| 1 |
============================= |
| 2 |
Examining Process Page Tables |
| 3 |
============================= |
| 4 |
|
| 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``. |
| 8 |
|
| 9 |
There are four components to pagemap: |
| 10 |
|
| 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): |
| 15 |
|
| 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) |
| 24 |
* Bit 58 pte is a guard region (since 6.15) (see madvise (2) man page) |
| 25 |
* Bits 59-60 zero |
| 26 |
* Bit 61 page is file-page or shared-anon (since 3.5) |
| 27 |
* Bit 62 page swapped |
| 28 |
* Bit 63 page present |
| 29 |
|
| 30 |
Since Linux 4.0 only users with the CAP_SYS_ADMIN capability can get PFNs. |
| 31 |
In 4.0 and 4.1 opens by unprivileged fail with -EPERM. Starting from |
| 32 |
4.2 the PFN field is zeroed if the user does not have CAP_SYS_ADMIN. |
| 33 |
Reason: information about PFNs helps in exploiting Rowhammer vulnerability. |
| 34 |
|
| 35 |
If the page is not present but in swap, then the PFN contains an |
| 36 |
encoding of the swap file number and the page's offset into the |
| 37 |
swap. Unmapped pages return a null PFN. This allows determining |
| 38 |
precisely which pages are mapped (or in swap) and comparing mapped |
| 39 |
pages between processes. |
| 40 |
|
| 41 |
Traditionally, bit 56 indicates that a page is mapped exactly once and bit |
| 42 |
56 is clear when a page is mapped multiple times, even when mapped in the |
| 43 |
same process multiple times. In some kernel configurations, the semantics |
| 44 |
for pages part of a larger allocation (e.g., THP) can differ: bit 56 is set |
| 45 |
if all pages part of the corresponding large allocation are *certainly* |
| 46 |
mapped in the same process, even if the page is mapped multiple times in that |
| 47 |
process. Bit 56 is clear when any page page of the larger allocation |
| 48 |
is *maybe* mapped in a different process. In some cases, a large allocation |
| 49 |
might be treated as "maybe mapped by multiple processes" even though this |
| 50 |
is no longer the case. |
| 51 |
|
| 52 |
Efficient users of this interface will use ``/proc/pid/maps`` to |
| 53 |
determine which areas of memory are actually mapped and llseek to |
| 54 |
skip over unmapped regions. |
| 55 |
|
| 56 |
* ``/proc/kpagecount``. This file contains a 64-bit count of the number of |
| 57 |
times each page is mapped, indexed by PFN. Some kernel configurations do |
| 58 |
not track the precise number of times a page part of a larger allocation |
| 59 |
(e.g., THP) is mapped. In these configurations, the average number of |
| 60 |
mappings per page in this larger allocation is returned instead. However, |
| 61 |
if any page of the large allocation is mapped, the returned value will |
| 62 |
be at least 1. |
| 63 |
|
| 64 |
The page-types tool in the tools/mm directory can be used to query the |
| 65 |
number of times a page is mapped. |
| 66 |
|
| 67 |
* ``/proc/kpageflags``. This file contains a 64-bit set of flags for each |
| 68 |
page, indexed by PFN. |
| 69 |
|
| 70 |
The flags are (from ``fs/proc/page.c``, above kpageflags_read): |
| 71 |
|
| 72 |
0. LOCKED |
| 73 |
1. ERROR |
| 74 |
2. REFERENCED |
| 75 |
3. UPTODATE |
| 76 |
4. DIRTY |
| 77 |
5. LRU |
| 78 |
6. ACTIVE |
| 79 |
7. SLAB |
| 80 |
8. WRITEBACK |
| 81 |
9. RECLAIM |
| 82 |
10. BUDDY |
| 83 |
11. MMAP |
| 84 |
12. ANON |
| 85 |
13. SWAPCACHE |
| 86 |
14. SWAPBACKED |
| 87 |
15. COMPOUND_HEAD |
| 88 |
16. COMPOUND_TAIL |
| 89 |
17. HUGE |
| 90 |
18. UNEVICTABLE |
| 91 |
19. HWPOISON |
| 92 |
20. NOPAGE |
| 93 |
21. KSM |
| 94 |
22. THP |
| 95 |
23. OFFLINE |
| 96 |
24. ZERO_PAGE |
| 97 |
25. IDLE |
| 98 |
26. PGTABLE |
| 99 |
|
| 100 |
* ``/proc/kpagecgroup``. This file contains a 64-bit inode number of the |
| 101 |
memory cgroup each page is charged to, indexed by PFN. Only available when |
| 102 |
CONFIG_MEMCG is set. |
| 103 |
|
| 104 |
Short descriptions to the page flags |
| 105 |
==================================== |
| 106 |
|
| 107 |
0 - LOCKED |
| 108 |
The page is being locked for exclusive access, e.g. by undergoing read/write |
| 109 |
IO. |
| 110 |
7 - SLAB |
| 111 |
The page is managed by the SLAB/SLUB kernel memory allocator. |
| 112 |
When compound page is used, either will only set this flag on the head |
| 113 |
page. |
| 114 |
10 - BUDDY |
| 115 |
A free memory block managed by the buddy system allocator. |
| 116 |
The buddy system organizes free memory in blocks of various orders. |
| 117 |
An order N block has 2^N physically contiguous pages, with the BUDDY flag |
| 118 |
set for all pages. |
| 119 |
Before 4.6 only the first page of the block had the flag set. |
| 120 |
15 - COMPOUND_HEAD |
| 121 |
A compound page with order N consists of 2^N physically contiguous pages. |
| 122 |
A compound page with order 2 takes the form of "HTTT", where H donates its |
| 123 |
head page and T donates its tail page(s). The major consumers of compound |
| 124 |
pages are hugeTLB pages (Documentation/admin-guide/mm/hugetlbpage.rst), |
| 125 |
the SLUB etc. memory allocators and various device drivers. |
| 126 |
However in this interface, only huge/giga pages are made visible |
| 127 |
to end users. |
| 128 |
16 - COMPOUND_TAIL |
| 129 |
A compound page tail (see description above). |
| 130 |
17 - HUGE |
| 131 |
This is an integral part of a HugeTLB page. |
| 132 |
19 - HWPOISON |
| 133 |
Hardware detected memory corruption on this page: don't touch the data! |
| 134 |
20 - NOPAGE |
| 135 |
No page frame exists at the requested address. |
| 136 |
21 - KSM |
| 137 |
Identical memory pages dynamically shared between one or more processes. |
| 138 |
22 - THP |
| 139 |
Contiguous pages which construct THP of any size and mapped by any granularity. |
| 140 |
23 - OFFLINE |
| 141 |
The page is logically offline. |
| 142 |
24 - ZERO_PAGE |
| 143 |
Zero page for pfn_zero or huge_zero page. |
| 144 |
25 - IDLE |
| 145 |
The page has not been accessed since it was marked idle (see |
| 146 |
Documentation/admin-guide/mm/idle_page_tracking.rst). |
| 147 |
Note that this flag may be stale in case the page was accessed via |
| 148 |
a PTE. To make sure the flag is up-to-date one has to read |
| 149 |
``/sys/kernel/mm/page_idle/bitmap`` first. |
| 150 |
26 - PGTABLE |
| 151 |
The page is in use as a page table. |
| 152 |
|
| 153 |
IO related page flags |
| 154 |
--------------------- |
| 155 |
|
| 156 |
1 - ERROR |
| 157 |
IO error occurred. |
| 158 |
3 - UPTODATE |
| 159 |
The page has up-to-date data. |
| 160 |
ie. for file backed page: (in-memory data revision >= on-disk one) |
| 161 |
4 - DIRTY |
| 162 |
The page has been written to, hence contains new data. |
| 163 |
i.e. for file backed page: (in-memory data revision > on-disk one) |
| 164 |
8 - WRITEBACK |
| 165 |
The page is being synced to disk. |
| 166 |
|
| 167 |
LRU related page flags |
| 168 |
---------------------- |
| 169 |
|
| 170 |
5 - LRU |
| 171 |
The page is in one of the LRU lists. |
| 172 |
6 - ACTIVE |
| 173 |
The page is in the active LRU list. |
| 174 |
18 - UNEVICTABLE |
| 175 |
The page is in the unevictable (non-)LRU list It is somehow pinned and |
| 176 |
not a candidate for LRU page reclaims, e.g. ramfs pages, |
| 177 |
shmctl(SHM_LOCK) and mlock() memory segments. |
| 178 |
2 - REFERENCED |
| 179 |
The page has been referenced since last LRU list enqueue/requeue. |
| 180 |
9 - RECLAIM |
| 181 |
The page will be reclaimed soon after its pageout IO completed. |
| 182 |
11 - MMAP |
| 183 |
A memory mapped page. |
| 184 |
12 - ANON |
| 185 |
A memory mapped page that is not part of a file. |
| 186 |
13 - SWAPCACHE |
| 187 |
The page is mapped to swap space, i.e. has an associated swap entry. |
| 188 |
14 - SWAPBACKED |
| 189 |
The page is backed by swap/RAM. |
| 190 |
|
| 191 |
The page-types tool in the tools/mm directory can be used to query the |
| 192 |
above flags. |
| 193 |
|
| 194 |
Exceptions for Shared Memory |
| 195 |
============================ |
| 196 |
|
| 197 |
Page table entries for shared pages are cleared when the pages are zapped or |
| 198 |
swapped out. This makes swapped out pages indistinguishable from never-allocated |
| 199 |
ones. |
| 200 |
|
| 201 |
In kernel space, the swap location can still be retrieved from the page cache. |
| 202 |
However, values stored only on the normal PTE get lost irretrievably when the |
| 203 |
page is swapped out (i.e. SOFT_DIRTY). |
| 204 |
|
| 205 |
In user space, whether the page is present, swapped or none can be deduced with |
| 206 |
the help of lseek and/or mincore system calls. |
| 207 |
|
| 208 |
lseek() can differentiate between accessed pages (present or swapped out) and |
| 209 |
holes (none/non-allocated) by specifying the SEEK_DATA flag on the file where |
| 210 |
the pages are backed. For anonymous shared pages, the file can be found in |
| 211 |
``/proc/pid/map_files/``. |
| 212 |
|
| 213 |
mincore() can differentiate between pages in memory (present, including swap |
| 214 |
cache) and out of memory (swapped out or none/non-allocated). |
| 215 |
|
| 216 |
Other notes |
| 217 |
=========== |
| 218 |
|
| 219 |
Reading from any of the files will return -EINVAL if you are not starting |
| 220 |
the read on an 8-byte boundary (e.g., if you sought an odd number of bytes |
| 221 |
into the file), or if the size of the read is not a multiple of 8 bytes. |
| 222 |
|
| 223 |
Before Linux 3.11 pagemap bits 55-60 were used for "page-shift" (which is |
| 224 |
always 12 at most architectures). Since Linux 3.11 their meaning changes |
| 225 |
after first clear of soft-dirty bits. Since Linux 4.2 they are used for |
| 226 |
flags unconditionally. |
| 227 |
|
| 228 |
Pagemap Scan IOCTL |
| 229 |
================== |
| 230 |
|
| 231 |
The ``PAGEMAP_SCAN`` IOCTL on the pagemap file can be used to get or optionally |
| 232 |
clear the info about page table entries. The following operations are supported |
| 233 |
in this IOCTL: |
| 234 |
|
| 235 |
- Scan the address range and get the memory ranges matching the provided criteria. |
| 236 |
This is performed when the output buffer is specified. |
| 237 |
- Write-protect the pages. The ``PM_SCAN_WP_MATCHING`` is used to write-protect |
| 238 |
the pages of interest. The ``PM_SCAN_CHECK_WPASYNC`` aborts the operation if |
| 239 |
non-Async Write Protected pages are found. The ``PM_SCAN_WP_MATCHING`` can be |
| 240 |
used with or without ``PM_SCAN_CHECK_WPASYNC``. |
| 241 |
- Both of those operations can be combined into one atomic operation where we can |
| 242 |
get and write protect the pages as well. |
| 243 |
|
| 244 |
Following flags about pages are currently supported: |
| 245 |
|
| 246 |
- ``PAGE_IS_WPALLOWED`` - Page has async-write-protection enabled |
| 247 |
- ``PAGE_IS_WRITTEN`` - Page has been written to from the time it was write protected |
| 248 |
- ``PAGE_IS_FILE`` - Page is file backed |
| 249 |
- ``PAGE_IS_PRESENT`` - Page is present in the memory |
| 250 |
- ``PAGE_IS_SWAPPED`` - Page is in swapped |
| 251 |
- ``PAGE_IS_PFNZERO`` - Page has zero PFN |
| 252 |
- ``PAGE_IS_HUGE`` - Page is PMD-mapped THP or Hugetlb backed |
| 253 |
- ``PAGE_IS_SOFT_DIRTY`` - Page is soft-dirty |
| 254 |
- ``PAGE_IS_GUARD`` - Page is a part of a guard region |
| 255 |
|
| 256 |
The ``struct pm_scan_arg`` is used as the argument of the IOCTL. |
| 257 |
|
| 258 |
1. The size of the ``struct pm_scan_arg`` must be specified in the ``size`` |
| 259 |
field. This field will be helpful in recognizing the structure if extensions |
| 260 |
are done later. |
| 261 |
2. The flags can be specified in the ``flags`` field. The ``PM_SCAN_WP_MATCHING`` |
| 262 |
and ``PM_SCAN_CHECK_WPASYNC`` are the only added flags at this time. The get |
| 263 |
operation is optionally performed depending upon if the output buffer is |
| 264 |
provided or not. |
| 265 |
3. The range is specified through ``start`` and ``end``. |
| 266 |
4. The walk can abort before visiting the complete range such as the user buffer |
| 267 |
can get full etc. The walk ending address is specified in``end_walk``. |
| 268 |
5. The output buffer of ``struct page_region`` array and size is specified in |
| 269 |
``vec`` and ``vec_len``. |
| 270 |
6. The optional maximum requested pages are specified in the ``max_pages``. |
| 271 |
7. The masks are specified in ``category_mask``, ``category_anyof_mask``, |
| 272 |
``category_inverted`` and ``return_mask``. |
| 273 |
|
| 274 |
Find pages which have been written and WP them as well:: |
| 275 |
|
| 276 |
struct pm_scan_arg arg = { |
| 277 |
.size = sizeof(arg), |
| 278 |
.flags = PM_SCAN_CHECK_WPASYNC | PM_SCAN_CHECK_WPASYNC, |
| 279 |
.. |
| 280 |
.category_mask = PAGE_IS_WRITTEN, |
| 281 |
.return_mask = PAGE_IS_WRITTEN, |
| 282 |
}; |
| 283 |
|
| 284 |
Find pages which have been written, are file backed, not swapped and either |
| 285 |
present or huge:: |
| 286 |
|
| 287 |
struct pm_scan_arg arg = { |
| 288 |
.size = sizeof(arg), |
| 289 |
.flags = 0, |
| 290 |
.. |
| 291 |
.category_mask = PAGE_IS_WRITTEN | PAGE_IS_SWAPPED, |
| 292 |
.category_inverted = PAGE_IS_SWAPPED, |
| 293 |
.category_anyof_mask = PAGE_IS_PRESENT | PAGE_IS_HUGE, |
| 294 |
.return_mask = PAGE_IS_WRITTEN | PAGE_IS_SWAPPED | |
| 295 |
PAGE_IS_PRESENT | PAGE_IS_HUGE, |
| 296 |
}; |
| 297 |
|
| 298 |
The ``PAGE_IS_WRITTEN`` flag can be considered as a better-performing alternative |
| 299 |
of soft-dirty flag. It doesn't get affected by VMA merging of the kernel and hence |
| 300 |
the user can find the true soft-dirty pages in case of normal pages. (There may |
| 301 |
still be extra dirty pages reported for THP or Hugetlb pages.) |
| 302 |
|
| 303 |
"PAGE_IS_WRITTEN" category is used with uffd write protect-enabled ranges to |
| 304 |
implement memory dirty tracking in userspace: |
| 305 |
|
| 306 |
1. The userfaultfd file descriptor is created with ``userfaultfd`` syscall. |
| 307 |
2. The ``UFFD_FEATURE_WP_UNPOPULATED`` and ``UFFD_FEATURE_WP_ASYNC`` features |
| 308 |
are set by ``UFFDIO_API`` IOCTL. |
| 309 |
3. The memory range is registered with ``UFFDIO_REGISTER_MODE_WP`` mode |
| 310 |
through ``UFFDIO_REGISTER`` IOCTL. |
| 311 |
4. Then any part of the registered memory or the whole memory region must |
| 312 |
be write protected using ``PAGEMAP_SCAN`` IOCTL with flag ``PM_SCAN_WP_MATCHING`` |
| 313 |
or the ``UFFDIO_WRITEPROTECT`` IOCTL can be used. Both of these perform the |
| 314 |
same operation. The former is better in terms of performance. |
| 315 |
5. Now the ``PAGEMAP_SCAN`` IOCTL can be used to either just find pages which |
| 316 |
have been written to since they were last marked and/or optionally write protect |
| 317 |
the pages as well. |