Wj
J U d dl Z d dlZd dlZd dlZd dlZd dlZd dlZd dlmZ d dl m
Z
d dl m Z m
Z
d dlZd dlZd dlmZ d dlmZ d dlmZ d dlmZ d d lmZ d d
lmZ d d l m Z d d l m Z m!Z!m"Z" d d
l#m$Z$m%Z% d dl&m'Z' ej( e) Z*dZ+e,e-d< dZ.e,e-d< dZ/e,e-d< e!j0 Z1e2e-d< dZ3e,e-d< dZ4e,e-d< ddiZ5d e2fd Z6 G d d e j7 Z8 G d d!e j7 Z9 e:h d" Z; G d# d$e j7 Z< G d% d&e j= e> Z?d'e2d(e2d e?fd)Z@ G d* d+ ZAd,e2d-e,dz d e2fd.ZB G d/ d0e ZCe1fd1e2d ejD fd2ZE e"jF d3eg e1e+fd1e2d4eGfd5 ZHed'e2d6e
fd7 ZIe jJ G d8 d9 ZKdYd'e2d:e2d;e2dz d eKfd<ZLd=ZMd>ZN ejO d? d@f ejO dA dBf ejO dC dDf ejO dE dFffZPdGe2dz d e2dz fdHZQdIe2dJe2d e2dz fdKZRdIe2dJe2d eSfdLZTd'e2dJe2d eSfdMZUdJe2d eSfdNZVdYd'e2dJe2d;e2dz d eSfdOZWd'e2dJe2d eSfdPZXd'e2dQe2fdRZYd'e2dSe2d1e2d;e2dz d4eGd eKf dTZZ dZd'e2dUe2dSe2d;e2dz dVe2dz d eKf dWZ[ dZd'e2dUe2d;e2dz dVe2dz d eKf
dXZ\dS )[ N)contextmanager)Path)urlsplit
urlunsplit)drop_privileges)docroot) HTTPAdapter)ChunkedEncodingError)
disable_quota)ReadTimeoutError)gettext)
exceptions
probe_targetsutils)canonical_variantsis_allowed_ini_path)verify_user_owns_domain
TIMEOUT _CONNECT_TIMEOUT _SWEEP_BUDGET_SECONDS _PIN_ADDRESSi _MAX_RESPONSE_BYTESi _CHUNK_BYTESz
User-AgentzjMozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9.2.13) Gecko/20101209 CentOS/3.6-2.el5.centos Firefox/3.6.13returnc @ t d t | z S )u The user-facing "website not accessible" text.
Shared by every terminal probe failure path so the message the tenant sees —
and its translation — is identical regardless of which failure produced it.
zUnable to detect php version for website because it is not accessible. Try again and contact an administrator if the issue persists. Original error: %s. )_str)errors R/opt/cloudlinux/venv/lib64/python3.11/site-packages/xray/internal/phpinfo_utils.py_unreachable_messager# G s+
E
c " e Zd ZdZ fdZ xZS )WebsiteNotRespondinguc Nothing usable answered the probe on this connect target and scheme.
``url`` and ``details`` drive the caller's scheme + connect-candidate
fallback. But this exception can also ESCAPE get_php_configuration — when
every candidate failed on both schemes — and the CLI then renders it with
``print(e)`` (console_utils/base_app.py), i.e. XRayError.__str__, which reads
attributes that ONLY XRayError.__init__ sets. So the base MUST be initialised:
without it `str(e)` raised AttributeError('type_flag') and the tenant got an
unhandled traceback instead of the error message.
The message is the same "website not accessible" text every other terminal
probe failure uses, so a caller that lets this escape produces exactly the
output an unreachable site produced before connect failures were routed
through this class.
c || _ || _ t t | d S N)urldetailssuper__init__r# )selfr) r* __class__s r" r, z WebsiteNotResponding.__init__f s8
-g6677777r$ )__name__
__module__ __qualname____doc__r,
__classcell__r. s @r" r& r& U sB 8 8 8 8 8 8 8 8 8r$ r& c e Zd ZdZdS )WebsiteTooSlowu A webserver accepted the connection on this address but did not finish
responding within the read timeout.
Kept distinct from WebsiteNotResponding (nothing answered at all) because the
two mean different things in a support log. It is NOT treated as terminal: a
timeout says a webserver answered, but says nothing about WHICH vhost did —
and on the very topology this module exists for, the loopback connection is
accepted by the address bucket's DEFAULT vhost. Aborting the sweep there would
reproduce the CLOS-6825 symptom whenever that default vhost is slow, so the
sweep advances and `_SWEEP_BUDGET_SECONDS` (not terminality) is what stops a
slow site from multiplying the stall across candidates.
Nr/ r0 r1 r2 r$ r" r6 r6 l r$ r6 >
phpversion include_path ini_scan_dir
php_sapi_namec e Zd ZdZdS )ProbeBodyNotUnderstoodu The fetch succeeded, but the body is not this probe's output.
A 200 does NOT prove the right vhost answered. On a per-IP-vhost host the
address bucket's default vhost may carry a front controller or catch-all that
serves its own page with status 200 instead of 404 — so the sweep must treat
an unparseable body as "this candidate failed" and keep going, or it stops on
a wrong-but-successful answer before ever reaching the account's real vhost
address (the CLOS-6825 symptom, from the other direction).
Raised instead of the bare ValueError/TypeError that `_parse_configuration`
used to leak: those are not XRayErrors, so the CLI could not render them.
Nr7 r8 r$ r" r? r? r9 r$ r? c e Zd ZdZdS )IniScanDirRejecteduF The probe's ini_scan_dir did not pass validation for the privileged write.
An XRayError so the CLI can RENDER it. This used to be a bare ValueError,
which `console_utils/base_app.py` neither prints nor reports: it catches only
XRayError, so a rejected ini_scan_dir surfaced as an unhandled traceback and
never reached `capture_exception`. That is what made CLOS-6825 look unfixed
after its connect-address fix shipped — the probe succeeded on the vhost's own
address, this gate discarded the result, and the newest renderable line left in
manager.log was the UNRELATED "failed against 127.0.0.1" INFO from the sweep's
first candidate. Support and the customer both read that as the original 404
bug persisting.
Deliberately XRayError and NOT XRayManagerError, even though every other error
on this path is the latter. `manager/cpanel.get_domain_info` wraps
`check_domain` — which is where `absolute_ini_scan_dir` is evaluated — in
`except XRayManagerError` and retries the whole thing against
`resolve_alias(domain)`. As an XRayManagerError this would be swallowed there:
the probe would re-run (a second temp-file write into a docroot plus a second
connect sweep) and the surfaced error would be the alias attempt's, not the
real cause — reintroducing the exact misdiagnosis this class exists to end.
XRayError renders identically (`XRayError.__str__` does the work; base_app
catches XRayManagerExit then XRayError) and the only other broad
`except XRayError`, in `manager/base.stop`, re-raises anything without the
`_internal_missing_domain` marker.
Still a ValueError as well: callers and tests that predate this class treat a
rejected ini_scan_dir as one, and `manager/base.get_ini_path` catches
ValueError around the neighbouring ini properties.
Nr7 r8 r$ r" rA rA s r$ rA usernamedetailc r t d| | t t d S )ug Log the technical detail, and return an error that does NOT carry it.
``detail`` keeps the exact wording these rejections have always had — a support
engineer greping manager.log for "outside allowed paths" or "not bound to user"
still finds it, and now finds it as a log line of its own rather than only
inside a traceback.
It deliberately does NOT reach the returned exception. ``ini_scan_dir`` is
attacker-controlled (the probe body is served from the tenant's own docroot, and
on a CGI/FastCGI SAPI PHP_INI_SCAN_DIR is reachable from the vhost's
environment), and the message is rendered to an unprivileged tenant by the panel
plugin. Echoing the resolved path back — and distinguishing "does not resolve to
an existing dir" from "outside allowed paths" from "not bound to user" — would
make the root daemon a path-existence oracle for arbitrary absolute paths. So
every reason yields ONE uniform message and the detail stays in the root-only
log. Same reasoning as the missing-vs-foreign domain unification in
`user_plugin_utils.user_mode_verification`.
z4Rejected probe-reported ini_scan_dir for user %s: %szUnable to detect php configuration for website because its PHP configuration directory is not supported. Contact an administrator if the issue persists. )loggerwarningrA r )rB rC s r" _reject_ini_scan_dirrG s@ & NNI8U[\\\
?
r$ c " e Zd ZdZdZdefdZdS )_BoundedResponsezDA drop-in replacement exposing `.text` from a bounded streamed read.textrK c || _ d S r( rJ )r- rK s r" r, z_BoundedResponse.__init__ s
r$ N)r/ r0 r1 r2 __slots__r r, r8 r$ r" rI rI s: NNIS r$ rI addressportc 0 d| v rd| dn| }||n| d| S )zGBuild the netloc for a pinned connect target, bracketing IPv6 literals.:[]r8 )rN rO hosts r" _pinned_netlocrU s7 G^^>w>>>>D<44%5%5t%5%55r$ c 6 e Zd ZdZefdef fd
Z fdZ xZS )_LocalAddressPinnedAdapteru$ requests adapter that pins every connection to one local address.
The vhost is hosted on this machine; we must not follow the
tenant-controlled public-DNS record. We rewrite the connect target to
``pin_address`` — loopback, or an address the caller has already verified is
configured on this machine — while keeping the original hostname in the Host
header so the local webserver routes to the correct vhost. The TLS SNI/cert
mismatch that results is tolerated because the caller already requests
verify=False.
pin_addressc H || _ t j |i | d S r( )rX r+ r, )r- rX argskwargsr. s r" r, z#_LocalAddressPinnedAdapter.__init__ s- &$)&)))))r$ c t |j }|j }|O|j |j d< t | t | j |j |_ t j |fi |S )NHost)netloc) r r) hostnamer^ headersr _replacerU rX rO r+ send)r- requestr[ split
original_hostr. s r" rb z _LocalAddressPinnedAdapter.send sy %%
$&+lGOF #$U^^>$JZ\a\f;g;g^%h%hiiGKuww|G..v...r$ ) r/ r0 r1 r2 r r r, rb r3 r4 s @r" rW rW sq +7 * *C * * * * * */ / / / / / / / /r$ rW rX c t j }t | }| d| | d| |S )Nhttp://https://)requestsSessionrW mount)rX sessionadapters r" _pinned_sessionrn sJ G(55G MM)W%%% MM*g&&& Nr$ read_timeoutc |t j vr%t j t d| d t | 5 }| | t |fdt dd 5 }|j sd|j
cxk rdk r n nt | d|j
| t }| t
D ]m}|st! | t! | z t" k r+t j t% d | t" fz | | n|j pd } | |d
}n'# t, $ r | d d
}Y nw xY wddd n # 1 swxY w Y ddd n # 1 swxY w Y n# t. j j t0 f$ rf} | j r sometimes error happens due to network issues/glitch
zrefusing to probe z*: not an address configured on this serverFT)timeoutverifyr` streamallow_redirectsi, i z probe returned redirect status )r* )
chunk_sizezVphpinfo response for %s exceeds %d bytes; refusing to buffer attacker-controlled body.zutf-8replace)errorsNr )!r local_addressesr XRayManagerErrorr# rn getr HEADERS is_redirect status_coder& raise_for_status bytearray iter_contentr lenr r extendencodingdecode LookupErrorri ConnectionErrorrZ
isinstancer r6 r ReadTimeoutRequestExceptionrI )
r) rX rp rl responsebufchunkr rK es
r" _request_urlr s -79999) !mk!m!m!m n n
NC
K ( (+ =,3 KK)<8 %
+ = #
rsh.B'H'H'H'HS'H'H'H'H'H +38pZbZn8p8pqqqq % % ' ' '++C!..,.GG
"
" s88c%jj(+>>>$5rss 345
5!!!! (3GH
=zz(9z==
=
=
= zz')z<<
=K+ = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + = + =X
/ A 8 8 8
6
:j,<==
: !!5a!8!899 9"3A7777 * 6 6 6 1! 4 4555 $ C C C)*>q*A*ABBBC D ! !!s G (G 3C*F) E65F)6!FF)FF) G )F- -G 0F- 1G 4 G GG GG J3'A!IJ3 I;;J3
!J..J3
document_rootc # p K d}dt j d d}||z }t | 5 t 5 | | d d d n # 1 swxY w Y d d d n # 1 swxY w Y |V | d S # | w xY w)Na
d| j v r
d| j S | | j S )Nz/opt/altzalt-php)r; r )r- r s r" get_full_php_versionz%PhpConfiguration.get_full_php_version s5 * * *5T 355 5 :$"8:::r$ c d| j v rt | j | j }n| j }|r t t | d }n-# t t f$ r t | j d| w xY wt | st | j d| t | j || j s t | j d| j d| |S )Nz link/confTstrictz2ini_scan_dir does not resolve to an existing dir: z$ini_scan_dir outside allowed paths: z ini_scan_dir not bound to user z: ) r< _resolve_ini_path_in_cagefsrB r r resolveOSError RuntimeErrorrG r _ini_scan_dir_belongs_to_userr )r- resolveds r" absolute_ini_scan_dirz&PhpConfiguration.absolute_ini_scan_dir s$ $+ + +24=$BSTTHH(H
tH~~ 5 5T 5 B BCC\*
*M#dX`#d#d
'x00
o*4=:mai:m:mnnn 1 $Jbcc
*M#dT]#d#dX`#d#d s 0A *B
c | j dk S )Nzfpm-fcgi)r= r s r"
is_php_fpmzPhpConfiguration.is_php_fpm s !Z//r$ ) r/ r0 r1 r __annotations__r propertyr r r r r8 r$ r" r r s MMMOOO '+t***
73 7 7 7 X7;3 ; ; ; ;
$ $ X$L 0 0 X0 0 0r$ r r r c
i }| d D ]R}| d \ }}}|s | }|t v r| ||< St | z
}|r?t t
dd t | t d| |d|S )u Parse the probe payload, or raise ProbeBodyNotUnderstood.
Deliberately lenient per line and strict overall: unusable lines are skipped
and only `_REQUIRED_PROBE_KEYS` decides whether this really is our payload.
That single gate replaces three ways this used to leak a non-XRayError the CLI
could not render — `line.split('=')` raising ValueError on a line with no `=`
(a default vhost's HTML) or with two (`include_path=.:/a=b`), and
`PhpConfiguration(**config)` raising TypeError on an unexpected key.
=zprobe response is missing z, )rB r r8 )
rd partitionstrip_REQUIRED_PROBE_KEYSkeysr? r# r sortedr ) rB r r configlinekey separatorvaluemissings r" _parse_configurationr s Ft$$
(
( $s 3 3 Y iikk & & & ++--F3K"V[[]]2G v$%9:sW[W`W`aghoapapWqWq:s:s%t%tuuu aXCV a aZ` a aar$ /var/cagefs/)z /etc/users/z/etc/configs/users/z^ea-php(?P\d{2})$z$/opt/cpanel/ea-php{v}/root/etc/php.dz^alt-php(?P\d{2})$z/opt/alt/php{v}/link/confz!^plesk-php(?P\d)(?P\d)$z$/opt/plesk/php/{maj}.{min}/etc/php.dz^php(?P\d{2})$z /usr/local/php{v}/lib/php.conf.dtrusted_handlerc | sdS t D ]J\ }}| t | }|r! |j di | c S KdS )u Build the single global INI_LOCATION for a TRUSTED, FULL PHP handler.
``trusted_handler`` is the account's REAL provisioned handler carrying its
distro family ('ea-php80', 'alt-php80', 'plesk-php80', 'php80'). Returns the
ONE server-wide per-distro+version dir that handler's PHP scans, keyed BY the
distro family — so the derived dir's family always equals the trusted
handler's family, never the family of whatever dir the probe reported.
Returns None when ``trusted_handler`` is empty or carries no distro family
(a bare number such as custom's '80'): the expected dir cannot be uniquely
derived, so the caller default-denies the global per-version accept branch.
Nr8 ) _HANDLER_INI_LOCATION_TEMPLATESmatchr format groupdict)r patterntemplater s r" _expected_ini_locationr 9 sx t < 8 8
c/ 2 233 8"8?77U__%6%677 7 7 7 8 4r$ rootr c t | D ]T}| d }||k r dS | |dz r |t | dz d c S UdS )a ``resolved`` relative to ``root`` ('' for root itself), or None if outside.
The trailing-separator boundary stops /var/cagefs/01/aliceEVIL from being
treated as inside alice's /var/cagefs/01/alice subtree.
``resolved`` is a realpath, so ``root`` is matched in its canonical spelling
as well as the literal one: on a host where /var/cagefs is a symlink (the
documented relocation layout) the tenant's dir resolves to /home/cagefs/...,
which the literal root would miss (CLOS-7045).
/r N)r rstrip
startswithr )r r candidates r" _relative_to_rootr O s (-- 2 2 $$S)) y 22 y3 / / 2C NNQ . 0 01 1 1 1 2 4r$ c & t | | duS )zETrue if ``resolved`` is ``root`` itself or a path strictly inside it.N)r )r r s r"
_is_withinr c s T8 , ,D 88r$ c t D ]6}t || }|"| d d | k r dS 7dS )u{ True if ``resolved`` is inside ``username``'s OWN per-domain ini scan dir.
One helper for every gate that has to answer this question — the phpinfo
binding gate below, `manager/custom.py._validate_ini_path` and
`types.Task._ini_dir_validator`'s write-time inode binding. They each used to
inline the same `/etc/users/` containment check, which is precisely how
/etc/configs/users ended up supported by none of them: an operator layout has
to be added in one place, not three. Compare the first relative path segment
to ``username``; never append caller data to a root before canonicalizing it.
Nr r TF)_PER_DOMAIN_INI_ROOTSr r )rB r r rels r" _is_own_per_domain_ini_dirr h sO & h// ?s}}S 1 1! 4 @ @44 5r$ c D t fdt D S )aH True if ``resolved`` is inside ANY per-domain ini scan root.
Recognises "this is some tenant's per-domain dir" without saying whose. A path
here that is not the owner's own subtree belongs to a FOREIGN tenant and must
be rejected outright, never fall through to a root-ownership check that would
accept it.
c 3 8 K | ]}t | V d S r( )r ).0r r s r" z0_is_under_per_domain_ini_root.. s- LLdz$))LLLLLLr$ )anyr )r s `r" _is_under_per_domain_ini_rootr z s) LLLL6KLLL L LLr$ c t | | rdS t | | rdS t | }|R t t | d }n# t t f$ r Y dS w xY w||k rdS dS )u Confirm a resolved (realpath'd) ini_scan_dir is one the tenant may drive.
``resolved`` has already been realpath'd and prefix-checked against
ALLOWED_INI_PREFIXES. That prefix list is intentionally broad and admits
global root-owned dirs, so this is the default-deny binding gate: accept
ONLY paths provably tied to the principal, reject everything else.
In phpinfo mode the legitimate ini_scan_dir is one of:
* the tenant's own CageFS subtree (the link/conf branch is re-rooted there
by _resolve_ini_path_in_cagefs(self.username, ...));
* the tenant's own per-domain ini dir, //... for any
root in _PER_DOMAIN_INI_ROOTS (cPanel's /etc/users, DirectAdmin's
/etc/configs/users);
* for a NON-CageFS tenant (ea-php / Plesk / native), the global, root-owned
per-distro+version INI_LOCATION for the account's PHP handler — e.g.
/opt/cpanel/ea-php80/root/etc/php.d (see xray-ini-placement.md). This is
legitimate, so rejecting it outright would regress those tenants.
The danger is that ``ini_scan_dir`` AND the probe ``phpversion`` are both
served from the tenant's own docroot, so a tenant could report a global dir
its account does NOT actually run. A bare two-digit version compare is not
enough: it is distro-BLIND, so an ea-php80 tenant could report the same-number
alt-php dir /opt/alt/php80/link/conf and steer the root xray.ini write into a
DIFFERENT distro family's server-wide, all-tenant-shared dir (cross-distro
spoof) — on a mixed alt-php+ea-php server that dir exists, so strict resolve
passes. To close that AND the cross-version spoof, the global branch accepts
ONLY the EXACT INI_LOCATION derived from ``trusted_php_version`` — here the
account's REAL provisioned FULL handler carrying its distro family (cPanel
domain['version'] 'ea-php80', Plesk resolve_lsphp_version(handler), DA
opts[...]['ver'] 'php80'), NOT the probe-reported version and NOT a
distro-blind number — and requires ``resolved`` to equal it.
Accepted:
* the tenant's OWN CageFS subtree (/var/cagefs///...);
* the tenant's OWN per-domain ini dir (//..., any root in
_PER_DOMAIN_INI_ROOTS);
* the SINGLE INI_LOCATION == _expected_ini_location(trusted_php_version).
Rejected (default-deny): global config dirs with no version segment
(/etc/php.d, /etc/php.scan.d, /etc/cl.php.d, /usr/share/cagefs[-skeleton]
roots), a per-version dir of a DIFFERENT distro family (cross-distro spoof) or
a DIFFERENT version (cross-version spoof) than the trusted handler, a trusted
handler that carries no distro family (a bare number — expected dir cannot be
derived), another tenant's CageFS subtree, another account's
per-domain ini dir under another account's // subtree, and any
other dir not matched above.
TNr F)_is_own_cagefs_subtreer r r r r r r )rB r r expecteds r" r r s b h11 t "(H55 t &&9::H 4>>111>>??HH & 55 x 4 5s 0A( (A=<A=c F t t | }|dS | d } t j | }n# t
$ r d}Y nw xY w|r+t
| dk o|d |k o |d | k S t
| dk o |d | k S )a True if ``resolved`` is inside the tenant's OWN /var/cagefs subtree.
Prefer the exact /var/cagefs// root when the server can
supply the CageFS prefix (cagefsctl). When the prefix is unavailable, fall
back to a structural check that still binds to the principal: the path must
be /var/cagefs//[/...], i.e. ``username`` must occupy the
per-user segment directly under a single shard directory. This never accepts
another tenant's subtree (the username segment is pinned and realpath has
already been applied) and avoids failing closed on a legitimate own-subtree
path when a transient cagefsctl lookup fails. Note such a /var/cagefs path
only reaches here when the prefix WAS resolvable in
_resolve_ini_path_in_cagefs, so the fallback covers a mid-request race, not
a routine "no prefix" case. Both the prefix and username are compared as
relative path segments so neither can alter the canonicalized root.
NFr r r r )r _CAGEFS_ROOTrd r cagefsctl_get_prefix Exceptionr )rB r r partsprefixs r" r r s L(
3
3C
{u IIcNNE+H55
O 5zzQN58v#5N%(h:NN u::? 3uQx8 33s A AApathc t j | n# t $ r Y dS w xY wd| d|g} t j |dt j }n# t j $ r Y dS w xY w| d rhttps answers the http probe with a 3xx (surfaced as
WebsiteNotResponding and never followed — F-01), so the probe is re-issued
directly over https against the SAME address.
Parsing happens HERE, inside the candidate loop, not after it. A 200 is not
proof that the account's vhost answered, so a body that is not our payload has
to count as this candidate failing — otherwise a default vhost that serves its
own page with status 200 ends the sweep before the right address is tried.
rg rh )r )r r& r rK )rB r rX r rp r s r" _probe_one_candidater sy &b ,D DkS_`` b b b
-E E{T`aab (-M` a a aas 88domain
panel_addressc t j | || }t d|| t j t z }d}|D ]}|t j z
} | dk r#t d|t n t | |||t t | }
t d|||
j |
j |
c S # t j $ r+} t d||| || }Y d} ~ d} ~ ww xY w||t j t d t d
) u Fetch and parse the probe file, trying each validated connect target in turn.
A failure against one address says nothing about the others. The defining
symptom of a per-IP-vhost host (CLOS-6825) is that the request DOES reach the
local webserver, but on an address whose bucket holds no matching vhost, so
that bucket's default vhost answers from the server's own DocumentRoot — with
a 404, or with its own page if it has a catch-all. An error status, a refused
connection, a read timeout and an unparseable body therefore all advance to
the next candidate: none of them tells us which vhost answered.
Two things bound the cost of that, since nothing is terminal any more:
* `_CONNECT_TIMEOUT` caps the connect phase per attempt — every candidate is
on-box, so a slow connect is a firewall DROP, not a slow site;
* `_SWEEP_BUDGET_SECONDS` caps the whole sweep's wall-clock, shrinking each
attempt's read budget to what is left. The temp probe file lives in the
tenant's docroot for this whole window, so the ceiling matters.
When no candidate serves the probe the FIRST error is re-raised, not the last:
loopback is the canonical attempt, so its error is the one a support ticket
can recognise. Surfacing the last one would report, say, "connection refused"
against some unrelated interface address on a multi-IP host.
)r z3phpinfo probe for %s will try connect candidates %sNr z@phpinfo probe for %s gave up after %ds; candidates left unprobed)rp zBphpinfo probe for %s succeeded against %s: php %s, ini_scan_dir %sz*phpinfo probe for %s failed against %s: %sz,no connect candidate could be probed within s)r connect_candidatesrE infotime monotonicr rF r minr r: r< r rz r# ) rB r r r r
candidatesdeadline first_errorrX remaining
configuration probe_errors r" _probe_candidatesr 4 s B 1(FR_```J KKEvzZZZ~ "77HK! ! ! t~ / // >> NNR%
E ! 0(# ) 4 4 M
KKT ( *
! * * * * KKDfk[f g g g" ) *
% dLa d d dee s 'C((D"7!D D"c 4 t || t j t j j t | t
t | d 5 }|dz |z }t | |||| cddd S # 1 swxY w Y dS )ac
Writes temporary phpinfo-like file to document root
and executes request to website to retrieve the current
php version and configuration
``trusted_php_version`` is the account's REAL provisioned PHP version as
known to the panel (server-side, not the probe). The caller passes it so a
global per-version ini_scan_dir can be version-bound in
absolute_ini_scan_dir; None disables that accept branch for the panel.
``panel_address`` is the address the panel bound THIS domain's vhost to, when
the caller can resolve it better than `probe_targets` can. DirectAdmin needs
this: it records the address in the PARENT domain's conf, and only the manager
knows that mapping for a subdomain or an alias. Still validated against this
machine's own addresses before use; None falls back to `probe_targets`' own
lookup.
r r )r r N) r urllib3disable_warningsr InsecureRequestWarningr r r r )rB r r r
php_info_filer s r" get_php_configurationr s > FH--- W / FGGG 40B+C+C D D
#)C<-#? ! $ 3'
s $ B
BBr( )NN)] dataclassesloggingr rer r r
contextlibr pathlibr urllib.parser r ri r
clcommon.clpwdr clcommon.cpapir requests.adaptersr requests.exceptionsr
secureior urllib3.exceptionsr xrayr
r
xray.internalr r r xray.internal.constantsr r xray.internal.user_plugin_utilsr getLoggerr/ rE r intr r r LOOPBACK_ADDRESSr r r r r| r# rz r& r6 frozensetr r? XRayErrorr rA rG rI rU rW rj rn retry_on_exceptionsfloatr r dataclassr r r r compiler r r boolr r r r r r r r r r8 r$ r" r, s
% % % % % % - - - - - - - - * * * * * * " " " " " " ) ) ) ) ) ) 4 4 4 4 4 4 " " " " " " / / / / / / : : : : : : : : : : K K K K K K K K C C C C C C 8 $ $ #
s "2 c 2 2 2
' S & & & c < 3 8 8 8 8 8:6 8 8 8. Z0 " !y!`!`!`aa Z 8 -z @3 8J : 6C 6sTz 6c 6 6 6 6 / / / / / / / /2 (4 8H 1 3 455)5W c" c"3 c"U c" c" c" 65c"L c $ 8
I0 I0 I0 I0 I0 I0 I0 I0Xb b3 b# bCRVJ bbr b b b b@ P ? 4 RZ'((*PQRZ())+FGRZ4557]^RZ$%%'IJ # C$J 3: ,C 3 3: (9S 9C 9D 9 9 9 9
$MC MD M M M MM MC M3 MUX[_U_ Mko M M M M`!4S !4C !4D !4 !4 !4 !4H# S 8 b b ! b b t b
b
b b b bF '+ $ S SS S "S t S
: S
S S S Sr '+ $ /
/
/
/
t/
: /
/
/
/
/
/
/
r$