1 =============================
2 Examining Process Page Tables
3 =============================
5 pagemap is a new (as of 2.6.25) set of interfaces in the kernel that allow
6 userspace programs to examine the page tables and related information by
7 reading files in ``/proc``.
9 There are four components to pagemap:
11 * ``/proc/pid/pagemap``. This file lets a userspace process find out which
12 physical frame each virtual page is mapped to. It contains one 64-bit
13 value for each virtual page, containing the following data (from
14 ``fs/proc/task_mmu.c``, above pagemap_read):
16 * Bits 0-54 page frame number (PFN) if present
17 * Bits 0-4 swap type if swapped
18 * Bits 5-54 swap offset if swapped
19 * Bit 55 pte is soft-dirty (see
20 Documentation/admin-guide/mm/soft-dirty.rst)
21 * Bit 56 page exclusively mapped (since 4.2)
22 * Bit 57 pte is uffd-wp write-protected (since 5.13) (see
23 Documentation/admin-guide/mm/userfaultfd.rst)
25 * Bit 61 page is file-page or shared-anon (since 3.5)
29 Since Linux 4.0 only users with the CAP_SYS_ADMIN capability can get PFNs.
30 In 4.0 and 4.1 opens by unprivileged fail with -EPERM. Starting from
31 4.2 the PFN field is zeroed if the user does not have CAP_SYS_ADMIN.
32 Reason: information about PFNs helps in exploiting Rowhammer vulnerability.
34 If the page is not present but in swap, then the PFN contains an
35 encoding of the swap file number and the page's offset into the
36 swap. Unmapped pages return a null PFN. This allows determining
37 precisely which pages are mapped (or in swap) and comparing mapped
38 pages between processes.
40 Efficient users of this interface will use ``/proc/pid/maps`` to
41 determine which areas of memory are actually mapped and llseek to
42 skip over unmapped regions.
44 * ``/proc/kpagecount``. This file contains a 64-bit count of the number of
45 times each page is mapped, indexed by PFN.
47 The page-types tool in the tools/mm directory can be used to query the
48 number of times a page is mapped.
50 * ``/proc/kpageflags``. This file contains a 64-bit set of flags for each
53 The flags are (from ``fs/proc/page.c``, above kpageflags_read):
83 * ``/proc/kpagecgroup``. This file contains a 64-bit inode number of the
84 memory cgroup each page is charged to, indexed by PFN. Only available when
87 Short descriptions to the page flags
88 ====================================
91 The page is being locked for exclusive access, e.g. by undergoing read/write
94 The page is managed by the SLAB/SLUB kernel memory allocator.
95 When compound page is used, either will only set this flag on the head
98 A free memory block managed by the buddy system allocator.
99 The buddy system organizes free memory in blocks of various orders.
100 An order N block has 2^N physically contiguous pages, with the BUDDY flag
101 set for and _only_ for the first page.
103 A compound page with order N consists of 2^N physically contiguous pages.
104 A compound page with order 2 takes the form of "HTTT", where H donates its
105 head page and T donates its tail page(s). The major consumers of compound
106 pages are hugeTLB pages (Documentation/admin-guide/mm/hugetlbpage.rst),
107 the SLUB etc. memory allocators and various device drivers.
108 However in this interface, only huge/giga pages are made visible
111 A compound page tail (see description above).
113 This is an integral part of a HugeTLB page.
115 Hardware detected memory corruption on this page: don't touch the data!
117 No page frame exists at the requested address.
119 Identical memory pages dynamically shared between one or more processes.
121 Contiguous pages which construct transparent hugepages.
123 The page is logically offline.
125 Zero page for pfn_zero or huge_zero page.
127 The page has not been accessed since it was marked idle (see
128 Documentation/admin-guide/mm/idle_page_tracking.rst).
129 Note that this flag may be stale in case the page was accessed via
130 a PTE. To make sure the flag is up-to-date one has to read
131 ``/sys/kernel/mm/page_idle/bitmap`` first.
133 The page is in use as a page table.
135 IO related page flags
136 ---------------------
141 The page has up-to-date data.
142 ie. for file backed page: (in-memory data revision >= on-disk one)
144 The page has been written to, hence contains new data.
145 i.e. for file backed page: (in-memory data revision > on-disk one)
147 The page is being synced to disk.
149 LRU related page flags
150 ----------------------
153 The page is in one of the LRU lists.
155 The page is in the active LRU list.
157 The page is in the unevictable (non-)LRU list It is somehow pinned and
158 not a candidate for LRU page reclaims, e.g. ramfs pages,
159 shmctl(SHM_LOCK) and mlock() memory segments.
161 The page has been referenced since last LRU list enqueue/requeue.
163 The page will be reclaimed soon after its pageout IO completed.
165 A memory mapped page.
167 A memory mapped page that is not part of a file.
169 The page is mapped to swap space, i.e. has an associated swap entry.
171 The page is backed by swap/RAM.
173 The page-types tool in the tools/mm directory can be used to query the
176 Using pagemap to do something useful
177 ====================================
179 The general procedure for using pagemap to find out about a process' memory
180 usage goes like this:
182 1. Read ``/proc/pid/maps`` to determine which parts of the memory space are
184 2. Select the maps you are interested in -- all of them, or a particular
185 library, or the stack or the heap, etc.
186 3. Open ``/proc/pid/pagemap`` and seek to the pages you would like to examine.
187 4. Read a u64 for each page from pagemap.
188 5. Open ``/proc/kpagecount`` and/or ``/proc/kpageflags``. For each PFN you
189 just read, seek to that entry in the file, and read the data you want.
191 For example, to find the "unique set size" (USS), which is the amount of
192 memory that a process is using that is not shared with any other process,
193 you can go through every map in the process, find the PFNs, look those up
194 in kpagecount, and tally up the number of pages that are only referenced
197 Exceptions for Shared Memory
198 ============================
200 Page table entries for shared pages are cleared when the pages are zapped or
201 swapped out. This makes swapped out pages indistinguishable from never-allocated
204 In kernel space, the swap location can still be retrieved from the page cache.
205 However, values stored only on the normal PTE get lost irretrievably when the
206 page is swapped out (i.e. SOFT_DIRTY).
208 In user space, whether the page is present, swapped or none can be deduced with
209 the help of lseek and/or mincore system calls.
211 lseek() can differentiate between accessed pages (present or swapped out) and
212 holes (none/non-allocated) by specifying the SEEK_DATA flag on the file where
213 the pages are backed. For anonymous shared pages, the file can be found in
214 ``/proc/pid/map_files/``.
216 mincore() can differentiate between pages in memory (present, including swap
217 cache) and out of memory (swapped out or none/non-allocated).
222 Reading from any of the files will return -EINVAL if you are not starting
223 the read on an 8-byte boundary (e.g., if you sought an odd number of bytes
224 into the file), or if the size of the read is not a multiple of 8 bytes.
226 Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is
227 always 12 at most architectures). Since Linux 3.11 their meaning changes
228 after first clear of soft-dirty bits. Since Linux 4.2 they are used for
229 flags unconditionally.
234 The ``PAGEMAP_SCAN`` IOCTL on the pagemap file can be used to get or optionally
235 clear the info about page table entries. The following operations are supported
238 - Scan the address range and get the memory ranges matching the provided criteria.
239 This is performed when the output buffer is specified.
240 - Write-protect the pages. The ``PM_SCAN_WP_MATCHING`` is used to write-protect
241 the pages of interest. The ``PM_SCAN_CHECK_WPASYNC`` aborts the operation if
242 non-Async Write Protected pages are found. The ``PM_SCAN_WP_MATCHING`` can be
243 used with or without ``PM_SCAN_CHECK_WPASYNC``.
244 - Both of those operations can be combined into one atomic operation where we can
245 get and write protect the pages as well.
247 Following flags about pages are currently supported:
249 - ``PAGE_IS_WPALLOWED`` - Page has async-write-protection enabled
250 - ``PAGE_IS_WRITTEN`` - Page has been written to from the time it was write protected
251 - ``PAGE_IS_FILE`` - Page is file backed
252 - ``PAGE_IS_PRESENT`` - Page is present in the memory
253 - ``PAGE_IS_SWAPPED`` - Page is in swapped
254 - ``PAGE_IS_PFNZERO`` - Page has zero PFN
255 - ``PAGE_IS_HUGE`` - Page is THP or Hugetlb backed
257 The ``struct pm_scan_arg`` is used as the argument of the IOCTL.
259 1. The size of the ``struct pm_scan_arg`` must be specified in the ``size``
260 field. This field will be helpful in recognizing the structure if extensions
262 2. The flags can be specified in the ``flags`` field. The ``PM_SCAN_WP_MATCHING``
263 and ``PM_SCAN_CHECK_WPASYNC`` are the only added flags at this time. The get
264 operation is optionally performed depending upon if the output buffer is
266 3. The range is specified through ``start`` and ``end``.
267 4. The walk can abort before visiting the complete range such as the user buffer
268 can get full etc. The walk ending address is specified in``end_walk``.
269 5. The output buffer of ``struct page_region`` array and size is specified in
270 ``vec`` and ``vec_len``.
271 6. The optional maximum requested pages are specified in the ``max_pages``.
272 7. The masks are specified in ``category_mask``, ``category_anyof_mask``,
273 ``category_inverted`` and ``return_mask``.
275 Find pages which have been written and WP them as well::
277 struct pm_scan_arg arg = {
279 .flags = PM_SCAN_CHECK_WPASYNC | PM_SCAN_CHECK_WPASYNC,
281 .category_mask = PAGE_IS_WRITTEN,
282 .return_mask = PAGE_IS_WRITTEN,
285 Find pages which have been written, are file backed, not swapped and either
288 struct pm_scan_arg arg = {
292 .category_mask = PAGE_IS_WRITTEN | PAGE_IS_SWAPPED,
293 .category_inverted = PAGE_IS_SWAPPED,
294 .category_anyof_mask = PAGE_IS_PRESENT | PAGE_IS_HUGE,
295 .return_mask = PAGE_IS_WRITTEN | PAGE_IS_SWAPPED |
296 PAGE_IS_PRESENT | PAGE_IS_HUGE,
299 The ``PAGE_IS_WRITTEN`` flag can be considered as a better-performing alternative
300 of soft-dirty flag. It doesn't get affected by VMA merging of the kernel and hence
301 the user can find the true soft-dirty pages in case of normal pages. (There may
302 still be extra dirty pages reported for THP or Hugetlb pages.)
304 "PAGE_IS_WRITTEN" category is used with uffd write protect-enabled ranges to
305 implement memory dirty tracking in userspace:
307 1. The userfaultfd file descriptor is created with ``userfaultfd`` syscall.
308 2. The ``UFFD_FEATURE_WP_UNPOPULATED`` and ``UFFD_FEATURE_WP_ASYNC`` features
309 are set by ``UFFDIO_API`` IOCTL.
310 3. The memory range is registered with ``UFFDIO_REGISTER_MODE_WP`` mode
311 through ``UFFDIO_REGISTER`` IOCTL.
312 4. Then any part of the registered memory or the whole memory region must
313 be write protected using ``PAGEMAP_SCAN`` IOCTL with flag ``PM_SCAN_WP_MATCHING``
314 or the ``UFFDIO_WRITEPROTECT`` IOCTL can be used. Both of these perform the
315 same operation. The former is better in terms of performance.
316 5. Now the ``PAGEMAP_SCAN`` IOCTL can be used to either just find pages which
317 have been written to since they were last marked and/or optionally write protect