The GUI has shipped a Russian locale for years; the README did not. A
translation of the English file with the same structure and links, its
terms taken from the GUI's ru.js; the language line of all three
files points at all three.
Its label changes between 就绪, 正在运行 and the hover text (启动/关闭),
so the group selector beside it moved every time and the two ended up
touching. Every label the button can show is laid out in the same grid
cell, visible or not, and the button takes the widest.
Both outbound chips were tonal and, on the default seed, the same
beige; the text view colours the outbound. A direct rule's chip is
outlined and neutral, a proxied one tonal in the primary colour, block
stays red.
The login bootstrap retried only network errors; a timeout or an HTTP
error on the first request returned without a dialog, a notice or a way
back short of reloading. Timeouts are retried like network errors, and
when the retries are spent a banner names the address and carries a
retry.
A settings save the backend rejected kept the rejected values in the
form, so the next save resent them; the form reloads what is stored.
The "members only" chip on the proxies page was not focusable. A
WebSocket frame that is not JSON threw inside the handler.
Keeping every stored field of a rule also kept upstream and domain, the
migrated aliases of server and domains that the generator prefers, so
an edit to the visible fields did not reach the running configuration.
The aliases fold into the form and are not sent back. The two SSR-only
locale keys are gone.
After a start from the app bar the confirmed state came from a touch,
but only its running flag was read: a core whose network is paused was
shown as running. The state is derived the way the dashboard derives
it.
ShadowsocksR is no longer supported by the service; the editor has no
SSR tab and the share-link codec no longer knows the scheme. A stored
node the service still lists cannot be opened for editing, which the
editor reports and closes.
The WireGuard form had DNS and Endpoint inputs that no share link and
no backend field carries, while workers, reserved and kernelMode —
which the backend parses and exports — were dropped on every edit. The
form now has the three real fields and the codec round-trips them.
When a node's link could not be loaded, the editor stayed open on a
default VMess model; saving replaced the node the user meant to edit.
The dialog now closes with the error.
The action rows on the settings page (RoutingA, ports, DNS, custom
inbounds, TUN, About) were list items with a click handler but no link
role, so they took no focus and Enter did nothing. They are links now.
The DNS dialog rebuilt each rule from its three editable fields, so a
save dropped the matchers the DNS module reads and only the API sets
(ip, queryType, clientIP, action, policy). The rows keep every field
they came with.
The log stream kept every line since the page opened; it keeps the last
20000. The GFWList hint condition was named as if it meant stale; it
means the local file's date is ahead of the remote one, which is what
the text explains.
The dashboard's touch rows do not carry a subscription server's index,
so the member dialog keyed every subscription row as subscription 0: a
member of the second subscription showed unchecked, nodes with the same
id in two subscriptions shared one checkbox, and Save re-pointed every
chosen node at the first subscription's node with that id.
A VLESS or VMess link with type=raw — xray's new name for the tcp
transport — was rejected with "unexpected transport type: raw"
(discussion #1999). A vmess JSON whose port, aid or v is a number was
dropped as "not a JSON object"; v2rayN writes strings but most other
exporters write numbers, and one public subscription lost 152 of its
1625 nodes to it (#1957).
Deleting nodes stored the renumbered connections, the subscriptions and
the servers in three commits; they are one transaction. Without a
catch-all DNS rule the module keeps its own fallback, as every earlier
release did. The login body is capped at 4 KiB, since it is now read
before the one-at-a-time semaphore. The apply-config error names a
failed restore, so every handler's message says whether the previous
value came back. The Windows system proxy reuses a snapshot pending
from an earlier run instead of refusing to start.
When elevated, saveProxyState captured the first readable profile's
proxy values and stopped, then the restore wrote those values into
every profile: users with different proxy settings got the first
user's, or none. The snapshot now holds one entry per SID (the current
user under "" when not elevated); restore writes each profile its own
values, a profile written at setup but not captured is left with the
proxy disabled, and a snapshot from an earlier release (one flat
profile) still restores the way it did.
PostImport, PutPorts, PutDnsRules, PutRoutingA, PutGFWList,
DeleteGFWList, PutDomainsExcluded, PutTproxyWhiteIpGroups, PostOutbound
and PutOutbound wrote the store and regenerated the core outside the
mutation turn, so two of them could interleave core restarts with each
other or with a latency test. They take it like the rest.
ParseSSURL and ParseVmessURL sliced past the end of a link shorter than
its scheme prefix ("ss:", "vmess:x"), a panic reachable from a
subscription line.
The core has no ShadowsocksR outbound and the plugin that once served
one was removed in May, so every SSR node has been dead since. Instead
of keeping a protocol that cannot connect: an ssr:// link is refused at
import with "ShadowsocksR is not supported" and the Clash converter
skips ssr entries. A node stored by an earlier release still decodes so
it keeps its place in the list and its group ordinal; it shows as
"SSR (unsupported)", cannot be configured and can be deleted. Both
READMEs list the protocols the current core actually speaks. The GUI
side of the removal is in the GUI change that follows.
A custom latency test URL that http.NewRequest rejects made a spawned
goroutine dereference nil and exit the process; the URL is validated
before any goroutine starts and the request error is handled. A
Shadowsocks link whose plugin field has no ";" panicked the parser, and
from the automatic subscription update that goroutine was unrecovered;
the parser accepts it and each updater recovers and logs.
The DNS module took the last non-bootstrap upstream as the fallback
while the template emitted upstreams in rule order with deduplication,
so a catch-all rule anywhere but last, or one whose upstream was
deduplicated, was not the fallback. The template names the fallback in
a new default_upstream field and the module honours it.
PUT /routingA stored the rules without regenerating the running core.
It goes through ApplyCoreConfig like the other config writers, with
restore on failure. UpdateSubscription logged a failed core reload and
answered success; it returns the error. PATCH /subscription joins the
mutation turn.
Subscription fetching retried through a direct connection when the
selected proxy failed, leaking the subscription URL; it returns the
error. Subscription documents are capped at 32 MiB and downloaded
assets at 256 MiB. PUT /gfwList validates the downloaded file as a
geosite list before replacing the working one.
The login handler took its one-at-a-time semaphore before reading the
body, so a client sending headers and no body blocked every login; the
body is read first and the HTTP server has header and idle timeouts.
Share links: a VLESS or VMess link without a numeric port is rejected
at parse time instead of producing port 0; an SSR link with a password
that is not base64 is rejected instead of importing empty; the SIP003
path gets its leading slash prepended, not a trailing one appended.
The macOS system proxy quoted shell arguments with strconv.Quote,
which leaves $(...) live; values are single-quoted. The Windows system
proxy snapshot includes AutoConfigURL, which teardown deleted and never
restored. Latency test responses with a non-2xx status are closed. The
version response carries the docker flag the GUI's port hints read.
Commit 0baefc45 (2026-05-03) removed the plugin layer under
service/pkg/plugin and the code that started a plugin per node, but
left the HTTP and ShadowsocksR node types producing what that layer
consumed: a socks outbound to 127.0.0.1:<plugin port> plus a plugin
chain nothing reads. Every http/https proxy node and every SSR node
imported since then connected to a closed local port.
HTTP and HTTPS nodes now produce the core's native http outbound with
the credentials and, for https, TLS to the server name; they no longer
ask for a plugin port. The core has no ShadowsocksR outbound, so an SSR
node's configuration returns an "unsupported" error the latency test
and the automatic selection already understand; the node still parses,
displays and exports so nothing is lost from the store.
Three structural changes that share files, so one commit:
Settings, ports, DNS rules and custom inbounds each carried their own
copy of "store the new value, regenerate the core config, put the old
value back when that fails", and the outbound setting path had no
restore at all. ApplyCoreConfig takes a snapshot function and a store
function and does the sequence once for all five callers; the outbound
setting path restores like the others, and the error names which of
the three steps failed.
configure.Which was the persisted discriminator, the key that locates a
node, and the response row with its latency, link and selected flag at
once. NodeRef {type, id, sub, outbound} is the identity used by the
locator, EqualTo, the connected lists and the group operations; Which
embeds it and keeps the status fields. The JSON stays byte for byte
what it was (a test marshals a fixture before and after), stored
connected lists decode unchanged, and deduplication keys by identity
instead of by the whole row.
db.List* dispatched on "bucket/key" string literals to reach the
servers and subscriptions tables. The bodies are now Servers* and
Subscriptions* functions that configure calls directly; the List*
wrappers remain for the migration.
OpenRC's default stop escalates after a few seconds; the service stops
its core and removes the transparent-proxy rules on SIGTERM, so it gets
30 s before KILL. stop_post kills a core the service could not stop (it
had been killed), which would otherwise keep the inbound ports and fail
the next start. The log and config directories are created with
checkpath, the service is ordered after the firewall units, and the
per-user script creates its config directory the same way.
A VLESS or VMess link with type=raw — xray's new name for the tcp
transport — was rejected with "unexpected transport type: raw"
(discussion #1999). A vmess JSON whose port, aid or v is a number was
dropped as "not a JSON object"; v2rayN writes strings but most other
exporters write numbers, and one public subscription lost 152 of its
1625 nodes to it (#1957).
Deleting nodes stored the renumbered connections, the subscriptions and
the servers in three commits; they are one transaction. Without a
catch-all DNS rule the module keeps its own fallback, as every earlier
release did. The login body is capped at 4 KiB, since it is now read
before the one-at-a-time semaphore. The apply-config error names a
failed restore, so every handler's message says whether the previous
value came back. The Windows system proxy reuses a snapshot pending
from an earlier run instead of refusing to start.
When elevated, saveProxyState captured the first readable profile's
proxy values and stopped, then the restore wrote those values into
every profile: users with different proxy settings got the first
user's, or none. The snapshot now holds one entry per SID (the current
user under "" when not elevated); restore writes each profile its own
values, a profile written at setup but not captured is left with the
proxy disabled, and a snapshot from an earlier release (one flat
profile) still restores the way it did.
PostImport, PutPorts, PutDnsRules, PutRoutingA, PutGFWList,
DeleteGFWList, PutDomainsExcluded, PutTproxyWhiteIpGroups, PostOutbound
and PutOutbound wrote the store and regenerated the core outside the
mutation turn, so two of them could interleave core restarts with each
other or with a latency test. They take it like the rest.
ParseSSURL and ParseVmessURL sliced past the end of a link shorter than
its scheme prefix ("ss:", "vmess:x"), a panic reachable from a
subscription line.
The core has no ShadowsocksR outbound and the plugin that once served
one was removed in May, so every SSR node has been dead since. Instead
of keeping a protocol that cannot connect: an ssr:// link is refused at
import with "ShadowsocksR is not supported" and the Clash converter
skips ssr entries. A node stored by an earlier release still decodes so
it keeps its place in the list and its group ordinal; it shows as
"SSR (unsupported)", cannot be configured and can be deleted. Both
READMEs list the protocols the current core actually speaks. The GUI
side of the removal is in the GUI change that follows.
A custom latency test URL that http.NewRequest rejects made a spawned
goroutine dereference nil and exit the process; the URL is validated
before any goroutine starts and the request error is handled. A
Shadowsocks link whose plugin field has no ";" panicked the parser, and
from the automatic subscription update that goroutine was unrecovered;
the parser accepts it and each updater recovers and logs.
The DNS module took the last non-bootstrap upstream as the fallback
while the template emitted upstreams in rule order with deduplication,
so a catch-all rule anywhere but last, or one whose upstream was
deduplicated, was not the fallback. The template names the fallback in
a new default_upstream field and the module honours it.
PUT /routingA stored the rules without regenerating the running core.
It goes through ApplyCoreConfig like the other config writers, with
restore on failure. UpdateSubscription logged a failed core reload and
answered success; it returns the error. PATCH /subscription joins the
mutation turn.
Subscription fetching retried through a direct connection when the
selected proxy failed, leaking the subscription URL; it returns the
error. Subscription documents are capped at 32 MiB and downloaded
assets at 256 MiB. PUT /gfwList validates the downloaded file as a
geosite list before replacing the working one.
The login handler took its one-at-a-time semaphore before reading the
body, so a client sending headers and no body blocked every login; the
body is read first and the HTTP server has header and idle timeouts.
Share links: a VLESS or VMess link without a numeric port is rejected
at parse time instead of producing port 0; an SSR link with a password
that is not base64 is rejected instead of importing empty; the SIP003
path gets its leading slash prepended, not a trailing one appended.
The macOS system proxy quoted shell arguments with strconv.Quote,
which leaves $(...) live; values are single-quoted. The Windows system
proxy snapshot includes AutoConfigURL, which teardown deleted and never
restored. Latency test responses with a non-2xx status are closed. The
version response carries the docker flag the GUI's port hints read.
Commit 0baefc45 (2026-05-03) removed the plugin layer under
service/pkg/plugin and the code that started a plugin per node, but
left the HTTP and ShadowsocksR node types producing what that layer
consumed: a socks outbound to 127.0.0.1:<plugin port> plus a plugin
chain nothing reads. Every http/https proxy node and every SSR node
imported since then connected to a closed local port.
HTTP and HTTPS nodes now produce the core's native http outbound with
the credentials and, for https, TLS to the server name; they no longer
ask for a plugin port. The core has no ShadowsocksR outbound, so an SSR
node's configuration returns an "unsupported" error the latency test
and the automatic selection already understand; the node still parses,
displays and exports so nothing is lost from the store.
Three structural changes that share files, so one commit:
Settings, ports, DNS rules and custom inbounds each carried their own
copy of "store the new value, regenerate the core config, put the old
value back when that fails", and the outbound setting path had no
restore at all. ApplyCoreConfig takes a snapshot function and a store
function and does the sequence once for all five callers; the outbound
setting path restores like the others, and the error names which of
the three steps failed.
configure.Which was the persisted discriminator, the key that locates a
node, and the response row with its latency, link and selected flag at
once. NodeRef {type, id, sub, outbound} is the identity used by the
locator, EqualTo, the connected lists and the group operations; Which
embeds it and keeps the status fields. The JSON stays byte for byte
what it was (a test marshals a fixture before and after), stored
connected lists decode unchanged, and deduplication keys by identity
instead of by the whole row.
db.List* dispatched on "bucket/key" string literals to reach the
servers and subscriptions tables. The bodies are now Servers* and
Subscriptions* functions that configure calls directly; the List*
wrappers remain for the migration.
v2raya.initd / v2raya.confd for the root service and v2raya-user.initd /
v2raya-user.confd for the per-user lite service, taken from the overlay
that has run them on Gentoo since 2021 (the maintainer of both is the
same person); the description and the option list in the conf.d files
match the current binary. The same files serve Alpine.
Alpine, Gentoo and the other OpenRC systems had no service script in
the repository; the one Gentoo ships is now here as
install/universal/v2raya.initd with its conf.d file, the same shape
as the systemd unit (log file under /var/log/v2raya, options through
V2RAYA_OPTS). The runit script names the current maintainer's address.
install/universal/v2raya.run is the runit run script users install by
hand on runit distributions (Void, Artix); it has no in-repository
caller by design. It went out with the dead code by mistake; back in.
systemd sends the whole cgroup SIGTERM, so the core died before the
service and was recorded as a crash; the next boot then refused to
restore the proxy ("exited abnormally the last time", discussion
#1922). The signal handler marks the manager as shutting down before
anything else, a core exit in that state keeps the running flag, and
the units use KillMode=mixed so only the service receives the signal.
A SQLite file that an earlier release's migration left empty next to an
intact bolt.db was taken as authoritative; a file without the system
settings a completed migration writes is discarded and the migration
runs again. A database migrated by that earlier release holds its
groups and outbound settings in the two tables the runtime never read;
opening it copies them into the keys it reads, for outbounds that have
none yet.
A start whose host-state recovery failed refused to run, and a start
that rolled back after its own stop tore the host state down a second
time; recovery is logged and retried at the next start, and a rollback
that stopped its process only clears the marker. A system-proxy
snapshot pending from an earlier run blocked every later system-proxy
start; it is reused as the original instead (Linux and macOS here; the
Windows side follows with its per-profile snapshot).
An empty /etc/resolv.conf was backed up as an empty file, which the
restore rejects as truncated; it is written as a marker and restored as
empty. db.Close resets the once so a reopen works.
Start persisted running=false, stopped the old core, published the new
process and unlocked, installed the firewall rules, DNS hijack, TUN
routes or system proxy, and only then persisted running=true. A crash in
that window made the next start skip the core and clean only TUN
residue, so redirect rules, a hijacked resolver or the system proxy kept
pointing at a dead listener.
Before it installs anything, Start now stores a host-state marker (the
transparent type, API port and TUN teardown settings) and clears it once
the start commits or is rolled back; startup and the next Start tear
that state down when they find the marker. Every host mutation and the
final commit check that the process they started is still the manager's
current one, so a core that exits during afterStart fails the start
instead of leaving the UI and the store saying running. The system-proxy
implementations persist the user's original proxy settings when they
capture them and restore from that copy when the in-memory one is gone.
A recovery that fails is logged and retried at the next start; it never
blocks the service.
The migration wrote each outbound's connected servers and setting into
the outbound_connections and outbound_settings tables, while the runtime
reads both from system_config under outbound.<name>:connectedServers and
outbound.<name>:setting. Every upgrade from a BoltDB release therefore
came up with no group members and default outbound settings. The
migration now stores the Bolt values under the keys the runtime reads
(the ordinals are the ones configure.Which resolves at runtime) and the
verification checks those keys. The two tables are gone from the schema;
a database that still has them is left alone.
The migration also built v2raya.db in place, so an error before the
rename left a partial database that made the next start skip the
migration and then fail on the remaining bolt.db. It builds a temporary
file and renames it in after verification; a bolt.db found next to an
existing v2raya.db is moved to a numbered backup instead of blocking
every start. The v4 import no longer copies boltv4.db to bolt.db, which
produced exactly that pair.
A subscription update rewrote the connected ordinals and then replaced
the subscription in two commits; they are one transaction.
Three service packages outside the executable's dependency graph
(common/antiPollution, pkg/util/gopeed, kernel/vmessInfo), a blank
import of miekg/dns the core already imports, the --core-type flag no
code read, the PutAccount handler whose route was commented out, two GUI
API wrappers without a caller, two .SRCINFO files the release regenerates
and an unreferenced runit sample. The workflow path filters watched a
root go.mod that does not exist and the PR build skipped core/.
The core hooks ran under exec.Command with no deadline while Start held
the manager's lock, so a hook that never returned left the proxy stopped
and every status read blocked. They now run with the startup timeout as
their deadline in their own process group.
The custom-inbound POST and DELETE handlers did not take the mutation
turn; two concurrent additions read the same list and the last write
won. They take it like every other state-changing handler.
A failed HTTP bind ended in log.Fatal before the deferred cleanup, so
the core and the transparent-proxy rules stayed behind; run() now
returns the error through the same cleanup the shutdown path uses.
The resolv.conf backup was written in place; a crash mid-write left a
truncated file that the next start restored as the original. It is
written to a temporary file, synced and renamed, and an empty backup is
ignored on restore.
build.sh and build.ps1 each built the GUI, the core and the service
their own way, and nothing in the repository ran them; build.sh passed a
with_gvisor tag no source file declares. build.sh now runs the same
commands as _build_web.yml and _build_binaries.yml, takes the version as
an argument or from git describe, and stops at the first failing step;
build.ps1 is the same sequence for PowerShell.
release_main.yml and beta_release_main.yml duplicated every release job
and had drifted (the beta looked the release up in a hard-coded
repository). Both now call _release.yml with the tag, a prerelease flag
and whether to publish to the downstream package channels.
One `updating` flag guarded thirteen handlers, reads included: while a
subscription update or a latency test ran, GET /touch answered
REQUEST_IN_PROGRESS, so refreshing the page during a test failed at
random. A touch is a database read and needs no turn.
The flag becomes a turn that only the state-changing handlers take, and
a handler that finds it taken waits up to five seconds before answering
REQUEST_IN_PROGRESS instead of refusing at once; a client that leaves
while waiting is not answered.
ListSet deleted a subscription's node rows and re-inserted them with one
autocommit per row. A crash between the delete and the last insert left
the subscription without nodes, and the latency test, which stores its
results this way, paid a commit per node. The rewrite now runs inside
ReadModifyWrite: all or nothing, and 1000 nodes take 9 ms instead of
25 ms (BenchmarkSetSubscription).