| Message ID | 20230503013608.2431726-3-nphamcs@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | cachestat: a new syscall for page cache state of files | expand |
On Tue, May 02, 2023 at 06:36:07PM -0700, Nhat Pham wrote: > There is currently no good way to query the page cache state of large > file sets and directory trees. There is mincore(), but it scales poorly: > the kernel writes out a lot of bitmap data that userspace has to > aggregate, when the user really doesn not care about per-page > information in that case. The user also needs to mmap and unmap each > file as it goes along, which can be quite slow as well. > > Some use cases where this information could come in handy: > * Allowing database to decide whether to perform an index scan or > direct table queries based on the in-memory cache state of the > index. > * Visibility into the writeback algorithm, for performance issues > diagnostic. > * Workload-aware writeback pacing: estimating IO fulfilled by page > cache (and IO to be done) within a range of a file, allowing for > more frequent syncing when and where there is IO capacity, and > batching when there is not. > * Computing memory usage of large files/directory trees, analogous to > the du tool for disk usage. > > More information about these use cases could be found in the following > thread: > > https://lore.kernel.org/lkml/20230315170934.GA97793@cmpxchg.org/ > > This patch implements a new syscall that queries cache state of a file > and summarizes the number of cached pages, number of dirty pages, number > of pages marked for writeback, number of (recently) evicted pages, etc. > in a given range. Currently, the syscall is only wired in for x86 > architecture. > > NAME > cachestat - query the page cache statistics of a file. > > SYNOPSIS > #include <sys/mman.h> > > struct cachestat_range { > __u64 off; > __u64 len; > }; > > struct cachestat { > __u64 nr_cache; > __u64 nr_dirty; > __u64 nr_writeback; > __u64 nr_evicted; > __u64 nr_recently_evicted; > }; > > int cachestat(unsigned int fd, struct cachestat_range *cstat_range, > struct cachestat *cstat, unsigned int flags); > > DESCRIPTION > cachestat() queries the number of cached pages, number of dirty > pages, number of pages marked for writeback, number of evicted > pages, number of recently evicted pages, in the bytes range given by > `off` and `len`. > > An evicted page is a page that is previously in the page cache but > has been evicted since. A page is recently evicted if its last > eviction was recent enough that its reentry to the cache would > indicate that it is actively being used by the system, and that > there is memory pressure on the system. > > These values are returned in a cachestat struct, whose address is > given by the `cstat` argument. > > The `off` and `len` arguments must be non-negative integers. If > `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` == > 0, we will query in the range from `off` to the end of the file. > > The `flags` argument is unused for now, but is included for future > extensibility. User should pass 0 (i.e no flag specified). > > Currently, hugetlbfs is not supported. > > Because the status of a page can change after cachestat() checks it > but before it returns to the application, the returned values may > contain stale information. > > RETURN VALUE > On success, cachestat returns 0. On error, -1 is returned, and errno > is set to indicate the error. > > ERRORS > EFAULT cstat or cstat_args points to an invalid address. > > EINVAL invalid flags. > > EBADF invalid file descriptor. > > EOPNOTSUPP file descriptor is of a hugetlbfs file > > Signed-off-by: Nhat Pham <nphamcs@gmail.com> Thanks for persisting through the pain. This looks great to me now. Like I've said before, I think this is sorely needed. The cache is frequently the biggest memory consumer in the system. We have a rich API for influencing it, but there is a glaring gap when it comes to introspection. It's difficult to design control loops without feedback. This proposes an intuitive, versatile and scalable interface to bridge that gap, and it integrates nicely with the existing VFS API for managing the cache. I would love to see this go in. I'd also love for the `mu' tool you wrote to make it into coreutils eventually. It would make debugging memory consumption and writeback issues on live systems, especially with complex and/or multiple workloads, so much easier. Acked-by: Johannes Weiner <hannes@cmpxchg.org>
On Wed, May 3, 2023 at 8:04 AM Johannes Weiner <hannes@cmpxchg.org> wrote: > > On Tue, May 02, 2023 at 06:36:07PM -0700, Nhat Pham wrote: > > There is currently no good way to query the page cache state of large > > file sets and directory trees. There is mincore(), but it scales poorly: > > the kernel writes out a lot of bitmap data that userspace has to > > aggregate, when the user really doesn not care about per-page > > information in that case. The user also needs to mmap and unmap each > > file as it goes along, which can be quite slow as well. > > > > Some use cases where this information could come in handy: > > * Allowing database to decide whether to perform an index scan or > > direct table queries based on the in-memory cache state of the > > index. > > * Visibility into the writeback algorithm, for performance issues > > diagnostic. > > * Workload-aware writeback pacing: estimating IO fulfilled by page > > cache (and IO to be done) within a range of a file, allowing for > > more frequent syncing when and where there is IO capacity, and > > batching when there is not. > > * Computing memory usage of large files/directory trees, analogous to > > the du tool for disk usage. > > > > More information about these use cases could be found in the following > > thread: > > > > https://lore.kernel.org/lkml/20230315170934.GA97793@cmpxchg.org/ > > > > This patch implements a new syscall that queries cache state of a file > > and summarizes the number of cached pages, number of dirty pages, number > > of pages marked for writeback, number of (recently) evicted pages, etc. > > in a given range. Currently, the syscall is only wired in for x86 > > architecture. > > > > NAME > > cachestat - query the page cache statistics of a file. > > > > SYNOPSIS > > #include <sys/mman.h> > > > > struct cachestat_range { > > __u64 off; > > __u64 len; > > }; > > > > struct cachestat { > > __u64 nr_cache; > > __u64 nr_dirty; > > __u64 nr_writeback; > > __u64 nr_evicted; > > __u64 nr_recently_evicted; > > }; > > > > int cachestat(unsigned int fd, struct cachestat_range *cstat_range, > > struct cachestat *cstat, unsigned int flags); > > > > DESCRIPTION > > cachestat() queries the number of cached pages, number of dirty > > pages, number of pages marked for writeback, number of evicted > > pages, number of recently evicted pages, in the bytes range given by > > `off` and `len`. > > > > An evicted page is a page that is previously in the page cache but > > has been evicted since. A page is recently evicted if its last > > eviction was recent enough that its reentry to the cache would > > indicate that it is actively being used by the system, and that > > there is memory pressure on the system. > > > > These values are returned in a cachestat struct, whose address is > > given by the `cstat` argument. > > > > The `off` and `len` arguments must be non-negative integers. If > > `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` == > > 0, we will query in the range from `off` to the end of the file. > > > > The `flags` argument is unused for now, but is included for future > > extensibility. User should pass 0 (i.e no flag specified). > > > > Currently, hugetlbfs is not supported. > > > > Because the status of a page can change after cachestat() checks it > > but before it returns to the application, the returned values may > > contain stale information. > > > > RETURN VALUE > > On success, cachestat returns 0. On error, -1 is returned, and errno > > is set to indicate the error. > > > > ERRORS > > EFAULT cstat or cstat_args points to an invalid address. > > > > EINVAL invalid flags. > > > > EBADF invalid file descriptor. > > > > EOPNOTSUPP file descriptor is of a hugetlbfs file > > > > Signed-off-by: Nhat Pham <nphamcs@gmail.com> > > Thanks for persisting through the pain. This looks great to me now. > > Like I've said before, I think this is sorely needed. The cache is > frequently the biggest memory consumer in the system. We have a rich > API for influencing it, but there is a glaring gap when it comes to > introspection. It's difficult to design control loops without > feedback. This proposes an intuitive, versatile and scalable interface > to bridge that gap, and it integrates nicely with the existing VFS API > for managing the cache. I would love to see this go in. > > I'd also love for the `mu' tool you wrote to make it into coreutils > eventually. It would make debugging memory consumption and writeback > issues on live systems, especially with complex and/or multiple > workloads, so much easier. I'd love to share this too! Let me clean it up and submit it separately. > > Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Hi Nhat, On Wed, May 3, 2023 at 3:38 AM Nhat Pham <nphamcs@gmail.com> wrote: > There is currently no good way to query the page cache state of large > file sets and directory trees. There is mincore(), but it scales poorly: > the kernel writes out a lot of bitmap data that userspace has to > aggregate, when the user really doesn not care about per-page > information in that case. The user also needs to mmap and unmap each > file as it goes along, which can be quite slow as well. > > Some use cases where this information could come in handy: > * Allowing database to decide whether to perform an index scan or > direct table queries based on the in-memory cache state of the > index. > * Visibility into the writeback algorithm, for performance issues > diagnostic. > * Workload-aware writeback pacing: estimating IO fulfilled by page > cache (and IO to be done) within a range of a file, allowing for > more frequent syncing when and where there is IO capacity, and > batching when there is not. > * Computing memory usage of large files/directory trees, analogous to > the du tool for disk usage. > > More information about these use cases could be found in the following > thread: > > https://lore.kernel.org/lkml/20230315170934.GA97793@cmpxchg.org/ > > This patch implements a new syscall that queries cache state of a file > and summarizes the number of cached pages, number of dirty pages, number > of pages marked for writeback, number of (recently) evicted pages, etc. > in a given range. Currently, the syscall is only wired in for x86 > architecture. > > NAME > cachestat - query the page cache statistics of a file. > > SYNOPSIS > #include <sys/mman.h> > > struct cachestat_range { > __u64 off; > __u64 len; > }; > > struct cachestat { > __u64 nr_cache; > __u64 nr_dirty; > __u64 nr_writeback; > __u64 nr_evicted; > __u64 nr_recently_evicted; > }; > > int cachestat(unsigned int fd, struct cachestat_range *cstat_range, > struct cachestat *cstat, unsigned int flags); > > DESCRIPTION > cachestat() queries the number of cached pages, number of dirty > pages, number of pages marked for writeback, number of evicted > pages, number of recently evicted pages, in the bytes range given by > `off` and `len`. > > An evicted page is a page that is previously in the page cache but > has been evicted since. A page is recently evicted if its last > eviction was recent enough that its reentry to the cache would > indicate that it is actively being used by the system, and that > there is memory pressure on the system. > > These values are returned in a cachestat struct, whose address is > given by the `cstat` argument. > > The `off` and `len` arguments must be non-negative integers. If > `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` == > 0, we will query in the range from `off` to the end of the file. > > The `flags` argument is unused for now, but is included for future > extensibility. User should pass 0 (i.e no flag specified). > > Currently, hugetlbfs is not supported. > > Because the status of a page can change after cachestat() checks it > but before it returns to the application, the returned values may > contain stale information. > > RETURN VALUE > On success, cachestat returns 0. On error, -1 is returned, and errno > is set to indicate the error. > > ERRORS > EFAULT cstat or cstat_args points to an invalid address. > > EINVAL invalid flags. > > EBADF invalid file descriptor. > > EOPNOTSUPP file descriptor is of a hugetlbfs file > > Signed-off-by: Nhat Pham <nphamcs@gmail.com> > --- > arch/x86/entry/syscalls/syscall_32.tbl | 1 + > arch/x86/entry/syscalls/syscall_64.tbl | 1 + This should be wired up on each and every architecture. Currently we're getting <stdin>:1567:2: warning: #warning syscall cachestat not implemented [-Wcpp] in linux-next for all the missing architectures. Gr{oetje,eeting}s, Geert
On Thu, May 4, 2023 at 10:26 AM Geert Uytterhoeven <geert@linux-m68k.org> wrote: > > Hi Nhat, > > On Wed, May 3, 2023 at 3:38 AM Nhat Pham <nphamcs@gmail.com> wrote: > > There is currently no good way to query the page cache state of large > > file sets and directory trees. There is mincore(), but it scales poorly: > > the kernel writes out a lot of bitmap data that userspace has to > > aggregate, when the user really doesn not care about per-page > > information in that case. The user also needs to mmap and unmap each > > file as it goes along, which can be quite slow as well. > > > > Some use cases where this information could come in handy: > > * Allowing database to decide whether to perform an index scan or > > direct table queries based on the in-memory cache state of the > > index. > > * Visibility into the writeback algorithm, for performance issues > > diagnostic. > > * Workload-aware writeback pacing: estimating IO fulfilled by page > > cache (and IO to be done) within a range of a file, allowing for > > more frequent syncing when and where there is IO capacity, and > > batching when there is not. > > * Computing memory usage of large files/directory trees, analogous to > > the du tool for disk usage. > > > > More information about these use cases could be found in the following > > thread: > > > > https://lore.kernel.org/lkml/20230315170934.GA97793@cmpxchg.org/ > > > > This patch implements a new syscall that queries cache state of a file > > and summarizes the number of cached pages, number of dirty pages, number > > of pages marked for writeback, number of (recently) evicted pages, etc. > > in a given range. Currently, the syscall is only wired in for x86 > > architecture. > > > > NAME > > cachestat - query the page cache statistics of a file. > > > > SYNOPSIS > > #include <sys/mman.h> > > > > struct cachestat_range { > > __u64 off; > > __u64 len; > > }; > > > > struct cachestat { > > __u64 nr_cache; > > __u64 nr_dirty; > > __u64 nr_writeback; > > __u64 nr_evicted; > > __u64 nr_recently_evicted; > > }; > > > > int cachestat(unsigned int fd, struct cachestat_range *cstat_range, > > struct cachestat *cstat, unsigned int flags); > > > > DESCRIPTION > > cachestat() queries the number of cached pages, number of dirty > > pages, number of pages marked for writeback, number of evicted > > pages, number of recently evicted pages, in the bytes range given by > > `off` and `len`. > > > > An evicted page is a page that is previously in the page cache but > > has been evicted since. A page is recently evicted if its last > > eviction was recent enough that its reentry to the cache would > > indicate that it is actively being used by the system, and that > > there is memory pressure on the system. > > > > These values are returned in a cachestat struct, whose address is > > given by the `cstat` argument. > > > > The `off` and `len` arguments must be non-negative integers. If > > `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` == > > 0, we will query in the range from `off` to the end of the file. > > > > The `flags` argument is unused for now, but is included for future > > extensibility. User should pass 0 (i.e no flag specified). > > > > Currently, hugetlbfs is not supported. > > > > Because the status of a page can change after cachestat() checks it > > but before it returns to the application, the returned values may > > contain stale information. > > > > RETURN VALUE > > On success, cachestat returns 0. On error, -1 is returned, and errno > > is set to indicate the error. > > > > ERRORS > > EFAULT cstat or cstat_args points to an invalid address. > > > > EINVAL invalid flags. > > > > EBADF invalid file descriptor. > > > > EOPNOTSUPP file descriptor is of a hugetlbfs file > > > > Signed-off-by: Nhat Pham <nphamcs@gmail.com> > > --- > > arch/x86/entry/syscalls/syscall_32.tbl | 1 + > > arch/x86/entry/syscalls/syscall_64.tbl | 1 + > > This should be wired up on each and every architecture. > Currently we're getting > > <stdin>:1567:2: warning: #warning syscall cachestat not implemented [-Wcpp] > > in linux-next for all the missing architectures. Hi Geert, I saw that there are several instances where we have separate patches to wire up a syscall to these architectures, so I was doing something similar. For e.g: ARM: wire up process_vm_writev and process_vm_readv syscalls (e5489847d6fc0ff176048b6e1cf5034507bf703a) MIPS: Hook up process_vm_readv and process_vm_writev system calls. (8ff8584e51d4d3fbe08ede413c4a221223766323) As for these non-x86 architecture wiring patches, I can give it a shot and cross-compile to see if it builds, but I have limited abilities for runtime tests as I don't have machines with these architectures. I would really appreciate it if there are arch people that could help wire it up. (cc-ing linux-arch as well) > > Gr{oetje,eeting}s, > > Geert > > -- > Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org > > In personal conversations with technical people, I call myself a hacker. But > when I'm talking to journalists I just say "programmer" or something like that. > -- Linus Torvalds
On Thu, 4 May 2023 19:26:11 +0200 Geert Uytterhoeven <geert@linux-m68k.org> wrote: > > arch/x86/entry/syscalls/syscall_32.tbl | 1 + > > arch/x86/entry/syscalls/syscall_64.tbl | 1 + > > This should be wired up on each and every architecture. > Currently we're getting > > <stdin>:1567:2: warning: #warning syscall cachestat not implemented [-Wcpp] > > in linux-next for all the missing architectures. Is that wise? We risk adding a syscall to an architecture without the arch maintainers and testers even knowing about it. The compile-time nag is there to inform the arch maintainers that a new syscall is available and that they should wire it up, run the selftest and then ship the code if they're happy with the result.
On Fri, May 5, 2023, at 22:34, Andrew Morton wrote: > On Thu, 4 May 2023 19:26:11 +0200 Geert Uytterhoeven > <geert@linux-m68k.org> wrote: > >> > arch/x86/entry/syscalls/syscall_32.tbl | 1 + >> > arch/x86/entry/syscalls/syscall_64.tbl | 1 + >> >> This should be wired up on each and every architecture. >> Currently we're getting >> >> <stdin>:1567:2: warning: #warning syscall cachestat not implemented [-Wcpp] >> >> in linux-next for all the missing architectures. > > Is that wise? We risk adding a syscall to an architecture without the > arch maintainers and testers even knowing about it. > > The compile-time nag is there to inform the arch maintainers that a new > syscall is available and that they should wire it up, run the selftest > and then ship the code if they're happy with the result. The usual approach is for the author of a new syscall to include a patch with all the architecture specific changes and Cc the architecture maintainers for that. Note that half the architectures get the entry from include/uapi/asm-generic/unistd.h, so adding it there does not necessarily trigger adding each maintainer from scripts/get_maintainer.pl. The only real risk in adding a new syscall is passing __u64 register arguments that behave differently across architectures, or using pointers to data structures that require a compat handler on some architectures. I watch out for those as they get sent to me or the linux-arch list, and this one is fine. Arnd
On Sat, May 6, 2023 at 10:35 AM Arnd Bergmann <arnd@arndb.de> wrote: > > On Fri, May 5, 2023, at 22:34, Andrew Morton wrote: > > On Thu, 4 May 2023 19:26:11 +0200 Geert Uytterhoeven > > <geert@linux-m68k.org> wrote: > > > >> > arch/x86/entry/syscalls/syscall_32.tbl | 1 + > >> > arch/x86/entry/syscalls/syscall_64.tbl | 1 + > >> > >> This should be wired up on each and every architecture. > >> Currently we're getting > >> > >> <stdin>:1567:2: warning: #warning syscall cachestat not implemented [-Wcpp] > >> > >> in linux-next for all the missing architectures. > > > > Is that wise? We risk adding a syscall to an architecture without the > > arch maintainers and testers even knowing about it. > > > > The compile-time nag is there to inform the arch maintainers that a new > > syscall is available and that they should wire it up, run the selftest > > and then ship the code if they're happy with the result. > > The usual approach is for the author of a new syscall to > include a patch with all the architecture specific changes > and Cc the architecture maintainers for that. > > Note that half the architectures get the entry from > include/uapi/asm-generic/unistd.h, so adding it there > does not necessarily trigger adding each maintainer > from scripts/get_maintainer.pl. > > The only real risk in adding a new syscall is passing __u64 > register arguments that behave differently across > architectures, or using pointers to data structures that > require a compat handler on some architectures. I watch out > for those as they get sent to me or the linux-arch list, > and this one is fine. > > Arnd I took a stab at wiring the new syscall in this follow-up patch: https://lore.kernel.org/lkml/20230510195806.2902878-1-nphamcs@gmail.com/ Let me know if I missed something! Review and/or suggestion is very much appreciated.
diff --git a/arch/x86/entry/syscalls/syscall_32.tbl b/arch/x86/entry/syscalls/syscall_32.tbl index 320480a8db4f..bc0a3c941b35 100644 --- a/arch/x86/entry/syscalls/syscall_32.tbl +++ b/arch/x86/entry/syscalls/syscall_32.tbl @@ -455,3 +455,4 @@ 448 i386 process_mrelease sys_process_mrelease 449 i386 futex_waitv sys_futex_waitv 450 i386 set_mempolicy_home_node sys_set_mempolicy_home_node +451 i386 cachestat sys_cachestat diff --git a/arch/x86/entry/syscalls/syscall_64.tbl b/arch/x86/entry/syscalls/syscall_64.tbl index c84d12608cd2..227538b0ce80 100644 --- a/arch/x86/entry/syscalls/syscall_64.tbl +++ b/arch/x86/entry/syscalls/syscall_64.tbl @@ -372,6 +372,7 @@ 448 common process_mrelease sys_process_mrelease 449 common futex_waitv sys_futex_waitv 450 common set_mempolicy_home_node sys_set_mempolicy_home_node +451 common cachestat sys_cachestat # # Due to a historical design error, certain syscalls are numbered differently diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h index 33a0ee3bcb2e..6648c07c4381 100644 --- a/include/linux/syscalls.h +++ b/include/linux/syscalls.h @@ -72,6 +72,8 @@ struct open_how; struct mount_attr; struct landlock_ruleset_attr; enum landlock_rule_type; +struct cachestat_range; +struct cachestat; #include <linux/types.h> #include <linux/aio_abi.h> @@ -1058,6 +1060,9 @@ asmlinkage long sys_memfd_secret(unsigned int flags); asmlinkage long sys_set_mempolicy_home_node(unsigned long start, unsigned long len, unsigned long home_node, unsigned long flags); +asmlinkage long sys_cachestat(unsigned int fd, + struct cachestat_range __user *cstat_range, + struct cachestat __user *cstat, unsigned int flags); /* * Architecture-specific system calls diff --git a/include/uapi/asm-generic/unistd.h b/include/uapi/asm-generic/unistd.h index 45fa180cc56a..cd639fae9086 100644 --- a/include/uapi/asm-generic/unistd.h +++ b/include/uapi/asm-generic/unistd.h @@ -886,8 +886,11 @@ __SYSCALL(__NR_futex_waitv, sys_futex_waitv) #define __NR_set_mempolicy_home_node 450 __SYSCALL(__NR_set_mempolicy_home_node, sys_set_mempolicy_home_node) +#define __NR_cachestat 451 +__SYSCALL(__NR_cachestat, sys_cachestat) + #undef __NR_syscalls -#define __NR_syscalls 451 +#define __NR_syscalls 452 /* * 32 bit systems traditionally used different diff --git a/include/uapi/linux/mman.h b/include/uapi/linux/mman.h index f55bc680b5b0..a246e11988d5 100644 --- a/include/uapi/linux/mman.h +++ b/include/uapi/linux/mman.h @@ -4,6 +4,7 @@ #include <asm/mman.h> #include <asm-generic/hugetlb_encode.h> +#include <linux/types.h> #define MREMAP_MAYMOVE 1 #define MREMAP_FIXED 2 @@ -41,4 +42,17 @@ #define MAP_HUGE_2GB HUGETLB_FLAG_ENCODE_2GB #define MAP_HUGE_16GB HUGETLB_FLAG_ENCODE_16GB +struct cachestat_range { + __u64 off; + __u64 len; +}; + +struct cachestat { + __u64 nr_cache; + __u64 nr_dirty; + __u64 nr_writeback; + __u64 nr_evicted; + __u64 nr_recently_evicted; +}; + #endif /* _UAPI_LINUX_MMAN_H */ diff --git a/init/Kconfig b/init/Kconfig index 32c24950c4ce..f7f65af4ee12 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -1771,6 +1771,16 @@ config RSEQ If unsure, say Y. +config CACHESTAT_SYSCALL + bool "Enable cachestat() system call" if EXPERT + default y + help + Enable the cachestat system call, which queries the page cache + statistics of a file (number of cached pages, dirty pages, + pages marked for writeback, (recently) evicted pages). + + If unsure say Y here. + config DEBUG_RSEQ default n bool "Enabled debugging of rseq() system call" if EXPERT diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c index 860b2dcf3ac4..04bfb1e4d377 100644 --- a/kernel/sys_ni.c +++ b/kernel/sys_ni.c @@ -299,6 +299,7 @@ COND_SYSCALL(set_mempolicy); COND_SYSCALL(migrate_pages); COND_SYSCALL(move_pages); COND_SYSCALL(set_mempolicy_home_node); +COND_SYSCALL(cachestat); COND_SYSCALL(perf_event_open); COND_SYSCALL(accept4); diff --git a/mm/filemap.c b/mm/filemap.c index a34abfe8c654..73b043240cef 100644 --- a/mm/filemap.c +++ b/mm/filemap.c @@ -22,6 +22,7 @@ #include <linux/mm.h> #include <linux/swap.h> #include <linux/swapops.h> +#include <linux/syscalls.h> #include <linux/mman.h> #include <linux/pagemap.h> #include <linux/file.h> @@ -58,6 +59,8 @@ #include <asm/mman.h> +#include "swap.h" + /* * Shared mappings implemented 30.11.1994. It's not fully working yet, * though. @@ -4119,3 +4122,172 @@ bool filemap_release_folio(struct folio *folio, gfp_t gfp) return try_to_free_buffers(folio); } EXPORT_SYMBOL(filemap_release_folio); + +#ifdef CONFIG_CACHESTAT_SYSCALL +/** + * filemap_cachestat() - compute the page cache statistics of a mapping + * @mapping: The mapping to compute the statistics for. + * @first_index: The starting page cache index. + * @last_index: The final page index (inclusive). + * @cs: the cachestat struct to write the result to. + * + * This will query the page cache statistics of a mapping in the + * page range of [first_index, last_index] (inclusive). The statistics + * queried include: number of dirty pages, number of pages marked for + * writeback, and the number of (recently) evicted pages. + */ +static void filemap_cachestat(struct address_space *mapping, + pgoff_t first_index, pgoff_t last_index, struct cachestat *cs) +{ + XA_STATE(xas, &mapping->i_pages, first_index); + struct folio *folio; + + rcu_read_lock(); + xas_for_each(&xas, folio, last_index) { + unsigned long nr_pages; + pgoff_t folio_first_index, folio_last_index; + + if (xas_retry(&xas, folio)) + continue; + + if (xa_is_value(folio)) { + /* page is evicted */ + void *shadow = (void *)folio; + bool workingset; /* not used */ + int order = xa_get_order(xas.xa, xas.xa_index); + + nr_pages = 1 << order; + /* rounds down to the nearest multiple of 2^order */ + folio_first_index = xas.xa_index >> order << order; + folio_last_index = folio_first_index + nr_pages - 1; + + /* Folios might straddle the range boundaries, only count covered pages */ + if (folio_first_index < first_index) + nr_pages -= first_index - folio_first_index; + + if (folio_last_index > last_index) + nr_pages -= folio_last_index - last_index; + + cs->nr_evicted += nr_pages; + +#ifdef CONFIG_SWAP /* implies CONFIG_MMU */ + if (shmem_mapping(mapping)) { + /* shmem file - in swap cache */ + swp_entry_t swp = radix_to_swp_entry(folio); + + shadow = get_shadow_from_swap_cache(swp); + } +#endif + if (workingset_test_recent(shadow, true, &workingset)) + cs->nr_recently_evicted += nr_pages; + + goto resched; + } + + nr_pages = folio_nr_pages(folio); + folio_first_index = folio_pgoff(folio); + folio_last_index = folio_first_index + nr_pages - 1; + + /* Folios might straddle the range boundaries, only count covered pages */ + if (folio_first_index < first_index) + nr_pages -= first_index - folio_first_index; + + if (folio_last_index > last_index) + nr_pages -= folio_last_index - last_index; + + /* page is in cache */ + cs->nr_cache += nr_pages; + + if (folio_test_dirty(folio)) + cs->nr_dirty += nr_pages; + + if (folio_test_writeback(folio)) + cs->nr_writeback += nr_pages; + +resched: + if (need_resched()) { + xas_pause(&xas); + cond_resched_rcu(); + } + } + rcu_read_unlock(); +} + +/* + * The cachestat(2) system call. + * + * cachestat() returns the page cache statistics of a file in the + * bytes range specified by `off` and `len`: number of cached pages, + * number of dirty pages, number of pages marked for writeback, + * number of evicted pages, and number of recently evicted pages. + * + * An evicted page is a page that is previously in the page cache + * but has been evicted since. A page is recently evicted if its last + * eviction was recent enough that its reentry to the cache would + * indicate that it is actively being used by the system, and that + * there is memory pressure on the system. + * + * `off` and `len` must be non-negative integers. If `len` > 0, + * the queried range is [`off`, `off` + `len`]. If `len` == 0, + * we will query in the range from `off` to the end of the file. + * + * The `flags` argument is unused for now, but is included for future + * extensibility. User should pass 0 (i.e no flag specified). + * + * Currently, hugetlbfs is not supported. + * + * Because the status of a page can change after cachestat() checks it + * but before it returns to the application, the returned values may + * contain stale information. + * + * return values: + * zero - success + * -EFAULT - cstat or cstat_range points to an illegal address + * -EINVAL - invalid flags + * -EBADF - invalid file descriptor + * -EOPNOTSUPP - file descriptor is of a hugetlbfs file + */ +SYSCALL_DEFINE4(cachestat, unsigned int, fd, + struct cachestat_range __user *, cstat_range, + struct cachestat __user *, cstat, unsigned int, flags) +{ + struct fd f = fdget(fd); + struct address_space *mapping; + struct cachestat_range csr; + struct cachestat cs; + pgoff_t first_index, last_index; + + if (!f.file) + return -EBADF; + + if (copy_from_user(&csr, cstat_range, + sizeof(struct cachestat_range))) { + fdput(f); + return -EFAULT; + } + + /* hugetlbfs is not supported */ + if (is_file_hugepages(f.file)) { + fdput(f); + return -EOPNOTSUPP; + } + + if (flags != 0) { + fdput(f); + return -EINVAL; + } + + first_index = csr.off >> PAGE_SHIFT; + last_index = + csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT; + memset(&cs, 0, sizeof(struct cachestat)); + mapping = f.file->f_mapping; + filemap_cachestat(mapping, first_index, last_index, &cs); + fdput(f); + + if (copy_to_user(cstat, &cs, sizeof(struct cachestat))) + return -EFAULT; + + return 0; +} +#endif /* CONFIG_CACHESTAT_SYSCALL */
There is currently no good way to query the page cache state of large file sets and directory trees. There is mincore(), but it scales poorly: the kernel writes out a lot of bitmap data that userspace has to aggregate, when the user really doesn not care about per-page information in that case. The user also needs to mmap and unmap each file as it goes along, which can be quite slow as well. Some use cases where this information could come in handy: * Allowing database to decide whether to perform an index scan or direct table queries based on the in-memory cache state of the index. * Visibility into the writeback algorithm, for performance issues diagnostic. * Workload-aware writeback pacing: estimating IO fulfilled by page cache (and IO to be done) within a range of a file, allowing for more frequent syncing when and where there is IO capacity, and batching when there is not. * Computing memory usage of large files/directory trees, analogous to the du tool for disk usage. More information about these use cases could be found in the following thread: https://lore.kernel.org/lkml/20230315170934.GA97793@cmpxchg.org/ This patch implements a new syscall that queries cache state of a file and summarizes the number of cached pages, number of dirty pages, number of pages marked for writeback, number of (recently) evicted pages, etc. in a given range. Currently, the syscall is only wired in for x86 architecture. NAME cachestat - query the page cache statistics of a file. SYNOPSIS #include <sys/mman.h> struct cachestat_range { __u64 off; __u64 len; }; struct cachestat { __u64 nr_cache; __u64 nr_dirty; __u64 nr_writeback; __u64 nr_evicted; __u64 nr_recently_evicted; }; int cachestat(unsigned int fd, struct cachestat_range *cstat_range, struct cachestat *cstat, unsigned int flags); DESCRIPTION cachestat() queries the number of cached pages, number of dirty pages, number of pages marked for writeback, number of evicted pages, number of recently evicted pages, in the bytes range given by `off` and `len`. An evicted page is a page that is previously in the page cache but has been evicted since. A page is recently evicted if its last eviction was recent enough that its reentry to the cache would indicate that it is actively being used by the system, and that there is memory pressure on the system. These values are returned in a cachestat struct, whose address is given by the `cstat` argument. The `off` and `len` arguments must be non-negative integers. If `len` > 0, the queried range is [`off`, `off` + `len`]. If `len` == 0, we will query in the range from `off` to the end of the file. The `flags` argument is unused for now, but is included for future extensibility. User should pass 0 (i.e no flag specified). Currently, hugetlbfs is not supported. Because the status of a page can change after cachestat() checks it but before it returns to the application, the returned values may contain stale information. RETURN VALUE On success, cachestat returns 0. On error, -1 is returned, and errno is set to indicate the error. ERRORS EFAULT cstat or cstat_args points to an invalid address. EINVAL invalid flags. EBADF invalid file descriptor. EOPNOTSUPP file descriptor is of a hugetlbfs file Signed-off-by: Nhat Pham <nphamcs@gmail.com> --- arch/x86/entry/syscalls/syscall_32.tbl | 1 + arch/x86/entry/syscalls/syscall_64.tbl | 1 + include/linux/syscalls.h | 5 + include/uapi/asm-generic/unistd.h | 5 +- include/uapi/linux/mman.h | 14 ++ init/Kconfig | 10 ++ kernel/sys_ni.c | 1 + mm/filemap.c | 172 +++++++++++++++++++++++++ 8 files changed, 208 insertions(+), 1 deletion(-)