Impact
Wasmtime's native implementation of WASIp1 suffers from a leak in the fdrenumber function where the file descriptor being renumbered to is not properly closed. Wasmtime's implementation erroneously only updated the table of descriptors for WASIp1 and didn't update the underlying table of descriptors used by the host. This behavior means that while fdrenumber works correctly from a guest's perspective it ends up leaking resources in the host that aren't cleaned up until the corresponding Store is destroyed. This means that guests can, in a loop, use fdrenumber to cause hosts to exhaust resources or exhaust file descriptors.
This bug only affects the native implementation of WASIp1. This means that only runtimes which load core wasm modules and expose fdrenumber are affected. Runtimes are additionally only affected if they expose the ability to acquire a file descriptor, such as opening a file. For runtimes that deny access to files they are unaffected.
Patches
Wasmtime 24.0.10, 36.0.11, 44.0.3, and 45.0.2 have been released which fix this issue.
Workarounds
There are no workarounds for this issue and hosts are recommended to update.
Summary
In wasmtime-wasi, when a filesystem preopen is given DirPerms::all() and FilePerms::READ without FilePerms::WRITE, this wasmtime-wasi enforced access control mechanism can be bypassed by using the wasip2 descriptor.open-at or wasip1 pathopen interfaces by opening a file with OpenFlags::TRUNCATE oflag only, for example:
rust dirdescriptor.openat( PathFlags::empty(), FILENAME, OpenFlags::TRUNCATE, DescriptorFlags::READ, )
rust wasip1::pathopen( dirfd, 0, FILENAME, wasip1::OFLAGSTRUNC, wasip1::RIGHTSFDREAD, 0, 0 )
The root cause is that the clause that considered OpenFlags::TRUNCATE did not set openmode |= OpenMode::WRITE;, used later in that function for the access control check against FilePerms for whether opening that file is permitted. With the bug corrected, these calls to open-at and pathopen fail with error-code.not-permitted and ERRNOPERM respectively.
The bug in crates/wasi/src/filesystem.rs, Dir::openat, lines 967–969:
rust if oflags.contains(OpenFlags::TRUNCATE) { opts.truncate(true).write(true); } and the single line fix is: rust if oflags.contains(OpenFlags::TRUNCATE) { opts.truncate(true).write(true); openmode |= OpenMode::WRITE; }
Only wasmtime-wasi embeddings that use a combination of DirPerms::MUTATE with FilePerms::READ are affected by this bug, e.g. those that use in the WasiCtxBuilder: rust builder.preopeneddir("readonly", "readonly", DirPerms::READ | DirPerms::MUTATE, FilePerms::READ);
In particular, the Wasmtime project's wasmtime-cli's use of wasmtime-wasi is not affected, because it always sets FilePerms::all() for all preopens.
Impact
Wasmtime's allocation logic for a WebAssembly table contained checked arithmetic which panicked on overflow. This overflow is possible to trigger, and thus panic, when a table with an extremely large size is allocated. This is possible with the WebAssembly memory64 proposal where tables can have sizes in the 64-bit range as opposed to the previous 32-bit range which would not overflow. The panic happens when attempting to create a very large table, such as when instantiating a WebAssembly module or component.
This bug does not affect the pooling allocator which limits tables sizes to much less than the required amount to trigger the overflow. This bug is only present for the on-demand instance allocator, which is Wasmtime's default allocator. This bug also requires the memory64 WebAssembly feature to be enabled, which is on-by-default.
Panicking in the host process is considered a denial-of-service vector for Wasmtime.
Patches
Wasmtime 36.0.8, 43.0.2, and 44.0.1 have all been released which fixes this issue.
Workarounds
Embeddings can switch to using the pooling allocator to work around this issue, or the memory64 WebAssembly proposal can be disabled. Otherwise there is no workaround and users are recommended to upgrade.
Impact
Wasmtime's implementation of transcoding strings between components contains a bug where the return value of a guest component's realloc is not validated before the host attempts to write through the pointer. This enables a guest to cause the host to write arbitrary transcoded string bytes to an arbitrary location up to 4GiB away from the base of linear memory. These writes on the host could hit unmapped memory or could corrupt host data structures depending on Wasmtime's configuration.
Wasmtime by default reserves 4GiB of virtual memory for a guest's linear memory meaning that this bug will by default on hosts cause the host to hit unmapped memory and abort the process due to an unhandled fault. Wasmtime can be configured, however, to reserve less memory for a guest and to remove all guard pages, so some configurations of Wasmtime may lead to corruption of data outside of a guest's linear memory, such as host data structures or other guests's linear memories.
Patches
Wasmtime 24.0.7, 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
There is no known workaround for this issue and affected hosts/embeddings are recommended to upgrade.
Impact
Wasmtime's Winch compiler backend contains a bug where translating the table.grow operator causes the result to be incorrectly typed. For 32-bit tables this means that the result of the operator, internally in Winch, is tagged as a 64-bit value instead of a 32-bit value. This invalid internal representation of Winch's compiler state compounds into further issues depending on how the value is consumed.
One example can be seen when the result of table.grow is used as the address of a load operation. The load operation is tricked into thinking the address is a 64-bit value, not a 32-bit value, which means that the final address to load from is calculated incorrectly. This can lead to a situation where the bytes before the start of linear memory can be loaded/stored to.
The primary consequence of this bug is that bytes in the host's address space can be stored/read from. This is only applicable to the 16 bytes before linear memory, however, as the only significant return value of table.grow that can be misinterpreted is -1. The bytes before linear memory are, by default, unmapped memory. Wasmtime will detect this fault and abort the process, however, because wasm should not be able to access these bytes.
Overall this this bug in Winch represents a DoS vector by crashing the host process, a correctness issue within Winch, and a possible leak of up to 16-bytes before linear memory. Wasmtime's default compiler is Cranelift, not Winch, and Wasmtime's default settings are to place guard pages before linear memory. This means that Wasmtime's default configuration is not affected by this issue, and when explicitly choosing Winch Wasmtime's otherwise default configuration leads to a DoS. Disabling guard pages before linear memory is required to possibly leak up to 16-bytes of host data.
Patches
Wasmtime 43.0.1, 42.0.2, and 36.0.7 have been released with fixes for this issue.
Workaround
There are no workarounds within the Winch compiler backend while using the affected versions. Users of Wasmtime are encouraged either to upgrade to patched versions or, if that is not possible, use the Cranelift compiler backend.
Impact
Wasmtime's implementation of its pooling allocator contains a bug where in certain configurations the contents of linear memory can be leaked from one instance to the next. The implementation of resetting the virtual memory permissions for linear memory used the wrong predicate to determine if resetting was necessary, where the compilation process used a different predicate. This divergence meant that the pooling allocator incorrectly deduced at runtime that resetting virtual memory permissions was not necessary while compile-time determine that virtual memory could be relied upon.
Exposing this bug requires specific configuration values to be used. If any of these configurations are not applicable then this bug does not happen:
The pooling allocator must be in use. The Config::memoryguardsize configuration option must be 0. The Config::memoryreservation configuration must be less than 4GiB. The pooling allocator must be configured with maxmemorysize the same as the memoryreservation value.
If all of these conditions are applicable then when a linear memory is reused the VM permissions of the previous iteration are not reset. This means that the compiled code, which is assuming out-of-bounds loads will segfault, will not actually segfault and can read the previous contents of linear memory if it was previously mapped.
This represents a data leakage vulnerability between guest WebAssembly instances which breaks WebAssembly's semantics and additionally breaks the sandbox that Wasmtime provides. Wasmtime is not vulnerable to this issue with its default settings, nor with the default settings of the pooling allocator, but embeddings are still allowed to configure these values to cause this vulnerability.
Patches
Wasmtime 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
All four conditions above must be met to be vulnerable to this bug, and users can work around this bug by adjusting any of the above conditions. For example it is strongly recommended that guard pages are configured for linear memories which would make this bug not applicable.
Impact
Wasmtime with its Winch (baseline) non-default compiler backend may allow properly constructed guest Wasm to access host memory outside of its linear-memory sandbox.
This vulnerability requires use of the Winch compiler (-Ccompiler=winch). By default, Wasmtime uses its Cranelift backend, not Winch. With Winch, the same incorrect assumption is present in theory on both aarch64 and x86-64. The aarch64 case has an observed-working proof of concept, while the x86-64 case is theoretical and may not be reachable in practice.
This Winch compiler bug can allow the Wasm guest to access memory before or after the linear-memory region, independently of whether pre- or post-guard regions are configured. The accessible range in the initial bug proof-of-concept is up to 32KiB before the start of memory, or ~4GiB after the start of memory, independently of the size of pre- or post-guard regions or the use of explicit or guard-region-based bounds checking. However, the underlying bug assumes a 32-bit memory offset stored in a 64-bit register has its upper bits cleared when it may not, and so closely related variants of the initial proof-of-concept may be able to access truly arbitrary memory in-process. This could result in a host process segmentation fault (DoS), an arbitrary data leak from the host process, or with a write, potentially an arbitrary RCE.
Patches
Wasmtime 43.0.1, 42.0.2, and 36.0.7 have been released with fixes for this issue.
Workaround
There are no workarounds within the Winch compiler backend while using the affected versions. Users of Wasmtime are encouraged either to upgrade to patched versions or, if that is not possible, use the Cranelift compiler backend.
Impact
In version 43.0.0 of the wasmtime crate, cloning a wasmtime::Linker is unsound and can result in use-after-free bugs.
This bug is not controllable by guest Wasm programs. It can only be triggered by a specific sequence of embedder API calls made by the host.
The typical symptom of this use-after-free bug is a segfault. It does not enable heap corruption or data leakage.
If you are using the wasmtime CLI, rather than the embedding API, you are not affected. If you are using the embedding API but are not calling wasmtime::Linker's Clone implementation, you are not affected.
Specifically, the following steps must occur to trigger the bug:
Clone a wasmtime::Linker Drop the original linker instance Use the new, cloned linker instance, resulting in a use-after-free
Patches
This bug has been patched in Wasmtime version 43.0.1
Workarounds
Wasmtime embedders are highly encouraged to upgrade their wasmtime crate dependency.
If upgrading is not an option, or as a temporary workaround before upgrading, you can avoid this bug by not cloning wasmtime::Linker and instead creating a new, empty wasmtime::Linker and manually reregistering the host APIs from the original linker:
rust use wasmtime::{Linker, Result, Store};
fn clonelinker<T>(linker: &Linker<T>, store: &mut Store<T>) -> Result<Linker<T>> { let mut cloned = Linker::new(); for (module, name, item) in linker.iter(store) { cloned.define(module, name, item)?; } Ok(cloned) }
References
This bug was introduced during an internal refactoring that was part of our efforts to robustly handle allocation failure in Wasmtime. This refactoring introduced an string-interning pool which had an unsound TryClone[^try-clone] implementation.
[^try-clone]: The TryClone trait is our version of the Rust standard library's Clone trait that allows for returning OutOfMemory errors.
The StringPool was introduced in https://github.com/bytecodealliance/wasmtime/pull/12536, at which time the bug in TryClone for StringPool was already present, although this code path was not yet used anywhere. wasmtime::Linker was refactored to internally use StringPool in https://github.com/bytecodealliance/wasmtime/pull/12537, at which time the buggy code path became accessible. This bug was originally reported to the Wasmtime maintainers as https://github.com/bytecodealliance/wasmtime/pull/12906
Impact
Wasmtime's Cranelift compilation backend contains a bug on aarch64 when performing a certain shape of heap accesses which means that the wrong address is accessed. When combined with explicit bounds checks a guest WebAssembly module this can create a situation where there are two diverging computations for the same address: one for the address to bounds-check and one for the address to load. This difference in address being operated on means that a guest module can pass a bounds check but then load a different address. Combined together this enables an arbitrary read/write primitive for guest WebAssembly when accesssing host memory. This is a sandbox escape as guests are able to read/write arbitrary host memory.
This vulnerability has a few ingredients, all of which must be met, for this situation to occur and bypass the sandbox restrictions:
This miscompiled shape of load only occurs on 64-bit WebAssembly linear memories, or when Config::wasmmemory64 is enabled. 32-bit WebAssembly is not affected. Spectre mitigations or signals-based-traps must be disabled. When spectre mitigations are enabled then the offending shape of load is not generated. When signals-based-traps are disabled then spectre mitigations are also automatically disabled.
The specific bug in Cranelift is a miscompile of a load of the shape load(iadd(base, ishl(index, amt))) where amt is a constant. The amt value is masked incorrectly to test if it's a certain value, and this incorrect mask means that Cranelift can pattern-match this lowering rule during instruction selection erroneously, diverging from WebAssembly's and Cranelift's semantics. This incorrect lowering would, for example, load an address much further away than intended as the correct address's computation would have wrapped around to a smaller value insetad.
Patches
Wasmtime 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
This bug only affects users of Cranelift on aarch64. Cranelift on other platforms is not affected. Additionally this only affects 64-bit WebAssembly linear memories, so if Config::wasmmemory64 is disabled then hosts are not affected. Note that Config::wasmmemory64 is enabled by default. If spectre mitigations are enabled, which are enabled by default, then hosts are not affected by this issue.
Impact
Wasmtime's Winch compiler contains a vulnerability where the compilation of the table.fill instruction can result in a host panic. This means that a valid guest can be compiled with Winch, on any architecture, and cause the host to panic. This represents a denial-of-service vulnerability in Wasmtime due to guests being able to trigger a panic.
The specific issue is that a historical refactoring, #11254, changed how compiled code referenced tables within the table. instructions. This refactoring forgot to update the Winch code paths associated as well, meaning that Winch was using the wrong indexing scheme. Due to the feature support of Winch the only problem that can result is tables being mixed up or nonexistent tables being used, meaning that the guest is limited to panicking the host (using a nonexistent table), or executing spec-incorrect behavior and modifying the wrong table.
Patches
Wasmtime 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
Users of Cranelift are not affected by this issue, but for users of Winch there is no workaround for this bug. Hosts are recommended to updated to a patched version of Wasmtime.
Impact
Wasmtime's Winch compiler contains a bug where a 64-bit table, part of the memory64 proposal of WebAssembly, incorrectly translated the table.size instruction. This bug could lead to disclosing data on the host's stack to WebAssembly guests. The host's stack can possibly contain sensitive data related to other host-originating operations which is not intended to be disclosed to guests.
This bug specifically arose from a mistake where the return value of table.size was statically typed as a 32-bit integer, as opposed to consulting the table's index type to see how large the returned register could be. When combined with details about Wnich's ABI, such as multi-value returns, this can be combined to read stack data from the host, within a guest. This information disclosure should not be possible in WebAssembly, violates spec semantics, and is a vulnerability in Wasmtime.
Patches
Wasmtime 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
Users of Cranelift are not affected by this issue, but users of Winch have no workarounds other than disabling the Config::wasmmemory64 proposal.
On x86-64 platforms with SSE3 disabled Wasmtime's compilation of the f64x2.splat WebAssembly instruction with Cranelift may load 8 more bytes than is necessary. When signals-based-traps are disabled this can result in a uncaught segfault due to loading from unmapped guard pages. With guard pages disabled it's possible for out-of-sandbox data to be loaded, but this data is not visible to WebAssembly guests.
Details
The f64x2.splat operator, when operating on a value loaded from a memory (for example with f64.load), compiles with Cranelift to code on x86-64 without SSE3 that loads 128 bits (16 bytes) rather than the expected 64 bits (8 bytes) from memory. When the address is in-bounds for a (correct) 8-byte load but not an (incorrect) 16-byte load, this can load beyond memory by up to 8 bytes. This can result in three different behaviors depending on Wasmtime's configuration:
1. If guard pages are disabled then this extra data will be loaded. The extra data is present in the upper bits of a register, but the upper bits are not visible to WebAssembly guests. Actually witnessing this data would require a different bug in Cranelift, of which none are known. Thus in this situation while it's something we're patching in Cranelift it's not a security issue. 2. If guard pages are enabled, and signals-based-traps are enabled, then this operation will result in a safe WebAssembly trap. The trap is incorrect because the load is not out-of-bounds as defined by WebAssembly, but this mistakenly widened load will load bytes from an unmapped guard page, causing a segfault which is caught and handled as a Wasm trap. In this situation this is not a security issue, but we're patching Cranelift to fix the WebAssembly behavior. 3. If guard pages are enabled, and signals-based-traps are disabled, then this operation results in an uncaught segfault. Like the previous case with guard pages enabled this will load from an unmapped guard page. Unlike before, however, signals-based-traps are disabled meaning that signal handlers aren't configured. The resulting segfault will, by default, terminate the process. This is a security issue from a DoS perspective, but does not represent an arbitrary read or write from WebAssembly, for example.
Wasmtime's default configuration is case (2) in this case. That means that Wasmtime, by default, incorrectly executes this WebAssembly instruction but does not have insecure behavior.
Impact
If signals-based-traps are disabled and guard pages are enabled then guests can trigger an uncaught segfault in the host, likely aborting the host process. This represents, for example, a DoS vector for WebAssembly guests.
This bug does not affect Wasmtime's default configuration and requires signals-based-traps to be disabled. This bug only affects the x86-64 target with the SSE3 feature disabled and the Cranelift backend (Wasmtime's default backend).
Patches
Wasmtime 24.0.7, 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
This bug only affects x86-64 hosts where SSE3 is disabled. If SSE3 is enabled or if a non-x86-64 host is used then hosts are not affect. Otherwise there are no known workarounds to this issue.
Impact
Wasmtime contains a possible panic which can happen when a flags-typed component model value is lifted with the Val type. If bits are set outside of the set of flags the component model specifies that these bits should be ignored but Wasmtime will panic when this value is lifted. This panic only affects wasmtime's implementation of lifting into Val, not when using the flags! macro. This additionally only affects flags-typed values which are part of a WIT interface.
This has the risk of being a guest-controlled panic within the host which Wasmtime considers a DoS vector.
Patches
Wasmtime 24.0.7, 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
There is no workaround for this bug if a host meets the criteria to be affected. To be affected a host must be using wasmtime::component::Val and possibly work with a flags type in the component model.
Impact
Wasmtime's implementation of transcoding strings into the Component Model's utf16 or latin1+utf16 encodings improperly verified the alignment of reallocated strings. This meant that unaligned pointers could be passed to the host for transcoding which would trigger a host panic. This panic is possible to trigger from malicious guests which transfer very specific strings across components with specific addresses.
Host panics are considered a DoS vector in Wasmtime as the panic conditions are controlled by the guest in this situation.
Patches
Wasmtime 24.0.7, 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime.
Workarounds
There is no workaround for this bug. Hosts are recommended to updated to a patched version of Wasmtime.
Summary
Wasmtime contains a vulnerability where when transcoding a UTF-16 string to the latin1+utf16 component-model encoding it would incorrectly validate the byte length of the input string when performing a bounds check. Specifically the number of code units were checked instead of the byte length, which is twice the size of the code units.
This vulnerability can cause the host to read beyond the end of a WebAssembly's linear memory in an attempt to transcode nonexistent bytes. In Wasmtime's default configuration this will read unmapped memory on a guard page, terminating the process with a segfault. Wasmtime can be configured, however, without guard pages which would mean that host memory beyond the end of linear memory may be read and interpreted as UTF-16.
A host segfault is a denial-of-service vulnerability in Wasmtime, and possibly being able to read beyond the end of linear memory is additionally a vulnerability. Note that reading beyond the end of linear memory requires nonstandard configuration of Wasmtime, specifically with guard pages disabled.
Impact
This is an out-of-bounds memory access. Any user running untrusted wasm components that use cross-component string passing (with UTF-16 source and latin1+utf16 destination encodings) is affected.
- With guard pages: Denial of service. The host process crashes with SIGBUS/SIGSEGV. - Without guard pages: Potential information disclosure. The guest can read host memory beyond its linear memory allocation.
Patches
Wasmtime 24.0.7, 36.0.7, 42.0.2, and 43.0.1 have been issued to fix this bug. Users are recommended to update to these patched versions of Wasmtime. Workarounds
There is no workaround for this bug. Hosts are recommended to updated to a patched version of Wasmtime.
Impact
Wasmtime's implementation of the wasi:http/types.fields resource is susceptible to panics when too many fields are added to the set of headers. Wasmtime's implementation in the wasmtime-wasi-http crate is backed by a data structure which panics when it reaches excessive capacity and this condition was not handled gracefully in Wasmtime. Panicking in a WASI implementation is a Denial of Service vector for embedders and is treated as a security vulnerability in Wasmtime.
Patches
Wasmtime 24.0.6, 36.0.6, 40.0.4, 41.0.4, and 42.0.0 patch this vulnerability and return a trap to the guest instead of panicking.
Workarounds
There are no known workarounds at this time, embedders are encouraged to update to a patched version of Wasmtime.
Resources
Limitations of http::HeaderMap
Impact
Wasmtime's implementation of WASI host interfaces are susceptible to guest-controlled resource exhaustion on the host. Wasmtime did not appropriately place limits on resource allocations requested by the guests. This serves as a Denial of Service vector where a guest can induce a range of crashing behaviors on the host such as:
Allocating arbitrarily large amounts of host memory. Causing an allocation failure on the host, which in Rust defaults to aborting the process. Causing a panic on the host due to over-large allocations being performed. Cause degredation in performance of the host by holding excessive host memory alive.
Wasmtime's security bug policy considers all of these behaviors a security vulnerability. Wasmtime's implementation of WASI has a number of different ways that resource exhaustion could happen, and fixing any one of them is insufficient from solving this vulnerability. A number of individual issues are grouped within this advisory and as a whole represent the known ways that guests can exhaust resources on the host.
An example of guest-controlled resource exhaustion within Wasmtime's implementation of WASI is guests could repeatedly allocate handles to themselves without limit. Some APIs also caused the host to perform a guest-controlled-sized allocation of a buffer on the host for I/O operations. Other APIs could force the host to buffer arbitrary amounts of data for the guest. Finally the guest could hand arbitrarily large allocations from itself to the host which could cause the host to perform an arbitrarily sized copy of memory which in some situations could result in quadratically sized allocations.
Wasmtime's implementations of WASIp1 and WASIp2 are affected by this vulnerability. Any host API modeled with the Component Model (or WIT) which operates on a string or list<T> type is also affected. Not all WIT and WASI APIs are affected by this issue, but that's more of an exception so it's recommended for all embedders to consider themselves affected.
To address this issue a number of mitigations are being applied to limit the behavior of a guest in WASI. All of these mitigations manifest in the form of a limit of some kind applied to various situations, and as such all of these mitigations are backwards-incompatible as they run the risk of breaking preexisting programs. To address this all backports to previous stable releases have these limits tuned to overly-large values. This ensures that preexisting guests do not break while still providing embedders the knobs to prevent this DoS vector as well. The limits added to Wasmtime are:
-Smax-resources=N or ResourceTable::setmaxcapacity - the maximum number of resources that a guest is allowed to allocate for itself. -Shostcall-fuel=N or Store::sethostcallfuel - the maximum amount of data that the guest may copy to the host in a single function call. -Smax-random-size=N or WasiCtxBuilder::maxrandomsize - the maximum size of the return value of get-random-bytes and get-insecure-random-bytes in the wasi:random implementations. -Smax-http-fields-size=N or WasiHttpCtx::setmaxfieldssize - the maximum size of headers for an HTTP request/response.
These settings are equally applicable to both WASIp1 and WASIp2. Wasmtime 41.0.x and prior previously did not limit these settings and the knobs being released are set to very large values by default to avoid any breaking behavior. Embedders will need to proactively tune these knobs as appropriate for their embeddings. The default settings in the unreleased Wasmtime 42.0.0 are 1M for max resources, 128MiB for hostcall fuel, 64MiB for max-random-size, and 32KiB for http fields size. Tuning is not expected for Wasmtime 42.0.0+.
Hosts/embedders affected by this issue are encouraged to audit and double-check their own host APIs they have implemented to see whether they are affected by this issue as well. The -Shostcall-fuel setting is intended to be a relatively coarse fix for many possible issues by limiting the amount of data for all host APIs at once, so many embedders may not need to take further action beyond updating Wasmtime and configuring it appropriately (if not updating to 42.0.0). Embedders should audit to see, however, if the guest is able to force the host to allocate on its behalf and ensure that the allocation is limited or tracked somehow.
Patches
Wasmtime 24.0.6, 36.0.6, 40.0.4, 41.0.4, and 42.0.0 have all been released with the fix for this issue. These versions do not prevent this issue in their default configuration to avoid breaking preexisting behaviors. All versions of Wasmtime have appropriate knobs to prevent this behavior, and Wasmtime 42.0.0-and-later will have these knobs tuned by default to prevent this issue from happening.
Workarounds
There are no known workarounds for this issue without upgrading. Embedders are recommended to upgrade and configure their embeddings as necessary to prevent possibly-malicious guests from triggering this issue.
Resources
Store::sethostcallfuel ResourceTable::setmaxcapacity WasiCtxBuilder::maxrandomsize Original PR showing resource exhaustion Issue about limiting max resource handles per-guest
The affected versions of Wasmtime can panic if the host embedder drops the future returned by wasmtime::component::[Typed]Func::callasync before it resolves.
Details
Starting with Wasmtime 39.0.0, the component-model-async feature became the default, which brought with it a new implementation of [Typed]Func::callasync which made it capable of calling async-typed guest export functions. However, that implementation had a bug leading to a panic under certain circumstances:
1. The host embedding calls [Typed]Func::callasync on a function exported by a component, polling the returned Future once. 2. The component function yields control to the async runtime (e.g. Tokio), e.g. due to a call to host function registered using LinkerInstance::funcwrapasync which yields, or due an epoch interruption. 3. The host embedding drops the Future after polling it once. This leaves the component instance in a non-reenterable state since the call never had a chance to complete. 4. The host embedding calls [Typed]Func::callasync again, polling the returned Future. Since the component instance cannot be entered at this point, the call traps, but not before allocating a task and thread for the call. 5. The host embedding ignores the trap and drops the Future. This panics due to the runtime attempting to dispose of the task created above, which panics since the thread has not yet exited.
Impact When a host embedder using the affected versions of Wasmtime calls wasmtime::component::[Typed]Func::callasync on a guest export and then drops the returned future without waiting for it to resolve, and then does so again with the same component instance, Wasmtime will panic. Embeddings that have the component-model-async compile-time feature disabled are unaffected.
Patches Wasmtime 40.0.4 and 41.0.4 have been patched to fix this issue. Versions 42.0.0 and later are not affected.
Workarounds If an embedding is not actually using any component-model-async features then disabling the component-model-async Cargo feature can work around this issue. This issue can also be worked around by either ensuring every callasync future is awaited until it completes or refraining from using the Store again after dropping a not-yet-resolved callasync future.
Resources This was first reported in https://bytecodealliance.zulipchat.com/#narrow/channel/206238-general/topic/Panic.20in.20Wasmtime.2041.2E0.2E3.20.28runtime.2Fconcurrent.2Fcomponent.29
On x86-64 platforms with AVX Wasmtime's compilation of the f64.copysign WebAssembly instruction with Cranelift may load 8 more bytes than is necessary. When [signals-based-traps] are disabled this can result in a uncaught segfault due to loading from unmapped guard pages. With guard pages disabled it's possible for out-of-sandbox data to be loaded, but unless there is another bug in Cranelift this data is not visible to WebAssembly guests.
Details
The f64.copysign operator, when operating on a value loaded from a memory (for example with f64.load), compiles with Cranelift to code on x86-64 with AVX that loads 128 bits (16 bytes) rather than the expected 64 bits (8 bytes) from memory. When the address is in-bounds for a (correct) 8-byte load but not an (incorrect) 16-byte load, this can load beyond memory by up to 8 bytes. This can result in three different behaviors depending on Wasmtime's configuration:
1. If guard pages are disabled then this extra data will be loaded. The extra data is present in the upper bits of a register, but the upper bits are not visible to WebAssembly guests. Actually witnessing this data would require a different bug in Cranelift, of which none are known. Thus in this situation while it's something we're patching in Cranelift it's not a security issue. 2. If guard pages are enabled, and [signals-based-traps] are enabled, then this operation will result in a safe WebAssembly trap. The trap is incorrect because the load is not out-of-bounds as defined by WebAssembly, but this mistakenly widened load will load bytes from an unmapped guard page, causing a segfault which is caught and handled as a Wasm trap. In this situation this is not a security issue, but we're patching Cranelift to fix the WebAssembly behavior. 3. If guard pages are enabled, and [signals-based-traps] are disabled, then this operation results in an uncaught segfault. Like the previous case with guard pages enabled this will load from an unmapped guard page. Unlike before, however, signals-based-traps are disabled meaning that signal handlers aren't configured. The resulting segfault will, by default, terminate the process. This is a security issue from a DoS perspective, but does not represent an arbitrary read or write from WebAssembly, for example.
Wasmtime's default configuration is case (2) in this case. That means that Wasmtime, by default, incorrectly executes this WebAssembly instruction but does not have insecure behavior.
Impact
If [signals-based-traps] are disabled and guard pages are enabled then guests can trigger an uncaught segfault in the host, likely aborting the host process. This represents, for example, a DoS vector for WebAssembly guests.
This bug does not affect Wasmtime's default configuration and requires [signals-based-traps] to be disabled. This bug only affects the x86-64 target with the AVX feature enabled and the Cranelift backend (Wasmtime's default backend).
Patches
Wasmtime 36.0.5, 40.0.3, and 41.0.1 have been released to fix this issue. Users are recommended to upgrade to the patched versions of Wasmtime. Other affected versions are not patched and users should updated to supported major version instead.
Workarounds
This bug can be worked around by enabling [signals-based-traps]. While disabling guard pages can be a quick fix in some situations, it's not recommended to disabled guard pages as it is a key defense-in-depth measure of Wasmtime.
Resources
[signals-based-traps configuration][signals-based-traps] guard pages configuration
[signals-based-traps]: https://docs.rs/wasmtime/latest/wasmtime/struct.Config.html#method.signalsbasedtraps
Impact
Wasmtime 37.0.0 and 37.0.1 have memory leaks in the C/C++ API when using bindings for the anyref or externref WebAssembly values. This is caused by a regression introduced during the development of 37.0.0 and all prior versions of Wasmtime are unaffected. If anyref or externref is not used in the C/C++ API then embeddings are also unaffected by the leaky behavior. The wasmtime Rust crate is unaffected by this leak.
Development of Wasmtime 37.0.0 included a refactoring in Rust of changing the old ManuallyRooted<T> type to a new OwnedRooted<T> type. This change was integrated into Wasmtime's C API but left the C API in a state which had memory leaks. Additionally the new ownership semantics around this type were not reflected into the C++ API, making it leak-prone. A short version of the change is that previously ManuallyRooted<T>, as the name implies, required manual calls to an "unroot" operation. If this was forgotten then the memory was still cleaned up when the wasmtimestoret itself was destroyed eventually. Documentation of when to "unroot" was sparse and there were already situations prior to 37.0.0 where memory would be leaked until the store was destroyed anyway. All memory, though, was always bound by the store, and destroying the store would guarantee that there were no memory leaks.
In migrating to OwnedRooted<T> the usage of the type in Rust changed. A manual "unroot" operation is no longer required and it happens naturally as a destructor of the OwnedRooted<T> type in Rust itself. These new resource ownership semantics were not fully integrated into the preexisting semantics of the C/C++ APIs in Wasmtime. A crucial distinction of OwnedRooted<T> vs ManuallyRooted<T> is that the OwnedRooted<T> type allocates host memory outside of the store. This means that if an OwnedRooted<T> is leaked then destroying a store does not release this memory and it's a permanent memory leak on the host. This led to a few distinct, but related, issues arising:
A typo in the wasmtimevalunroot function in the C API meant that it did not actually unroot anything. This meant that even if embedders faithfully call the function then memory will be leaked. If a host-defined function returned a wasmtime{externref,anyref}t value then the value was never unrooted. The C/C++ API no longer has access to the value and the Rust implementation did not unroot. This meant that any values returned this way were never unrooted. The goal of the C++ API of Wasmtime is to encode automatic memory management in the type system, but the C++ API was not updated when OwnedRooted<T> was added. This meant that idiomatic usage of the C++ API would leak memory due to a lack of destructors on values.
These issues have all been fixed in a 37.0.2 release of Wasmtime. The implementation of the C and C++ APIs have been updated accordingly and respectively to account for the changes of ownership here. For example wasmtimevalunroot has been fixed to unroot, the Rust-side implementation of calling an embedder-defined function will unroot return values, and the C++ API now has destructors on the ExternRef, AnyRef, and Val types. These changes have been made to the 37.0.x release branch in a non-API-breaking fashion. Changes to the 38.0.0 release branch (and main in the Wasmtime repository) include minor API updates to better accomodate the API semantic changes.
Patches
Wasmtime 37.0.2 has been released which fixes these issues. Users of 37.0.0 and 37.0.1 are recommended to upgrade. Users of 36.0.x and prior are unaffected and need not upgrade.
Workarounds
The only known workaround at this time is to avoid using externref and anyref in the C/C++ API of Wasmtime. If avoiding those types is not possible then it's required for users to update to mitigate the leak issue.
Remediations
The Wasmtime project will be looking more closely into improving the situation around testing the C and C++ APIs of Wasmtime in the near future. For example we plan to integrate ASAN testing to discover leaks and errors earlier on in the development process. We're also going to look into expanding the test coverage of the C/C++ APIs to catch issues like this earlier on in development in the future.
Impact
There was an invalid free and out-of-bounds read and write bug when running Wasm that uses externrefs in Wasmtime.
To trigger this bug, Wasmtime needs to be running Wasm that uses externrefs, the host creates non-null externrefs, Wasmtime performs a garbage collection (GC), and there has to be a Wasm frame on the stack that is at a GC safepoint where
there are no live references at this safepoint, and there is a safepoint with live references earlier in this frame's function.
Under this scenario, Wasmtime would incorrectly use the GC stack map for the safepoint from earlier in the function instead of the empty safepoint. This would result in Wasmtime treating arbitrary stack slots as externrefs that needed to be rooted for GC. At the next GC, it would be determined that nothing was referencing these bogus externrefs (because nothing could ever reference them, because they are not really externrefs) and then Wasmtime would deallocate them and run <ExternRef as Drop>::drop on them. This results in a free of memory that is not necessarily on the heap (and shouldn't be freed at this moment even if it was), as well as potential out-of-bounds reads and writes.
Even though support for externrefs (via the reference types proposal) is enabled by default, unless you are creating non-null externrefs in your host code or explicitly triggering GCs, you cannot be affected by this bug.
We have reason to believe that the effective impact of this bug is relatively small because usage of externref is currently quite rare.
Patches
This bug has been patched and users should upgrade to Wasmtime version 0.30.0.
Additionally, we have updated our primary externref fuzz target such that it better exercises these code paths and we can have greater confidence in their correctness going forward.
Workarounds
If you cannot upgrade Wasmtime at this time, you can avoid this bug by disabling the reference types proposal by passing false to wasmtime::Config::wasmreferencetypes
References
The Wasm reference types proposal, which introduces externref
For more information
If you have any questions or comments about this advisory:
Reach out to us on the Bytecode Alliance Zulip chat Open an issue in the bytecodealliance/wasmtime repository