Linaro-mm-sig

linaro-mm-sig@lists.linaro.org

2 participants
2906 discussions

Re: [PATCH] epoll: try to be a _bit_ better about file lifetimes

by Christian König

Am 08.05.24 um 10:23 schrieb Christian Brauner: > On Tue, May 07, 2024 at 07:45:02PM +0200, Christian König wrote: >> Am 07.05.24 um 18:46 schrieb Linus Torvalds: >>> On Tue, 7 May 2024 at 04:03, Daniel Vetter <daniel(a)ffwll.ch> wrote: >>>> It's really annoying that on some distros/builds we don't have that, and >>>> for gpu driver stack reasons we _really_ need to know whether a fd is the >>>> same as another, due to some messy uniqueness requirements on buffer >>>> objects various drivers have. >>> It's sad that such a simple thing would require two other horrid >>> models (EPOLL or KCMP). >>> >>> There'[s a reason that KCMP is a config option - *some* of that is >>> horrible code - but the "compare file descriptors for equality" is not >>> that reason. >>> >>> Note that KCMP really is a broken mess. It's also a potential security >>> hole, even for the simple things, because of how it ends up comparing >>> kernel pointers (ie it doesn't just say "same file descriptor", it >>> gives an ordering of them, so you can use KCMP to sort things in >>> kernel space). >>> >>> And yes, it orders them after obfuscating the pointer, but it's still >>> not something I would consider sane as a baseline interface. It was >>> designed for checkpoint-restore, it's the wrong thing to use for some >>> "are these file descriptors the same". >>> >>> The same argument goes for using EPOLL for that. Disgusting hack. >>> >>> Just what are the requirements for the GPU stack? Is one of the file >>> descriptors "trusted", IOW, you know what kind it is? >>> >>> Because dammit, it's *so* easy to do. You could just add a core DRM >>> ioctl for it. Literally just >>> >>> struct fd f1 = fdget(fd1); >>> struct fd f2 = fdget(fd2); >>> int same; >>> >>> same = f1.file && f1.file == f2.file; >>> fdput(fd1); >>> fdput(fd2); >>> return same; >>> >>> where the only question is if you also woudl want to deal with O_PATH >>> fd's, in which case the "fdget()" would be "fdget_raw()". >>> >>> Honestly, adding some DRM ioctl for this sounds hacky, but it sounds >>> less hacky than relying on EPOLL or KCMP. >>> >>> I'd be perfectly ok with adding a generic "FISAME" VFS level ioctl >>> too, if this is possibly a more common thing. and not just DRM wants >>> it. >>> >>> Would something like that work for you? >> Well the generic approach yes, the DRM specific one maybe. IIRC we need to >> be able to compare both DRM as well as DMA-buf file descriptors. >> >> The basic problem userspace tries to solve is that drivers might get the >> same fd through two different code paths. >> >> For example application using OpenGL/Vulkan for rendering and VA-API for >> video decoding/encoding at the same time. >> >> Both APIs get a fd which identifies the device to use. It can be the same, >> but it doesn't have to. >> >> If it's the same device driver connection (or in kernel speak underlying >> struct file) then you can optimize away importing and exporting of buffers >> for example. >> >> Additional to that it makes cgroup accounting much easier because you don't >> count things twice because they are shared etc... > One thing to keep in mind is that a generic VFS level comparing function > will only catch the obvious case where you have dup() equivalency as > outlined above by Linus. That's what most people are interested in and > that could easily replace most kcmp() use-cases for comparing fds. > > But, of course there's the case where you have two file descriptors > referring to two different files that reference the same underlying > object (usually stashed in file->private_data). > > For most cases that problem can ofc be solved by comparing the > underlying inode. But that doesn't work for drivers using the generic > anonymous inode infrastructure because it uses the same inode for > everything or for cases where the same underlying object can even be > represented by different inodes. > > So for such cases a driver specific ioctl() to compare two fds will > be needed in addition to the generic helper. At least for the DRM we already have some solution for that, see drmGetPrimaryDeviceNameFromFd() for an example. Basically the file->private_data is still something different, but we use this to figure out if we have two file descriptors (with individual struct files underneath) pointing to the same hw driver. This is important if you need to know if just importing/exporting of DMA-buf handles between the two file descriptors is enough or if you deal with two different hw devices and need to do stuff like format conversion etc... Regards, Christian.

1 year, 1 month

Re: get_file() unsafe under epoll (was Re: [syzbot] [fs?] [io-uring?] general protection fault in __ep_remove)

by Linus Torvalds

On Mon, 6 May 2024 at 10:46, Stefan Metzmacher <metze(a)samba.org> wrote: > > I think it's a very important detail that epoll does not take > real references. Otherwise an application level 'close()' on a socket > would not trigger a tcp disconnect, when an fd is still registered with > epoll. Yes, exactly. epoll() ends up actually wanting the lifetime of the ep item be the lifetime of the file _descriptor_, not the lifetime of the file itself. We approximate that - badly - with epoll not taking a reference on the file pointer, and then at final fput() it tears things down. But that has two real issues, and one of them is that "oh, now epoll has file pointers that are actually dead" that caused this thread. The other issue is that "approximates" thing, where it means that duplicating the file pointer (dup*() and fork() end unix socket file sending etc) will not mean that the epoll ref is also out of sync with the lifetime of the file descriptor. That's why I suggested that "clean up epoll references at file_close_fd() time instead: https://lore.kernel.org/all/CAHk-=wj6XL9MGCd_nUzRj6SaKeN0TsyTTZDFpGdW34R+zM… because it would actually really *fix* the lifetime issue of ep items. In the process, it would make it possible to actually take a f_count reference at EPOLL_CTL_ADD time, since now the lifetime of the EP wouldn't be tied to the lifetime of the 'struct file *' pointer, it would be properly tied to the lifetime of the actual file descriptor that you are adding. So it would be a huge conceptual cleanup too. (Of course - at that point EPOLL_CTL_ADD still doesn't actually _need_ a reference to the file, since the file being open in itself is already that reference - but the point here being that there would *be* a reference that the epoll code would effectively have, and you'd never be in the situation we were in where there would be stale "dead" file pointers that just haven't gone through the cleanup yet). But I'd rather not touch the epoll code more than I have to. Which is why I applied the minimal patch for just "refcount over vfs_poll()", and am just mentioning my suggestion in the hope that some eager beaver would like to see how painful it would do to make the bigger surgery... Linus

1 year, 1 month

Re: [PATCH] epoll: try to be a _bit_ better about file lifetimes

by David Laight

From: Christian Brauner > Sent: 06 May 2024 09:45 > > > The fact is, it's not dma-buf that is violating any rules. It's epoll. > > I agree that epoll() not taking a reference on the file is at least > unexpected and contradicts the usual code patterns for the sake of > performance and that it very likely is the case that most callers of > f_op->poll() don't know this. > > Note, I cleary wrote upthread that I'm ok to do it like you suggested > but raised two concerns a) there's currently only one instance of > prolonged @file lifetime in f_op->poll() afaict and b) that there's > possibly going to be some performance impact on epoll(). > > So it's at least worth discussing what's more important because epoll() > is very widely used and it's not that we haven't favored performance > before. > > But you've already said that you aren't concerned with performance on > epoll() upthread. So afaict then there's really not a lot more to > discuss other than take the patch and see whether we get any complaints. Surely there isn't a problem with epoll holding a reference to the file structure - it isn't really any different to a dup(). 'All' that needs to happen is that the 'magic' that makes epoll() remove files on the last fput happen when the close is done. I'm sure there are horrid locking issues it that code (separate from it calling ->poll() after ->release()) eg if you call close() concurrently with EPOLL_CTL_ADD. I'm not at all sure it would have mattered if epoll kept the file open. But it can't do that because it is documented not to. As well as poll/select holding a reference to all their fd for the duration of the system call, a successful mmap() holds a reference until the pages are all unmapped - usually by process exit. We (dayjob) have code that uses epoll() to monitor large numbers of UDP sockets. I was doing some tests (trying to) receive RTP (audio) data concurrently on 10000 sockets with typically one packet every 20ms. There are 10000 associated RCTP sockets that are usually idle. A more normal limit would be 1000 RTP sockets. All the data needs to go into a single (multithreaded) process. Just getting all the packets queued on the sockets was non-trivial. epoll is about the only way to actually read the data. (That needed multiple epoll fd so each thread could process all the events from one epoll fd then look for another unprocessed fd.) David - Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

1 year, 1 month

Re: Safety of opening up /dev/dma_heap/* to physically present users (udev uaccess tag) ?

by Daniel Vetter

On Mon, May 06, 2024 at 02:05:12PM +0200, Maxime Ripard wrote: > Hi, > > On Mon, May 06, 2024 at 01:49:17PM GMT, Hans de Goede wrote: > > Hi dma-buf maintainers, et.al., > > > > Various people have been working on making complex/MIPI cameras work OOTB > > with mainline Linux kernels and an opensource userspace stack. > > > > The generic solution adds a software ISP (for Debayering and 3A) to > > libcamera. Libcamera's API guarantees that buffers handed to applications > > using it are dma-bufs so that these can be passed to e.g. a video encoder. > > > > In order to meet this API guarantee the libcamera software ISP allocates > > dma-bufs from userspace through one of the /dev/dma_heap/* heaps. For > > the Fedora COPR repo for the PoC of this: > > https://hansdegoede.dreamwidth.org/28153.html > > For the record, we're also considering using them for ARM KMS devices, > so it would be better if the solution wasn't only considering v4l2 > devices. > > > I have added a simple udev rule to give physically present users access > > to the dma_heap-s: > > > > KERNEL=="system", SUBSYSTEM=="dma_heap", TAG+="uaccess" > > > > (and on Rasperry Pi devices any users in the video group get access) > > > > This was just a quick fix for the PoC. Now that we are ready to move out > > of the PoC phase and start actually integrating this into distributions > > the question becomes if this is an acceptable solution; or if we need some > > other way to deal with this ? > > > > Specifically the question is if this will have any negative security > > implications? I can certainly see this being used to do some sort of > > denial of service attack on the system (1). This is especially true for > > the cma heap which generally speaking is a limited resource. > > There's plenty of other ways to exhaust CMA, like allocating too much > KMS or v4l2 buffers. I'm not sure we should consider dma-heaps > differently than those if it's part of our threat model. So generally for an arm soc where your display needs cma, your render node doesn't. And user applications only have access to the later, while only the compositor gets a kms fd through logind. At least in drm aside from vc4 there's really no render driver that just gives you access to cma and allows you to exhaust that, you need to be a compositor with drm master access to the display. Which means we're mostly protected against bad applications, and that's not a threat the "user physically sits in front of the machine accounts for", and which giving cma access to everyone would open up. And with flathub/snaps/... this is very much an issue. So you need more, either: - cgroups limits on dma-buf and dma-buf heaps. This has been bikeshedded for years and is just not really moving. - An allocator service which checks whether you're allowed to allocate these special buffers. Android does that through binder. Probably also some way to nuke applications that refuse to release buffers when they're no longer the right application. cgroups is a lot more convenient for that. -Sima > > But devices tagged for uaccess are only opened up to users who are > > physcially present behind the machine and those can just hit > > the powerbutton, so I don't believe that any *on purpose* DOS is part of > > the thread model. > > How would that work for headless devices? > > Maxime -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

1 year, 1 month

Re: Safety of opening up /dev/dma_heap/* to physically present users (udev uaccess tag) ?

by Daniel Vetter

On Tue, May 07, 2024 at 04:15:05PM +0100, Bryan O'Donoghue wrote: > On 07/05/2024 16:09, Dmitry Baryshkov wrote: > > Ah, I see. Then why do you require the DMA-ble buffer at all? If you are > > providing data to VPU or DRM, then you should be able to get the buffer > > from the data-consuming device. > > Because we don't necessarily know what the consuming device is, if any. Well ... that's an entirely different issue. And it's unsolved. Currently the approach is to allocate where the constraints are usually most severe (like display, if you need that, or the camera module for input) and then just pray the stack works out without too much copying. All userspace (whether the generic glue or the userspace driver depends a bit upon the exact api) does need to have a copy fallback for these sharing cases, ideally with the copying accelerated by hw. If you try to solve this by just preemptive allocating everything as cma buffers, then you'll make the situation substantially worse (because now you're wasting tons of cma memory where you might not even need it). And without really solving the problem, since for some gpus that memory might be unusable (because you cannot scan that out on any discrete gpu, and sometimes not even on an integrated one). -Sima > Could be VPU, could be Zoom/Hangouts via pipewire, could for argument sake > be GPU or DSP. > > Also if we introduce a dependency on another device to allocate the output > buffers - say always taking the output buffer from the GPU, then we've added > another dependency which is more difficult to guarantee across different > arches. > > --- > bod -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

1 year, 1 month

Re: Safety of opening up /dev/dma_heap/* to physically present users (udev uaccess tag) ?

by Dmitry Baryshkov

On Tue, May 07, 2024 at 04:34:24PM +0200, Hans de Goede wrote: > Hi Dmitry, > > On 5/7/24 3:32 PM, Dmitry Baryshkov wrote: > > On Mon, May 06, 2024 at 01:49:17PM +0200, Hans de Goede wrote: > >> Hi dma-buf maintainers, et.al., > >> > >> Various people have been working on making complex/MIPI cameras work OOTB > >> with mainline Linux kernels and an opensource userspace stack. > >> > >> The generic solution adds a software ISP (for Debayering and 3A) to > >> libcamera. Libcamera's API guarantees that buffers handed to applications > >> using it are dma-bufs so that these can be passed to e.g. a video encoder. > >> > >> In order to meet this API guarantee the libcamera software ISP allocates > >> dma-bufs from userspace through one of the /dev/dma_heap/* heaps. For > >> the Fedora COPR repo for the PoC of this: > >> https://hansdegoede.dreamwidth.org/28153.html > > > > Is there any reason for allocating DMA buffers for libcamera through > > /dev/dma_heap/ rather than allocating them via corresponding media > > device and then giving them away to DRM / VPU / etc? > > > > At least this should solve the permission usecase: if the app can open > > camera device, it can allocate a buffer. > > This is with a software ISP, the input buffers with raw bayer data > come from a v4l2 device, but the output buffers with the processed > data are purely userspace managed in this case. Ah, I see. Then why do you require the DMA-ble buffer at all? If you are providing data to VPU or DRM, then you should be able to get the buffer from the data-consuming device. If the data is further processed by a userspace app, then it should not require DMA memory at all. My main concern is that dma-heaps is both too generic and too limiting for the generic library implementation. -- With best wishes Dmitry

1 year, 1 month

Re: Safety of opening up /dev/dma_heap/* to physically present users (udev uaccess tag) ?

by Dmitry Baryshkov

On Mon, May 06, 2024 at 04:01:42PM +0200, Hans de Goede wrote: > Hi Sima, > > On 5/6/24 3:38 PM, Daniel Vetter wrote: > > On Mon, May 06, 2024 at 02:05:12PM +0200, Maxime Ripard wrote: > >> Hi, > >> > >> On Mon, May 06, 2024 at 01:49:17PM GMT, Hans de Goede wrote: > >>> Hi dma-buf maintainers, et.al., > >>> > >>> Various people have been working on making complex/MIPI cameras work OOTB > >>> with mainline Linux kernels and an opensource userspace stack. > >>> > >>> The generic solution adds a software ISP (for Debayering and 3A) to > >>> libcamera. Libcamera's API guarantees that buffers handed to applications > >>> using it are dma-bufs so that these can be passed to e.g. a video encoder. > >>> > >>> In order to meet this API guarantee the libcamera software ISP allocates > >>> dma-bufs from userspace through one of the /dev/dma_heap/* heaps. For > >>> the Fedora COPR repo for the PoC of this: > >>> https://hansdegoede.dreamwidth.org/28153.html > >> > >> For the record, we're also considering using them for ARM KMS devices, > >> so it would be better if the solution wasn't only considering v4l2 > >> devices. > >> > >>> I have added a simple udev rule to give physically present users access > >>> to the dma_heap-s: > >>> > >>> KERNEL=="system", SUBSYSTEM=="dma_heap", TAG+="uaccess" > >>> > >>> (and on Rasperry Pi devices any users in the video group get access) > >>> > >>> This was just a quick fix for the PoC. Now that we are ready to move out > >>> of the PoC phase and start actually integrating this into distributions > >>> the question becomes if this is an acceptable solution; or if we need some > >>> other way to deal with this ? > >>> > >>> Specifically the question is if this will have any negative security > >>> implications? I can certainly see this being used to do some sort of > >>> denial of service attack on the system (1). This is especially true for > >>> the cma heap which generally speaking is a limited resource. > >> > >> There's plenty of other ways to exhaust CMA, like allocating too much > >> KMS or v4l2 buffers. I'm not sure we should consider dma-heaps > >> differently than those if it's part of our threat model. > > > > So generally for an arm soc where your display needs cma, your render node > > doesn't. And user applications only have access to the later, while only > > the compositor gets a kms fd through logind. At least in drm aside from > > vc4 there's really no render driver that just gives you access to cma and > > allows you to exhaust that, you need to be a compositor with drm master > > access to the display. > > > > Which means we're mostly protected against bad applications, and that's > > not a threat the "user physically sits in front of the machine accounts > > for", and which giving cma access to everyone would open up. And with > > flathub/snaps/... this is very much an issue. > > I agree that bad applications are an issue, but not for the flathub / snaps > case. Flatpacks / snaps run sandboxed and don't have access to a full /dev > so those should not be able to open /dev/dma_heap/* independent of > the ACLs on /dev/dma_heap/*. The plan is for cameras using the > libcamera software ISP to always be accessed through pipewire and > the camera portal, so in this case pipewere is taking the place of > the compositor in your kms vs render node example. > > So this reduces the problem to bad apps packaged by regular distributions > and if any of those misbehave the distros should fix that. > > So I think that for the denial of service side allowing physical > present users (but not sandboxed apps running as those users) to > access /dev/dma_heap/* should be ok. What about an app built by the user? The machines can still be multi-seat or have multiple users. > My bigger worry is if dma_heap (u)dma-bufs can be abused in other > ways then causing a denial of service. > > I guess that the answer there is that causing other security issues > should not be possible ? > > Regards, > > Hans > -- With best wishes Dmitry

1 year, 1 month

Re: Safety of opening up /dev/dma_heap/* to physically present users (udev uaccess tag) ?

by Dmitry Baryshkov

On Mon, May 06, 2024 at 01:49:17PM +0200, Hans de Goede wrote: > Hi dma-buf maintainers, et.al., > > Various people have been working on making complex/MIPI cameras work OOTB > with mainline Linux kernels and an opensource userspace stack. > > The generic solution adds a software ISP (for Debayering and 3A) to > libcamera. Libcamera's API guarantees that buffers handed to applications > using it are dma-bufs so that these can be passed to e.g. a video encoder. > > In order to meet this API guarantee the libcamera software ISP allocates > dma-bufs from userspace through one of the /dev/dma_heap/* heaps. For > the Fedora COPR repo for the PoC of this: > https://hansdegoede.dreamwidth.org/28153.html Is there any reason for allocating DMA buffers for libcamera through /dev/dma_heap/ rather than allocating them via corresponding media device and then giving them away to DRM / VPU / etc? At least this should solve the permission usecase: if the app can open camera device, it can allocate a buffer. > I have added a simple udev rule to give physically present users access > to the dma_heap-s: > > KERNEL=="system", SUBSYSTEM=="dma_heap", TAG+="uaccess" > > (and on Rasperry Pi devices any users in the video group get access) > > This was just a quick fix for the PoC. Now that we are ready to move out > of the PoC phase and start actually integrating this into distributions > the question becomes if this is an acceptable solution; or if we need some > other way to deal with this ? > > Specifically the question is if this will have any negative security > implications? I can certainly see this being used to do some sort of > denial of service attack on the system (1). This is especially true for > the cma heap which generally speaking is a limited resource. > > But devices tagged for uaccess are only opened up to users who are > physcially present behind the machine and those can just hit > the powerbutton, so I don't believe that any *on purpose* DOS is part of > the thread model. Any accidental DOS would be a userspace stack bug. > > Do you foresee any other negative security implications from allowing > physically present non root users to create (u)dma-bufs ? > > Regards, > > Hans > > > 1) There are some limits in drivers/dma-buf/udmabuf.c and distributions > could narrow these. > > -- With best wishes Dmitry

1 year, 1 month

Re: [PATCH] epoll: try to be a _bit_ better about file lifetimes

by Daniel Vetter

On Mon, May 06, 2024 at 04:46:54PM +0200, Christian Brauner wrote: > On Mon, May 06, 2024 at 02:47:23PM +0200, Daniel Vetter wrote: > > On Sun, May 05, 2024 at 01:53:48PM -0700, Linus Torvalds wrote: > > > On Sun, 5 May 2024 at 13:30, Al Viro <viro(a)zeniv.linux.org.uk> wrote: > > > > > > > > 0. special-cased ->f_count rule for ->poll() is a wart and it's > > > > better to get rid of it. > > > > > > > > 1. fs/eventpoll.c is a steaming pile of shit and I'd be glad to see > > > > git rm taken to it. Short of that, by all means, let's grab reference > > > > in there around the call of vfs_poll() (see (0)). > > > > > > Agreed on 0/1. > > > > > > > 2. having ->poll() instances grab extra references to file passed > > > > to them is not something that should be encouraged; there's a plenty > > > > of potential problems, and "caller has it pinned, so we are fine with > > > > grabbing extra refs" is nowhere near enough to eliminate those. > > > > > > So it's not clear why you hate it so much, since those extra > > > references are totally normal in all the other VFS paths. > > > > > > I mean, they are perhaps not the *common* case, but we have a lot of > > > random get_file() calls sprinkled around in various places when you > > > end up passing a file descriptor off to some asynchronous operation > > > thing. > > > > > > Yeah, I think most of them tend to be special operations (eg the tty > > > TIOCCONS ioctl to redirect the console), but it's not like vfs_ioctl() > > > is *that* different from vfs_poll. Different operation, not somehow > > > "one is more special than the other". > > > > > > cachefiles and backing-file does it for regular IO, and drop it at IO > > > completion - not that different from what dma-buf does. It's in > > > ->read_iter() rather than ->poll(), but again: different operations, > > > but not "one of them is somehow fundamentally different". > > > > > > > 3. dma-buf uses of get_file() are probably safe (epoll shite aside), > > > > but they do look fishy. That has nothing to do with epoll. > > > > > > Now, what dma-buf basically seems to do is to avoid ref-counting its > > > own fundamental data structure, and replaces that by refcounting the > > > 'struct file' that *points* to it instead. > > > > > > And it is a bit odd, but it actually makes some amount of sense, > > > because then what it passes around is that file pointer (and it allows > > > passing it around from user space *as* that file). > > > > > > And honestly, if you look at why it then needs to add its refcount to > > > it all, it actually makes sense. dma-bufs have this notion of > > > "fences" that are basically completion points for the asynchronous > > > DMA. Doing a "poll()" operation will add a note to the fence to get > > > that wakeup when it's done. > > > > > > And yes, logically it takes a ref to the "struct dma_buf", but because > > > of how the lifetime of the dma_buf is associated with the lifetime of > > > the 'struct file', that then turns into taking a ref on the file. > > > > > > Unusual? Yes. But not illogical. Not obviously broken. Tying the > > > lifetime of the dma_buf to the lifetime of a file that is passed along > > > makes _sense_ for that use. > > > > > > I'm sure dma-bufs could add another level of refcounting on the > > > 'struct dma_buf' itself, and not make it be 1:1 with the file, but > > > it's not clear to me what the advantage would really be, or why it > > > would be wrong to re-use a refcount that is already there. > > > > So there is generally another refcount, because dma_buf is just the > > cross-driver interface to some kind of real underlying buffer object from > > the various graphics related subsystems we have. > > > > And since it's a pure file based api thing that ceases to serve any > > function once the fd/file is gone we tied all the dma_buf refcounting to > > the refcount struct file already maintains. But the underlying buffer > > object can easily outlive the dma_buf, and over the lifetime of an > > underlying buffer object you might actually end up creating different > > dma_buf api wrappers for it (but at least in drm we guarantee there's at > > most one, hence why vmwgfx does the atomic_inc_unless_zero trick, which I > > don't particularly like and isn't really needed). > > > > But we could add another refcount, it just means we have 3 of those then > > when only really 2 are needed. > > Fwiw, the TTM thing described upthread and in the other thread really > tries hard to work around the dma_buf == file lifetime choice by hooking > into the dma-buf specific release function so it can access the dmabuf > and then the file. All that seems like a pretty error prone thing to me. > So a separate refcount for dma_buf wouldn't be the worst as that would > allow that TTM thing to benefit and remove that nasty hacking into your > generic dma_buf ops. But maybe I'm the only one who sees it that way and > I'm certainly not familiar enough with dma-buf. So the tricky part is the uniqueness requirement drm has for buffer objects (and hence dma_buf wrappers), which together with the refcounting makes dma_buf quite tricky: - dma_buf needs to hold some reference onto the underlying object, or it wont work - but you're not allowed to just create a new dma_buf every time someone exports an underlying object to a dma_buf, because that would break the uniqueness requirement. Which means the underlying object must also hold some kind of reference to its dma_buf, if it exists. So that on buffer export it can just increment the refcount for that and return it, instead of creating a new one. Which would be a reference loop that never gets freed, so you need one of two tricks: - Either a weak reference, i.e. just a pointer plus atomic_inc_unless_zero trickery like ttm does. Splitting that refcount into more refcounts doesn't fundamentally solve the problem, it just adds even more refcounts. - Or you do what all other drm drivers do in drm_prime.c do and careful clean up the dma_buf re-export cache when the userspace references (but not all kernel internal ones) disappear, to unbreak that reference loop. This needs to be done with extreme care and took a lot of screaming to get right, because if you have a race you might end up breaking the uniqueness requirement and have two dma_buf floating around. So neither of these solutions really are simple, but I agree with you that the atomic_inc_unless_zero trickery is less simple. It's definitely not cool that it's done by digging around in struct file internals. -Sima -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch

1 year, 1 month

get_file() unsafe under epoll (was Re: [syzbot] [fs?] [io-uring?] general protection fault in __ep_remove)

by Kees Cook

On Fri, May 03, 2024 at 06:54:22PM +0700, Bui Quang Minh wrote: > [...] > Root cause: > AFAIK, eventpoll (epoll) does not increase the registered file's reference. > To ensure the safety, when the registered file is deallocated in __fput, > eventpoll_release is called to unregister the file from epoll. When calling > poll on epoll, epoll will loop through registered files and call vfs_poll on > these files. In the file's poll, file is guaranteed to be alive, however, as > epoll does not increase the registered file's reference, the file may be > dying > and it's not safe the get the file for later use. And dma_buf_poll violates > this. In the dma_buf_poll, it tries to get_file to use in the callback. This > leads to a race where the dmabuf->file can be fput twice. > > Here is the race occurs in the above proof-of-concept > > close(dmabuf->file) > __fput_sync (f_count == 1, last ref) > f_count-- (f_count == 0 now) > __fput > epoll_wait > vfs_poll(dmabuf->file) > get_file(dmabuf->file)(f_count == 1) > eventpoll_release > dmabuf->file deallocation > fput(dmabuf->file) (f_count == 1) > f_count-- > dmabuf->file deallocation > > I am not familiar with the dma_buf so I don't know the proper fix for the > issue. About the rule that don't get the file for later use in poll callback > of > file, I wonder if it is there when only select/poll exist or just after > epoll > appears. > > I hope the analysis helps us to fix the issue. Thanks for doing this analysis! I suspect at least a start of a fix would be this: diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c index 8fe5aa67b167..15e8f74ee0f2 100644 --- a/drivers/dma-buf/dma-buf.c +++ b/drivers/dma-buf/dma-buf.c @@ -267,9 +267,8 @@ static __poll_t dma_buf_poll(struct file *file, poll_table *poll) if (events & EPOLLOUT) { /* Paired with fput in dma_buf_poll_cb */ - get_file(dmabuf->file); - - if (!dma_buf_poll_add_cb(resv, true, dcb)) + if (!atomic_long_inc_not_zero(&dmabuf->file) && + !dma_buf_poll_add_cb(resv, true, dcb)) /* No callback queued, wake up any other waiters */ dma_buf_poll_cb(NULL, &dcb->cb); else @@ -290,9 +289,8 @@ static __poll_t dma_buf_poll(struct file *file, poll_table *poll) if (events & EPOLLIN) { /* Paired with fput in dma_buf_poll_cb */ - get_file(dmabuf->file); - - if (!dma_buf_poll_add_cb(resv, false, dcb)) + if (!atomic_long_inc_not_zero(&dmabuf->file) && + !dma_buf_poll_add_cb(resv, false, dcb)) /* No callback queued, wake up any other waiters */ dma_buf_poll_cb(NULL, &dcb->cb); else But this ends up leaving "active" non-zero, and at close time it runs into: BUG_ON(dmabuf->cb_in.active || dmabuf->cb_out.active); But the bottom line is that get_file() is unsafe to use in some places, one of which appears to be in the poll handler. There are maybe some other fragile places too, like in drivers/gpu/drm/vmwgfx/ttm_object.c: static bool __must_check get_dma_buf_unless_doomed(struct dma_buf *dmabuf) { return atomic_long_inc_not_zero(&dmabuf->file->f_count) != 0L; } Which I also note involves a dmabuf... Due to this issue I've proposed fixing get_file() to detect pathological states: https://lore.kernel.org/lkml/20240502222252.work.690-kees@kernel.org/ But that has run into some push-back. I'm hoping that seeing this epoll example will help illustrate what needs fixing a little better. I think the best current proposal is to just WARN sooner instead of a full refcount_t implementation: diff --git a/include/linux/fs.h b/include/linux/fs.h index 8dfd53b52744..e09107d0a3d6 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -1040,7 +1040,8 @@ struct file_handle { static inline struct file *get_file(struct file *f) { - atomic_long_inc(&f->f_count); + long prior = atomic_long_fetch_inc_relaxed(&f->f_count); + WARN_ONCE(!prior, "struct file::f_count incremented from zero; use-after-free condition present!\n"); return f; } What's the right way to deal with the dmabuf situation? (And I suspect it applies to get_dma_buf_unless_doomed() as well...) -Kees -- Kees Cook

1 year, 1 month

Jump to page:

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

Linaro-mm-sig