151478 - https wyciwyg page is cached

Status: VERIFIED FIXED

(Core :: Networking: Cache, defect, P2)

Description

[Zdenek Cerny]

From Bugzilla Helper:
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.0.0) Gecko/20020530
BuildID:    2002053012

dynamic rendered pages gets protocol wyciwyg so they is cached even if source
come through https connection (if memory cache is full I supouse)

Reproducible: Always
Steps to Reproduce:
1.open any https page with dynamic scripts
2.close mozilla
3.see your cache

Comment 1

[John Morrison]

Yep.

Comment 2

[Mitchell Stoltz (not reading bugmail)]

Reassigning. Can we ensure that wyciwyg pages (created by document.write) obey
the no-cache rules of the page that writes them?

Comment 3

[gordon]

Hey Radha, is this your cache client storing this?

Comment 4

[Radha on family leave (not reading bugmail)]

Does the secure page have a no-cache directive? Even if it did, cache will
continue to store no-cache pages so that back/forward to these pages will
continue to work. The only thing that is not done for secure,no-cache pages is
that, their form element state (layout history) is  not saved, so that if you go
back/forward to them, text fields and selection boxes will not restore their
previous values. 

Comment 5

[Zdenek Cerny]

after i close mozilla in my cache stay couple files with
<meta http-equiv="Pragma" content="no-cache"> directive.
In header of this document isnt such directive(any of anti caching directives),
only meta in head of page.
back / forward button will not work either well (language code of page is not
stored correctly, in multi frame page back button will back frames in very
cofusing way, links to images broken etc.), may wyciwyg free some memory
consumation, but IMHO this protocol is useless.

Comment 6

[Darin Fisher]

the problem here is that the wyciwyg data is being stored in the disk cache
instead of the memory cache.  the wyciwyg implementation needs to use the same
cache policy as the underlying channel.

-> me

Comment 7

[Darin Fisher]

this has minor security implications since our policy for HTTPS content is not
to store it in the disk cache.  however, this is not a huge security problem IMO.

Comment 8

[Darin Fisher]

i looked at trying to learn the security state of the main document channel, but
after spending some time looking through nsHTMLDocument, it isn't obvious to me
how to determine the channel used to load the original document.  so, that leads
me to consider the simpler solution of just always caching wyciwyg data in the
memory cache.  how does that strike folks?  certainly wyciwyg data is not valid
after restarting the browser, so having it in the disk cache isn't that helpful.
 however, having it in the memory cache obviously means that it will be more
likely to be evicted.  but, fortunately wyciwyg data is typically small, so such
documents would be less likely to be evicted.  thoughts?

the best solution would still be to transfer the load flags from the original
channel to the wyciwyg channel and make the wyciwyg channel honor the
INHIBIT_PERSISTENT_CACHING load flag.

Comment 9

[Darin Fisher]

Attachment: v1 patch - stop gap measure

this patch is a stop gap solution.  it simply prevents wyciwyg data from being
stored on disk.  not optimal for the reasons i gave above, but should be
sufficient.  if we can't do better soon, then i think we should run with this
for moz 1.2.

Comment 10

[Darin Fisher]

Attachment: v2 patch - potentially better solution

of course, just after submitting the v1 patch i thought up a way to solve this.
 basically, all we have to do is record the load flags whenever we record the 
channel's principal.  nsDocument::Reset is where the principal is recorded, so
i added code to record the load flags there.  as a result, nsHTMLDocument can
easily apply those load flags to the newly created nsWyciwygChannel.  this
gives the nsWyciwygChannel context enough (by checking for the
INHIBIT_PERSISTENT_CACHING load flag) to determine whether to use the disk
cache or the memory cache.  of course, i'm new to nsDocument and friends, so i
half expect to hear that there is yet another better way.

Comment 11

[gordon]

Comment on attachment 102264 [details] [diff] [review]
v1 patch - stop gap measure

r=gordon for the v1 patch.

Comment 12

[Daniel Veditz [:dveditz]]

Want this in Blackbird I think

Comment 13

[Mitchell Stoltz (not reading bugmail)]

Comment on attachment 102271 [details] [diff] [review]
v2 patch - potentially better solution

r=mstoltz.

Comment 14

[rpotts (gone)]

Comment on attachment 102271 [details] [diff] [review]
v2 patch - potentially better solution

sr=[e-mail]

Comment 15

[Darin Fisher]

Comment on attachment 102264 [details] [diff] [review]
v1 patch - stop gap measure

marking this patch obsolete.

in addition to the reviews from rick and mitch, i've also bounced the second
patch off of jkeiser, radha, and vidur.

Comment 16

[Asa Dotzler [:asa]]

Comment on attachment 102271 [details] [diff] [review]
v2 patch - potentially better solution

a=asa for checkin to 1.2beta (on behalf of drivers)

Comment 17

[Darin Fisher]

fixed-on-trunk

Comment 18

[Randell Jesup [:jesup] (needinfo me)]

a=[e-mail] for 1.0 branch; change mozilla1.0.2+ to fixed1.0.2 when
checked in

Comment 19

[Daniel Veditz [:dveditz]]

adt+ approval for the 1.0 branch

Comment 20

[Darin Fisher]

fixed1.0.2

Comment 21

[Tom Everingham]

verified trunk - 10/23/02 builds - linux rh6, mac osX, winNT4

still need to test branch

Comment 22

[Tom Everingham]

verified 1.0.2 branch - page sent via https not stored in disk cache anymore

winNT4, linux rh6, mac osX   10/25/02 br