for
// which `measurex` defines an `OpParallel` API where `Op` is the
// name of an operation implemented by `measurex`). If parallel's value
// is zero or negative, we use a reasonably small default.
//
// - URL is the URL to measure.
//
// - header contains the HTTP headers for the request.
//
// - cookies contains the cookies we should use for measuring
// this URL and possibly future redirections.
//
// To create an empty set of cookies, use NewCookieJar. It's
// normal to have empty cookies at the beginning. If we follow
// extra redirections after this run then the cookie jar will
// contain the cookies for following the next redirection.
//
// We need cookies because a small amount of URLs does not
// redirect properly without cookies. This has been
// documented at https://github.com/ooni/probe/issues/1727.
func (mx *Measurer) MeasureURL(
ctx context.Context, parallelism int, URL string, headers http.Header,
cookies http.CookieJar) (*URLMeasurement, error) {
mx.Logger.Infof("MeasureURL url=%s", URL)
m := &URLMeasurement{URL: URL}
begin := time.Now()
defer func() { m.TotalRuntime = time.Since(begin) }()
parsed, err := url.Parse(URL)
if err != nil {
return nil, err
}
if len(mx.Resolvers) < 1 {
return nil, errors.New("measurer: no configured resolver")
}
dnsBegin := time.Now()
for dns := range mx.LookupURLHostParallel(ctx, parallelism, parsed, mx.Resolvers...) {
m.DNS = append(m.DNS, dns)
}
m.DNSRuntime = time.Since(dnsBegin)
epnts, err := AllHTTPEndpointsForURL(parsed, headers, m.DNS...)
if err != nil {
return nil, err
}
if mx.MeasureURLHelper != nil {
thBegin := time.Now()
extraEpnts, thMeasurement, _ := mx.MeasureURLHelper.LookupExtraHTTPEndpoints(
ctx, parsed, headers, epnts...)
m.THRuntime = time.Since(thBegin)
epnts = removeDuplicateHTTPEndpoints(append(epnts, extraEpnts...)...)
m.TH = thMeasurement
mx.enforceAllowedHeadersOnly(epnts)
}
epntRuntime := time.Now()
for epnt := range mx.HTTPEndpointGetParallel(ctx, parallelism, cookies, epnts...) {
m.Endpoints = append(m.Endpoints, epnt)
}
switch parsed.Scheme {
case "https":
mx.maybeQUICFollowUp(ctx, parallelism, m, cookies, epnts...)
default:
// nothing to do
}
m.EpntsRuntime = time.Since(epntRuntime)
m.fillRedirects()
return m, nil
}
// maybeQUICFollowUp checks whether we need to use Alt-Svc to check
// for QUIC. We query for HTTPSSvc but currently only Cloudflare
// implements this proposed standard. So, this function is
// where we take care of all the other servers implementing QUIC.
func (mx *Measurer) maybeQUICFollowUp(ctx context.Context, parallelism int,
m *URLMeasurement, cookies http.CookieJar, epnts ...*HTTPEndpoint) {
altsvc := []string{}
for _, epnt := range m.Endpoints {
// Check whether we have a QUIC handshake. If so, then
// HTTPSSvc worked and we can stop here.
if epnt.QUICHandshake != nil {
return
}
for _, rtrip := range epnt.HTTPRoundTrip {
if v := rtrip.ResponseHeaders.Get("alt-svc"); v != "" {
altsvc = append(altsvc, v)
}
}
}
// syntax:
//
// Alt-Svc: clear
// Alt-Svc: <protocol-id>=<alt-authority>; ma=<max-age>
// Alt-Svc: <protocol-id>=<alt-authority>; ma=<max-age>; persist=1
//
// multiple entries may be separated by comma.
//
// See https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Alt-Svc
for _, header := range altsvc {
entries := strings.Split(header, ",")
if len(entries) < 1 {
continue
}
for _, entry := range entries {
parts := strings.Split(entry, ";")
if len(parts) < 1 {
continue
}
if parts[0] == "h3=\":443\"" {
mx.doQUICFollowUp(ctx, parallelism, m, cookies, epnts...)
return
}
}
}
}
// doQUICFollowUp runs when we know there's QUIC support via Alt-Svc.
func (mx *Measurer) doQUICFollowUp(ctx context.Context, parallelism int,
m *URLMeasurement, cookies http.CookieJar, epnts ...*HTTPEndpoint) {
quicEpnts := []*HTTPEndpoint{}
// do not mutate the existing list rather create a new one
for _, epnt := range epnts {
quicEpnts = append(quicEpnts, &HTTPEndpoint{
Domain: epnt.Domain,
Network: NetworkUDP,
Address: epnt.Address,
SNI: epnt.SNI,
ALPN: []string{"h3"},
URL: epnt.URL,
Header: epnt.Header,
})
}
for mquic := range mx.HTTPEndpointGetParallel(ctx, parallelism, cookies, quicEpnts...) {
m.Endpoints = append(m.Endpoints, mquic)
}
}
func (mx *Measurer) enforceAllowedHeadersOnly(epnts []*HTTPEndpoint) {
for _, epnt := range epnts {
epnt.Header = mx.keepOnlyAllowedHeaders(epnt.Header)
}
}
func (mx *Measurer) keepOnlyAllowedHeaders(header http.Header) (out http.Header) {
out = http.Header{}
for k, vv := range header {
switch strings.ToLower(k) {
case "accept", "accept-language", "cookie", "user-agent":
for _, v := range vv {
out.Add(k, v)
}
default:
// ignore all the other headers
}
}
return
}
// redirectionQueue is the type we use to manage the redirection
// queue and to follow a reasonable number of redirects.
type redirectionQueue struct {
q []string
cnt int
}
func (r *redirectionQueue) append(URL ...string) {
r.q = append(r.q, URL...)
r.cnt++
}
func (r *redirectionQueue) popleft() (URL string) {
URL = r.q[0]
r.q = r.q[1:]
return
}
func (r *redirectionQueue) empty() bool {
return len(r.q) <= 0
}
func (r *redirectionQueue) redirectionsCount() int {
return r.cnt
}
// MeasureURLAndFollowRedirections is like MeasureURL except
// that it _also_ follows all the HTTP redirections.
func (mx *Measurer) MeasureURLAndFollowRedirections(ctx context.Context, parallelism int,
URL string, headers http.Header, cookies http.CookieJar) <-chan *URLMeasurement {
out := make(chan *URLMeasurement)
go func() {
defer close(out)
meas, err := mx.MeasureURL(ctx, parallelism, URL, headers, cookies)
if err != nil {
mx.Logger.Warnf("mx.MeasureURL failed: %s", err.Error())
return
}
out <- meas
rq := &redirectionQueue{q: meas.RedirectURLs}
const maxRedirects = 7
for !rq.empty() && rq.redirectionsCount() < maxRedirects {
URL = rq.popleft()
meas, err = mx.MeasureURL(ctx, parallelism, URL, headers, cookies)
if err != nil {
mx.Logger.Warnf("mx.MeasureURL failed: %s", err.Error())
return
}
out <- meas
rq.append(meas.RedirectURLs...)
}
}()
return out
}