bool
huge_update_hashes(shared_file_t *sf,
const struct sha1 *sha1, const struct tth *tth)
{
struct sha1_cache_entry *cached;
filestat_t sb;
shared_file_check(sf);
g_return_val_if_fail(sha1, FALSE);
/*
* Make sure the file's timestamp is still accurate.
*/
if (-1 == stat(shared_file_path(sf), &sb)) {
g_warning("discarding SHA1 for file \"%s\": can't stat(): %m",
shared_file_path(sf));
shared_file_remove(sf);
return TRUE;
}
if (sb.st_mtime != shared_file_modification_time(sf)) {
g_warning("file \"%s\" was modified whilst SHA1 was computed",
shared_file_path(sf));
shared_file_set_modification_time(sf, sb.st_mtime);
request_sha1(sf); /* Retry! */
return TRUE;
}
if (huge_spam_check(sf, sha1)) {
shared_file_remove(sf);
return FALSE;
}
shared_file_set_sha1(sf, sha1);
shared_file_set_tth(sf, tth);
/* Update cache */
cached = hikset_lookup(sha1_cache, shared_file_path(sf));
if (cached) {
update_volatile_cache(cached, shared_file_size(sf),
shared_file_modification_time(sf), sha1, tth);
cache_dirty = TRUE;
/* Dump the cache at most about once per minute. */
if (!cache_dumped || delta_time(tm_time(), cache_dumped) > 60) {
dump_cache(FALSE);
}
} else {
add_volatile_cache_entry(shared_file_path(sf),
shared_file_size(sf), shared_file_modification_time(sf),
sha1, tth, TRUE);
add_persistent_cache_entry(shared_file_path(sf),
shared_file_size(sf), shared_file_modification_time(sf),
sha1, tth);
}
return TRUE;
}
/**
* External interface to call for getting the hash for a shared_file.
*/
void
request_sha1(shared_file_t *sf)
{
struct sha1_cache_entry *cached;
shared_file_check(sf);
if (!shared_file_indexed(sf))
return; /* "stale" shared file, has been superseded or removed */
cached = hikset_lookup(sha1_cache, shared_file_path(sf));
if (cached && cached_entry_up_to_date(cached, sf)) {
cache_dirty = TRUE;
cached->shared = TRUE;
shared_file_set_sha1(sf, cached->sha1);
shared_file_set_tth(sf, cached->tth);
request_tigertree(sf, NULL == cached->tth);
} else {
if (GNET_PROPERTY(share_debug) > 1) {
if (cached)
g_debug("cached SHA1 entry for \"%s\" outdated: "
"had mtime %lu, now %lu",
shared_file_path(sf),
(ulong) cached->mtime,
(ulong) shared_file_modification_time(sf));
else
g_debug("queuing \"%s\" for SHA1 computation",
shared_file_path(sf));
}
queue_shared_file_for_sha1_computation(sf);
}
}
static bool
huge_verify_callback(const struct verify *ctx, enum verify_status status,
void *user_data)
{
shared_file_t *sf = user_data;
shared_file_check(sf);
switch (status) {
case VERIFY_START:
if (!huge_need_sha1(sf))
return FALSE;
gnet_prop_set_boolean_val(PROP_SHA1_REBUILDING, TRUE);
return TRUE;
case VERIFY_PROGRESS:
return 0 != (SHARE_F_INDEXED & shared_file_flags(sf));
case VERIFY_DONE:
huge_update_hashes(sf, verify_sha1_digest(ctx), NULL);
request_tigertree(sf, TRUE);
/* FALL THROUGH */
case VERIFY_ERROR:
case VERIFY_SHUTDOWN:
gnet_prop_set_boolean_val(PROP_SHA1_REBUILDING, FALSE);
shared_file_unref(&sf);
return TRUE;
case VERIFY_INVALID:
break;
}
g_assert_not_reached();
return FALSE;
}
void
request_tigertree(shared_file_t *sf, bool high_priority)
{
int inserted;
verify_tth_init();
shared_file_check(sf);
g_return_if_fail(shared_file_is_finished(sf));
if (!shared_file_is_servable(sf))
return; /* "stale" shared file, has been superseded or removed */
/*
* This routine can be called when the VERIFY_DONE event is received by
* huge_verify_callback(). We may have already shutdown the TTH
* verification thread.
*/
if G_UNLIKELY(NULL == verify_tth.verify)
return;
sf = shared_file_ref(sf);
inserted = verify_enqueue(verify_tth.verify, high_priority,
shared_file_path(sf), 0, shared_file_size(sf),
request_tigertree_callback, sf);
if (!inserted)
shared_file_unref(&sf);
}
static bool
request_tigertree_callback(const struct verify *ctx, enum verify_status status,
void *user_data)
{
shared_file_t *sf = user_data;
shared_file_check(sf);
switch (status) {
case VERIFY_START:
if (!(SHARE_F_INDEXED & shared_file_flags(sf))) {
/*
* After a rescan, there might be files in the queue which are
* no longer shared.
*/
if (GNET_PROPERTY(verify_debug) > 1) {
g_debug("skipping TTH computation for %s: no longer shared",
shared_file_path(sf));
}
return FALSE;
}
if (
shared_file_tth(sf) &&
tth_cache_lookup(shared_file_tth(sf), shared_file_size(sf)) > 0
) {
if (
GNET_PROPERTY(tigertree_debug) > 1 ||
GNET_PROPERTY(verify_debug) > 1
) {
g_debug("TTH for %s is already cached (%s)",
shared_file_path(sf), tth_base32(shared_file_tth(sf)));
}
return FALSE;
}
gnet_prop_set_boolean_val(PROP_TTH_REBUILDING, TRUE);
return TRUE;
case VERIFY_PROGRESS:
return 0 != (SHARE_F_INDEXED & shared_file_flags(sf));
case VERIFY_DONE:
{
const struct tth *tth = verify_tth_digest(ctx);
huge_update_hashes(sf, shared_file_sha1(sf), tth);
tth_cache_insert(tth, verify_tth_leaves(ctx),
verify_tth_leave_count(ctx));
}
/* FALL THROUGH */
case VERIFY_ERROR:
case VERIFY_SHUTDOWN:
shared_file_unref(&sf);
gnet_prop_set_boolean_val(PROP_TTH_REBUILDING, FALSE);
return TRUE;
case VERIFY_INVALID:
break;
}
g_assert_not_reached();
return FALSE;
}
/**
* Put the shared file on the stack of the things to do.
*
* We first begin with the computation of the SHA1, and when completed we
* will continue with the TTH computation.
*/
static void
queue_shared_file_for_sha1_computation(shared_file_t *sf)
{
int inserted;
shared_file_check(sf);
inserted = verify_sha1_enqueue(FALSE, shared_file_path(sf),
shared_file_size(sf), huge_verify_callback,
shared_file_ref(sf));
if (!inserted)
shared_file_unref(&sf);
}
void
request_tigertree(shared_file_t *sf, bool high_priority)
{
int inserted;
verify_tth_init();
g_return_if_fail(sf);
shared_file_check(sf);
g_return_if_fail(!shared_file_is_partial(sf));
sf = shared_file_ref(sf);
inserted = verify_enqueue(verify_tth.verify, high_priority,
shared_file_path(sf), 0, shared_file_size(sf),
request_tigertree_callback, sf);
if (!inserted)
shared_file_unref(&sf);
}
/**
* Look whether we still need to compute the SHA1 of the given shared file
* by looking into our in-core cache to see whether the entry we have is
* up-to-date.
*
* @param sf the shared file for which we want to compute the SHA1
*
* @return TRUE if the file need SHA1 recomputation.
*/
static bool
huge_need_sha1(shared_file_t *sf)
{
struct sha1_cache_entry *cached;
shared_file_check(sf);
/*
* After a rescan, there might be files in the queue which are
* no longer shared.
*/
if (!shared_file_indexed(sf))
return FALSE;
if G_UNLIKELY(NULL == sha1_cache)
return FALSE; /* Shutdown occurred (processing TEQ event?) */
cached = hikset_lookup(sha1_cache, shared_file_path(sf));
if (cached != NULL) {
filestat_t sb;
if (-1 == stat(shared_file_path(sf), &sb)) {
g_warning("ignoring SHA1 recomputation request for \"%s\": %m",
shared_file_path(sf));
return FALSE;
}
if (
cached->size + (fileoffset_t) 0 == sb.st_size + (filesize_t) 0 &&
cached->mtime == sb.st_mtime
) {
if (GNET_PROPERTY(share_debug) > 1) {
g_warning("ignoring duplicate SHA1 work for \"%s\"",
shared_file_path(sf));
}
return FALSE;
}
}
return TRUE;
}
/**
* Add file to the current query hit.
*
* @return TRUE if we kept the file, FALSE if we did not include it in the hit.
*/
static bool
g2_build_qh2_add(struct g2_qh2_builder *ctx, const shared_file_t *sf)
{
const sha1_t *sha1;
g2_tree_t *h, *c;
shared_file_check(sf);
/*
* Make sure the file is still in the library.
*/
if (0 == shared_file_index(sf))
return FALSE;
/*
* On G2, the H/URN child is required, meaning we need the SHA1 at least.
*/
if (!sha1_hash_available(sf))
return FALSE;
/*
* Do not send duplicates, as determined by the SHA1 of the resource.
*
* A user may share several files with different names but the same SHA1,
* and if all of them are hits, we only want to send one instance.
*
* When generating hits for host-browsing, we do not care about duplicates
* and ctx->hs is NULL then.
*/
sha1 = shared_file_sha1(sf); /* This is an atom */
if (ctx->hs != NULL) {
if (hset_contains(ctx->hs, sha1))
return FALSE;
hset_insert(ctx->hs, sha1);
}
/*
* Create the "H" child and attach it to the current tree.
*/
if (NULL == ctx->t)
g2_build_qh2_start(ctx);
h = g2_tree_alloc_empty("H");
g2_tree_add_child(ctx->t, h);
/*
* URN -- Universal Resource Name
*
* If there is a known TTH, then we can generate a bitprint, otherwise
* we just convey the SHA1.
*/
{
const tth_t * const tth = shared_file_tth(sf);
char payload[SHA1_RAW_SIZE + TTH_RAW_SIZE + sizeof G2_URN_BITPRINT];
char *p = payload;
if (NULL == tth) {
p = mempcpy(p, G2_URN_SHA1, sizeof G2_URN_SHA1);
p += clamp_memcpy(p, sizeof payload - ptr_diff(p, payload),
sha1, SHA1_RAW_SIZE);
} else {
p = mempcpy(p, G2_URN_BITPRINT, sizeof G2_URN_BITPRINT);
p += clamp_memcpy(p, sizeof payload - ptr_diff(p, payload),
sha1, SHA1_RAW_SIZE);
p += clamp_memcpy(p, sizeof payload - ptr_diff(p, payload),
tth, TTH_RAW_SIZE);
}
g_assert(ptr_diff(p, payload) <= sizeof payload);
c = g2_tree_alloc_copy("URN", payload, ptr_diff(p, payload));
g2_tree_add_child(h, c);
}
/*
* URL -- empty to indicate that we share the file via uri-res.
*/
if (ctx->flags & QHIT_F_G2_URL) {
uint known;
uint16 csc;
c = g2_tree_alloc_empty("URL");
g2_tree_add_child(h, c);
/*
* CSC -- if we know alternate sources, indicate how many in "CSC".
*
* This child is only emitted when they requested "URL".
*/
known = dmesh_count(sha1);
csc = MIN(known, MAX_INT_VAL(uint16));
if (csc != 0) {
char payload[2];
poke_le16(payload, csc);
c = g2_tree_alloc_copy("CSC", payload, sizeof payload);
g2_tree_add_child(h, c);
}
/*
* PART -- if we only have a partial file, indicate how much we have.
*
* This child is only emitted when they requested "URL".
*/
if (shared_file_is_partial(sf) && !shared_file_is_finished(sf)) {
filesize_t available = shared_file_available(sf);
char payload[8]; /* If we have to encode file size as 64-bit */
uint32 av32;
time_t mtime = shared_file_modification_time(sf);
c = g2_tree_alloc_empty("PART");
g2_tree_add_child(h, c);
av32 = available;
if (av32 == available) {
/* Fits within a 32-bit quantity */
poke_le32(payload, av32);
g2_tree_set_payload(c, payload, sizeof av32, TRUE);
} else {
/* Encode as a 64-bit quantity then */
poke_le64(payload, available);
g2_tree_set_payload(c, payload, sizeof payload, TRUE);
}
/*
* GTKG extension: encode the last modification time of the
* partial file in an "MT" child. This lets the other party
* determine whether the host is still able to actively complete
* the file.
*/
poke_le32(payload, (uint32) mtime);
g2_tree_add_child(c,
g2_tree_alloc_copy("MT", payload, sizeof(uint32)));
}
/*
* CT -- creation time of the resource (GTKG extension).
*/
{
time_t create_time = shared_file_creation_time(sf);
if ((time_t) -1 != create_time) {
char payload[8];
int n;
create_time = MAX(0, create_time);
n = vlint_encode(create_time, payload);
g2_tree_add_child(h,
g2_tree_alloc_copy("CT", payload, n)); /* No trailing 0s */
}
}
}
/*
* DN -- distinguished name.
*
* Note that the presence of DN also governs the presence of SZ if the
* file length does not fit a 32-bit unsigned quantity.
*/
if (ctx->flags & QHIT_F_G2_DN) {
char payload[8]; /* If we have to encode file size as 64-bit */
uint32 fs32;
filesize_t fs = shared_file_size(sf);
const char *name;
const char *rp;
c = g2_tree_alloc_empty("DN");
fs32 = fs;
if (fs32 == fs) {
/* Fits within a 32-bit quantity */
poke_le32(payload, fs32);
g2_tree_set_payload(c, payload, sizeof fs32, TRUE);
} else {
/* Does not fit a 32-bit quantity, emit a SZ child */
poke_le64(payload, fs);
g2_tree_add_child(h,
g2_tree_alloc_copy("SZ", payload, sizeof payload));
}
name = shared_file_name_nfc(sf);
g2_tree_append_payload(c, name, shared_file_name_nfc_len(sf));
g2_tree_add_child(h, c);
/*
* GTKG extension: if there is a file path, expose it as a "P" child
* under the DN node.
*/
rp = shared_file_relative_path(sf);
if (rp != NULL) {
g2_tree_add_child(c, g2_tree_alloc_copy("P", rp, strlen(rp)));
}
}
/*
* GTKG extension: if they requested alt-locs in the /Q2/I with "A", then
* send them some known alt-locs in an "ALT" child.
*
* Note that these alt-locs can be for Gnutella hosts: since both Gnutella
* and G2 share a common HTTP-based file transfer mechanism with compatible
* extra headers, there is no need to handle them separately.
*/
if (ctx->flags & QHIT_F_G2_ALT) {
gnet_host_t hvec[G2_BUILD_QH2_MAX_ALT];
int hcnt = 0;
hcnt = dmesh_fill_alternate(sha1, hvec, N_ITEMS(hvec));
if (hcnt > 0) {
int i;
c = g2_tree_alloc_empty("ALT");
for (i = 0; i < hcnt; i++) {
host_addr_t addr;
uint16 port;
addr = gnet_host_get_addr(&hvec[i]);
port = gnet_host_get_port(&hvec[i]);
if (host_addr_is_ipv4(addr)) {
char payload[6];
host_ip_port_poke(payload, addr, port, NULL);
g2_tree_append_payload(c, payload, sizeof payload);
}
}
/*
* If the payload is still empty, then drop the "ALT" child.
* Otherwise, attach it to the "H" node.
*/
if (NULL == g2_tree_node_payload(c, NULL)) {
g2_tree_free_null(&c);
} else {
g2_tree_add_child(h, c);
}
}
}
/*
* Update the size of the query hit we're generating.
*/
ctx->current_size += g2_frame_serialize(h, NULL, 0);
return TRUE;
}
static bool
request_tigertree_callback(const struct verify *ctx, enum verify_status status,
void *user_data)
{
shared_file_t *sf = user_data;
shared_file_check(sf);
switch (status) {
case VERIFY_START:
if (!shared_file_is_servable(sf)) {
/*
* After a rescan, there might be files in the queue which are
* no longer shared.
*/
if (GNET_PROPERTY(verify_debug) > 1) {
g_debug("skipping TTH computation for %s: not a servable file",
shared_file_path(sf));
}
return FALSE;
}
if (shared_file_tth_is_available(sf)) {
if (
GNET_PROPERTY(tigertree_debug) > 1 ||
GNET_PROPERTY(verify_debug) > 1
) {
g_debug("TTH for %s is already cached (%s)",
shared_file_path(sf), tth_base32(shared_file_tth(sf)));
}
return FALSE;
}
gnet_prop_set_boolean_val(PROP_TTH_REBUILDING, TRUE);
return TRUE;
case VERIFY_PROGRESS:
/*
* Processing can continue whilst the library file is indexed or the
* completed file is still beeing seeded.
*/
return shared_file_is_servable(sf);
case VERIFY_DONE:
{
const struct tth *tth = verify_tth_digest(ctx);
size_t n_leaves = verify_tth_leave_count(ctx);
if (GNET_PROPERTY(verify_debug)) {
g_debug("%s(): computed TTH %s (%zu lea%s) for %s",
G_STRFUNC, tth_base32(tth),
n_leaves, plural_f(n_leaves),
shared_file_path(sf));
}
/*
* Write the TTH to the cache first, before updating the hashes.
* That way, the logic behind huge_update_hashes() can rely on
* the fact that the TTH is persisted already.
*
* This is important for seeded files for which we re-compute
* the TTH once they are completed (to make sure we can serve
* THEX requests at the proper good depth). In order to update
* the GUI information, we'll need to probe the cache to determine
* how large the TTH is exactly, since all we pass back to the
* routines is the TTH root hash.
* --RAM, 2017-10-20
*/
tth_cache_insert(tth, verify_tth_leaves(ctx), n_leaves);
huge_update_hashes(sf, shared_file_sha1(sf), tth);
}
goto done;
case VERIFY_ERROR:
if (GNET_PROPERTY(verify_debug)) {
g_debug("%s(): unable to compute TTH for %s",
G_STRFUNC, shared_file_path(sf));
}
/* FALL THROUGH */
case VERIFY_SHUTDOWN:
goto done;
case VERIFY_INVALID:
break;
}
g_assert_not_reached();
return FALSE;
done:
shared_file_unref(&sf);
gnet_prop_set_boolean_val(PROP_TTH_REBUILDING, FALSE);
return TRUE;
}
