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 116The userland application should set the feature flags it intends to use 117when invoking the ``UFFDIO_API`` ioctl, to request that those features be 118enabled if supported. 119 120Once the ``userfaultfd`` API has been enabled the ``UFFDIO_REGISTER`` 121ioctl should be invoked (if present in the returned ``uffdio_api.ioctls`` 122bitmask) to register a memory range in the ``userfaultfd`` by setting the 123uffdio_register structure accordingly. The ``uffdio_register.mode`` 124bitmask will specify to the kernel which kind of faults to track for 125the range. The ``UFFDIO_REGISTER`` ioctl will return the 126``uffdio_register.ioctls`` bitmask of ioctls that are suitable to resolve 127userfaults on the range registered. Not all ioctls will necessarily be 128supported for all memory types (e.g. anonymous memory vs. shmem vs. 129hugetlbfs), or all types of intercepted faults. 130 131Userland can use the ``uffdio_register.ioctls`` to manage the virtual 132address space in the background (to add or potentially also remove 133memory from the ``userfaultfd`` registered range). This means a userfault 134could be triggering just before userland maps in the background the 135user-faulted page. 136 137Resolving Userfaults 138-------------------- 139 140There are three basic ways to resolve userfaults: 141 142- ``UFFDIO_COPY`` atomically copies some existing page contents from 143 userspace. 144 145- ``UFFDIO_ZEROPAGE`` atomically zeros the new page. 146 147- ``UFFDIO_CONTINUE`` maps an existing, previously-populated page. 148 149These operations are atomic in the sense that they guarantee nothing can 150see a half-populated page, since readers will keep userfaulting until the 151operation has finished. 152 153By default, these wake up userfaults blocked on the range in question. 154They support a ``UFFDIO_*_MODE_DONTWAKE`` ``mode`` flag, which indicates 155that waking will be done separately at some later time. 156 157Which ioctl to choose depends on the kind of page fault, and what we'd 158like to do to resolve it: 159 160- 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. 165 166- 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``. 171 172Notes: 173 174- 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. 177 178- 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. 181 182- 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. 187 188- Be sure to test for all errors including 189 (``pollfd[0].revents & POLLERR``). This can happen, e.g. when ranges 190 supplied were incorrect. 191 192Write Protect Notifications 193--------------------------- 194 195This is equivalent to (but faster than) using mprotect and a SIGSEGV 196signal handler. 197 198Firstly you need to register a range with ``UFFDIO_REGISTER_MODE_WP``. 199Instead of using mprotect(2) you use 200``ioctl(uffd, UFFDIO_WRITEPROTECT, struct *uffdio_writeprotect)`` 201while ``mode = UFFDIO_WRITEPROTECT_MODE_WP`` 202in the struct passed in. The range does not default to and does not 203have to be identical to the range you registered with. You can write 204protect as many ranges as you like (inside the registered range). 205Then, 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)`` 208again while ``pagefault.mode`` does not have ``UFFDIO_WRITEPROTECT_MODE_WP`` 209set. This wakes up the thread which will continue to run with writes. This 210allows you to do the bookkeeping about the write in the uffd reading 211thread before the ioctl. 212 213If you registered with both ``UFFDIO_REGISTER_MODE_MISSING`` and 214``UFFDIO_REGISTER_MODE_WP`` then you need to think about the sequence in 215which you supply a page and undo write protect. Note that there is a 216difference between writes into a WP area and into a !WP area. The 217former will have ``UFFD_PAGEFAULT_FLAG_WP`` set, the latter 218``UFFD_PAGEFAULT_FLAG_WRITE``. The latter did not fail on protection but 219you still need to supply a page when ``UFFDIO_REGISTER_MODE_MISSING`` was 220used. 221 222Userfaultfd write-protect mode currently behave differently on none ptes 223(when e.g. page is missing) over different types of memories. 224 225For anonymous memory, ``ioctl(UFFDIO_WRITEPROTECT)`` will ignore none ptes 226(e.g. when pages are missing and not populated). For file-backed memories 227like shmem and hugetlbfs, none ptes will be write protected just like a 228present pte. In other words, there will be a userfaultfd write fault 229message generated when writing to a missing page on file typed memories, 230as long as the page range was write-protected before. Such a message will 231not be generated on anonymous memories by default. 232 233If the application wants to be able to write protect none ptes on anonymous 234memory, one can pre-populate the memory with e.g. MADV_POPULATE_READ. On 235newer kernels, one can also detect the feature UFFD_FEATURE_WP_UNPOPULATED 236and set the feature bit in advance to make sure none ptes will also be 237write protected even upon anonymous memory. 238 239When using ``UFFDIO_REGISTER_MODE_WP`` in combination with either 240``UFFDIO_REGISTER_MODE_MISSING`` or ``UFFDIO_REGISTER_MODE_MINOR``, when 241resolving missing / minor faults with ``UFFDIO_COPY`` or ``UFFDIO_CONTINUE`` 242respectively, it may be desirable for the new page / mapping to be 243write-protected (so future writes will also result in a WP fault). These ioctls 244support a mode flag (``UFFDIO_COPY_MODE_WP`` or ``UFFDIO_CONTINUE_MODE_WP`` 245respectively) to configure the mapping this way. 246 247If the userfaultfd context has ``UFFD_FEATURE_WP_ASYNC`` feature bit set, 248any vma registered with write-protection will work in async mode rather 249than the default sync mode. 250 251In async mode, there will be no message generated when a write operation 252happens, meanwhile the write-protection will be resolved automatically by 253the kernel. It can be seen as a more accurate version of soft-dirty 254tracking and it can be different in a few ways: 255 256 - The dirty result will not be affected by vma changes (e.g. vma 257 merging) because the dirty is only tracked by the pte. 258 259 - It supports range operations by default, so one can enable tracking on 260 any range of memory as long as page aligned. 261 262 - Dirty information will not get lost if the pte was zapped due to 263 various reasons (e.g. during split of a shmem transparent huge page). 264 265 - Due to a reverted meaning of soft-dirty (page clean when uffd-wp bit 266 set; dirty when uffd-wp bit cleared), it has different semantics on 267 some of the memory operations. For example: ``MADV_DONTNEED`` on 268 anonymous (or ``MADV_REMOVE`` on a file mapping) will be treated as 269 dirtying of memory by dropping uffd-wp bit during the procedure. 270 271The user app can collect the "written/dirty" status by looking up the 272uffd-wp bit for the pages being interested in /proc/pagemap. 273 274The page will not be under track of uffd-wp async mode until the page is 275explicitly write-protected by ``ioctl(UFFDIO_WRITEPROTECT)`` with the mode 276flag ``UFFDIO_WRITEPROTECT_MODE_WP`` set. Trying to resolve a page fault 277that was tracked by async mode userfaultfd-wp is invalid. 278 279When userfaultfd-wp async mode is used alone, it can be applied to all 280kinds of memory. 281 282Memory Poisioning Emulation 283--------------------------- 284 285In response to a fault (either missing or minor), an action userspace can 286take to "resolve" it is to issue a ``UFFDIO_POISON``. This will cause any 287future faulters to either get a SIGBUS, or in KVM's case the guest will 288receive an MCE as if there were hardware memory poisoning. 289 290This is used to emulate hardware memory poisoning. Imagine a VM running on a 291machine which experiences a real hardware memory error. Later, we live migrate 292the VM to another physical machine. Since we want the migration to be 293transparent to the guest, we want that same address range to act as if it was 294still poisoned, even though it's on a new physical host which ostensibly 295doesn't have a memory error in the exact same spot. 296 297QEMU/KVM 298======== 299 300QEMU/KVM is using the ``userfaultfd`` syscall to implement postcopy live 301migration. Postcopy live migration is one form of memory 302externalization consisting of a virtual machine running with part or 303all of its memory residing on a different node in the cloud. The 304``userfaultfd`` abstraction is generic enough that not a single line of 305KVM kernel code had to be modified in order to add postcopy live 306migration to QEMU. 307 308Guest async page faults, ``FOLL_NOWAIT`` and all other ``GUP*`` features work 309just fine in combination with userfaults. Userfaults trigger async 310page faults in the guest scheduler so those guest processes that 311aren't waiting for userfaults (i.e. network bound) can keep running in 312the guest vcpus. 313 314It is generally beneficial to run one pass of precopy live migration 315just before starting postcopy live migration, in order to avoid 316generating userfaults for readonly guest regions. 317 318The implementation of postcopy live migration currently uses one 319single bidirectional socket but in the future two different sockets 320will be used (to reduce the latency of the userfaults to the minimum 321possible without having to decrease ``/proc/sys/net/ipv4/tcp_wmem``). 322 323The QEMU in the source node writes all pages that it knows are missing 324in the destination node, into the socket, and the migration thread of 325the QEMU running in the destination node runs ``UFFDIO_COPY|ZEROPAGE`` 326ioctls on the ``userfaultfd`` in order to map the received pages into the 327guest (``UFFDIO_ZEROCOPY`` is used if the source page was a zero page). 328 329A different postcopy thread in the destination node listens with 330poll() to the ``userfaultfd`` in parallel. When a ``POLLIN`` event is 331generated after a userfault triggers, the postcopy thread read() from 332the ``userfaultfd`` and receives the fault address (or ``-EAGAIN`` in case the 333userfault was already resolved and waken by a ``UFFDIO_COPY|ZEROPAGE`` run 334by the parallel QEMU migration thread). 335 336After the QEMU postcopy thread (running in the destination node) gets 337the userfault address it writes the information about the missing page 338into the socket. The QEMU source node receives the information and 339roughly "seeks" to that page address and continues sending all 340remaining missing pages from that new page offset. Soon after that 341(just the time to flush the tcp_wmem queue through the network) the 342migration thread in the QEMU running in the destination node will 343receive the page that triggered the userfault and it'll map it as 344usual with the ``UFFDIO_COPY|ZEROPAGE`` (without actually knowing if it 345was spontaneously sent by the source or if it was an urgent page 346requested through a userfault). 347 348By the time the userfaults start, the QEMU in the destination node 349doesn't need to keep any per-page state bitmap relative to the live 350migration around and a single per-page bitmap has to be maintained in 351the QEMU running in the source node to know which pages are still 352missing in the destination node. The bitmap in the source node is 353checked to find which missing pages to send in round robin and we seek 354over it when receiving incoming userfaults. After sending each page of 355course the bitmap is updated accordingly. It's also useful to avoid 356sending the same page twice (in case the userfault is read by the 357postcopy thread just before ``UFFDIO_COPY|ZEROPAGE`` runs in the migration 358thread). 359 360Non-cooperative userfaultfd 361=========================== 362 363When the ``userfaultfd`` is monitored by an external manager, the manager 364must be able to track changes in the process virtual memory 365layout. Userfaultfd can notify the manager about such changes using 366the same read(2) protocol as for the page fault notifications. The 367manager has to explicitly enable these events by setting appropriate 368bits in ``uffdio_api.features`` passed to ``UFFDIO_API`` ioctl: 369 370``UFFD_FEATURE_EVENT_FORK`` 371 enable ``userfaultfd`` hooks for fork(). When this feature is 372 enabled, the ``userfaultfd`` context of the parent process is 373 duplicated into the newly created process. The manager 374 receives ``UFFD_EVENT_FORK`` with file descriptor of the new 375 ``userfaultfd`` context in the ``uffd_msg.fork``. 376 377``UFFD_FEATURE_EVENT_REMAP`` 378 enable notifications about mremap() calls. When the 379 non-cooperative process moves a virtual memory area to a 380 different location, the manager will receive 381 ``UFFD_EVENT_REMAP``. The ``uffd_msg.remap`` will contain the old and 382 new addresses of the area and its original length. 383 384``UFFD_FEATURE_EVENT_REMOVE`` 385 enable notifications about madvise(MADV_REMOVE) and 386 madvise(MADV_DONTNEED) calls. The event ``UFFD_EVENT_REMOVE`` will 387 be generated upon these calls to madvise(). The ``uffd_msg.remove`` 388 will contain start and end addresses of the removed area. 389 390``UFFD_FEATURE_EVENT_UNMAP`` 391 enable notifications about memory unmapping. The manager will 392 get ``UFFD_EVENT_UNMAP`` with ``uffd_msg.remove`` containing start and 393 end addresses of the unmapped area. 394 395Although the ``UFFD_FEATURE_EVENT_REMOVE`` and ``UFFD_FEATURE_EVENT_UNMAP`` 396are pretty similar, they quite differ in the action expected from the 397``userfaultfd`` manager. In the former case, the virtual memory is 398removed, but the area is not, the area remains monitored by the 399``userfaultfd``, and if a page fault occurs in that area it will be 400delivered to the manager. The proper resolution for such page fault is 401to zeromap the faulting address. However, in the latter case, when an 402area is unmapped, either explicitly (with munmap() system call), or 403implicitly (e.g. during mremap()), the area is removed and in turn the 404``userfaultfd`` context for such area disappears too and the manager will 405not get further userland page faults from the removed area. Still, the 406notification is required in order to prevent manager from using 407``UFFDIO_COPY`` on the unmapped area. 408 409Unlike userland page faults which have to be synchronous and require 410explicit or implicit wakeup, all the events are delivered 411asynchronously and the non-cooperative process resumes execution as 412soon as manager executes read(). The ``userfaultfd`` manager should 413carefully synchronize calls to ``UFFDIO_COPY`` with the events 414processing. To aid the synchronization, the ``UFFDIO_COPY`` ioctl will 415return ``-ENOSPC`` when the monitored process exits at the time of 416``UFFDIO_COPY``, and ``-ENOENT``, when the non-cooperative process has changed 417its virtual memory layout simultaneously with outstanding ``UFFDIO_COPY`` 418operation. 419 420The current asynchronous model of the event delivery is optimal for 421single threaded non-cooperative ``userfaultfd`` manager implementations. A 422synchronous event delivery model can be added later as a new 423``userfaultfd`` feature to facilitate multithreading enhancements of the 424non cooperative manager, for example to allow ``UFFDIO_COPY`` ioctls to 425run in parallel to the event reception. Single threaded 426implementations should continue to use the current async event 427delivery model instead. 428