No boundaries: Exfiltration of non-public knowledge by session-replay scripts

That is the primary submit in our “No Boundaries” sequence, wherein we reveal how third-party scripts on web sites have been extracting private info in more and more intrusive methods. [0]
by Steven Englehardt, Gunes Acar, and Arvind Narayanan

You might know that almost all web sites have third-party analytics scripts that file which pages you go to and the searches you make.  However currently, increasingly more websites use “session replay” scripts. These scripts file your keystrokes, mouse actions, and scrolling conduct, together with your entire contents of the pages you go to, and ship them to third-party servers. In contrast to typical analytics providers that present mixture statistics, these scripts are supposed for the recording and playback of particular person shopping classes, as if somebody is trying over your shoulder.

The acknowledged function of this knowledge assortment contains gathering insights into how customers work together with web sites and discovering damaged or complicated pages. Nevertheless the extent of knowledge collected by these providers far exceeds person expectations [1]; textual content typed into varieties is collected earlier than the person submits the shape, and exact mouse actions are saved, all with none visible indication to the person. This knowledge can’t fairly be anticipated to be saved nameless. The truth is, some corporations permit publishers to explicitly hyperlink recordings to a person’s actual identification.

For this examine we analyzed seven of the highest session replay corporations (primarily based on their relative recognition in our measurements [2]). The providers studied are Yandex, FullStory, Hotjar, UserReplay, Smartlook, Clicktale, and SessionCam. We discovered these providers in use on 482 of the Alexa prime 50,000 websites.


This video exhibits the co-browse” function of 1 firm, the place the writer can watch person classes dwell.

What can go unsuitable? Briefly, quite a bit.

Assortment of web page content material by third-party replay scripts could trigger delicate info comparable to medical situations, bank card particulars and different private info displayed on a web page to leak to the third-party as a part of the recording. This will likely expose customers to identification theft, on-line scams, and different undesirable conduct. The identical is true for the gathering of person inputs throughout checkout and registration processes.

The replay providers supply a mixture of handbook and computerized redaction instruments that permit publishers to exclude delicate info from recordings. Nevertheless, to ensure that leaks to be prevented, publishers would want to diligently verify and scrub all pages which show or settle for person info. For dynamically generated websites, this course of would contain inspecting the underlying net software’s server-side code. Additional, this course of would have to be repeated each time a website is up to date or the online software that powers the location is modified.

An intensive redaction course of is definitely a requirement for a number of of the recording providers, which explicitly forbid the gathering of person knowledge. This negates the core premise of those session replay scripts, who market themselves as plug and play. For instance, Hotjar’s homepage advertises: “Arrange Hotjar with one script in a matter of seconds” and Smartlook’s sign-up process options their script tag subsequent to a timer with the tagline “each minute you lose is a number of video”.

To raised perceive the effectiveness of those redaction practices, we arrange check pages and put in replay scripts from six of the seven corporations [3]. From the outcomes of those assessments, in addition to an evaluation of various dwell websites, we spotlight 4 sorts of vulnerabilities beneath:

1. Passwords are included in session recordings. All the providers studied try to stop password leaks by robotically excluding password enter fields from recordings. Nevertheless, mobile-friendly login packing containers that use textual content inputs to retailer unmasked passwords usually are not redacted by this rule, except the writer manually provides redaction tags to exclude them. We discovered not less than one web site the place the password entered right into a registration type leaked to SessionCam, even when the shape isn’t submitted.

2. Delicate person inputs are redacted in a partial and imperfect manner. As customers work together with a website they’ll present delicate knowledge throughout account creation, whereas making a purchase order, or whereas looking the location. Session recording scripts can use keystroke or enter component loggers to gather this knowledge.

All the corporations studied supply some mitigation by means of automated redaction, however the protection supplied varies significantly by supplier. UserReplay and SessionCam change all person enter with an equal size masking textual content, whereas FullStory, Hotjar, and Smartlook exclude particular enter fields by kind. We summarize the redaction of different fields within the desk beneath.

summary of automated redaction features offered by each service

Abstract of the automated redaction options for type inputs enabled by default from every firm.
Stuffed circle: Knowledge is excluded; Half-filled circle: equal size masking; Empty circle: Knowledge is distributed within the clear
* UserReplay sends the final four digits of the bank card area in plain textual content
† Hotjar masks the road deal with portion of the deal with area.

Automated redaction is imperfect; fields are redacted by enter component kind or heuristics, which can not at all times match the implementation utilized by publishers. For instance, FullStory redacts bank card fields with the `autocomplete` attribute set to `cc-number`, however will gather any bank card numbers included in varieties with out this attribute.

Credit card data leaking on Bonobos checkout page

To complement automated redaction, a number of of the session recording corporations, together with Smartlook, Yandex, FullStory, SessionCam, and Hotjar permit websites to additional specify inputs components to be excluded from the recording. To successfully deploy these mitigations a writer might want to actively audit each enter component to find out if it accommodates private knowledge. That is difficult, error susceptible and dear, particularly as a website or the underlying net software code modifications over time. For example, the monetary service website constancy.com has a number of redaction guidelines for Clicktale that contain nested tables and little one components referenced by their index. Within the subsequent part we additional discover these challenges.

A safer strategy could be to masks or redact all inputs by default, as is finished by UserReplay and SessionCam, and permit whitelisting of known-safe values. Even absolutely masked inputs present imperfect safety. For instance, the masking utilized by UserReplay and Smartlook leaks the size of the person’s password

three. Guide redaction of personally figuring out info displayed on a web page is a basically insecure mannequin. Along with gathering person inputs, the session recording corporations additionally gather rendered web page content material. In contrast to person enter recording, not one of the corporations seem to offer automated redaction of displayed content material by default; all displayed content material in our assessments ended up leaking.

As a substitute, session recording corporations count on websites to manually label all personally figuring out info included in a rendered web page. Delicate person knowledge has various avenues to finish up in recordings, and small leaks over a number of pages can result in a big accumulation of non-public knowledge in a single session recording.

For recordings to be fully free of non-public info, a website’s net software builders would want to work with the location’s advertising and analytics groups to iteratively scrub personally figuring out info from recordings because it’s found. Any change to the location design, comparable to a change within the class attribute of a component containing delicate info or a choice to load non-public knowledge into a special kind of component requires a overview of the redaction guidelines.

As a case examine, we look at the pharmacy part of Walgreens.com, which embeds FullStory. Walgreens makes in depth use of handbook redaction for each displayed and enter knowledge. Regardless of this, we discover that delicate info together with medical situations and prescriptions are leaked to FullStory alongside the names of customers.

Walgreens prescription request page leaks prescription information

Walgreens health history page leaks health conditions

Walgreens identity verification page leaks answers to questions

We don’t current the above examples to level fingers at a sure web site. As a substitute, we intention to indicate that the redaction course of can fail even for a big writer with a robust, authorized incentive to guard person knowledge. We noticed comparable private info leaks on different web sites, together with on the checkout pages of Lenovo [5]. Websites with much less sources or much less experience are much more more likely to fail.

four. Recording providers could fail to guard person knowledge. Recording providers enhance the publicity to knowledge breaches, as private knowledge will inevitably find yourself in recordings. These providers should deal with recording knowledge with the identical safety practices with which a writer could be anticipated to deal with person knowledge.

We offer a selected instance of how recording providers can fail to take action. As soon as a session recording is full, publishers can overview it utilizing a dashboard supplied by the recording service. The writer dashboards for Yandex, Hotjar, and Smartlook all ship playbacks inside an HTTP web page, even for recordings which occur on HTTPS pages. This enables an energetic man-in-the-middle to injecting a script into the playback web page and extract all the recording knowledge. Worse but, Yandex and Hotjar ship the writer web page content material over HTTP — knowledge that was beforehand protected by HTTPS is now susceptible to passive community surveillance.

The vulnerabilities we spotlight above are inherent to full-page session recording. That’s to not say the particular examples can’t be mounted — certainly, the publishers we examined can patch their leaks of person knowledge and passwords. The recording providers can all use HTTPS throughout playbacks. However so long as the safety of person knowledge depends on publishers absolutely redacting their websites, these underlying vulnerabilities will live on.

Does monitoring safety assist?

Two generally used ad-blocking lists EasyList and EasyPrivacy don’t block FullStory, Smartlook, or UserReplay scripts. EasyPrivacy has filter guidelines that block Yandex, Hotjar, ClickTale and SessionCam.

At the very least one of many 5 corporations we studied (UserReplay) permits publishers to disable knowledge assortment from customers who’ve Do Not Observe (DNT) set of their browsers. We scanned the configuration settings of the Alexa prime 1 million publishers utilizing UserReplay on their homepages, and located that none of them selected to honor the DNT sign.

Enhancing person expertise is a crucial process for publishers. Nevertheless it shouldn’t come on the expense of person privateness.


Finish notes:

[0] We use the time period ‘exfiltrate’ on this sequence to consult with the third-party knowledge assortment that we examine. The time period ‘leakage’ is usually used, however we eschew it, as a result of it suggests an unintended assortment ensuing from a bug. Quite, our analysis means that whereas not essentially malicious, the gathering of delicate private knowledge by the third events that we examine is inherent of their operation and is well-known to most if not all of those entities. Additional, there is a component of furtiveness; these knowledge flows usually are not public information and neither publishers nor third events usually are not clear about them.

[1] A current evaluation of the corporate Navistone, accomplished by Hill and Mattu for Gizmodo, explores how knowledge assortment previous to type submission exceeds person expectations. On this examine, we present how analytics corporations gather much more person knowledge with minimal disclosure to the person. The truth is, some providers counsel the primary social gathering websites merely embrace a disclaimer of their website’s privateness coverage or phrases of service.

[2] We used OpenWPM to crawl the Alexa prime 50,000 websites, visiting the homepage and 5 further inside pages on every website. We use a two-step strategy to detect analytics providers which gather web page content material.

First, we inject a singular worth into the HTML of the web page and seek for proof of that worth being despatched to a 3rd social gathering within the web page site visitors. To detect values that could be encoded or hashed we use a detection methodology much like earlier work on e mail monitoring. After filtering out leak recipients, we isolate pages on which not less than one third social gathering receives a considerable amount of knowledge through the go to, however for which we don’t detect a singular ID. On these websites, we carry out a follow-up crawl which injects a 200KB chunk of knowledge into the web page and verify if we observe a corresponding bump within the measurement of the info despatched to the third social gathering.

We discovered 482 websites on which both the distinctive marker was leaked to a set endpoint from one of many providers or on which we noticed an information assortment enhance roughly equal to the compressed size of the injected chunk. We consider this worth is a decrease certain since lots of the recording providers supply the power to pattern web page visits, which is compounded by our two-step methodology.

[3] One firm (Clicktale) was excluded as a result of we have been unable to make the sensible preparations to investigate script’s performance at scale.

[4] FullStory’s phrases and situations explicitly classify well being or medical info, or some other info lined by HIPAA as delicate knowledge and asks clients to “not present any Delicate Knowledge to FullStory.”

[5] Lenovo.com is one other instance of a website which leaks person knowledge in session recordings.

Lenovo's checkout process leaks shipping and payment information.

[6] We used the default scripts obtainable to new accounts for five of the 6 suppliers. For UserReplay, we used a script taken from a dwell website and verified that the configuration choices match the commonest choices discovered on the internet.

Powered by WPeMatico

I by no means signed up for this! Privateness implications of e-mail monitoring

On this put up I focus on a new paper that may seem at PETS 2018, authored on my own, Jeffrey Han, and Arvind Narayanan.

What occurs once you open an e-mail and permit it to show embedded pictures and pixels? Chances are you’ll count on the sender to study that you just’ve learn the e-mail, and which gadget you used to learn it. However in a new paper we discover that privateness dangers of e-mail monitoring lengthen far past senders realizing when emails are seen. Opening an e-mail can set off requests to tens of third events, and lots of of those requests include your e-mail deal with. This permits these third events to trace you throughout the net and join your on-line actions to your e-mail deal with, quite than simply to a pseudonymous cookie.

Illustrative instance. Think about an e-mail from the offers web site LivingSocial (see particulars of the instance e-mail). When the e-mail is opened, consumer will make requests to 24 third events throughout 29 third-party domains.[1] A complete of 10 third events obtain an MD5 hash of the consumer’s e-mail deal with, together with main information brokers Datalogix and Acxiom. Almost the entire third events (22 of the 24) set or obtain cookies with their requests. In a webmail consumer the cookies are the identical browser cookies used to trace customers on the net, and certainly many main internet trackers (together with domains belonging to Google, comScore, Adobe, and AOL) are loaded when the e-mail is opened. Whereas this instance e-mail has numerous trackers relative to the typical e-mail in our corpus, nearly all of emails (70%) embed a minimum of one tracker.

The way it works. E-mail monitoring is feasible as a result of trendy graphical e-mail purchasers enable rendering a subset of HTML. JavaScript is invariably stripped, however embedded pictures and stylesheets are allowed. These are downloaded and rendered by the e-mail consumer when the consumer views the e-mail.[2] Crucially, many e-mail purchasers, and virtually all internet browsers, within the case of webmail, ship third-party cookies with these requests. The e-mail deal with is leaked by being encoded as a parameter into these third-party URLs.

Diagram showing the process of tracking with email address

When the consumer opens the e-mail, a monitoring pixel from “tracker.com” is loaded. The consumer’s e-mail deal with is included as a parameter inside the pixel’s URL. The e-mail consumer here’s a internet browser, so it mechanically sends the monitoring cookies for “tracker.com” together with the request. This permits the tracker to create a hyperlink between the consumer’s cookie and her e-mail deal with. Later, when the consumer browses a information web site, the browser sends the identical cookie, and thus the brand new exercise may be related again to the e-mail deal with. E-mail addresses are usually distinctive and chronic identifiers. So email-based monitoring can be utilized for concentrating on on-line advertisements based mostly on offline exercise (say, to buyers who used a loyalty card linked to an e-mail deal with) and for linking completely different gadgets belonging to the identical consumer.

Measuring e-mail monitoring at scale. To grasp the privateness implications of viewing and interacting with emails we assembled a group of messages from mailing lists on the highest websites.[3] Utilizing OpenWPM, an online measurement platform developed at Princeton, we simulated a consumer opening every e-mail and clicking hyperlinks from inside a webmail consumer that masses distant content material.  We discovered that 85% of emails in our corpus include embedded third-party content material, and 70% include sources categorized as trackers by widespread tracking-protection lists. Many of those third events, together with 7 of the highest 10, even have a big internet presence.

When “nameless” internet monitoring isn’t. About 29% of emails leak the consumer’s e-mail deal with to a minimum of one third social gathering when the e-mail is opened, and about 19% of senders despatched a minimum of one e-mail that had such a leak. Nearly all of these leaks (62%) are intentional.[4] If the leaked e-mail deal with is related to a monitoring cookie, as it will be in lots of webmail purchasers, the privateness danger to customers is drastically amplified. Since a monitoring cookie may be shared with conventional internet trackers, e-mail deal with can enable these trackers to hyperlink monitoring profiles from earlier than and after a consumer clears their cookies. If a consumer reads their e-mail on a number of gadgets, trackers can use that deal with as an identifier to hyperlink monitoring information cross-device.

A lot of the prime leak recipients, together with LiveIntent, Acxiom, Conversant Media, and Neustar, are concerned in “people-based” advertising and marketing. These third events obtain leaked e-mail addresses from between 24 to 68 of the 902 e-mail senders studied. Individuals-based advertising and marketing is outlined by Acxiom as “the flexibility to carry out concentrating on and measurement on the stage of actual individuals, not simply gadgets, by resolving id throughout digital and offline channels.” In different phrases, it’s a time period used to explain a set of companies which permit entrepreneurs to make use of monitoring information collected throughout any of a consumer’s gadgets, in addition to offline information, to focus on that consumer on any of their gadgets. As mentioned above, this might embrace offline information similar to purchases made utilizing a loyalty card at a grocery retailer, if that information is out there related to the purchaser’s e-mail deal with (or a hash of it).

Whereas our information doesn’t allow us to measure how the businesses use leaked e-mail addresses they obtain when a consumer views an e-mail, we will get some perception into potential makes use of by analyzing their product pages. The advertising and marketing supplies and privateness insurance policies of the 4 corporations talked about above element their use of e-mail addresses for cross-device concentrating on and/or information onboarding merchandise.[5]

Are leaks of hashed e-mail addresses much less of a privateness concern? In lots of instances the leaked e-mail deal with is hashed; in truth, 68% of all leaks which happen whereas viewing emails are hashed, one-third of which additionally embrace the area portion of the e-mail deal with in plaintext. Hashed e-mail is taken into account by some leak recipients to not be personally figuring out data.[6]

From a pc science perspective, the declare hashed e-mail deal with is just not personally figuring out is patently false. When consumer information in a database are keyed by hashed e-mail deal with, trying up the file for a given e-mail deal with is trivial: merely hash it first and look it up (certainly, that is the entire level of storing hashed e-mail addresses in any respect). What when you’ve got information related to a hash of an unknown e-mail deal with and wish to recuperate the unique deal with? It’s surprisingly straightforward: you’ll be able to lease a multi-GPU digital machine for $14.40 an hour[7] , which supplies you 73 billion MD5 hash computations per second based mostly on revealed benchmarks. Trendy strategies have gotten actually good at enumerating believable sequences of characters and numbers in passwords, and we imagine these strategies will lengthen to e-mail addresses. In the event that they do, it will imply that e-mail deal with hashes may be damaged far more effectively than by means of brute forcing (i.e., attempting all potential mixtures of characters). We posit that with a trillion guesses — a value of 6 US cents — it ought to be potential to enumerate nearly all of e-mail deal with in use.

Extra leaks happen when customers click on on hyperlinks in emails. When an e-mail hyperlink is clicked the URL is usually handed over to the consumer’s browser, or to a brand new tab within the consumer’s browser, within the case of webmail. E-mail addresses and different identifiers could also be embedded in these hyperlinks, and will in the end trigger the consumer’s e-mail deal with to leak to third-parties on the net. We discovered that about 11% of hyperlinks include requests that leak the consumer’s e-mail deal with to a third-party and about 12% of all emails include such a hyperlink. The biggest recipients of those leaks are Google, Fb, and Twitter, and the highest recipients total are similar to the prime third-party trackers on the net.

Leaks in hyperlink clicks may enable e-mail trackers to work round privateness protections in emails purchasers that strip cookies from distant sources (like Apple Mail) or in people who proxy distant sources (like Gmail). For the reason that clicked hyperlink is opened within the consumer’s browser, the tracker could make the express hyperlink between the consumer’s cookie and the leaked e-mail deal with whereas the ensuing web page is loaded.

What can customers do? The entire privateness dangers mentioned in our paper stem from distant sources, so customers can use mail purchasers which help blocking pictures by default to fully keep away from the issue. Nevertheless, that may typically lead to emails that are unreadable; that is notably true for advertising and marketing emails.

A diagram with a mail client rendering all-image email with images disabledBlocking pictures by default supplies full safety from monitoring when emails are seen, however can typically lead to unreadable emails.

In Part 6.2 of the paper we survey 16 mail purchasers and discover patchwork of privateness options are employed, however that no setup presents full safety from the threats we determine. Mail purchasers that block cookies by default, like Apple Mail, supply some stage of safety. In these purchasers it’s harder for a tracker to trace customers throughout mailing lists, because the mail consumer doesn’t present a persistent identifier. The identical is true for webmail purchasers which proxy pictures, like Gmail and Yandex. Content material proxying has the additional benefit of stopping a tracker from having the ability to hyperlink the browser’s cookies to any identifiers acquired when an e-mail is opened.

Even with the defenses employed by the purchasers we studied, trackers which obtain the consumer’s leaked e-mail deal with will proceed to have the ability to monitor and goal customers in these purchasers and on the net. For example, LiveIntent’s advertising and marketing materials reassures purchasers that it’s going to proceed to work in Gmail since “concentrating on is based totally across the e-mail deal with’s [sic] MD5 hash”. Whatever the defenses deployed by the consumer, management of monitoring is handed off to the consumer’s browser when e-mail hyperlinks are clicked.

We discovered that the monitoring safety lists EasyList and EasyPrivacy scale back the variety of e-mail leaks that happen when an e-mail is seen by 87%. Maybe the best choice for privacy-conscious customers in the present day is to make use of webmail and set up monitoring safety instruments, similar to uBlock Origin or Ghostery. Customers who wish to use a standalone consumer should discover one which helps privateness extensions; of the purchasers we studied, the one one which helps such extensions is Thunderbird. Having monitoring safety instruments put in within the browser can even present safety when e-mail hyperlinks are clicked. In Part 7 of the paper we prototyped a server-side filtering function which makes use of the monitoring safety lists to filter the HTML physique of emails earlier than they attain the consumer. We discovered it to be almost as efficient as a monitoring blocker working within the consumer’s browser.

Information, code, and paper launch

You may learn the paper right here. We’re additionally releasing the code and information publicly, together with the the entire uncooked and parsed e-mail our bodies and crawls of all HTML emails. We hope that this dataset will spur further analysis on this space.

Eager about listening to extra from me? Comply with me on Twitter @s_englehardt.

Because of Arvind Narayanan and Gunes Acar for his or her useful feedback on this weblog put up.


[1] The total checklist of third events embedded within the LivingSocial instance e-mail given above are as follows:
Events receiving an MD5 hash of the consumer’s e-mail deal with: American Record Counsel (alcmpn.com), LiveIntent (liadm.com), Datalogix (nexac.com), Acxiom (rlcdn.com, pippio.com, acxiom-online.com), Criteo (criteo.com, emailretargeting.com), Conversant Media (dotomi.com), V12 Information (v12group.com), VideoAmp (videoamp.com), Neustar (agkn.com), and alocdn.com. Except for emailretargeting.com and agkn.com the entire earlier domains additionally set or obtain cookies.
Extra events setting or receiving cookies: MediaMath (mathtag.com), TapAd (tapad.com), IPONWEB (bidswitch.web), AOL (promoting.com), Centro (sitescout.com), The Commerce Desk (adsrvr.org), Adobe (demdex.web), OpenX (openx.web), comScore (scorecardresearch.com, voicefive.com), Oracle (bluekai.com), Google (doubleclick.web), Realtime Concentrating on Aps (mojn.com).
Third-party domains requested with out cookies or e-mail hash: LiveIntent (licasd.com), Google (2mdn.web), Akamai (akamai.web).

[2] Until they’re proxied by the consumer’s e-mail server; of the suppliers we studied (Part 6.2 within the paper), solely Gmail and Yandex accomplish that.

[3] Our e-mail corpus was compiled by mechanically signing up for mailing lists on the highest 14,700 of the Alexa prime 1 million websites, along with the Alexa prime 500 purchasing and prime 500 information websites. In whole, we acquired 12,618 emails from 902 senders.

[4] We classify the intentionality of leaks utilizing the methodology detailed in Part four.1 of the paper.

[5] LiveIntent’s advertising and marketing materials touts the advantages of email-address-based monitoring over cookies. Specifically they spotlight that e-mail hash permits “Communication with purchasers throughout all screens and gadgets: Not like the cookie, which represents an nameless consumer, the e-mail deal with represents a identified buyer. It’s distinctive to that particular person, and stays persistent throughout all gadgets, apps and browsers.” Equally, LiveIntent additionally explains how concentrating on customers with hashed e-mail addresses permits them to proceed to serve focused advertisements in Gmail regardless of Gmail’s picture proxy.

Neustar’s privateness coverage states: “[The onboarding process] permits advertisers to make use of their offline details about buyer preferences (CRM information) … within the on-line atmosphere. … We use de-identified data similar to a hashed e-mail deal with supplied by our promoting consumer, to create a hyperlink between that de-identified CRM information and a Cookie ID, Cellular Promoting ID, or different persistent identifier assigned to a novel however de-identified consumer. That data can then be used to ship focused promoting…”. and “We additionally create and retailer linkages between and amongst family or particular person stage identifiers similar to Cookie IDs, Cellular Promoting IDs, hashed e-mail addresses and/or different persistent IDs which were assigned to a novel however de-identified consumer. This course of is typically referred to as ‘cross gadget linking’.”

Acxiom’s Information Service API helps information queries on an MD5 or SHA1 hash of an e-mail deal with.

Conversant Media’s advertising and marketing materials implies that they use e-mail deal with, along with buy information, to match consumer information throughout gadgets.

[6] For instance, LiveIntent’s privateness coverage states: “We might accumulate identifiers which are utilized by our promoting companions to determine a particular particular person … To de-identify this data, both we or our enterprise companions carry out a mathematical course of (generally referred to as hashing) to transform the data right into a code.”

[7] A GPU is a kind of processor optimized for extremely parallel duties, and is usually used for graphics processing. GPUs may very effectively compute hashes. On this put up, we offer value quotes for Amazon’s `p2.16xlarge` EC2 cloud occasion.

Picture property from the Noun Venture used on this put up: “Browser” by Designify.me, “Database” by Aybige, “Picture” by Alfa Design, “HTML File” by Burak Kucukparmaksiz, “Pc Tower” by Melvin.

Powered by WPeMatico