1=========== 2Userfaultfd 3=========== 4 5Objective 6========= 7 8Userfaults allow the implementation of on-demand paging from userland 9and more generally they allow userland to take control of various 10memory page faults, something otherwise only the kernel code could do. 11 12For example userfaults allows a proper and more optimal implementation 13of the ``PROT_NONE+SIGSEGV`` trick. 14 15Design 16====== 17 18Userspace creates a new userfaultfd, initializes it, and registers one or more 19regions of virtual memory with it. Then, any page faults which occur within the 20region(s) result in a message being delivered to the userfaultfd, notifying 21userspace of the fault. 22 23The ``userfaultfd`` (aside from registering and unregistering virtual 24memory ranges) provides two primary functionalities: 25 261) ``read/POLLIN`` protocol to notify a userland thread of the faults 27 happening 28 292) 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 33 34The real advantage of userfaults if compared to regular virtual memory 35management of mremap/mprotect is that the userfaults in all their 36operations never involve heavyweight structures like vmas (in fact the 37``userfaultfd`` runtime load never takes the mmap_lock for writing). 38Vmas are not suitable for page- (or hugepage) granular fault tracking 39when dealing with virtual address spaces that could span 40Terabytes. Too many vmas would be needed for that. 41 42The ``userfaultfd``, once created, can also be 43passed using unix domain sockets to a manager process, so the same 44manager process could handle the userfaults of a multitude of 45different processes without them being aware about what is going on 46(well of course unless they later try to use the ``userfaultfd`` 47themselves on the same region the manager is already tracking, which 48is a corner case that would currently return ``-EBUSY``). 49 50API 51=== 52 53Creating a userfaultfd 54---------------------- 55 56There are two ways to create a new userfaultfd, each of which provide ways to 57restrict access to this functionality (since historically userfaultfds which 58handle kernel page faults have been a useful tool for exploiting the kernel). 59 60The first way, supported since userfaultfd was introduced, is the 61userfaultfd(2) syscall. Access to this is controlled in several ways: 62 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. 66 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 70 is set to 0. 71 72The 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 74yields equivalent userfaultfds to the userfaultfd(2) syscall. 75 76Unlike userfaultfd(2), access to /dev/userfaultfd is controlled via normal 77filesystem permissions (user/group/mode), which gives fine grained access to 78userfaultfd specifically, without also granting other unrelated privileges at 79the same time (as e.g. granting CAP_SYS_PTRACE would do). Users who have access 80to /dev/userfaultfd can always create userfaultfds that trap kernel page faults; 81vm.unprivileged_userfaultfd is not considered. 82 83Initializing a userfaultfd 84-------------------------- 85 86When first opened the ``userfaultfd`` must be enabled invoking the 87``UFFDIO_API`` ioctl specifying a ``uffdio_api.api`` value set to ``UFFD_API`` (or 88a later API version) which will specify the ``read/POLLIN`` protocol 89userland intends to speak on the ``UFFD`` and the ``uffdio_api.features`` 90userland requires. The ``UFFDIO_API`` ioctl if successful (i.e. if the 91requested ``uffdio_api.api`` is spoken also by the running kernel and the 92requested features are going to be enabled) will return into 93``uffdio_api.features`` and ``uffdio_api.ioctls`` two 64bit bitmasks of 94respectively all the available features of the read(2) protocol and 95the generic ioctl available. 96 97The ``uffdio_api.features`` bitmask returned by the ``UFFDIO_API`` ioctl 98defines what memory types are supported by the ``userfaultfd`` and what 99events, except page fault notifications, may be generated: 100 101- 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. 104 105- ``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. 110 111- ``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. 115 116- ``UFFD_FEATURE_MOVE`` indicates that the kernel supports moving an 117 existing page contents from userspace. 118 119The userland application should set the feature flags it intends to use 120when invoking the ``UFFDIO_API`` ioctl, to request that those features be 121enabled if supported. 122 123Once the ``userfaultfd`` API has been enabled the ``UFFDIO_REGISTER`` 124ioctl should be invoked (if present in the returned ``uffdio_api.ioctls`` 125bitmask) to register a memory range in the ``userfaultfd`` by setting the 126uffdio_register structure accordingly. The ``uffdio_register.mode`` 127bitmask will specify to the kernel which kind of faults to track for 128the range. The ``UFFDIO_REGISTER`` ioctl will return the 129``uffdio_register.ioctls`` bitmask of ioctls that are suitable to resolve 130userfaults on the range registered. Not all ioctls will necessarily be 131supported for all memory types (e.g. anonymous memory vs. shmem vs. 132hugetlbfs), or all types of intercepted faults. 133 134Userland can use the ``uffdio_register.ioctls`` to manage the virtual 135address space in the background (to add or potentially also remove 136memory from the ``userfaultfd`` registered range). This means a userfault 137could be triggering just before userland maps in the background the 138user-faulted page. 139 140Resolving Userfaults 141-------------------- 142 143There are three basic ways to resolve userfaults: 144 145- ``UFFDIO_COPY`` atomically copies some existing page contents from 146 userspace. 147 148- ``UFFDIO_ZEROPAGE`` atomically zeros the new page. 149 150- ``UFFDIO_CONTINUE`` maps an existing, previously-populated page. 151 152These operations are atomic in the sense that they guarantee nothing can 153see a half-populated page, since readers will keep userfaulting until the 154operation has finished. 155 156By default, these wake up userfaults blocked on the range in question. 157They support a ``UFFDIO_*_MODE_DONTWAKE`` ``mode`` flag, which indicates 158that waking will be done separately at some later time. 159 160Which ioctl to choose depends on the kind of page fault, and what we'd 161like to do to resolve it: 162 163- For ``UFFDIO_REGISTER_MODE_MISSING`` faults, the fault needs to be 164 resolved by either providing a new page (``UFFDIO_COPY``), or mapping 165 the zero page (``UFFDIO_ZEROPAGE``). By default, the kernel would map 166 the zero page for a missing fault. With userfaultfd, userspace can 167 decide what content to provide before the faulting thread continues. 168 169- For ``UFFDIO_REGISTER_MODE_MINOR`` faults, there is an existing page (in 170 the page cache). Userspace has the option of modifying the page's 171 contents before resolving the fault. Once the contents are correct 172 (modified or not), userspace asks the kernel to map the page and let the 173 faulting thread continue with ``UFFDIO_CONTINUE``. 174 175Notes: 176 177- You can tell which kind of fault occurred by examining 178 ``pagefault.flags`` within the ``uffd_msg``, checking for the 179 ``UFFD_PAGEFAULT_FLAG_*`` flags. 180 181- None of the page-delivering ioctls default to the range that you 182 registered with. You must fill in all fields for the appropriate 183 ioctl struct including the range. 184 185- You get the address of the access that triggered the missing page 186 event out of a struct uffd_msg that you read in the thread from the 187 uffd. You can supply as many pages as you want with these IOCTLs. 188 Keep in mind that unless you used DONTWAKE then the first of any of 189 those IOCTLs wakes up the faulting thread. 190 191- Be sure to test for all errors including 192 (``pollfd[0].revents & POLLERR``). This can happen, e.g. when ranges 193 supplied were incorrect. 194 195Write Protect Notifications 196--------------------------- 197 198This is equivalent to (but faster than) using mprotect and a SIGSEGV 199signal handler. 200 201Firstly you need to register a range with ``UFFDIO_REGISTER_MODE_WP``. 202Instead of using mprotect(2) you use 203``ioctl(uffd, UFFDIO_WRITEPROTECT, struct *uffdio_writeprotect)`` 204while ``mode = UFFDIO_WRITEPROTECT_MODE_WP`` 205in the struct passed in. The range does not default to and does not 206have to be identical to the range you registered with. You can write 207protect as many ranges as you like (inside the registered range). 208Then, in the thread reading from uffd the struct will have 209``msg.arg.pagefault.flags & UFFD_PAGEFAULT_FLAG_WP`` set. Now you send 210``ioctl(uffd, UFFDIO_WRITEPROTECT, struct *uffdio_writeprotect)`` 211again while ``pagefault.mode`` does not have ``UFFDIO_WRITEPROTECT_MODE_WP`` 212set. This wakes up the thread which will continue to run with writes. This 213allows you to do the bookkeeping about the write in the uffd reading 214thread before the ioctl. 215 216If you registered with both ``UFFDIO_REGISTER_MODE_MISSING`` and 217``UFFDIO_REGISTER_MODE_WP`` then you need to think about the sequence in 218which you supply a page and undo write protect. Note that there is a 219difference between writes into a WP area and into a !WP area. The 220former will have ``UFFD_PAGEFAULT_FLAG_WP`` set, the latter 221``UFFD_PAGEFAULT_FLAG_WRITE``. The latter did not fail on protection but 222you still need to supply a page when ``UFFDIO_REGISTER_MODE_MISSING`` was 223used. 224 225Userfaultfd write-protect mode currently behave differently on none ptes 226(when e.g. page is missing) over different types of memories. 227 228For anonymous memory, ``ioctl(UFFDIO_WRITEPROTECT)`` will ignore none ptes 229(e.g. when pages are missing and not populated). For file-backed memories 230like shmem and hugetlbfs, none ptes will be write protected just like a 231present pte. In other words, there will be a userfaultfd write fault 232message generated when writing to a missing page on file typed memories, 233as long as the page range was write-protected before. Such a message will 234not be generated on anonymous memories by default. 235 236If the application wants to be able to write protect none ptes on anonymous 237memory, one can pre-populate the memory with e.g. MADV_POPULATE_READ. On 238newer kernels, one can also detect the feature UFFD_FEATURE_WP_UNPOPULATED 239and set the feature bit in advance to make sure none ptes will also be 240write protected even upon anonymous memory. 241 242When using ``UFFDIO_REGISTER_MODE_WP`` in combination with either 243``UFFDIO_REGISTER_MODE_MISSING`` or ``UFFDIO_REGISTER_MODE_MINOR``, when 244resolving missing / minor faults with ``UFFDIO_COPY`` or ``UFFDIO_CONTINUE`` 245respectively, it may be desirable for the new page / mapping to be 246write-protected (so future writes will also result in a WP fault). These ioctls 247support a mode flag (``UFFDIO_COPY_MODE_WP`` or ``UFFDIO_CONTINUE_MODE_WP`` 248respectively) to configure the mapping this way. 249 250If the userfaultfd context has ``UFFD_FEATURE_WP_ASYNC`` feature bit set, 251any vma registered with write-protection will work in async mode rather 252than the default sync mode. 253 254In async mode, there will be no message generated when a write operation 255happens, meanwhile the write-protection will be resolved automatically by 256the kernel. It can be seen as a more accurate version of soft-dirty 257tracking and it can be different in a few ways: 258 259 - The dirty result will not be affected by vma changes (e.g. vma 260 merging) because the dirty is only tracked by the pte. 261 262 - It supports range operations by default, so one can enable tracking on 263 any range of memory as long as page aligned. 264 265 - Dirty information will not get lost if the pte was zapped due to 266 various reasons (e.g. during split of a shmem transparent huge page). 267 268 - Due to a reverted meaning of soft-dirty (page clean when uffd-wp bit 269 set; dirty when uffd-wp bit cleared), it has different semantics on 270 some of the memory operations. For example: ``MADV_DONTNEED`` on 271 anonymous (or ``MADV_REMOVE`` on a file mapping) will be treated as 272 dirtying of memory by dropping uffd-wp bit during the procedure. 273 274The user app can collect the "written/dirty" status by looking up the 275uffd-wp bit for the pages being interested in /proc/pagemap. 276 277The page will not be under track of uffd-wp async mode until the page is 278explicitly write-protected by ``ioctl(UFFDIO_WRITEPROTECT)`` with the mode 279flag ``UFFDIO_WRITEPROTECT_MODE_WP`` set. Trying to resolve a page fault 280that was tracked by async mode userfaultfd-wp is invalid. 281 282When userfaultfd-wp async mode is used alone, it can be applied to all 283kinds of memory. 284 285Memory Poisioning Emulation 286--------------------------- 287 288In response to a fault (either missing or minor), an action userspace can 289take to "resolve" it is to issue a ``UFFDIO_POISON``. This will cause any 290future faulters to either get a SIGBUS, or in KVM's case the guest will 291receive an MCE as if there were hardware memory poisoning. 292 293This is used to emulate hardware memory poisoning. Imagine a VM running on a 294machine which experiences a real hardware memory error. Later, we live migrate 295the VM to another physical machine. Since we want the migration to be 296transparent to the guest, we want that same address range to act as if it was 297still poisoned, even though it's on a new physical host which ostensibly 298doesn't have a memory error in the exact same spot. 299 300QEMU/KVM 301======== 302 303QEMU/KVM is using the ``userfaultfd`` syscall to implement postcopy live 304migration. Postcopy live migration is one form of memory 305externalization consisting of a virtual machine running with part or 306all of its memory residing on a different node in the cloud. The 307``userfaultfd`` abstraction is generic enough that not a single line of 308KVM kernel code had to be modified in order to add postcopy live 309migration to QEMU. 310 311Guest async page faults, ``FOLL_NOWAIT`` and all other ``GUP*`` features work 312just fine in combination with userfaults. Userfaults trigger async 313page faults in the guest scheduler so those guest processes that 314aren't waiting for userfaults (i.e. network bound) can keep running in 315the guest vcpus. 316 317It is generally beneficial to run one pass of precopy live migration 318just before starting postcopy live migration, in order to avoid 319generating userfaults for readonly guest regions. 320 321The implementation of postcopy live migration currently uses one 322single bidirectional socket but in the future two different sockets 323will be used (to reduce the latency of the userfaults to the minimum 324possible without having to decrease ``/proc/sys/net/ipv4/tcp_wmem``). 325 326The QEMU in the source node writes all pages that it knows are missing 327in the destination node, into the socket, and the migration thread of 328the QEMU running in the destination node runs ``UFFDIO_COPY|ZEROPAGE`` 329ioctls on the ``userfaultfd`` in order to map the received pages into the 330guest (``UFFDIO_ZEROCOPY`` is used if the source page was a zero page). 331 332A different postcopy thread in the destination node listens with 333poll() to the ``userfaultfd`` in parallel. When a ``POLLIN`` event is 334generated after a userfault triggers, the postcopy thread read() from 335the ``userfaultfd`` and receives the fault address (or ``-EAGAIN`` in case the 336userfault was already resolved and waken by a ``UFFDIO_COPY|ZEROPAGE`` run 337by the parallel QEMU migration thread). 338 339After the QEMU postcopy thread (running in the destination node) gets 340the userfault address it writes the information about the missing page 341into the socket. The QEMU source node receives the information and 342roughly "seeks" to that page address and continues sending all 343remaining missing pages from that new page offset. Soon after that 344(just the time to flush the tcp_wmem queue through the network) the 345migration thread in the QEMU running in the destination node will 346receive the page that triggered the userfault and it'll map it as 347usual with the ``UFFDIO_COPY|ZEROPAGE`` (without actually knowing if it 348was spontaneously sent by the source or if it was an urgent page 349requested through a userfault). 350 351By the time the userfaults start, the QEMU in the destination node 352doesn't need to keep any per-page state bitmap relative to the live 353migration around and a single per-page bitmap has to be maintained in 354the QEMU running in the source node to know which pages are still 355missing in the destination node. The bitmap in the source node is 356checked to find which missing pages to send in round robin and we seek 357over it when receiving incoming userfaults. After sending each page of 358course the bitmap is updated accordingly. It's also useful to avoid 359sending the same page twice (in case the userfault is read by the 360postcopy thread just before ``UFFDIO_COPY|ZEROPAGE`` runs in the migration 361thread). 362 363Non-cooperative userfaultfd 364=========================== 365 366When the ``userfaultfd`` is monitored by an external manager, the manager 367must be able to track changes in the process virtual memory 368layout. Userfaultfd can notify the manager about such changes using 369the same read(2) protocol as for the page fault notifications. The 370manager has to explicitly enable these events by setting appropriate 371bits in ``uffdio_api.features`` passed to ``UFFDIO_API`` ioctl: 372 373``UFFD_FEATURE_EVENT_FORK`` 374 enable ``userfaultfd`` hooks for fork(). When this feature is 375 enabled, the ``userfaultfd`` context of the parent process is 376 duplicated into the newly created process. The manager 377 receives ``UFFD_EVENT_FORK`` with file descriptor of the new 378 ``userfaultfd`` context in the ``uffd_msg.fork``. 379 380``UFFD_FEATURE_EVENT_REMAP`` 381 enable notifications about mremap() calls. When the 382 non-cooperative process moves a virtual memory area to a 383 different location, the manager will receive 384 ``UFFD_EVENT_REMAP``. The ``uffd_msg.remap`` will contain the old and 385 new addresses of the area and its original length. 386 387``UFFD_FEATURE_EVENT_REMOVE`` 388 enable notifications about madvise(MADV_REMOVE) and 389 madvise(MADV_DONTNEED) calls. The event ``UFFD_EVENT_REMOVE`` will 390 be generated upon these calls to madvise(). The ``uffd_msg.remove`` 391 will contain start and end addresses of the removed area. 392 393``UFFD_FEATURE_EVENT_UNMAP`` 394 enable notifications about memory unmapping. The manager will 395 get ``UFFD_EVENT_UNMAP`` with ``uffd_msg.remove`` containing start and 396 end addresses of the unmapped area. 397 398Although the ``UFFD_FEATURE_EVENT_REMOVE`` and ``UFFD_FEATURE_EVENT_UNMAP`` 399are pretty similar, they quite differ in the action expected from the 400``userfaultfd`` manager. In the former case, the virtual memory is 401removed, but the area is not, the area remains monitored by the 402``userfaultfd``, and if a page fault occurs in that area it will be 403delivered to the manager. The proper resolution for such page fault is 404to zeromap the faulting address. However, in the latter case, when an 405area is unmapped, either explicitly (with munmap() system call), or 406implicitly (e.g. during mremap()), the area is removed and in turn the 407``userfaultfd`` context for such area disappears too and the manager will 408not get further userland page faults from the removed area. Still, the 409notification is required in order to prevent manager from using 410``UFFDIO_COPY`` on the unmapped area. 411 412Unlike userland page faults which have to be synchronous and require 413explicit or implicit wakeup, all the events are delivered 414asynchronously and the non-cooperative process resumes execution as 415soon as manager executes read(). The ``userfaultfd`` manager should 416carefully synchronize calls to ``UFFDIO_COPY`` with the events 417processing. To aid the synchronization, the ``UFFDIO_COPY`` ioctl will 418return ``-ENOSPC`` when the monitored process exits at the time of 419``UFFDIO_COPY``, and ``-ENOENT``, when the non-cooperative process has changed 420its virtual memory layout simultaneously with outstanding ``UFFDIO_COPY`` 421operation. 422 423The current asynchronous model of the event delivery is optimal for 424single threaded non-cooperative ``userfaultfd`` manager implementations. A 425synchronous event delivery model can be added later as a new 426``userfaultfd`` feature to facilitate multithreading enhancements of the 427non cooperative manager, for example to allow ``UFFDIO_COPY`` ioctls to 428run in parallel to the event reception. Single threaded 429implementations should continue to use the current async event 430delivery model instead. 431