8 Userfaults allow the implementation of on-demand paging from userland
9 and more generally they allow userland to take control of various
10 memory page faults, something otherwise only the kernel code could do.
12 For example userfaults allows a proper and more optimal implementation
13 of the ``PROT_NONE+SIGSEGV`` trick.
18 Userspace creates a new userfaultfd, initializes it, and registers one or more
19 regions of virtual memory with it. Then, any page faults which occur within the
20 region(s) result in a message being delivered to the userfaultfd, notifying
21 userspace of the fault.
23 The ``userfaultfd`` (aside from registering and unregistering virtual
24 memory ranges) provides two primary functionalities:
26 1) ``read/POLLIN`` protocol to notify a userland thread of the faults
29 2) various ``UFFDIO_*`` ioctls that can manage the virtual memory regions
30 registered in the ``userfaultfd`` that allows userland to efficiently
31 resolve the userfaults it receives via 1) or to manage the virtual
32 memory in the background
34 The real advantage of userfaults if compared to regular virtual memory
35 management of mremap/mprotect is that the userfaults in all their
36 operations never involve heavyweight structures like vmas (in fact the
37 ``userfaultfd`` runtime load never takes the mmap_lock for writing).
38 Vmas are not suitable for page- (or hugepage) granular fault tracking
39 when dealing with virtual address spaces that could span
40 Terabytes. Too many vmas would be needed for that.
42 The ``userfaultfd``, once created, can also be
43 passed using unix domain sockets to a manager process, so the same
44 manager process could handle the userfaults of a multitude of
45 different processes without them being aware about what is going on
46 (well of course unless they later try to use the ``userfaultfd``
47 themselves on the same region the manager is already tracking, which
48 is a corner case that would currently return ``-EBUSY``).
53 Creating a userfaultfd
54 ----------------------
56 There are two ways to create a new userfaultfd, each of which provide ways to
57 restrict access to this functionality (since historically userfaultfds which
58 handle kernel page faults have been a useful tool for exploiting the kernel).
60 The first way, supported since userfaultfd was introduced, is the
61 userfaultfd(2) syscall. Access to this is controlled in several ways:
63 - Any user can always create a userfaultfd which traps userspace page faults
64 only. Such a userfaultfd can be created using the userfaultfd(2) syscall
65 with the flag UFFD_USER_MODE_ONLY.
67 - In order to also trap kernel page faults for the address space, either the
68 process needs the CAP_SYS_PTRACE capability, or the system must have
69 vm.unprivileged_userfaultfd set to 1. By default, vm.unprivileged_userfaultfd
72 The second way, added to the kernel more recently, is by opening
73 /dev/userfaultfd and issuing a USERFAULTFD_IOC_NEW ioctl to it. This method
74 yields equivalent userfaultfds to the userfaultfd(2) syscall.
76 Unlike userfaultfd(2), access to /dev/userfaultfd is controlled via normal
77 filesystem permissions (user/group/mode), which gives fine grained access to
78 userfaultfd specifically, without also granting other unrelated privileges at
79 the same time (as e.g. granting CAP_SYS_PTRACE would do). Users who have access
80 to /dev/userfaultfd can always create userfaultfds that trap kernel page faults;
81 vm.unprivileged_userfaultfd is not considered.
83 Initializing a userfaultfd
84 --------------------------
86 When first opened the ``userfaultfd`` must be enabled invoking the
87 ``UFFDIO_API`` ioctl specifying a ``uffdio_api.api`` value set to ``UFFD_API`` (or
88 a later API version) which will specify the ``read/POLLIN`` protocol
89 userland intends to speak on the ``UFFD`` and the ``uffdio_api.features``
90 userland requires. The ``UFFDIO_API`` ioctl if successful (i.e. if the
91 requested ``uffdio_api.api`` is spoken also by the running kernel and the
92 requested features are going to be enabled) will return into
93 ``uffdio_api.features`` and ``uffdio_api.ioctls`` two 64bit bitmasks of
94 respectively all the available features of the read(2) protocol and
95 the generic ioctl available.
97 The ``uffdio_api.features`` bitmask returned by the ``UFFDIO_API`` ioctl
98 defines what memory types are supported by the ``userfaultfd`` and what
99 events, except page fault notifications, may be generated:
101 - The ``UFFD_FEATURE_EVENT_*`` flags indicate that various other events
102 other than page faults are supported. These events are described in more
103 detail below in the `Non-cooperative userfaultfd`_ section.
105 - ``UFFD_FEATURE_MISSING_HUGETLBFS`` and ``UFFD_FEATURE_MISSING_SHMEM``
106 indicate that the kernel supports ``UFFDIO_REGISTER_MODE_MISSING``
107 registrations for hugetlbfs and shared memory (covering all shmem APIs,
108 i.e. tmpfs, ``IPCSHM``, ``/dev/zero``, ``MAP_SHARED``, ``memfd_create``,
109 etc) virtual memory areas, respectively.
111 - ``UFFD_FEATURE_MINOR_HUGETLBFS`` indicates that the kernel supports
112 ``UFFDIO_REGISTER_MODE_MINOR`` registration for hugetlbfs virtual memory
113 areas. ``UFFD_FEATURE_MINOR_SHMEM`` is the analogous feature indicating
114 support for shmem virtual memory areas.
116 The userland application should set the feature flags it intends to use
117 when invoking the ``UFFDIO_API`` ioctl, to request that those features be
118 enabled if supported.
120 Once the ``userfaultfd`` API has been enabled the ``UFFDIO_REGISTER``
121 ioctl should be invoked (if present in the returned ``uffdio_api.ioctls``
122 bitmask) to register a memory range in the ``userfaultfd`` by setting the
123 uffdio_register structure accordingly. The ``uffdio_register.mode``
124 bitmask will specify to the kernel which kind of faults to track for
125 the range. The ``UFFDIO_REGISTER`` ioctl will return the
126 ``uffdio_register.ioctls`` bitmask of ioctls that are suitable to resolve
127 userfaults on the range registered. Not all ioctls will necessarily be
128 supported for all memory types (e.g. anonymous memory vs. shmem vs.
129 hugetlbfs), or all types of intercepted faults.
131 Userland can use the ``uffdio_register.ioctls`` to manage the virtual
132 address space in the background (to add or potentially also remove
133 memory from the ``userfaultfd`` registered range). This means a userfault
134 could be triggering just before userland maps in the background the
140 There are three basic ways to resolve userfaults:
142 - ``UFFDIO_COPY`` atomically copies some existing page contents from
145 - ``UFFDIO_ZEROPAGE`` atomically zeros the new page.
147 - ``UFFDIO_CONTINUE`` maps an existing, previously-populated page.
149 These operations are atomic in the sense that they guarantee nothing can
150 see a half-populated page, since readers will keep userfaulting until the
151 operation has finished.
153 By default, these wake up userfaults blocked on the range in question.
154 They support a ``UFFDIO_*_MODE_DONTWAKE`` ``mode`` flag, which indicates
155 that waking will be done separately at some later time.
157 Which ioctl to choose depends on the kind of page fault, and what we'd
158 like to do to resolve it:
160 - For ``UFFDIO_REGISTER_MODE_MISSING`` faults, the fault needs to be
161 resolved by either providing a new page (``UFFDIO_COPY``), or mapping
162 the zero page (``UFFDIO_ZEROPAGE``). By default, the kernel would map
163 the zero page for a missing fault. With userfaultfd, userspace can
164 decide what content to provide before the faulting thread continues.
166 - For ``UFFDIO_REGISTER_MODE_MINOR`` faults, there is an existing page (in
167 the page cache). Userspace has the option of modifying the page's
168 contents before resolving the fault. Once the contents are correct
169 (modified or not), userspace asks the kernel to map the page and let the
170 faulting thread continue with ``UFFDIO_CONTINUE``.
174 - You can tell which kind of fault occurred by examining
175 ``pagefault.flags`` within the ``uffd_msg``, checking for the
176 ``UFFD_PAGEFAULT_FLAG_*`` flags.
178 - None of the page-delivering ioctls default to the range that you
179 registered with. You must fill in all fields for the appropriate
180 ioctl struct including the range.
182 - You get the address of the access that triggered the missing page
183 event out of a struct uffd_msg that you read in the thread from the
184 uffd. You can supply as many pages as you want with these IOCTLs.
185 Keep in mind that unless you used DONTWAKE then the first of any of
186 those IOCTLs wakes up the faulting thread.
188 - Be sure to test for all errors including
189 (``pollfd[0].revents & POLLERR``). This can happen, e.g. when ranges
190 supplied were incorrect.
192 Write Protect Notifications
193 ---------------------------
195 This is equivalent to (but faster than) using mprotect and a SIGSEGV
198 Firstly you need to register a range with ``UFFDIO_REGISTER_MODE_WP``.
199 Instead of using mprotect(2) you use
200 ``ioctl(uffd, UFFDIO_WRITEPROTECT, struct *uffdio_writeprotect)``
201 while ``mode = UFFDIO_WRITEPROTECT_MODE_WP``
202 in the struct passed in. The range does not default to and does not
203 have to be identical to the range you registered with. You can write
204 protect as many ranges as you like (inside the registered range).
205 Then, in the thread reading from uffd the struct will have
206 ``msg.arg.pagefault.flags & UFFD_PAGEFAULT_FLAG_WP`` set. Now you send
207 ``ioctl(uffd, UFFDIO_WRITEPROTECT, struct *uffdio_writeprotect)``
208 again while ``pagefault.mode`` does not have ``UFFDIO_WRITEPROTECT_MODE_WP``
209 set. This wakes up the thread which will continue to run with writes. This
210 allows you to do the bookkeeping about the write in the uffd reading
211 thread before the ioctl.
213 If you registered with both ``UFFDIO_REGISTER_MODE_MISSING`` and
214 ``UFFDIO_REGISTER_MODE_WP`` then you need to think about the sequence in
215 which you supply a page and undo write protect. Note that there is a
216 difference between writes into a WP area and into a !WP area. The
217 former will have ``UFFD_PAGEFAULT_FLAG_WP`` set, the latter
218 ``UFFD_PAGEFAULT_FLAG_WRITE``. The latter did not fail on protection but
219 you still need to supply a page when ``UFFDIO_REGISTER_MODE_MISSING`` was
222 Userfaultfd write-protect mode currently behave differently on none ptes
223 (when e.g. page is missing) over different types of memories.
225 For anonymous memory, ``ioctl(UFFDIO_WRITEPROTECT)`` will ignore none ptes
226 (e.g. when pages are missing and not populated). For file-backed memories
227 like shmem and hugetlbfs, none ptes will be write protected just like a
228 present pte. In other words, there will be a userfaultfd write fault
229 message generated when writing to a missing page on file typed memories,
230 as long as the page range was write-protected before. Such a message will
231 not be generated on anonymous memories by default.
233 If the application wants to be able to write protect none ptes on anonymous
234 memory, one can pre-populate the memory with e.g. MADV_POPULATE_READ. On
235 newer kernels, one can also detect the feature UFFD_FEATURE_WP_UNPOPULATED
236 and set the feature bit in advance to make sure none ptes will also be
237 write protected even upon anonymous memory.
239 When using ``UFFDIO_REGISTER_MODE_WP`` in combination with either
240 ``UFFDIO_REGISTER_MODE_MISSING`` or ``UFFDIO_REGISTER_MODE_MINOR``, when
241 resolving missing / minor faults with ``UFFDIO_COPY`` or ``UFFDIO_CONTINUE``
242 respectively, it may be desirable for the new page / mapping to be
243 write-protected (so future writes will also result in a WP fault). These ioctls
244 support a mode flag (``UFFDIO_COPY_MODE_WP`` or ``UFFDIO_CONTINUE_MODE_WP``
245 respectively) to configure the mapping this way.
247 Memory Poisioning Emulation
248 ---------------------------
250 In response to a fault (either missing or minor), an action userspace can
251 take to "resolve" it is to issue a ``UFFDIO_POISON``. This will cause any
252 future faulters to either get a SIGBUS, or in KVM's case the guest will
253 receive an MCE as if there were hardware memory poisoning.
255 This is used to emulate hardware memory poisoning. Imagine a VM running on a
256 machine which experiences a real hardware memory error. Later, we live migrate
257 the VM to another physical machine. Since we want the migration to be
258 transparent to the guest, we want that same address range to act as if it was
259 still poisoned, even though it's on a new physical host which ostensibly
260 doesn't have a memory error in the exact same spot.
265 QEMU/KVM is using the ``userfaultfd`` syscall to implement postcopy live
266 migration. Postcopy live migration is one form of memory
267 externalization consisting of a virtual machine running with part or
268 all of its memory residing on a different node in the cloud. The
269 ``userfaultfd`` abstraction is generic enough that not a single line of
270 KVM kernel code had to be modified in order to add postcopy live
273 Guest async page faults, ``FOLL_NOWAIT`` and all other ``GUP*`` features work
274 just fine in combination with userfaults. Userfaults trigger async
275 page faults in the guest scheduler so those guest processes that
276 aren't waiting for userfaults (i.e. network bound) can keep running in
279 It is generally beneficial to run one pass of precopy live migration
280 just before starting postcopy live migration, in order to avoid
281 generating userfaults for readonly guest regions.
283 The implementation of postcopy live migration currently uses one
284 single bidirectional socket but in the future two different sockets
285 will be used (to reduce the latency of the userfaults to the minimum
286 possible without having to decrease ``/proc/sys/net/ipv4/tcp_wmem``).
288 The QEMU in the source node writes all pages that it knows are missing
289 in the destination node, into the socket, and the migration thread of
290 the QEMU running in the destination node runs ``UFFDIO_COPY|ZEROPAGE``
291 ioctls on the ``userfaultfd`` in order to map the received pages into the
292 guest (``UFFDIO_ZEROCOPY`` is used if the source page was a zero page).
294 A different postcopy thread in the destination node listens with
295 poll() to the ``userfaultfd`` in parallel. When a ``POLLIN`` event is
296 generated after a userfault triggers, the postcopy thread read() from
297 the ``userfaultfd`` and receives the fault address (or ``-EAGAIN`` in case the
298 userfault was already resolved and waken by a ``UFFDIO_COPY|ZEROPAGE`` run
299 by the parallel QEMU migration thread).
301 After the QEMU postcopy thread (running in the destination node) gets
302 the userfault address it writes the information about the missing page
303 into the socket. The QEMU source node receives the information and
304 roughly "seeks" to that page address and continues sending all
305 remaining missing pages from that new page offset. Soon after that
306 (just the time to flush the tcp_wmem queue through the network) the
307 migration thread in the QEMU running in the destination node will
308 receive the page that triggered the userfault and it'll map it as
309 usual with the ``UFFDIO_COPY|ZEROPAGE`` (without actually knowing if it
310 was spontaneously sent by the source or if it was an urgent page
311 requested through a userfault).
313 By the time the userfaults start, the QEMU in the destination node
314 doesn't need to keep any per-page state bitmap relative to the live
315 migration around and a single per-page bitmap has to be maintained in
316 the QEMU running in the source node to know which pages are still
317 missing in the destination node. The bitmap in the source node is
318 checked to find which missing pages to send in round robin and we seek
319 over it when receiving incoming userfaults. After sending each page of
320 course the bitmap is updated accordingly. It's also useful to avoid
321 sending the same page twice (in case the userfault is read by the
322 postcopy thread just before ``UFFDIO_COPY|ZEROPAGE`` runs in the migration
325 Non-cooperative userfaultfd
326 ===========================
328 When the ``userfaultfd`` is monitored by an external manager, the manager
329 must be able to track changes in the process virtual memory
330 layout. Userfaultfd can notify the manager about such changes using
331 the same read(2) protocol as for the page fault notifications. The
332 manager has to explicitly enable these events by setting appropriate
333 bits in ``uffdio_api.features`` passed to ``UFFDIO_API`` ioctl:
335 ``UFFD_FEATURE_EVENT_FORK``
336 enable ``userfaultfd`` hooks for fork(). When this feature is
337 enabled, the ``userfaultfd`` context of the parent process is
338 duplicated into the newly created process. The manager
339 receives ``UFFD_EVENT_FORK`` with file descriptor of the new
340 ``userfaultfd`` context in the ``uffd_msg.fork``.
342 ``UFFD_FEATURE_EVENT_REMAP``
343 enable notifications about mremap() calls. When the
344 non-cooperative process moves a virtual memory area to a
345 different location, the manager will receive
346 ``UFFD_EVENT_REMAP``. The ``uffd_msg.remap`` will contain the old and
347 new addresses of the area and its original length.
349 ``UFFD_FEATURE_EVENT_REMOVE``
350 enable notifications about madvise(MADV_REMOVE) and
351 madvise(MADV_DONTNEED) calls. The event ``UFFD_EVENT_REMOVE`` will
352 be generated upon these calls to madvise(). The ``uffd_msg.remove``
353 will contain start and end addresses of the removed area.
355 ``UFFD_FEATURE_EVENT_UNMAP``
356 enable notifications about memory unmapping. The manager will
357 get ``UFFD_EVENT_UNMAP`` with ``uffd_msg.remove`` containing start and
358 end addresses of the unmapped area.
360 Although the ``UFFD_FEATURE_EVENT_REMOVE`` and ``UFFD_FEATURE_EVENT_UNMAP``
361 are pretty similar, they quite differ in the action expected from the
362 ``userfaultfd`` manager. In the former case, the virtual memory is
363 removed, but the area is not, the area remains monitored by the
364 ``userfaultfd``, and if a page fault occurs in that area it will be
365 delivered to the manager. The proper resolution for such page fault is
366 to zeromap the faulting address. However, in the latter case, when an
367 area is unmapped, either explicitly (with munmap() system call), or
368 implicitly (e.g. during mremap()), the area is removed and in turn the
369 ``userfaultfd`` context for such area disappears too and the manager will
370 not get further userland page faults from the removed area. Still, the
371 notification is required in order to prevent manager from using
372 ``UFFDIO_COPY`` on the unmapped area.
374 Unlike userland page faults which have to be synchronous and require
375 explicit or implicit wakeup, all the events are delivered
376 asynchronously and the non-cooperative process resumes execution as
377 soon as manager executes read(). The ``userfaultfd`` manager should
378 carefully synchronize calls to ``UFFDIO_COPY`` with the events
379 processing. To aid the synchronization, the ``UFFDIO_COPY`` ioctl will
380 return ``-ENOSPC`` when the monitored process exits at the time of
381 ``UFFDIO_COPY``, and ``-ENOENT``, when the non-cooperative process has changed
382 its virtual memory layout simultaneously with outstanding ``UFFDIO_COPY``
385 The current asynchronous model of the event delivery is optimal for
386 single threaded non-cooperative ``userfaultfd`` manager implementations. A
387 synchronous event delivery model can be added later as a new
388 ``userfaultfd`` feature to facilitate multithreading enhancements of the
389 non cooperative manager, for example to allow ``UFFDIO_COPY`` ioctls to
390 run in parallel to the event reception. Single threaded
391 implementations should continue to use the current async event
392 delivery model instead.
projects / releases.git / blob