xref: /linux/rust/kernel/mm/virt.rs (revision 0326440c3545c86b6501c7c636fcf018d6e87b8c)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 // Copyright (C) 2024 Google LLC.
4 
5 //! Virtual memory.
6 //!
7 //! This module deals with managing a single VMA in the address space of a userspace process. Each
8 //! VMA corresponds to a region of memory that the userspace process can access, and the VMA lets
9 //! you control what happens when userspace reads or writes to that region of memory.
10 //!
11 //! The module has several different Rust types that all correspond to the C type called
12 //! `vm_area_struct`. The different structs represent what kind of access you have to the VMA, e.g.
13 //! [`VmaRef`] is used when you hold the mmap or vma read lock. Using the appropriate struct
14 //! ensures that you can't, for example, accidentally call a function that requires holding the
15 //! write lock when you only hold the read lock.
16 
17 use crate::{
18     bindings,
19     error::{code::EINVAL, to_result, Result},
20     mm::MmWithUser,
21     page::Page,
22     types::Opaque,
23 };
24 
25 use core::ops::Deref;
26 
27 /// A wrapper for the kernel's `struct vm_area_struct` with read access.
28 ///
29 /// It represents an area of virtual memory.
30 ///
31 /// # Invariants
32 ///
33 /// The caller must hold the mmap read lock or the vma read lock.
34 #[repr(transparent)]
35 pub struct VmaRef {
36     vma: Opaque<bindings::vm_area_struct>,
37 }
38 
39 // Methods you can call when holding the mmap or vma read lock (or stronger). They must be usable
40 // no matter what the vma flags are.
41 impl VmaRef {
42     /// Access a virtual memory area given a raw pointer.
43     ///
44     /// # Safety
45     ///
46     /// Callers must ensure that `vma` is valid for the duration of 'a, and that the mmap or vma
47     /// read lock (or stronger) is held for at least the duration of 'a.
48     #[inline]
49     pub unsafe fn from_raw<'a>(vma: *const bindings::vm_area_struct) -> &'a Self {
50         // SAFETY: The caller ensures that the invariants are satisfied for the duration of 'a.
51         unsafe { &*vma.cast() }
52     }
53 
54     /// Returns a raw pointer to this area.
55     #[inline]
56     pub fn as_ptr(&self) -> *mut bindings::vm_area_struct {
57         self.vma.get()
58     }
59 
60     /// Access the underlying `mm_struct`.
61     #[inline]
62     pub fn mm(&self) -> &MmWithUser {
63         // SAFETY: By the type invariants, this `vm_area_struct` is valid and we hold the mmap/vma
64         // read lock or stronger. This implies that the underlying mm has a non-zero value of
65         // `mm_users`.
66         unsafe { MmWithUser::from_raw((*self.as_ptr()).vm_mm) }
67     }
68 
69     /// Returns the flags associated with the virtual memory area.
70     ///
71     /// The possible flags are a combination of the constants in [`flags`].
72     #[inline]
73     pub fn flags(&self) -> vm_flags_t {
74         // SAFETY: By the type invariants, the caller holds at least the mmap read lock, so this
75         // access is not a data race.
76         unsafe { (*self.as_ptr()).__bindgen_anon_2.vm_flags }
77     }
78 
79     /// Returns the (inclusive) start address of the virtual memory area.
80     #[inline]
81     pub fn start(&self) -> usize {
82         // SAFETY: By the type invariants, the caller holds at least the mmap read lock, so this
83         // access is not a data race.
84         unsafe { (*self.as_ptr()).__bindgen_anon_1.__bindgen_anon_1.vm_start }
85     }
86 
87     /// Returns the (exclusive) end address of the virtual memory area.
88     #[inline]
89     pub fn end(&self) -> usize {
90         // SAFETY: By the type invariants, the caller holds at least the mmap read lock, so this
91         // access is not a data race.
92         unsafe { (*self.as_ptr()).__bindgen_anon_1.__bindgen_anon_1.vm_end }
93     }
94 
95     /// Zap pages in the given page range.
96     ///
97     /// This clears page table mappings for the range at the leaf level, leaving all other page
98     /// tables intact, and freeing any memory referenced by the VMA in this range. That is,
99     /// anonymous memory is completely freed, file-backed memory has its reference count on page
100     /// cache folio's dropped, any dirty data will still be written back to disk as usual.
101     ///
102     /// It may seem odd that we clear at the leaf level, this is however a product of the page
103     /// table structure used to map physical memory into a virtual address space - each virtual
104     /// address actually consists of a bitmap of array indices into page tables, which form a
105     /// hierarchical page table level structure.
106     ///
107     /// As a result, each page table level maps a multiple of page table levels below, and thus
108     /// span ever increasing ranges of pages. At the leaf or PTE level, we map the actual physical
109     /// memory.
110     ///
111     /// It is here where a zap operates, as it the only place we can be certain of clearing without
112     /// impacting any other virtual mappings. It is an implementation detail as to whether the
113     /// kernel goes further in freeing unused page tables, but for the purposes of this operation
114     /// we must only assume that the leaf level is cleared.
115     #[inline]
116     pub fn zap_vma_range(&self, address: usize, size: usize) {
117         let (end, did_overflow) = address.overflowing_add(size);
118         if did_overflow || address < self.start() || self.end() < end {
119             // TODO: call WARN_ONCE once Rust version of it is added
120             return;
121         }
122 
123         // SAFETY: By the type invariants, the caller has read access to this VMA, which is
124         // sufficient for this method call. This method has no requirements on the vma flags. The
125         // address range is checked to be within the vma.
126         unsafe { bindings::zap_vma_range(self.as_ptr(), address, size) };
127     }
128 
129     /// If the [`VM_MIXEDMAP`] flag is set, returns a [`VmaMixedMap`] to this VMA, otherwise
130     /// returns `None`.
131     ///
132     /// This can be used to access methods that require [`VM_MIXEDMAP`] to be set.
133     ///
134     /// [`VM_MIXEDMAP`]: flags::MIXEDMAP
135     #[inline]
136     pub fn as_mixedmap_vma(&self) -> Option<&VmaMixedMap> {
137         if self.flags() & flags::MIXEDMAP != 0 {
138             // SAFETY: We just checked that `VM_MIXEDMAP` is set. All other requirements are
139             // satisfied by the type invariants of `VmaRef`.
140             Some(unsafe { VmaMixedMap::from_raw(self.as_ptr()) })
141         } else {
142             None
143         }
144     }
145 }
146 
147 /// A wrapper for the kernel's `struct vm_area_struct` with read access and [`VM_MIXEDMAP`] set.
148 ///
149 /// It represents an area of virtual memory.
150 ///
151 /// This struct is identical to [`VmaRef`] except that it must only be used when the
152 /// [`VM_MIXEDMAP`] flag is set on the vma.
153 ///
154 /// # Invariants
155 ///
156 /// The caller must hold the mmap read lock or the vma read lock. The `VM_MIXEDMAP` flag must be
157 /// set.
158 ///
159 /// [`VM_MIXEDMAP`]: flags::MIXEDMAP
160 #[repr(transparent)]
161 pub struct VmaMixedMap {
162     vma: VmaRef,
163 }
164 
165 // Make all `VmaRef` methods available on `VmaMixedMap`.
166 impl Deref for VmaMixedMap {
167     type Target = VmaRef;
168 
169     #[inline]
170     fn deref(&self) -> &VmaRef {
171         &self.vma
172     }
173 }
174 
175 impl VmaMixedMap {
176     /// Access a virtual memory area given a raw pointer.
177     ///
178     /// # Safety
179     ///
180     /// Callers must ensure that `vma` is valid for the duration of 'a, and that the mmap read lock
181     /// (or stronger) is held for at least the duration of 'a. The `VM_MIXEDMAP` flag must be set.
182     #[inline]
183     pub unsafe fn from_raw<'a>(vma: *const bindings::vm_area_struct) -> &'a Self {
184         // SAFETY: The caller ensures that the invariants are satisfied for the duration of 'a.
185         unsafe { &*vma.cast() }
186     }
187 
188     /// Maps a single page at the given address within the virtual memory area.
189     ///
190     /// This operation does not take ownership of the page.
191     #[inline]
192     pub fn vm_insert_page(&self, address: usize, page: &Page) -> Result {
193         // SAFETY: By the type invariant of `Self` caller has read access and has verified that
194         // `VM_MIXEDMAP` is set. By invariant on `Page` the page has order 0.
195         to_result(unsafe { bindings::vm_insert_page(self.as_ptr(), address, page.as_ptr()) })
196     }
197 }
198 
199 /// A configuration object for setting up a VMA in an `f_ops->mmap()` hook.
200 ///
201 /// The `f_ops->mmap()` hook is called when a new VMA is being created, and the hook is able to
202 /// configure the VMA in various ways to fit the driver that owns it. Using `VmaNew` indicates that
203 /// you are allowed to perform operations on the VMA that can only be performed before the VMA is
204 /// fully initialized.
205 ///
206 /// # Invariants
207 ///
208 /// For the duration of 'a, the referenced vma must be undergoing initialization in an
209 /// `f_ops->mmap()` hook.
210 #[repr(transparent)]
211 pub struct VmaNew {
212     vma: VmaRef,
213 }
214 
215 // Make all `VmaRef` methods available on `VmaNew`.
216 impl Deref for VmaNew {
217     type Target = VmaRef;
218 
219     #[inline]
220     fn deref(&self) -> &VmaRef {
221         &self.vma
222     }
223 }
224 
225 impl VmaNew {
226     /// Access a virtual memory area given a raw pointer.
227     ///
228     /// # Safety
229     ///
230     /// Callers must ensure that `vma` is undergoing initial vma setup for the duration of 'a.
231     #[inline]
232     pub unsafe fn from_raw<'a>(vma: *mut bindings::vm_area_struct) -> &'a Self {
233         // SAFETY: The caller ensures that the invariants are satisfied for the duration of 'a.
234         unsafe { &*vma.cast() }
235     }
236 
237     /// Internal method for updating the vma flags.
238     ///
239     /// # Safety
240     ///
241     /// This must not be used to set the flags to an invalid value.
242     #[inline]
243     unsafe fn update_flags(&self, set: vm_flags_t, unset: vm_flags_t) {
244         let mut flags = self.flags();
245         flags |= set;
246         flags &= !unset;
247 
248         // SAFETY: This is not a data race: the vma is undergoing initial setup, so it's not yet
249         // shared. Additionally, `VmaNew` is `!Sync`, so it cannot be used to write in parallel.
250         // The caller promises that this does not set the flags to an invalid value.
251         unsafe { (*self.as_ptr()).__bindgen_anon_2.vm_flags = flags };
252     }
253 
254     /// Set the `VM_MIXEDMAP` flag on this vma.
255     ///
256     /// This enables the vma to contain both `struct page` and pure PFN pages. Returns a reference
257     /// that can be used to call `vm_insert_page` on the vma.
258     #[inline]
259     pub fn set_mixedmap(&self) -> &VmaMixedMap {
260         // SAFETY: We don't yet provide a way to set VM_PFNMAP, so this cannot put the flags in an
261         // invalid state.
262         unsafe { self.update_flags(flags::MIXEDMAP, 0) };
263 
264         // SAFETY: We just set `VM_MIXEDMAP` on the vma.
265         unsafe { VmaMixedMap::from_raw(self.vma.as_ptr()) }
266     }
267 
268     /// Set the `VM_IO` flag on this vma.
269     ///
270     /// This is used for memory mapped IO and similar. The flag tells other parts of the kernel to
271     /// avoid looking at the pages. For memory mapped IO this is useful as accesses to the pages
272     /// could have side effects.
273     #[inline]
274     pub fn set_io(&self) {
275         // SAFETY: Setting the VM_IO flag is always okay.
276         unsafe { self.update_flags(flags::IO, 0) };
277     }
278 
279     /// Set the `VM_DONTEXPAND` flag on this vma.
280     ///
281     /// This prevents the vma from being expanded with `mremap()`.
282     #[inline]
283     pub fn set_dontexpand(&self) {
284         // SAFETY: Setting the VM_DONTEXPAND flag is always okay.
285         unsafe { self.update_flags(flags::DONTEXPAND, 0) };
286     }
287 
288     /// Set the `VM_DONTCOPY` flag on this vma.
289     ///
290     /// This prevents the vma from being copied on fork. This option is only permanent if `VM_IO`
291     /// is set.
292     #[inline]
293     pub fn set_dontcopy(&self) {
294         // SAFETY: Setting the VM_DONTCOPY flag is always okay.
295         unsafe { self.update_flags(flags::DONTCOPY, 0) };
296     }
297 
298     /// Set the `VM_DONTDUMP` flag on this vma.
299     ///
300     /// This prevents the vma from being included in core dumps. This option is only permanent if
301     /// `VM_IO` is set.
302     #[inline]
303     pub fn set_dontdump(&self) {
304         // SAFETY: Setting the VM_DONTDUMP flag is always okay.
305         unsafe { self.update_flags(flags::DONTDUMP, 0) };
306     }
307 
308     /// Returns whether `VM_READ` is set.
309     ///
310     /// This flag indicates whether userspace is mapping this vma as readable.
311     #[inline]
312     pub fn readable(&self) -> bool {
313         (self.flags() & flags::READ) != 0
314     }
315 
316     /// Try to clear the `VM_MAYREAD` flag, failing if `VM_READ` is set.
317     ///
318     /// This flag indicates whether userspace is allowed to make this vma readable with
319     /// `mprotect()`.
320     ///
321     /// Note that this operation is irreversible. Once `VM_MAYREAD` has been cleared, it can never
322     /// be set again.
323     #[inline]
324     pub fn try_clear_mayread(&self) -> Result {
325         if self.readable() {
326             return Err(EINVAL);
327         }
328         // SAFETY: Clearing `VM_MAYREAD` is okay when `VM_READ` is not set.
329         unsafe { self.update_flags(0, flags::MAYREAD) };
330         Ok(())
331     }
332 
333     /// Returns whether `VM_WRITE` is set.
334     ///
335     /// This flag indicates whether userspace is mapping this vma as writable.
336     #[inline]
337     pub fn writable(&self) -> bool {
338         (self.flags() & flags::WRITE) != 0
339     }
340 
341     /// Try to clear the `VM_MAYWRITE` flag, failing if `VM_WRITE` is set.
342     ///
343     /// This flag indicates whether userspace is allowed to make this vma writable with
344     /// `mprotect()`.
345     ///
346     /// Note that this operation is irreversible. Once `VM_MAYWRITE` has been cleared, it can never
347     /// be set again.
348     #[inline]
349     pub fn try_clear_maywrite(&self) -> Result {
350         if self.writable() {
351             return Err(EINVAL);
352         }
353         // SAFETY: Clearing `VM_MAYWRITE` is okay when `VM_WRITE` is not set.
354         unsafe { self.update_flags(0, flags::MAYWRITE) };
355         Ok(())
356     }
357 
358     /// Returns whether `VM_EXEC` is set.
359     ///
360     /// This flag indicates whether userspace is mapping this vma as executable.
361     #[inline]
362     pub fn executable(&self) -> bool {
363         (self.flags() & flags::EXEC) != 0
364     }
365 
366     /// Try to clear the `VM_MAYEXEC` flag, failing if `VM_EXEC` is set.
367     ///
368     /// This flag indicates whether userspace is allowed to make this vma executable with
369     /// `mprotect()`.
370     ///
371     /// Note that this operation is irreversible. Once `VM_MAYEXEC` has been cleared, it can never
372     /// be set again.
373     #[inline]
374     pub fn try_clear_mayexec(&self) -> Result {
375         if self.executable() {
376             return Err(EINVAL);
377         }
378         // SAFETY: Clearing `VM_MAYEXEC` is okay when `VM_EXEC` is not set.
379         unsafe { self.update_flags(0, flags::MAYEXEC) };
380         Ok(())
381     }
382 }
383 
384 /// The integer type used for vma flags.
385 #[doc(inline)]
386 pub use bindings::vm_flags_t;
387 
388 /// All possible flags for [`VmaRef`].
389 pub mod flags {
390     use super::vm_flags_t;
391     use crate::bindings;
392 
393     /// No flags are set.
394     pub const NONE: vm_flags_t = bindings::VM_NONE as vm_flags_t;
395 
396     /// Mapping allows reads.
397     pub const READ: vm_flags_t = bindings::VM_READ as vm_flags_t;
398 
399     /// Mapping allows writes.
400     pub const WRITE: vm_flags_t = bindings::VM_WRITE as vm_flags_t;
401 
402     /// Mapping allows execution.
403     pub const EXEC: vm_flags_t = bindings::VM_EXEC as vm_flags_t;
404 
405     /// Mapping is shared.
406     pub const SHARED: vm_flags_t = bindings::VM_SHARED as vm_flags_t;
407 
408     /// Mapping may be updated to allow reads.
409     pub const MAYREAD: vm_flags_t = bindings::VM_MAYREAD as vm_flags_t;
410 
411     /// Mapping may be updated to allow writes.
412     pub const MAYWRITE: vm_flags_t = bindings::VM_MAYWRITE as vm_flags_t;
413 
414     /// Mapping may be updated to allow execution.
415     pub const MAYEXEC: vm_flags_t = bindings::VM_MAYEXEC as vm_flags_t;
416 
417     /// Mapping may be updated to be shared.
418     pub const MAYSHARE: vm_flags_t = bindings::VM_MAYSHARE as vm_flags_t;
419 
420     /// Page-ranges managed without `struct page`, just pure PFN.
421     pub const PFNMAP: vm_flags_t = bindings::VM_PFNMAP as vm_flags_t;
422 
423     /// Memory mapped I/O or similar.
424     pub const IO: vm_flags_t = bindings::VM_IO as vm_flags_t;
425 
426     /// Do not copy this vma on fork.
427     pub const DONTCOPY: vm_flags_t = bindings::VM_DONTCOPY as vm_flags_t;
428 
429     /// Cannot expand with mremap().
430     pub const DONTEXPAND: vm_flags_t = bindings::VM_DONTEXPAND as vm_flags_t;
431 
432     /// Lock the pages covered when they are faulted in.
433     pub const LOCKONFAULT: vm_flags_t = bindings::VM_LOCKONFAULT as vm_flags_t;
434 
435     /// Is a VM accounted object.
436     pub const ACCOUNT: vm_flags_t = bindings::VM_ACCOUNT as vm_flags_t;
437 
438     /// Should the VM suppress accounting.
439     pub const NORESERVE: vm_flags_t = bindings::VM_NORESERVE as vm_flags_t;
440 
441     /// Huge TLB Page VM.
442     pub const HUGETLB: vm_flags_t = bindings::VM_HUGETLB as vm_flags_t;
443 
444     /// Synchronous page faults. (DAX-specific)
445     pub const SYNC: vm_flags_t = bindings::VM_SYNC as vm_flags_t;
446 
447     /// Architecture-specific flag.
448     pub const ARCH_1: vm_flags_t = bindings::VM_ARCH_1 as vm_flags_t;
449 
450     /// Wipe VMA contents in child on fork.
451     pub const WIPEONFORK: vm_flags_t = bindings::VM_WIPEONFORK as vm_flags_t;
452 
453     /// Do not include in the core dump.
454     pub const DONTDUMP: vm_flags_t = bindings::VM_DONTDUMP as vm_flags_t;
455 
456     /// Not soft dirty clean area.
457     pub const SOFTDIRTY: vm_flags_t = bindings::VM_SOFTDIRTY as vm_flags_t;
458 
459     /// Can contain `struct page` and pure PFN pages.
460     pub const MIXEDMAP: vm_flags_t = bindings::VM_MIXEDMAP as vm_flags_t;
461 
462     /// MADV_HUGEPAGE marked this vma.
463     pub const HUGEPAGE: vm_flags_t = bindings::VM_HUGEPAGE as vm_flags_t;
464 
465     /// MADV_NOHUGEPAGE marked this vma.
466     pub const NOHUGEPAGE: vm_flags_t = bindings::VM_NOHUGEPAGE as vm_flags_t;
467 
468     /// KSM may merge identical pages.
469     pub const MERGEABLE: vm_flags_t = bindings::VM_MERGEABLE as vm_flags_t;
470 }
471