mm: rearrange madvise code to allow for reuse
Subsystems (10)
Suren Baghdasaryan
reviewerPATCH v11 1/3 mm: rearrange madvise code to allow for reuse
From: Colin Cross <[email protected]>
Refactor the madvise syscall to allow for parts of it to be reused by a prctl syscall that affects vmas.
Move the code that walks vmas in a virtual address range into a function that takes a function pointer as a parameter. The only caller for now is sys_madvise, which uses it to call madvise_vma_behavior on each vma, but the next patch will add an additional caller.
Move handling all vma behaviors inside madvise_behavior, and rename it to madvise_vma_behavior.
Move the code that updates the flags on a vma, including splitting or merging the vma as necessary, into a new function called madvise_update_vma. The next patch will add support for updating a new anon_name field as well.
---
previous version at:
https://lore.kernel.org/all/[email protected]/
changes in v11
- Fixed misspelling, per Rolf Eike Beer
mm/madvise.c | 338 +++++++++++++++++++++++++++------------------------ 1 file changed, 178 insertions(+), 160 deletions(-)
mm/madvise.c +178 -160
@@ -63,76 +63,20 @@ static int madvise_need_mmap_write(int behavior) }/* -* We can potentially split a vm area into separate -* areas, each area with its own behavior. +* Update the vm_flags on region of a vma, splitting it or merging it as +* necessary. Must be called with mmap_sem held for writing; */ -static long madvise_behavior (struct vm_area_struct *vma, -struct vm_area_struct **prev, -unsigned long start,unsigned long end,int behavior) +static int madvise_update_vma (struct vm_area_struct *vma, +struct vm_area_struct **prev,unsigned long start, +unsigned long end,unsigned long new_flags) {struct mm_struct *mm = vma->vm_mm; -int error =0 ; +int error;pgoff_t pgoff; -unsigned long new_flags = vma->vm_flags; - -switch (behavior) { -case MADV_NORMAL : - new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ ; -break ; -case MADV_SEQUENTIAL : - new_flags = (new_flags & ~VM_RAND_READ ) |VM_SEQ_READ ; -break ; -case MADV_RANDOM : - new_flags = (new_flags & ~VM_SEQ_READ ) |VM_RAND_READ ; -break ; -case MADV_DONTFORK : - new_flags |=VM_DONTCOPY ; -break ; -case MADV_DOFORK : -if (vma->vm_flags &VM_IO ) { - error = -EINVAL ; - goto out; - } - new_flags &= ~VM_DONTCOPY ; -break ; -case MADV_WIPEONFORK : -/* MADV_WIPEONFORK is only supported on anonymous memory. */ -if (vma->vm_file || vma->vm_flags &VM_SHARED ) { - error = -EINVAL ; - goto out; - } - new_flags |=VM_WIPEONFORK ; -break ; -case MADV_KEEPONFORK : - new_flags &= ~VM_WIPEONFORK ; -break ; -case MADV_DONTDUMP : - new_flags |=VM_DONTDUMP ; -break ; -case MADV_DODUMP : -if (!is_vm_hugetlb_page (vma) && new_flags &VM_SPECIAL ) { - error = -EINVAL ; - goto out; - } - new_flags &= ~VM_DONTDUMP ; -break ; -case MADV_MERGEABLE : -case MADV_UNMERGEABLE : - error =ksm_madvise (vma, start, end, behavior, &new_flags); -if (error) - goto out_convert_errno; -break ; -case MADV_HUGEPAGE : -case MADV_NOHUGEPAGE : - error =hugepage_madvise (vma, &new_flags, behavior); -if (error) - goto out_convert_errno; -break ; - }if (new_flags == vma->vm_flags) { *prev = vma; - goto out; +return 0 ; } pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT );
@@ -147,23 +91,19 @@ static long madvise_behavior(struct vm_area_struct *vma, *prev = vma;if (start != vma->vm_start) { -if (unlikely (mm->map_count >= sysctl_max_map_count)) { - error = -ENOMEM ; - goto out; - } +if (unlikely (mm->map_count >= sysctl_max_map_count)) +return -ENOMEM ; error =__split_vma (mm, vma, start,1 );if (error) - goto out_convert_errno; +return error; }if (end != vma->vm_end) { -if (unlikely (mm->map_count >= sysctl_max_map_count)) { - error = -ENOMEM ; - goto out; - } +if (unlikely (mm->map_count >= sysctl_max_map_count)) +return -ENOMEM ; error =__split_vma (mm, vma, end,0 );if (error) - goto out_convert_errno; +return error; } success:
@@ -172,15 +112,7 @@ static long madvise_behavior(struct vm_area_struct *vma,*/ vma->vm_flags = new_flags; -out_convert_errno: -/* -* madvise() returns EAGAIN if kernel resources, such as -* slab, are temporarily unavailable. -*/ -if (error == -ENOMEM ) - error = -EAGAIN ; -out: -return error; +return 0 ; }#ifdef CONFIG_SWAP
@@ -930,6 +862,94 @@ static long madvise_remove(struct vm_area_struct *vma,return error; } +/* +* Apply an madvise behavior to a region of a vma. madvise_update_vma +* will handle splitting a vm area into separate areas, each area with its own +* behavior. +*/ +static int madvise_vma_behavior (struct vm_area_struct *vma, +struct vm_area_struct **prev, +unsigned long start,unsigned long end, +unsigned long behavior) +{ +int error; +unsigned long new_flags = vma->vm_flags; + +switch (behavior) { +case MADV_REMOVE : +return madvise_remove (vma, prev, start, end); +case MADV_WILLNEED : +return madvise_willneed (vma, prev, start, end); +case MADV_COLD : +return madvise_cold (vma, prev, start, end); +case MADV_PAGEOUT : +return madvise_pageout (vma, prev, start, end); +case MADV_FREE : +case MADV_DONTNEED : +return madvise_dontneed_free (vma, prev, start, end, behavior); +case MADV_POPULATE_READ : +case MADV_POPULATE_WRITE : +return madvise_populate (vma, prev, start, end, behavior); +case MADV_NORMAL : + new_flags = new_flags & ~VM_RAND_READ & ~VM_SEQ_READ ; +break ; +case MADV_SEQUENTIAL : + new_flags = (new_flags & ~VM_RAND_READ ) |VM_SEQ_READ ; +break ; +case MADV_RANDOM : + new_flags = (new_flags & ~VM_SEQ_READ ) |VM_RAND_READ ; +break ; +case MADV_DONTFORK : + new_flags |=VM_DONTCOPY ; +break ; +case MADV_DOFORK : +if (vma->vm_flags &VM_IO ) +return -EINVAL ; + new_flags &= ~VM_DONTCOPY ; +break ; +case MADV_WIPEONFORK : +/* MADV_WIPEONFORK is only supported on anonymous memory. */ +if (vma->vm_file || vma->vm_flags &VM_SHARED ) +return -EINVAL ; + new_flags |=VM_WIPEONFORK ; +break ; +case MADV_KEEPONFORK : + new_flags &= ~VM_WIPEONFORK ; +break ; +case MADV_DONTDUMP : + new_flags |=VM_DONTDUMP ; +break ; +case MADV_DODUMP : +if (!is_vm_hugetlb_page (vma) && new_flags &VM_SPECIAL ) +return -EINVAL ; + new_flags &= ~VM_DONTDUMP ; +break ; +case MADV_MERGEABLE : +case MADV_UNMERGEABLE : + error =ksm_madvise (vma, start, end, behavior, &new_flags); +if (error) + goto out; +break ; +case MADV_HUGEPAGE : +case MADV_NOHUGEPAGE : + error =hugepage_madvise (vma, &new_flags, behavior); +if (error) + goto out; +break ; + } + + error =madvise_update_vma (vma, prev, start, end, new_flags); + +out: +/* +* madvise() returns EAGAIN if kernel resources, such as +* slab, are temporarily unavailable. +*/ +if (error == -ENOMEM ) + error = -EAGAIN ; +return error; +} +#ifdef CONFIG_MEMORY_FAILURE /* * Error injection support for memory error handling.
@@ -978,30 +998,6 @@ static int madvise_inject_error(int behavior, }#endif -static long -madvise_vma (struct vm_area_struct *vma,struct vm_area_struct **prev, -unsigned long start,unsigned long end,int behavior) -{ -switch (behavior) { -case MADV_REMOVE : -return madvise_remove (vma, prev, start, end); -case MADV_WILLNEED : -return madvise_willneed (vma, prev, start, end); -case MADV_COLD : -return madvise_cold (vma, prev, start, end); -case MADV_PAGEOUT : -return madvise_pageout (vma, prev, start, end); -case MADV_FREE : -case MADV_DONTNEED : -return madvise_dontneed_free (vma, prev, start, end, behavior); -case MADV_POPULATE_READ : -case MADV_POPULATE_WRITE : -return madvise_populate (vma, prev, start, end, behavior); -default : -return madvise_behavior (vma, prev, start, end, behavior); - } -} -static bool madvise_behavior_valid (int behavior) {
@@ -1055,6 +1051,73 @@ process_madvise_behavior_valid(int behavior) } } +/* +* Walk the vmas in range [start,end), and call the visit function on each one. +* The visit function will get start and end parameters that cover the overlap +* between the current vma and the original range. Any unmapped regions in the +* original range will result in this function returning -ENOMEM while still +* calling the visit function on all of the existing vmas in the range. +* Must be called with the mmap_lock held for reading or writing. +*/ +static +int madvise_walk_vmas (struct mm_struct *mm,unsigned long start, +unsigned long end,unsigned long arg, +int (*visit)(struct vm_area_struct *vma, +struct vm_area_struct **prev,unsigned long start, +unsigned long end,unsigned long arg)) +{ +struct vm_area_struct *vma; +struct vm_area_struct *prev; +unsigned long tmp; +int unmapped_error =0 ; + +/* +* If the interval [start,end) covers some unmapped address +* ranges, just ignore them, but return -ENOMEM at the end. +* - different from the way of handling in mlock etc. +*/ + vma =find_vma_prev (mm, start, &prev); +if (vma && start > vma->vm_start) + prev = vma; + +for (;;) { +int error; + +/* Still start < end. */ +if (!vma) +return -ENOMEM ; + +/* Here start < (end|vma->vm_end). */ +if (start < vma->vm_start) { + unmapped_error = -ENOMEM ; + start = vma->vm_start; +if (start >= end) +break ; + } + +/* Here vma->vm_start <= start < (end|vma->vm_end) */ + tmp = vma->vm_end; +if (end < tmp) + tmp = end; + +/* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ + error =visit (vma, &prev, start, tmp, arg); +if (error) +return error; + start = tmp; +if (prev && start < prev->vm_end) + start = prev->vm_end; +if (start >= end) +break ; +if (prev) + vma = prev->vm_next; +else /* madvise_remove dropped mmap_lock */ + vma =find_vma (mm, start); + } + +return unmapped_error; +} +/* * The madvise(2) system call. *
@@ -1127,10 +1190,8 @@ process_madvise_behavior_valid(int behavior)*/ int do_madvise (struct mm_struct *mm,unsigned long start,size_t len_in,int behavior) { -unsigned long end, tmp; -struct vm_area_struct *vma, *prev; -int unmapped_error =0 ; -int error = -EINVAL ; +unsigned long end; +int error;int write;size_t len;struct blk_plug plug;
@@ -1138,23 +1199,22 @@ int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int beh start =untagged_addr (start);if (!madvise_behavior_valid (behavior)) -return error; +return -EINVAL ;if (!PAGE_ALIGNED (start)) -return error; +return -EINVAL ; len =PAGE_ALIGN (len_in);/* Check to see whether len was rounded up from small -ve to zero */ if (len_in && !len) -return error; +return -EINVAL ; end = start + len;if (end < start) -return error; +return -EINVAL ; - error =0 ;if (end == start) -return error; +return 0 ;#ifdef CONFIG_MEMORY_FAILURE if (behavior ==MADV_HWPOISON || behavior ==MADV_SOFT_OFFLINE )
@@ -1169,51 +1229,9 @@ int do_madvise(struct mm_struct *mm, unsigned long start, size_t len_in, int behmmap_read_lock (mm); } -/* -* If the interval [start,end) covers some unmapped address -* ranges, just ignore them, but return -ENOMEM at the end. -* - different from the way of handling in mlock etc. -*/ - vma =find_vma_prev (mm, start, &prev); -if (vma && start > vma->vm_start) - prev = vma; -blk_start_plug (&plug); -for (;;) { -/* Still start < end. */ - error = -ENOMEM ; -if (!vma) - goto out; - -/* Here start < (end|vma->vm_end). */ -if (start < vma->vm_start) { - unmapped_error = -ENOMEM ; - start = vma->vm_start; -if (start >= end) - goto out; - } - -/* Here vma->vm_start <= start < (end|vma->vm_end) */ - tmp = vma->vm_end; -if (end < tmp) - tmp = end; - -/* Here vma->vm_start <= start < tmp <= (end|vma->vm_end). */ - error =madvise_vma (vma, &prev, start, tmp, behavior); -if (error) - goto out; - start = tmp; -if (prev && start < prev->vm_end) - start = prev->vm_end; - error = unmapped_error; -if (start >= end) - goto out; -if (prev) - vma = prev->vm_next; -else /* madvise_remove dropped mmap_lock */ - vma =find_vma (mm, start); - } -out: + error =madvise_walk_vmas (mm, start, end, behavior, + madvise_vma_behavior);blk_finish_plug (&plug);if (write)mmap_write_unlock (mm);
Signature · 1 line
2.33.0.1079.g6e70778dc9-goog
Suren Baghdasaryan
reviewerPATCH v11 2/3 mm: add a field to store names for private anonymous memory
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in Documentation/vm/pagemap.txt to walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).
If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to /proc/pid/maps and /proc/pid/smaps to show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in /proc/pid/maps and /proc/pid/smaps as [anon:<name>].
Userspace can set the name for a region of memory by calling prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading /proc/pid/maps or /proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.
The name is stored in a pointer in the shared union in vm_area_struct that points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAME kernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.
The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from vm_area_struct.
One big concern is about fork() performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string during fork() or when splitting a vma it increments the refcount.
[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/
Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the
size specified in arg4. arg5 specifies the value of the attribute
to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.
Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including space), except '[',']','\','$' and '`'.
This feature is available only if the kernel is built with the CONFIG_ANON_VMA_NAME option enabled.
[surenb: rebased over v5.15-rc6, replaced userpointer with a kernel copy, added input sanitization and CONFIG_ANON_VMA_NAME config. The bulk of the work here was done by Colin Cross, therefore, with his permission, keeping him as the author]
---
previous version at:
https://lore.kernel.org/all/[email protected]/
changes in v11
- Introduced CONFIG_ANON_VMA_NAME, per Andrew Morton
- Amended patch description to explain the reason for string names, per Michal Hocko
- Fixed build error when CONFIG_ADVISE_SYSCALLS=n, per kernel test robot
Documentation/filesystems/proc.rst | 2 + fs/proc/task_mmu.c | 12 ++- fs/userfaultfd.c | 7 +- include/linux/mm.h | 13 ++- include/linux/mm_types.h | 64 +++++++++++++- include/uapi/linux/prctl.h | 3 + kernel/fork.c | 2 + kernel/sys.c | 63 ++++++++++++++ mm/Kconfig | 14 ++++ mm/madvise.c | 129 ++++++++++++++++++++++++++++- mm/mempolicy.c | 3 +- mm/mlock.c | 2 +- mm/mmap.c | 38 +++++---- mm/mprotect.c | 2 +- 14 files changed, 322 insertions(+), 32 deletions(-)
Documentation/filesystems/proc.rst +2 -0
@@ -431,6 +431,8 @@ is not associated with a file: [stack] the stack of the main process [vdso] the "virtual dynamic shared object", the kernel system call handler +[anon:<name>] an anonymous mapping that has been + named by userspace ======= ==================================== or if empty, the mapping is anonymous.
fs/proc/task_mmu.c +11 -1
@@ -308,6 +308,8 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma) name =arch_vma_name (vma);if (!name) { +const char *anon_name; +if (!mm) { name ="[vdso]" ; goto done;
@@ -319,8 +321,16 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma) goto done; } -if (is_stack (vma)) +if (is_stack (vma)) { name ="[stack]" ; + goto done; + } + + anon_name =vma_anon_name (vma); +if (anon_name) { +seq_pad (m,' ' ); +seq_printf (m,"[anon:%s]" , anon_name); + } } done:
fs/userfaultfd.c +4 -3
@@ -877,7 +877,7 @@ static int userfaultfd_release(struct inode *inode, struct file *file) new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), - NULL_VM_UFFD_CTX); + NULL_VM_UFFD_CTX, vma_anon_name(vma));if (prev) vma = prev;else
@@ -1436,7 +1436,8 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx, prev =vma_merge (mm, prev, start, vma_end, new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), - ((struct vm_userfaultfd_ctx ){ ctx })); + ((struct vm_userfaultfd_ctx ){ ctx }), +vma_anon_name (vma));if (prev) { vma = prev; goto next;
@@ -1613,7 +1614,7 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx, prev =vma_merge (mm, prev, start, vma_end, new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), -NULL_VM_UFFD_CTX ); +NULL_VM_UFFD_CTX ,vma_anon_name (vma));if (prev) { vma = prev; goto next;
include/linux/mm.h +12 -1
@@ -2548,7 +2548,7 @@ static inline int vma_adjust(struct vm_area_struct *vma, unsigned long start,extern struct vm_area_struct *vma_merge (struct mm_struct *,struct vm_area_struct *prev,unsigned long addr,unsigned long end,unsigned long vm_flags,struct anon_vma *,struct file *,pgoff_t , -struct mempolicy *,struct vm_userfaultfd_ctx ); +struct mempolicy *,struct vm_userfaultfd_ctx ,const char *);extern struct anon_vma *find_mergeable_anon_vma (struct vm_area_struct *);extern int __split_vma (struct mm_struct *,struct vm_area_struct *,unsigned long addr,int new_below);
@@ -3284,5 +3284,16 @@ static inline int seal_check_future_write(int seals, struct vm_area_struct *vma)return 0 ; } +#ifdef CONFIG_ANON_VMA_NAME +int madvise_set_anon_name (struct mm_struct *mm,unsigned long start, +unsigned long len_in,const char *name); +#else +static inline int +madvise_set_anon_name (struct mm_struct *mm,unsigned long start, +unsigned long len_in,const char *name) { +return 0 ; +} +#endif +#endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */
include/linux/mm_types.h +60 -4
75 lines not on this page · raw message
include/uapi/linux/prctl.h +3 -0
@@ -269,4 +269,7 @@ struct prctl_mm_map {# define PR_SCHED_CORE_SHARE_FROM 3/* pull core_sched cookie to pid */ # define PR_SCHED_CORE_MAX 4 +#define PR_SET_VMA 0x53564d41 +# define PR_SET_VMA_ANON_NAME 0 +#endif /* _LINUX_PRCTL_H */
kernel/fork.c +2 -0
@@ -366,12 +366,14 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) *new =data_race (*orig);INIT_LIST_HEAD (&new->anon_vma_chain); new->vm_next = new->vm_prev =NULL ; +dup_vma_anon_name (orig, new); }return new; }void vm_area_free (struct vm_area_struct *vma) { +free_vma_anon_name (vma);kmem_cache_free (vm_area_cachep, vma); }
kernel/sys.c +63 -0
75 lines not on this page · raw message
mm/Kconfig +14 -0
@@ -894,6 +894,20 @@ config IO_MAPPING config SECRETMEM def_bool ARCH_HAS_SET_DIRECT_MAP && !EMBEDDED +config ANON_VMA_NAME + bool "Anonymous VMA name support" + depends on PROC_FS && ADVISE_SYSCALLS && MMU + + help + Allow naming anonymous virtual memory areas. + + This feature allows assigning names to virtual memory areas. Assigned + names can be later retrieved from /proc/pid/maps and /proc/pid/smaps + and help identifying individual anonymous memory areas. + Assigning a name to anonymous virtual memory area might prevent that + area from being merged with adjacent virtual memory areas due to the + difference in their name. + source "mm/damon/Kconfig" endmenu
mm/madvise.c +125 -4
176 lines not on this page · raw message
mm/mempolicy.c +2 -1
@@ -810,7 +810,8 @@ static int mbind_range(struct mm_struct *mm, unsigned long start, ((vmstart - vma->vm_start) >>PAGE_SHIFT ); prev =vma_merge (mm, prev, vmstart, vmend, vma->vm_flags, vma->anon_vma, vma->vm_file, pgoff, - new_pol, vma->vm_userfaultfd_ctx); + new_pol, vma->vm_userfaultfd_ctx, +vma_anon_name (vma));if (prev) { vma = prev; next = vma->vm_next;
mm/mlock.c +1 -1
@@ -511,7 +511,7 @@ static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT ); *prev =vma_merge (mm, *prev, start, end, newflags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (*prev) { vma = *prev; goto success;
mm/mmap.c +22 -16
@@ -1029,7 +1029,8 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,*/ static inline int is_mergeable_vma (struct vm_area_struct *vma,struct file *file,unsigned long vm_flags, -struct vm_userfaultfd_ctx vm_userfaultfd_ctx) +struct vm_userfaultfd_ctx vm_userfaultfd_ctx, +const char *anon_name) {/* * VM_SOFTDIRTY should not prevent from VMA merging, if we
@@ -1047,6 +1048,8 @@ static inline int is_mergeable_vma(struct vm_area_struct *vma,return 0 ;if (!is_mergeable_vm_userfaultfd_ctx (vma, vm_userfaultfd_ctx))return 0 ; +if (!is_same_vma_anon_name (vma, anon_name)) +return 0 ;return 1 ; }
@@ -1079,9 +1082,10 @@ static int can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { - if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && + if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { if (vma->vm_pgoff == vm_pgoff) return 1;
@@ -1100,9 +1104,10 @@ static int can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { - if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && + if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { pgoff_t vm_pglen; vm_pglen =vma_pages (vma);
@@ -1113,9 +1118,9 @@ can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, }/* -* Given a mapping request (addr,end,vm_flags,file,pgoff), figure out -* whether that can be merged with its predecessor or its successor. -* Or both (it neatly fills a hole). +* Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), +* figure out whether that can be merged with its predecessor or its +* successor. Or both (it neatly fills a hole). * * In most cases - when called for mmap, brk or mremap - [addr,end) is * certain not to be mapped by the time vma_merge is called; but when
@@ -1160,7 +1165,8 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm, unsigned long end, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t pgoff, struct mempolicy *policy, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { pgoff_t pglen = (end - addr) >> PAGE_SHIFT;struct vm_area_struct *area, *next;
@@ -1190,7 +1196,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,mpol_equal (vma_policy (prev), policy) &&can_vma_merge_after (prev, vm_flags, anon_vma, file, pgoff, - vm_userfaultfd_ctx)) { + vm_userfaultfd_ctx, anon_name)) {/* * OK, it can. Can we now merge in the successor as well? */
@@ -1199,7 +1205,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,can_vma_merge_before (next, vm_flags, anon_vma, file, pgoff+pglen, - vm_userfaultfd_ctx) && + vm_userfaultfd_ctx, anon_name) &&is_mergeable_anon_vma (prev->anon_vma, next->anon_vma, NULL)) {/* cases 1, 6 */
@@ -1222,7 +1228,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,mpol_equal (policy,vma_policy (next)) &&can_vma_merge_before (next, vm_flags, anon_vma, file, pgoff+pglen, - vm_userfaultfd_ctx)) { + vm_userfaultfd_ctx, anon_name)) { if (prev && addr < prev->vm_end) /* case 4 */ err = __vma_adjust(prev, prev->vm_start, addr, prev->vm_pgoff,NULL , next);
@@ -1755,7 +1761,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,* Can we just expand an old mapping? */ vma =vma_merge (mm, prev, addr, addr + len, vm_flags, -NULL , file, pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL , file, pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (vma) goto out;
@@ -1804,7 +1810,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,*/ if (unlikely (vm_flags != vma->vm_flags && prev)) { merge =vma_merge (mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, -NULL , vma->vm_file, vma->vm_pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL , vma->vm_file, vma->vm_pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (merge) {/* ->mmap() can change vma->vm_file and fput the original file. So * fput the vma->vm_file here or we would add an extra fput for file
@@ -3057,7 +3063,7 @@ static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long fla/* Can we just expand an old private anonymous mapping? */ vma =vma_merge (mm, prev, addr, addr + len, flags, -NULL ,NULL , pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL ,NULL , pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (vma) goto out;
@@ -3250,7 +3256,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,return NULL ;/* should never get here */ new_vma =vma_merge (mm, prev, addr, addr + len, vma->vm_flags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (new_vma) {/* * Source vma may have been merged into new_vma
mm/mprotect.c +1 -1
@@ -464,7 +464,7 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT ); *pprev =vma_merge (mm, *pprev, start, end, newflags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (*pprev) { vma = *pprev;VM_WARN_ON ((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY );
Signature · 1 line
2.33.0.1079.g6e70778dc9-goog
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 112 lines · Suren Baghdasaryan
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in
Documentation/vm/pagemap.txtto walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to
/proc/pid/mapsand/proc/pid/smapsto show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in/proc/pid/mapsand/proc/pid/smapsas [anon:<name>].Userspace can set the name for a region of memory by calling
prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading/proc/pid/mapsor/proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.The name is stored in a pointer in the shared union in
vm_area_structthat points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAMEkernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from
vm_area_struct.
One big concern is aboutfork()performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string duringfork()or when splitting a vma it increments the refcount.[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the size specified in arg4. arg5 specifies the value of the attribute to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including space), except '[',']','\','$' and '`'.
Quoted 2 lines
This feature is available only if the kernel is built with the
CONFIG_ANON_VMA_NAMEoption enabled.
As Andrew suggested, I checked the image sizes with allnoconfig builds:
unpatched Linus' ToT
text data bss dec hex filename
1324759 32 73928 1398719 1557bf vmlinux
After the first patch is applied (madvise refactoring)
text data bss dec hex filename
1322346 32 73928 1396306 154e52 vmlinux
Quoted 1 line
Quoted 1 line
Quoted 1 line
2413 bytes decrease vs ToT <<<
After all patches applied with CONFIG_ANON_VMA_NAME=n
text data bss dec hex filename
1322337 32 73928 1396297 154e49 vmlinux
Quoted 1 line
Quoted 1 line
Quoted 1 line
2422 bytes decrease vs ToT <<<
After all patches applied with CONFIG_ANON_VMA_NAME=y
text data bss dec hex filename
1325228 32 73928 1399188 155994 vmlinux
Quoted 724 lines
Quoted 1 line
Quoted 1 line
469 bytes increase vs ToT <<<
Signed-off-by Colin Cross <[email protected]>[surenb: rebased over v5.15-rc6, replaced userpointer with a kernel copy, added input sanitization and
CONFIG_ANON_VMA_NAMEconfig. The bulk of the work here was done by Colin Cross, therefore, with his permission, keeping him as the author]Signed-off-by Suren Baghdasaryan <[email protected]>Reviewed-by Kees Cook <[email protected]>---
previous version at:
https://lore.kernel.org/all/[email protected]/changes in v11
- IntroducedCONFIG_ANON_VMA_NAME, per Andrew Morton
- Amended patch description to explain the reason for string names, per Michal Hocko
- Fixed build error whenCONFIG_ADVISE_SYSCALLS=n, per kernel test robotDocumentation/filesystems/proc.rst | 2 + fs/proc/task_mmu.c | 12 ++- fs/userfaultfd.c | 7 +- include/linux/mm.h | 13 ++- include/linux/mm_types.h | 64 +++++++++++++- include/uapi/linux/prctl.h | 3 + kernel/fork.c | 2 + kernel/sys.c | 63 ++++++++++++++ mm/Kconfig | 14 ++++ mm/madvise.c | 129 ++++++++++++++++++++++++++++- mm/mempolicy.c | 3 +- mm/mlock.c | 2 +- mm/mmap.c | 38 +++++---- mm/mprotect.c | 2 +- 14 files changed, 322 insertions(+), 32 deletions(-)Documentation/filesystems/proc.rst +2
-0@@ -431,6 +431,8 @@ is not associated with a file: [stack] the stack of the main process [vdso] the "virtual dynamic shared object", the kernel system call handler +[anon:<name>] an anonymous mapping that has been + named by userspace ======= ==================================== or if empty, the mapping is anonymous.fs/proc/task_mmu.c +11
-1@@ -308,6 +308,8 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma) name =arch_vma_name (vma);if (!name) { +const char *anon_name; +if (!mm) { name ="[vdso]" ; goto done;@@ -319,8 +321,16 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma) goto done; } -if (is_stack (vma)) +if (is_stack (vma)) { name ="[stack]" ; + goto done; + } + + anon_name =vma_anon_name (vma); +if (anon_name) { +seq_pad (m,' ' ); +seq_printf (m,"[anon:%s]" , anon_name); + } } done:fs/userfaultfd.c +4
-3@@ -877,7 +877,7 @@ static int userfaultfd_release(struct inode *inode, struct file *file) new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), - NULL_VM_UFFD_CTX); + NULL_VM_UFFD_CTX, vma_anon_name(vma));if (prev) vma = prev;else @@ -1436,7 +1436,8 @@ static int userfaultfd_register(struct userfaultfd_ctx *ctx, prev =vma_merge (mm, prev, start, vma_end, new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), - ((struct vm_userfaultfd_ctx ){ ctx })); + ((struct vm_userfaultfd_ctx ){ ctx }), +vma_anon_name (vma));if (prev) { vma = prev; goto next;@@ -1613,7 +1614,7 @@ static int userfaultfd_unregister(struct userfaultfd_ctx *ctx, prev =vma_merge (mm, prev, start, vma_end, new_flags, vma->anon_vma, vma->vm_file, vma->vm_pgoff,vma_policy (vma), -NULL_VM_UFFD_CTX ); +NULL_VM_UFFD_CTX ,vma_anon_name (vma));if (prev) { vma = prev; goto next;include/linux/mm.h +12
-1@@ -2548,7 +2548,7 @@ static inline int vma_adjust(struct vm_area_struct *vma, unsigned long start,extern struct vm_area_struct *vma_merge (struct mm_struct *,struct vm_area_struct *prev,unsigned long addr,unsigned long end,unsigned long vm_flags,struct anon_vma *,struct file *,pgoff_t , -struct mempolicy *,struct vm_userfaultfd_ctx ); +struct mempolicy *,struct vm_userfaultfd_ctx ,const char *);extern struct anon_vma *find_mergeable_anon_vma (struct vm_area_struct *);extern int __split_vma (struct mm_struct *,struct vm_area_struct *,unsigned long addr,int new_below);@@ -3284,5 +3284,16 @@ static inline int seal_check_future_write(int seals, struct vm_area_struct *vma)return 0 ; } +#ifdef CONFIG_ANON_VMA_NAME +int madvise_set_anon_name (struct mm_struct *mm,unsigned long start, +unsigned long len_in,const char *name); +#else +static inline int +madvise_set_anon_name (struct mm_struct *mm,unsigned long start, +unsigned long len_in,const char *name) { +return 0 ; +} +#endif +#endif /* __KERNEL__ */ #endif /* _LINUX_MM_H */ include/linux/mm_types.h +60
-4@@ -350,11 +350,19 @@ struct vm_area_struct {/* * For areas with an address space and backing store, * linkage into the address_space->i_mmap interval tree. +* +* For private anonymous mappings, a pointer to a null terminated string +* containing the name given to the vma, or NULL if unnamed. */ -struct { -struct rb_node rb; -unsigned long rb_subtree_last; - } shared; + +union { +struct { +struct rb_node rb; +unsigned long rb_subtree_last; + } shared; +/* Serialized by mmap_sem. */ +char *anon_name; + };/* * A file's MAP_PRIVATE vma can be in both i_mmap tree and anon_vma @@ -809,4 +817,52 @@ typedef struct {unsigned long val; } swp_entry_t; +#ifdef CONFIG_ANON_VMA_NAME +/* +* mmap_lock should be read-locked when calling vma_anon_name() and while using +* the returned pointer. +*/ +extern const char *vma_anon_name (struct vm_area_struct *vma); + +/* +* mmap_lock should be read-locked for orig_vma->vm_mm. +* mmap_lock should be write-locked for new_vma->vm_mm or new_vma should be +* isolated. +*/ +extern void dup_vma_anon_name (struct vm_area_struct *orig_vma, +struct vm_area_struct *new_vma); + +/* +* mmap_lock should be write-locked or vma should have been isolated under +* write-locked mmap_lock protection. +*/ +extern void free_vma_anon_name (struct vm_area_struct *vma); + +/* mmap_lock should be read-locked */ +static inline bool is_same_vma_anon_name (struct vm_area_struct *vma, +const char *name) +{ +const char *vma_name =vma_anon_name (vma); + +/* either both NULL, or pointers to same string */ +if (vma_name == name) +return true; + +return name && vma_name && !strcmp (name, vma_name); +} +#else /* CONFIG_ANON_VMA_NAME */ +static inline const char *vma_anon_name (struct vm_area_struct *vma) +{ +return NULL ; +} +static inline void dup_vma_anon_name (struct vm_area_struct *orig_vma, +struct vm_area_struct *new_vma) {} +static inline void free_vma_anon_name (struct vm_area_struct *vma) {} +static inline bool is_same_vma_anon_name (struct vm_area_struct *vma, +const char *name) +{ +return true; +} +#endif /* CONFIG_ANON_VMA_NAME */ +#endif /* _LINUX_MM_TYPES_H */ include/uapi/linux/prctl.h +3
-0@@ -269,4 +269,7 @@ struct prctl_mm_map {# define PR_SCHED_CORE_SHARE_FROM 3/* pull core_sched cookie to pid */ # define PR_SCHED_CORE_MAX 4 +#define PR_SET_VMA 0x53564d41 +# define PR_SET_VMA_ANON_NAME 0 +#endif /* _LINUX_PRCTL_H */ kernel/fork.c +2
-0@@ -366,12 +366,14 @@ struct vm_area_struct *vm_area_dup(struct vm_area_struct *orig) *new =data_race (*orig);INIT_LIST_HEAD (&new->anon_vma_chain); new->vm_next = new->vm_prev =NULL ; +dup_vma_anon_name (orig, new); }return new; }void vm_area_free (struct vm_area_struct *vma) { +free_vma_anon_name (vma);kmem_cache_free (vm_area_cachep, vma); }kernel/sys.c +63
-0@@ -2261,6 +2261,66 @@ int __weak arch_prctl_spec_ctrl_set(struct task_struct *t, unsigned long which,#define PR_IO_FLUSHER (PF_MEMALLOC_NOIO | PF_LOCAL_THROTTLE) +#ifdef CONFIG_ANON_VMA_NAME + +#define ANON_VMA_NAME_MAX_LEN 80 +#define ANON_VMA_NAME_INVALID_CHARS "\\`$[]" + +static inline bool is_valid_name_char (char ch) +{ +/* printable ascii characters, excluding ANON_VMA_NAME_INVALID_CHARS */ +return ch >0x1f && ch <0x7f && + !strchr (ANON_VMA_NAME_INVALID_CHARS , ch); +} + +static int prctl_set_vma (unsigned long opt,unsigned long addr, +unsigned long size,unsigned long arg) +{ +struct mm_struct *mm = current->mm; +const char __user *uname; +char *name, *pch; +int error; + +switch (opt) { +case PR_SET_VMA_ANON_NAME : + uname = (const char __user *)arg; +if (uname) { + name =strndup_user (uname,ANON_VMA_NAME_MAX_LEN ); + +if (IS_ERR (name)) +return PTR_ERR (name); + +for (pch = name; *pch !='\0' ; pch++) { +if (!is_valid_name_char (*pch)) { +kfree (name); +return -EINVAL ; + } + } + }else { +/* Reset the name */ + name =NULL ; + } + +mmap_write_lock (mm); + error =madvise_set_anon_name (mm, addr, size, name); +mmap_write_unlock (mm); +kfree (name); +break ; +default : + error = -EINVAL ; + } + +return error; +} + +#else /* CONFIG_ANON_VMA_NAME */ +static int prctl_set_vma (unsigned long opt,unsigned long start, +unsigned long size,unsigned long arg) +{ +return -EINVAL ; +} +#endif /* CONFIG_ANON_VMA_NAME */ + SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, unsigned long, arg4, unsigned long, arg5) {@@ -2530,6 +2590,9 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, error =sched_core_share_pid (arg2, arg3, arg4, arg5);break ;#endif +case PR_SET_VMA : + error =prctl_set_vma (arg2, arg3, arg4, arg5); +break ;default : error = -EINVAL ;break ;mm/Kconfig +14
-0@@ -894,6 +894,20 @@ config IO_MAPPING config SECRETMEM def_bool ARCH_HAS_SET_DIRECT_MAP && !EMBEDDED +config ANON_VMA_NAME + bool "Anonymous VMA name support" + depends on PROC_FS && ADVISE_SYSCALLS && MMU + + help + Allow naming anonymous virtual memory areas. + + This feature allows assigning names to virtual memory areas. Assigned + names can be later retrieved from /proc/pid/maps and /proc/pid/smaps + and help identifying individual anonymous memory areas. + Assigning a name to anonymous virtual memory area might prevent that + area from being merged with adjacent virtual memory areas due to the + difference in their name. + source "mm/damon/Kconfig" endmenumm/madvise.c +125
-4@@ -18,6 +18,7 @@#include <linux/fadvise.h> #include <linux/sched.h> #include <linux/sched/mm.h> +#include <linux/string.h> #include <linux/uio.h> #include <linux/ksm.h> #include <linux/fs.h> @@ -62,19 +63,84 @@ static int madvise_need_mmap_write(int behavior) } } +#ifdef CONFIG_ANON_VMA_NAME +static inline bool has_vma_anon_name (struct vm_area_struct *vma) +{ +return !vma->vm_file && vma->anon_name; +} + +const char *vma_anon_name (struct vm_area_struct *vma) +{ +if (!has_vma_anon_name (vma)) +return NULL ; + +mmap_assert_locked (vma->vm_mm); + +return vma->anon_name; +} + +void dup_vma_anon_name (struct vm_area_struct *orig_vma, +struct vm_area_struct *new_vma) +{ +if (!has_vma_anon_name (orig_vma)) +return ; + + new_vma->anon_name =kstrdup (orig_vma->anon_name,GFP_KERNEL ); +} + +void free_vma_anon_name (struct vm_area_struct *vma) +{ +if (!has_vma_anon_name (vma)) +return ; + +kfree (vma->anon_name); + vma->anon_name =NULL ; +} + +/* mmap_lock should be write-locked */ +static int replace_vma_anon_name (struct vm_area_struct *vma,const char *name) +{ +if (!name) { +free_vma_anon_name (vma); +return 0 ; + } + +if (vma->anon_name) { +/* Same name, nothing to do here */ +if (!strcmp (name, vma->anon_name)) +return 0 ; + +free_vma_anon_name (vma); + } + vma->anon_name =kstrdup (name,GFP_KERNEL ); +if (!vma->anon_name) +return -ENOMEM ; + +return 0 ; +} +#else /* CONFIG_ANON_VMA_NAME */ +static int replace_vma_anon_name (struct vm_area_struct *vma,const char *name) +{ +if (name) +return -EINVAL ; + +return 0 ; +} +#endif /* CONFIG_ANON_VMA_NAME */ /* * Update the vm_flags on region of a vma, splitting it or merging it as * necessary. Must be called with mmap_sem held for writing; */ static int madvise_update_vma (struct vm_area_struct *vma,struct vm_area_struct **prev,unsigned long start, -unsigned long end,unsigned long new_flags) +unsigned long end,unsigned long new_flags, +const char *name) {struct mm_struct *mm = vma->vm_mm;int error;pgoff_t pgoff; -if (new_flags == vma->vm_flags) { +if (new_flags == vma->vm_flags &&is_same_vma_anon_name (vma, name)) { *prev = vma;return 0 ; }@@ -82,7 +148,7 @@ static int madvise_update_vma(struct vm_area_struct *vma, pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT ); *prev =vma_merge (mm, *prev, start, end, new_flags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx, name);if (*prev) { vma = *prev; goto success;@@ -111,6 +177,11 @@ static int madvise_update_vma(struct vm_area_struct *vma,* vm_flags is protected by the mmap_lock held in write mode. */ vma->vm_flags = new_flags; +if (!vma->vm_file) { + error =replace_vma_anon_name (vma, name); +if (error) +return error; + }return 0 ; }@@ -938,7 +1009,8 @@ static int madvise_vma_behavior(struct vm_area_struct *vma,break ; } - error =madvise_update_vma (vma, prev, start, end, new_flags); + error =madvise_update_vma (vma, prev, start, end, new_flags, +vma_anon_name (vma)); out:/* @@ -1118,6 +1190,55 @@ int madvise_walk_vmas(struct mm_struct *mm, unsigned long start,return unmapped_error; } +#ifdef CONFIG_ANON_VMA_NAME +static int madvise_vma_anon_name (struct vm_area_struct *vma, +struct vm_area_struct **prev, +unsigned long start,unsigned long end, +unsigned long name) +{ +int error; + +/* Only anonymous mappings can be named */ +if (vma->vm_file) +return -EBADF ; + + error =madvise_update_vma (vma, prev, start, end, vma->vm_flags, + (const char *)name); + +/* +* madvise() returns EAGAIN if kernel resources, such as +* slab, are temporarily unavailable. +*/ +if (error == -ENOMEM ) + error = -EAGAIN ; +return error; +} + +int madvise_set_anon_name (struct mm_struct *mm,unsigned long start, +unsigned long len_in,const char *name) +{ +unsigned long end; +unsigned long len; + +if (start & ~PAGE_MASK ) +return -EINVAL ; + len = (len_in + ~PAGE_MASK ) &PAGE_MASK ; + +/* Check to see whether len was rounded up from small -ve to zero */ +if (len_in && !len) +return -EINVAL ; + + end = start + len; +if (end < start) +return -EINVAL ; + +if (end == start) +return 0 ; + +return madvise_walk_vmas (mm, start, end, (unsigned long )name, + madvise_vma_anon_name); +} +#endif /* CONFIG_ANON_VMA_NAME */ /* * The madvise(2) system call. * mm/mempolicy.c +2
-1@@ -810,7 +810,8 @@ static int mbind_range(struct mm_struct *mm, unsigned long start, ((vmstart - vma->vm_start) >>PAGE_SHIFT ); prev =vma_merge (mm, prev, vmstart, vmend, vma->vm_flags, vma->anon_vma, vma->vm_file, pgoff, - new_pol, vma->vm_userfaultfd_ctx); + new_pol, vma->vm_userfaultfd_ctx, +vma_anon_name (vma));if (prev) { vma = prev; next = vma->vm_next;mm/mlock.c +1
-1@@ -511,7 +511,7 @@ static int mlock_fixup(struct vm_area_struct *vma, struct vm_area_struct **prev, pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT ); *prev =vma_merge (mm, *prev, start, end, newflags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (*prev) { vma = *prev; goto success;mm/mmap.c +22
-16@@ -1029,7 +1029,8 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,*/ static inline int is_mergeable_vma (struct vm_area_struct *vma,struct file *file,unsigned long vm_flags, -struct vm_userfaultfd_ctx vm_userfaultfd_ctx) +struct vm_userfaultfd_ctx vm_userfaultfd_ctx, +const char *anon_name) {/* * VM_SOFTDIRTY should not prevent from VMA merging, if we @@ -1047,6 +1048,8 @@ static inline int is_mergeable_vma(struct vm_area_struct *vma,return 0 ;if (!is_mergeable_vm_userfaultfd_ctx (vma, vm_userfaultfd_ctx))return 0 ; +if (!is_same_vma_anon_name (vma, anon_name)) +return 0 ;return 1 ; }@@ -1079,9 +1082,10 @@ static int can_vma_merge_before(struct vm_area_struct *vma, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { - if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && + if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { if (vma->vm_pgoff == vm_pgoff) return 1;@@ -1100,9 +1104,10 @@ static int can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t vm_pgoff, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { - if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx) && + if (is_mergeable_vma(vma, file, vm_flags, vm_userfaultfd_ctx, anon_name) && is_mergeable_anon_vma(anon_vma, vma->anon_vma, vma)) { pgoff_t vm_pglen; vm_pglen =vma_pages (vma);@@ -1113,9 +1118,9 @@ can_vma_merge_after(struct vm_area_struct *vma, unsigned long vm_flags, }/* -* Given a mapping request (addr,end,vm_flags,file,pgoff), figure out -* whether that can be merged with its predecessor or its successor. -* Or both (it neatly fills a hole). +* Given a mapping request (addr,end,vm_flags,file,pgoff,anon_name), +* figure out whether that can be merged with its predecessor or its +* successor. Or both (it neatly fills a hole). * * In most cases - when called for mmap, brk or mremap - [addr,end) is * certain not to be mapped by the time vma_merge is called; but when @@ -1160,7 +1165,8 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm, unsigned long end, unsigned long vm_flags, struct anon_vma *anon_vma, struct file *file, pgoff_t pgoff, struct mempolicy *policy, - struct vm_userfaultfd_ctx vm_userfaultfd_ctx) + struct vm_userfaultfd_ctx vm_userfaultfd_ctx, + const char *anon_name) { pgoff_t pglen = (end - addr) >> PAGE_SHIFT;struct vm_area_struct *area, *next;@@ -1190,7 +1196,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,mpol_equal (vma_policy (prev), policy) &&can_vma_merge_after (prev, vm_flags, anon_vma, file, pgoff, - vm_userfaultfd_ctx)) { + vm_userfaultfd_ctx, anon_name)) {/* * OK, it can. Can we now merge in the successor as well? */ @@ -1199,7 +1205,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,can_vma_merge_before (next, vm_flags, anon_vma, file, pgoff+pglen, - vm_userfaultfd_ctx) && + vm_userfaultfd_ctx, anon_name) &&is_mergeable_anon_vma (prev->anon_vma, next->anon_vma, NULL)) {/* cases 1, 6 */ @@ -1222,7 +1228,7 @@ struct vm_area_struct *vma_merge(struct mm_struct *mm,mpol_equal (policy,vma_policy (next)) &&can_vma_merge_before (next, vm_flags, anon_vma, file, pgoff+pglen, - vm_userfaultfd_ctx)) { + vm_userfaultfd_ctx, anon_name)) { if (prev && addr < prev->vm_end) /* case 4 */ err = __vma_adjust(prev, prev->vm_start, addr, prev->vm_pgoff,NULL , next);@@ -1755,7 +1761,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,* Can we just expand an old mapping? */ vma =vma_merge (mm, prev, addr, addr + len, vm_flags, -NULL , file, pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL , file, pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (vma) goto out;@@ -1804,7 +1810,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,*/ if (unlikely (vm_flags != vma->vm_flags && prev)) { merge =vma_merge (mm, prev, vma->vm_start, vma->vm_end, vma->vm_flags, -NULL , vma->vm_file, vma->vm_pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL , vma->vm_file, vma->vm_pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (merge) {/* ->mmap() can change vma->vm_file and fput the original file. So * fput the vma->vm_file here or we would add an extra fput for file @@ -3057,7 +3063,7 @@ static int do_brk_flags(unsigned long addr, unsigned long len, unsigned long fla/* Can we just expand an old private anonymous mapping? */ vma =vma_merge (mm, prev, addr, addr + len, flags, -NULL ,NULL , pgoff,NULL ,NULL_VM_UFFD_CTX ); +NULL ,NULL , pgoff,NULL ,NULL_VM_UFFD_CTX ,NULL );if (vma) goto out;@@ -3250,7 +3256,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,return NULL ;/* should never get here */ new_vma =vma_merge (mm, prev, addr, addr + len, vma->vm_flags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (new_vma) {/* * Source vma may have been merged into new_vma mm/mprotect.c +1
-2@@ -464,7 +464,7 @@ mprotect_fixup(struct vm_area_struct *vma, struct vm_area_struct **pprev, pgoff = vma->vm_pgoff + ((start - vma->vm_start) >>PAGE_SHIFT ); *pprev =vma_merge (mm, *pprev, start, end, newflags, vma->anon_vma, vma->vm_file, pgoff,vma_policy (vma), - vma->vm_userfaultfd_ctx); + vma->vm_userfaultfd_ctx,vma_anon_name (vma));if (*pprev) { vma = *pprev;VM_WARN_ON ((vma->vm_flags ^ newflags) & ~VM_SOFTDIRTY ); --2.33.0.1079.g6e70778dc9-goog
Andrew Morton
maintainerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 11 lines · On Tue, 19 Oct 2021 14:58:36 -0700 Suren Baghdasaryan <[email protected]> wrote:
As Andrew suggested, I checked the image sizes with allnoconfig builds:
unpatched Linus' ToT
text data bss dec hex filename
1324759 32 73928 1398719 1557bf vmlinuxAfter the first patch is applied (madvise refactoring)
text data bss dec hex filename
1322346 32 73928 1396306 154e52 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
2413 bytes decrease vs ToT <<<
Quoted 9 lines
After all patches applied with
CONFIG_ANON_VMA_NAME=n
text data bss dec hex filename
1322337 32 73928 1396297 154e49 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
2422 bytes decrease vs ToT <<<
After all patches applied with
CONFIG_ANON_VMA_NAME=y
text data bss dec hex filename
1325228 32 73928 1399188 155994 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
469 bytes increase vs ToT <<<
Nice. Presumably there are memory savings from no longer duplicating the vma names?
I fudged up a [0/n] changelog (please don't forget this) and merged it all for testing.
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 19 lines · Andrew Morton
Quoted 16 lines · On Tue, 19 Oct 2021 14:58:36 -0700 Suren Baghdasaryan <[email protected]> wrote:
As Andrew suggested, I checked the image sizes with allnoconfig builds:
unpatched Linus' ToT
text data bss dec hex filename
1324759 32 73928 1398719 1557bf vmlinuxAfter the first patch is applied (madvise refactoring)
text data bss dec hex filename
1322346 32 73928 1396306 154e52 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
2413 bytes decrease vs ToT <<<
After all patches applied with
CONFIG_ANON_VMA_NAME=n
text data bss dec hex filename
1322337 32 73928 1396297 154e49 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
2422 bytes decrease vs ToT <<<
Quoted 7 lines
Quoted 4 lines
After all patches applied with
CONFIG_ANON_VMA_NAME=y
text data bss dec hex filename
1325228 32 73928 1399188 155994 vmlinuxQuoted 1 line
Quoted 1 line
Quoted 1 line
469 bytes increase vs ToT <<<
Nice. Presumably there are memory savings from no longer duplicating the vma names?
The third patch does have this effect.
Quoted 3 lines
I fudged up a [0/n] changelog (please don't forget this) and merged it all for testing.
Thanks!
Re: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
In reply to Suren BaghdasaryanQuoted 113 lines · Suren Baghdasaryan
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in
Documentation/vm/pagemap.txtto walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to
/proc/pid/mapsand/proc/pid/smapsto show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in/proc/pid/mapsand/proc/pid/smapsas [anon:<name>].Userspace can set the name for a region of memory by calling
prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading/proc/pid/mapsor/proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.The name is stored in a pointer in the shared union in
vm_area_structthat points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAMEkernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from
vm_area_struct.
One big concern is aboutfork()performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string duringfork()or when splitting a vma it increments the refcount.[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the
size specified in arg4. arg5 specifies the value of the attribute
to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including
space), except '[',']','\','$' and '`'.
Quoted 2 lines
This feature is available only if the kernel is built with the
CONFIG_ANON_VMA_NAMEoption enabled.
For what it’s worth, it’s definitely interesting to see this going upstream. In particular, we would use it for high-level grouping of the data in production profiling when proper symbolization is not available:
* JVM could associate a name with the memory regions it uses for the JIT code so that Linux perf data are associated with a high level name like "Java JIT" even if the proper Java JIT profiling is not enabled.
* Similar for other JIT engines like v8 - they could annotate the memory regions they manage and use as well.
* Traditional memory allocators like tcmalloc can use this as well so that the associated name is used in data access profiling via Linux perf.
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 121 lines · Alexey Alexandrov
Quoted 114 lines · Suren Baghdasaryan
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in
Documentation/vm/pagemap.txtto walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to
/proc/pid/mapsand/proc/pid/smapsto show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in/proc/pid/mapsand/proc/pid/smapsas [anon:<name>].Userspace can set the name for a region of memory by calling
prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading/proc/pid/mapsor/proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.The name is stored in a pointer in the shared union in
vm_area_structthat points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAMEkernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from
vm_area_struct.
One big concern is aboutfork()performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string duringfork()or when splitting a vma it increments the refcount.[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the size specified in arg4. arg5 specifies the value of the attribute to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including
space), except '[',']','\','$' and '`'.Quoted 2 lines
This feature is available only if the kernel is built with the
CONFIG_ANON_VMA_NAMEoption enabled.For what it’s worth, it’s definitely interesting to see this going upstream. In particular, we would use it for high-level grouping of the data in production profiling when proper symbolization is not available:
Quoted 7 lines
* JVM could associate a name with the memory regions it uses for the JIT code so that Linux perf data are associated with a high level name like "Java JIT" even if the proper Java JIT profiling is not enabled.
* Similar for other JIT engines like v8 - they could annotate the memory regions they manage and use as well.
* Traditional memory allocators like tcmalloc can use this as well so that the associated name is used in data access profiling via Linux perf.
Hi Alexey,
Thanks for providing your feedback! Nice to hear that this can be useful outside of Android.
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 131 lines · Suren Baghdasaryan
Quoted 121 lines · Alexey Alexandrov
Quoted 114 lines · Suren Baghdasaryan
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in
Documentation/vm/pagemap.txtto walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to
/proc/pid/mapsand/proc/pid/smapsto show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in/proc/pid/mapsand/proc/pid/smapsas [anon:<name>].Userspace can set the name for a region of memory by calling
prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading/proc/pid/mapsor/proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.The name is stored in a pointer in the shared union in
vm_area_structthat points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAMEkernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from
vm_area_struct.
One big concern is aboutfork()performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string duringfork()or when splitting a vma it increments the refcount.[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the size specified in arg4. arg5 specifies the value of the attribute to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including
space), except '[',']','\','$' and '`'.Quoted 2 lines
This feature is available only if the kernel is built with the
CONFIG_ANON_VMA_NAMEoption enabled.For what it’s worth, it’s definitely interesting to see this going upstream. In particular, we would use it for high-level grouping of the data in production profiling when proper symbolization is not available:
Quoted 7 lines
* JVM could associate a name with the memory regions it uses for the JIT code so that Linux perf data are associated with a high level name like "Java JIT" even if the proper Java JIT profiling is not enabled.
* Similar for other JIT engines like v8 - they could annotate the memory regions they manage and use as well.
* Traditional memory allocators like tcmalloc can use this as well so that the associated name is used in data access profiling via Linux perf.
Quoted 3 lines
Hi Alexey,
Thanks for providing your feedback! Nice to hear that this can be useful outside of Android.
Folks, it has been almost two weeks since I posted this v11 patchset. Is there anything else I can do to advance it towards merging?
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 133 lines · Suren Baghdasaryan
Quoted 131 lines · Suren Baghdasaryan
Quoted 128 lines · Alexey Alexandrov
Quoted 114 lines · Suren Baghdasaryan
From: Colin Cross <[email protected]>
In many userspace applications, and especially in VM based applications like Android uses heavily, there are multiple different allocators in use. At a minimum there is libc malloc and the stack, and in many cases there are libc malloc, the stack, direct syscalls to mmap anonymous memory, and multiple VM heaps (one for small objects, one for big objects, etc.). Each of these layers usually has its own tools to inspect its usage; malloc by compiling a debug version, the VM through heap inspection tools, and for direct syscalls there is usually no way to track them.
On Android we heavily use a set of tools that use an extended version of the logic covered in
Documentation/vm/pagemap.txtto walk all pages mapped in userspace and slice their usage by process, shared (COW) vs. unique mappings, backing, etc. This can account for real physical memory usage even in cases like fork without exec (which Android uses heavily to share as many private COW pages as possible between processes), Kernel SamePage Merging, and clean zero pages. It produces a measurement of the pages that only exist in that process (USS, for unique), and a measurement of the physical memory usage of that process with the cost of shared pages being evenly split between processes that share them (PSS).If all anonymous memory is indistinguishable then figuring out the real physical memory usage (PSS) of each heap requires either a pagemap walking tool that can understand the heap debugging of every layer, or for every layer's heap debugging tools to implement the pagemap walking logic, in which case it is hard to get a consistent view of memory across the whole system.
Tracking the information in userspace leads to all sorts of problems. It either needs to be stored inside the process, which means every process has to have an API to export its current heap information upon request, or it has to be stored externally in a filesystem that somebody needs to clean up on crashes. It needs to be readable while the process is still running, so it has to have some sort of synchronization with every layer of userspace. Efficiently tracking the ranges requires reimplementing something like the kernel vma trees, and linking to it from every layer of userspace. It requires more memory, more syscalls, more runtime cost, and more complexity to separately track regions that the kernel is already tracking.
This patch adds a field to
/proc/pid/mapsand/proc/pid/smapsto show a userspace-provided name for anonymous vmas. The names of named anonymous vmas are shown in/proc/pid/mapsand/proc/pid/smapsas [anon:<name>].Userspace can set the name for a region of memory by calling
prctl(PR_SET_VMA, PR_SET_VMA_ANON_NAME, start, len, (unsigned long)name); Setting the name to NULL clears it. The name length limit is 80 bytes including NUL-terminator and is checked to contain only printable ascii characters (including space), except '[',']','\','$' and '`'. Ascii strings are being used to have a descriptive identifiers for vmas, which can be understood by the users reading/proc/pid/mapsor/proc/pid/smaps. Names can be standardized for a given system and they can include some variable parts such as the name of the allocator or a library, tid of the thread using it, etc.The name is stored in a pointer in the shared union in
vm_area_structthat points to a null terminated string. Anonymous vmas with the same name (equivalent strings) and are otherwise mergeable will be merged. The name pointers are not shared between vmas even if they contain the same name. The name pointer is stored in a union with fields that are only used on file-backed mappings, so it does not increase memory usage.
CONFIG_ANON_VMA_NAMEkernel configuration is introduced to enable this feature. It keeps the feature disabled by default to prevent any additional memory overhead and to avoid confusing procfs parsers on systems which are not ready to support named anonymous vmas.The patch is based on the original patch developed by Colin Cross, more specifically on its latest version [1] posted upstream by Sumit Semwal. It used a userspace pointer to store vma names. In that design, name pointers could be shared between vmas. However during the last upstreaming attempt, Kees Cook raised concerns [2] about this approach and suggested to copy the name into kernel memory space, perform validity checks [3] and store as a string referenced from
vm_area_struct.
One big concern is aboutfork()performance which would need to strdup anonymous vma names. Dave Hansen suggested experimenting with worst-case scenario of forking a process with 64k vmas having longest possible names [4]. I ran this experiment on an ARM64 Android device and recorded a worst-case regression of almost 40% when forking such a process. This regression is addressed in the followup patch which replaces the pointer to a name with a refcounted structure that allows sharing the name pointer between vmas of the same name. Instead of duplicating the string duringfork()or when splitting a vma it increments the refcount.[1] https://lore.kernel.org/linux-mm/[email protected]/
[2] https://lore.kernel.org/linux-mm/202009031031.D32EF57ED@keescook/
[3] https://lore.kernel.org/linux-mm/202009031022.3834F692@keescook/
[4] https://lore.kernel.org/linux-mm/[email protected]/Changes for prctl(2) manual page (in the options section):
PR_SET_VMA
Sets an attribute specified in arg2 for virtual memory areas starting from the address specified in arg3 and spanning the size specified in arg4. arg5 specifies the value of the attribute to be set. Note that assigning an attribute to a virtual memory area might prevent it from being merged with adjacent virtual memory areas due to the difference in that attribute's value.Currently, arg2 must be one of:
PR_SET_VMA_ANON_NAME
Set a name for anonymous virtual memory areas. arg5 should be a pointer to a null-terminated string containing the name. The name length including null byte cannot exceed 80 bytes. If arg5 is NULL, the name of the appropriate anonymous virtual memory areas will be reset. The name can contain only printable ascii characters (including
space), except '[',']','\','$' and '`'.Quoted 2 lines
This feature is available only if the kernel is built with the
CONFIG_ANON_VMA_NAMEoption enabled.For what it’s worth, it’s definitely interesting to see this going upstream. In particular, we would use it for high-level grouping of the data in production profiling when proper symbolization is not available:
* JVM could associate a name with the memory regions it uses for the JIT code so that Linux perf data are associated with a high level name like "Java JIT" even if the proper Java JIT profiling is not enabled.
* Similar for other JIT engines like v8 - they could annotate the memory regions they manage and use as well.
* Traditional memory allocators like tcmalloc can use this as well so that the associated name is used in data access profiling via Linux perf.
Quoted 6 lines
Quoted 3 lines
Hi Alexey,
Thanks for providing your feedback! Nice to hear that this can be useful outside of Android.Folks, it has been almost two weeks since I posted this v11 patchset. Is there anything else I can do to advance it towards merging?
Hi Andrew,
I haven't seen any feedback on my patchset for some time now. I think I addressed all the questions and comments (please correct me if I missed anything). Can it be accepted as is or is there something I should address further?
From the feedback, I see that there are several interested parties in this patchset (albeit all from different teams at Google) but maybe when it's merged more users will start using it. I believe I've done everything I could to ensure no/minimal impact on the users who don't use this feature. Please advise.
Thanks,
Suren.
Michal Hocko
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
On Mon 15-11-21 10:59:20, Suren Baghdasaryan wrote:
[...]
Quoted 4 lines
Hi Andrew,
I haven't seen any feedback on my patchset for some time now. I think I addressed all the questions and comments (please correct me if I missed anything).
I believe the strings vs. ids have been mostly hand waved away. The biggest argument for the former was convenience for developers to have something human readable. There was no actual proposal about the naming convention so we are relying on some unwritten rules or knowledge of the code to be debugged to make human readable string human understandable ones. I believe this has never been properly resolved except for - this has been used in Android and working just fine. I am not convinced TBH.
So in the end we are adding a user interface that brings a runtime and resource overhead that will be hard to change in the future. Reference counting handles a part of that and that is nice but ids simply do not have any of that.
Quoted 2 lines
Can it be accepted as is or is there something I should address further?
Is the above reason to nack it? No, I do not think so. I just do not feel like I want to ack it either. Concerns have been expressed and I have to say that I would like a minimalistic approach much more. Also extending ids into string is always possible. The other way around is not possible.
Signature · 2 lines
Michal Hocko SUSE Labs
Suren Baghdasaryan
reviewerRe: [PATCH v11 2/3] mm: add a field to store names for private anonymous memory
Quoted 16 lines · Michal Hocko
On Mon 15-11-21 10:59:20, Suren Baghdasaryan wrote:
[...]Quoted 4 lines
Hi Andrew,
I haven't seen any feedback on my patchset for some time now. I think I addressed all the questions and comments (please correct me if I missed anything).I believe the strings vs. ids have been mostly hand waved away. The biggest argument for the former was convenience for developers to have something human readable. There was no actual proposal about the naming convention so we are relying on some unwritten rules or knowledge of the code to be debugged to make human readable string human understandable ones. I believe this has never been properly resolved except for - this has been used in Android and working just fine. I am not convinced TBH.
Quoted 4 lines
So in the end we are adding a user interface that brings a runtime and resource overhead that will be hard to change in the future. Reference counting handles a part of that and that is nice but ids simply do not have any of that.
I explained the way this interface is used and why ids would not work
for us in https://lore.kernel.org/all/CAJuCfpESeM_Xd8dhCj_okNggtDUXx3Nn9FpL_f9qsKXKZzCKpA@mail.gmail.com.
I explored all the proposed alternatives, all of which would be
prohibitive for our needs due to performance costs compared to this
solution. I wish I could come up with something simpler but a simpler
solution simply does not meet our needs.
It's true that this approach does not formalize how VMAs are named but I don't see why kernel should impose a naming convention. I can see some systems defining more formal conventions but I believe it should be up to the userspace to do that.
Quoted 9 lines
Quoted 3 lines
Can it be accepted as is or is there something I should address further?
Is the above reason to nack it? No, I do not think so. I just do not feel like I want to ack it either. Concerns have been expressed and I have to say that I would like a minimalistic approach much more. Also extending ids into string is always possible. The other way around is not possible.
Unfortunately, extending ids into strings comes with the cost we can't afford in Android. Therefore I don't see a point for me to upstream such a solution which I can't use.
Thanks,
Suren.
Quoted 4 lines
--
Michal Hocko
SUSE Labs
Suren Baghdasaryan
reviewerPATCH v11 3/3 mm: add anonymous vma name refcounting
In reply to Suren BaghdasaryanWhile forking a process with high number (64K) of named anonymous vmas the overhead caused by strdup() is noticeable. Experiments with ARM64 Android device show up to 40% performance regression when forking a process with 64k unpopulated anonymous vmas using the max name lengths vs the same process with the same number of anonymous vmas having no name. Introduce anon_vma_name refcounted structure to avoid the overhead of copying vma names during fork() and when splitting named anonymous vmas. When a vma is duplicated, instead of copying the name we increment the refcount of this structure. Multiple vmas can point to the same anon_vma_name as long as they increment the refcount. The name member of anon_vma_name structure is assigned at structure allocation time and is never changed. If vma name changes then the refcount of the original structure is dropped, a new anon_vma_name structure is allocated to hold the new name and the vma pointer is updated to point to the new structure.
With this approach the fork() performance regressions is reduced 3-4x times and with usecases using more reasonable number of VMAs (a few thousand) the regressions is not measurable.
---
previous version at:
https://lore.kernel.org/all/[email protected]/
include/linux/mm_types.h | 9 ++++++++- mm/madvise.c | 42 ++++++++++++++++++++++++++++++++++------ 2 files changed, 44 insertions(+), 7 deletions(-)
include/linux/mm_types.h +8 -1
@@ -5,6 +5,7 @@#include <linux/mm_types_task.h> #include <linux/auxvec.h> +#include <linux/kref.h> #include <linux/list.h> #include <linux/spinlock.h> #include <linux/rbtree.h>
@@ -310,6 +311,12 @@ struct vm_userfaultfd_ctx {struct vm_userfaultfd_ctx {};#endif /* CONFIG_USERFAULTFD */ +struct anon_vma_name { +struct kref kref; +/* The name needs to be at the end because it is dynamically sized. */ +char name[]; +}; +/* * This struct describes a virtual memory area. There is one of these * per VM-area/task. A VM area is any part of the process virtual memory
@@ -361,7 +368,7 @@ struct vm_area_struct {unsigned long rb_subtree_last; } shared;/* Serialized by mmap_sem. */ -char *anon_name; +struct anon_vma_name *anon_name; };/*
mm/madvise.c +36 -6
83 lines not on this page · raw message
Signature · 1 line
2.33.0.1079.g6e70778dc9-goog