fs/sync.c: sync_file_range(2) may use WB_SYNC_ALL writeback
Subsystems (2)
PATCH 1/1 fs/sync.c: sync_file_range(2) may use WB_SYNC_ALL writeback
Commit 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback") claims that sync_file_range(2) syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) to WB_SYNC_NONE.
This claim is only partially true. Is is true for users that use the flag SYNC_FILE_RANGE_WRITE by itself, as does PostgreSQL, the user that was the reason for changing to WB_SYNC_NONE writeback.
However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages. sync_file_range(2) man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.
Re-brand that flag combination as SYNC_FILE_RANGE_WRITE_AND_WAIT and use the helper filemap_write_and_wait_range(), that uses WB_SYNC_ALL writeback, to perform the range sync request.
One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR);
write(fd, newdata, count);
sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0);
linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of sync_file_range(2), even with WB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).
---
Hi folks,
I was debating with myself whether I should change behavior of sync_file_range(2) do to what documentation says it does and flag names suggest that they do or to introduce a new flag for the old/new behavior.
At the moment, myself won with restoring old behavior without a new flag, but with re-branding of existing flags. I do not see how there could be users relying on existing behavior (TM).
An argument in favor of new flag is that with existing flags, application doesn't know if kernel provides the guaranty or not.
An argument is favor of existing flags is to not further complicate the man page, that is complicated enough as it is.
With this patch in upstream and stable kernels, the only thing I would add to man page is:
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
+ (or SYNC_FILE_RANGE_WRITE_AND_WAIT)
This is a write-for-data-integrity operation that will ensure that all pages in the specified range which were dirty when sync_file_range() was called are
- committed to disk. + written to disk. It should be noted that disk caches are not flushed by this + call, so there are no guarantees here that the data will be available on disk + after a crash.
Thought?
Amir.
fs/sync.c | 25 ++++++++++++++++++------- include/uapi/linux/fs.h | 3 +++ 2 files changed, 21 insertions(+), 7 deletions(-)
fs/sync.c +18 -7
@@ -18,8 +18,8 @@#include <linux/backing-dev.h> #include "internal.h" -#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ - SYNC_FILE_RANGE_WAIT_AFTER) +#define VALID_FLAGS (SYNC_FILE_RANGE_WRITE | SYNC_FILE_RANGE_WRITE_AND_WAIT | \ + SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WAIT_AFTER)/* * Do the filesystem syncing work. For simple filesystems
@@ -235,9 +235,9 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd) }/* -* sys_sync_file_range() permits finely controlled syncing over a segment of +* ksys_sync_file_range() permits finely controlled syncing over a segment of * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is -* zero then sys_sync_file_range() will operate from offset out to EOF. +* zero then ksys_sync_file_range() will operate from offset out to EOF. * * The flag bits are: *
@@ -254,7 +254,7 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* Useful combinations of the flag bits are: * * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages -* in the range which were dirty on entry to sys_sync_file_range() are placed +* in the range which were dirty on entry to ksys_sync_file_range() are placed * under writeout. This is a start-write-for-data-integrity operation. * * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
@@ -266,10 +266,13 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait * for that operation to complete and to return the result. * -* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: +* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER +* (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): * a traditional sync() operation. This is a write-for-data-integrity operation * which will ensure that all pages in the range which were dirty on entry to -* sys_sync_file_range() are committed to disk. +* ksys_sync_file_range() are written to disk. It should be noted that disk +* caches are not flushed by this call, so there are no guarantees here that the +* data will be available on disk after a crash. * * * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
@@ -338,6 +341,14 @@ int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, mapping = f.file->f_mapping; ret =0 ; +if ((flags &SYNC_FILE_RANGE_WRITE_AND_WAIT ) == +SYNC_FILE_RANGE_WRITE_AND_WAIT ) { +/* Unlike SYNC_FILE_RANGE_WRITE alone uses WB_SYNC_ALL */ + ret =filemap_write_and_wait_range (mapping, offset, endbyte); +if (ret <0 ) + goto out_put; + } +if (flags &SYNC_FILE_RANGE_WAIT_BEFORE ) { ret =file_fdatawait_range (f.file, offset, endbyte);if (ret <0 )
include/uapi/linux/fs.h +3 -0
@@ -320,6 +320,9 @@ struct fscrypt_key {#define SYNC_FILE_RANGE_WAIT_BEFORE 1#define SYNC_FILE_RANGE_WRITE 2#define SYNC_FILE_RANGE_WAIT_AFTER 4 +#define SYNC_FILE_RANGE_WRITE_AND_WAIT (SYNC_FILE_RANGE_WRITE | \ + SYNC_FILE_RANGE_WAIT_BEFORE | \ + SYNC_FILE_RANGE_WAIT_AFTER)/* * Flags for preadv2/pwritev2:
Signature · 1 line
2.17.1
Jan Kara
reviewerQuoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);
Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).
I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Rather I'd modify the manpage to not say that SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WRITE | SYNC_FILE_RANGE_WAIT_AFTER is a write-for-data-integrity.
Honza
Quoted 126 lines
Fixes 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE")Signed-off-by Amir Goldstein <[email protected]>---
Hi folks,
I was debating with myself whether I should change behavior of
sync_file_range(2)do to what documentation says it does and flag names suggest that they do or to introduce a new flag for the old/new behavior.At the moment, myself won with restoring old behavior without a new flag, but with re-branding of existing flags. I do not see how there could be users relying on existing behavior (TM).
An argument in favor of new flag is that with existing flags, application doesn't know if kernel provides the guaranty or not.
An argument is favor of existing flags is to not further complicate the man page, that is complicated enough as it is.
With this patch in upstream and stable kernels, the only thing I would add to man page is:SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER
+ (orSYNC_FILE_RANGE_WRITE_AND_WAIT)
This is a write-for-data-integrity operation that will ensure that all pages in the specified range which were dirty whensync_file_range()was called are(quoted diff) +3
-1- committed to disk. + written to disk. It should be noted that disk caches are not flushed by this + call, so there are no guarantees here that the data will be available on disk + after a crash.Thought?
Amir.
fs/sync.c | 25 ++++++++++++++++++------- include/uapi/linux/fs.h | 3 +++ 2 files changed, 21 insertions(+), 7 deletions(-)fs/sync.c +18
-7@@ -18,8 +18,8 @@#include <linux/backing-dev.h> #include "internal.h" -#define VALID_FLAGS (SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE| \ - SYNC_FILE_RANGE_WAIT_AFTER) +#define VALID_FLAGS (SYNC_FILE_RANGE_WRITE | SYNC_FILE_RANGE_WRITE_AND_WAIT | \ + SYNC_FILE_RANGE_WAIT_BEFORE | SYNC_FILE_RANGE_WAIT_AFTER)/* * Do the filesystem syncing work. For simple filesystems @@ -235,9 +235,9 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd) }/* -* sys_sync_file_range() permits finely controlled syncing over a segment of +* ksys_sync_file_range() permits finely controlled syncing over a segment of * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is -* zero then sys_sync_file_range() will operate from offset out to EOF. +* zero then ksys_sync_file_range() will operate from offset out to EOF. * * The flag bits are: * @@ -254,7 +254,7 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* Useful combinations of the flag bits are: * * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages -* in the range which were dirty on entry to sys_sync_file_range() are placed +* in the range which were dirty on entry to ksys_sync_file_range() are placed * under writeout. This is a start-write-for-data-integrity operation. * * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which @@ -266,10 +266,13 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait * for that operation to complete and to return the result. * -* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: +* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER +* (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): * a traditional sync() operation. This is a write-for-data-integrity operation * which will ensure that all pages in the range which were dirty on entry to -* sys_sync_file_range() are committed to disk. +* ksys_sync_file_range() are written to disk. It should be noted that disk +* caches are not flushed by this call, so there are no guarantees here that the +* data will be available on disk after a crash. * * * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any @@ -338,6 +341,14 @@ int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, mapping = f.file->f_mapping; ret =0 ; +if ((flags &SYNC_FILE_RANGE_WRITE_AND_WAIT ) == +SYNC_FILE_RANGE_WRITE_AND_WAIT ) { +/* Unlike SYNC_FILE_RANGE_WRITE alone uses WB_SYNC_ALL */ + ret =filemap_write_and_wait_range (mapping, offset, endbyte); +if (ret <0 ) + goto out_put; + } +if (flags &SYNC_FILE_RANGE_WAIT_BEFORE ) { ret =file_fdatawait_range (f.file, offset, endbyte);if (ret <0 )include/uapi/linux/fs.h +3
-0@@ -320,6 +320,9 @@ struct fscrypt_key {#define SYNC_FILE_RANGE_WAIT_BEFORE 1#define SYNC_FILE_RANGE_WRITE 2#define SYNC_FILE_RANGE_WAIT_AFTER 4 +#define SYNC_FILE_RANGE_WRITE_AND_WAIT (SYNC_FILE_RANGE_WRITE | \ + SYNC_FILE_RANGE_WAIT_BEFORE | \ + SYNC_FILE_RANGE_WAIT_AFTER)/* * Flags for preadv2/pwritev2: Signature · 1 line
2.17.1
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 51 lines · Jan Kara
Quoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
Quoted 7 lines
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Rather I'd modify the manpage to not say that
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTERis a write-for-data-integrity.
OK. I do agree that manpage is misleading and that the language describing this flag combination should be toned down. I do not object to adding more and more disclaimers about this API not being a data integrity API.
But the requirement I have is a real life workload and fdatasync(2) is not at all a viable option when application is creating many files per second.
I need to export the functionality of data writeback to userspace. Objecting to expose this functionality via the interface that has been documented to expose it for years and used to expose it in the past (until the Fixes commit) is a bit weird IMO, even though I do completely relate to your concern.
I don't mind removing the "intended use case" part of commit message if that helps reducing the chance that people misunderstand the guaranties of the API.
Another thing I could do is introduce a new flag for sync_file_range() that will really mean what I want it to mean (all data will be written back and all related inode metadata changes will be committed to journal before the next inode metadata change is committed). For the sake of argument let's call it SYNC_FILE_RANGE_DATA_DEPENDENCY.
I could then extend the ->fsync() f_op 'datasync' argument to 'sync_flags' and let sync_file_range() call the filesystem to request the data dependency. By default, no filesystem will support the new flag, but xfs and ext4 could implement flag support simply by calling filemap_write_and_wait_range().
This way, the contract between the application and filesystem is drawn in a way that filesystem is still able to make writeback optimizations without breaking the contract. For example, if ext4 changes in a way that filemap_write_and_wait_range() no longer gives the data dependency guaranty, or if ext4 is mounted with no journal, then ext4 could refuse the data dependency request or it could implement it differently.
Does that sound any better? If so, any suggestion for a less horid name for this new flag?
Thanks,
Amir.
Jan Kara
reviewerQuoted 55 lines · Amir Goldstein
Quoted 54 lines · Jan Kara
Quoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Quoted 7 lines
Quoted 3 lines
Rather I'd modify the manpage to not say that
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTERis a write-for-data-integrity.OK. I do agree that manpage is misleading and that the language describing this flag combination should be toned down. I do not object to adding more and more disclaimers about this API not being a data integrity API.
I don't think we need more disclaimers, I'd just reword that "write-for-data-integrity" bit.
Quoted 13 lines
But the requirement I have is a real life workload and fdatasync(2) is not at all a viable option when application is creating many files per second.
I need to export the functionality of data writeback to userspace. Objecting to expose this functionality via the interface that has been documented to expose it for years and used to expose it in the past (until the Fixes commit) is a bit weird IMO, even though I do completely relate to your concern.
I don't mind removing the "intended use case" part of commit message if that helps reducing the chance that people misunderstand the guaranties of the API.
Quoted 5 lines
Another thing I could do is introduce a new flag for
sync_file_range()that will really mean what I want it to mean (all data will be written back and all related inode metadata changes will be committed to journal before the next inode metadata change is committed). For the sake of argument let's call itSYNC_FILE_RANGE_DATA_DEPENDENCY.
So there are two parts to your requirements:
1) Data writeback for all file pages has been submitted (+ completed).
2) What happens with related metadata.
sync_file_range() is fine for 1) although I agree that WB_SYNC_NONE mode can still end up skipping some pages due to unrelated writeback / lock contention so having sync_file_range() mode doing WB_SYNC_ALL writeback makes some sense to me.
What happens with 2) is pretty much up to the filesystem. You are speaking about the journal but that is very much a filesystem specific detail. What if the filesystem doesn't have journal at all? E.g. ext2? We cannot really expose a generic API that silently fails to work based on fs internal details.
I guess I fail to see what kind of guarantees from the fs you're looking for? When you speak about journal, I guess you're looking for some kind of ordering but it's hard to be sure about exact requirements. So can you please spell those out? Once we know requirements, we can speak about the API.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 89 lines · Jan Kara
Quoted 55 lines · Amir Goldstein
Quoted 54 lines · Jan Kara
Quoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Quoted 7 lines
Quoted 3 lines
Rather I'd modify the manpage to not say that
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTERis a write-for-data-integrity.OK. I do agree that manpage is misleading and that the language describing this flag combination should be toned down. I do not object to adding more and more disclaimers about this API not being a data integrity API.
I don't think we need more disclaimers, I'd just reword that "write-for-data-integrity" bit.
Quoted 13 lines
But the requirement I have is a real life workload and fdatasync(2) is not at all a viable option when application is creating many files per second.
I need to export the functionality of data writeback to userspace. Objecting to expose this functionality via the interface that has been documented to expose it for years and used to expose it in the past (until the Fixes commit) is a bit weird IMO, even though I do completely relate to your concern.
I don't mind removing the "intended use case" part of commit message if that helps reducing the chance that people misunderstand the guaranties of the API.
Quoted 5 lines
Another thing I could do is introduce a new flag for
sync_file_range()that will really mean what I want it to mean (all data will be written back and all related inode metadata changes will be committed to journal before the next inode metadata change is committed). For the sake of argument let's call itSYNC_FILE_RANGE_DATA_DEPENDENCY.So there are two parts to your requirements:
Quoted 8 lines
1) Data writeback for all file pages has been submitted (+ completed).
2) What happens with related metadata.
sync_file_range()is fine for 1) although I agree thatWB_SYNC_NONEmode can still end up skipping some pages due to unrelated writeback / lock contention so havingsync_file_range()mode doingWB_SYNC_ALLwriteback makes some sense to me.
OK. does it make enough sense for you to ACK my patch if the commit message was toned down to:
Commit 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback") claims that sync_file_range(2) syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) to WB_SYNC_NONE.
This claim is only partially true. It is true for users that use the flag SYNC_FILE_RANGE_WRITE by itself, as does PostgreSQL, the user that was the reason for changing to WB_SYNC_NONE writeback.
However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
Re-brand that flag combination as SYNC_FILE_RANGE_WRITE_AND_WAIT and use the helper filemap_write_and_wait_range(), that uses WB_SYNC_ALL writeback, to perform the range sync request.
Quoted 6 lines
What happens with 2) is pretty much up to the filesystem. You are speaking about the journal but that is very much a filesystem specific detail. What if the filesystem doesn't have journal at all? E.g. ext2? We cannot really expose a generic API that silently fails to work based on fs internal details.
Quoted 6 lines
I guess I fail to see what kind of guarantees from the fs you're looking for? When you speak about journal, I guess you're looking for some kind of ordering but it's hard to be sure about exact requirements. So can you please spell those out? Once we know requirements, we can speak about the API.
The requirement is quite easy to explain with an example, less easy to generalize. I illustrated the requirement with the example:
O_TMPFILE;write();DATA_BARRIER;linkat()
The requirement of the DATA_BARRIER operation is that if the effects of linkat() are observed after crash, then the effects of write() must also be observed after crash.
At this point one may say that fdatasync(2) meets the requirements of a DATA_BARRIER operation, so I need to introduce a non-requirement. The non-requirement is that DATA_BARRIER operation does not need to be a DATA_INTEGRITY operation, meaning that after crash, there is no requirement to be able to find or access the written data.
Now it should be clear to filesystem developers why the DATA_BARRIER requirement is easier to meet with far less performance penalty than fdatasync(2) on xfs and ext4.
Here is a clumsy attempt to generalize the guaranty of such operation, without referring to internal fs implementation details:
"Make sure that all file data hits persistent storage before the next
file metadata change hits persistent storage".
Does that make sense to you?
Is the requirement clear?
How does this sound for an API:sync_file_range(fd, SYNC_FILE_RANGE_WRITE_DATA_BARRIER, 0, 0);
which individual filesystems could support or EOPNOTSUPP via the ->fsync() file_operation?
Thanks,
Amir.
Jan Kara
reviewerQuoted 117 lines · Amir Goldstein
Quoted 89 lines · Jan Kara
Quoted 55 lines · Amir Goldstein
Quoted 54 lines · Jan Kara
Quoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Quoted 7 lines
Quoted 3 lines
Rather I'd modify the manpage to not say that
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTERis a write-for-data-integrity.OK. I do agree that manpage is misleading and that the language describing this flag combination should be toned down. I do not object to adding more and more disclaimers about this API not being a data integrity API.
I don't think we need more disclaimers, I'd just reword that "write-for-data-integrity" bit.
Quoted 13 lines
But the requirement I have is a real life workload and fdatasync(2) is not at all a viable option when application is creating many files per second.
I need to export the functionality of data writeback to userspace. Objecting to expose this functionality via the interface that has been documented to expose it for years and used to expose it in the past (until the Fixes commit) is a bit weird IMO, even though I do completely relate to your concern.
I don't mind removing the "intended use case" part of commit message if that helps reducing the chance that people misunderstand the guaranties of the API.
Quoted 5 lines
Another thing I could do is introduce a new flag for
sync_file_range()that will really mean what I want it to mean (all data will be written back and all related inode metadata changes will be committed to journal before the next inode metadata change is committed). For the sake of argument let's call itSYNC_FILE_RANGE_DATA_DEPENDENCY.So there are two parts to your requirements:
Quoted 8 lines
1) Data writeback for all file pages has been submitted (+ completed).
2) What happens with related metadata.
sync_file_range()is fine for 1) although I agree thatWB_SYNC_NONEmode can still end up skipping some pages due to unrelated writeback / lock contention so havingsync_file_range()mode doingWB_SYNC_ALLwriteback makes some sense to me.OK. does it make enough sense for you to ACK my patch if the commit message was toned down to:
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. It is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
Quoted 6 lines
Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.Fixes 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE")Signed-off-by Amir Goldstein <[email protected]>
Yes, I'd be fine with that.
Quoted 13 lines
Quoted 6 lines
What happens with 2) is pretty much up to the filesystem. You are speaking about the journal but that is very much a filesystem specific detail. What if the filesystem doesn't have journal at all? E.g. ext2? We cannot really expose a generic API that silently fails to work based on fs internal details.
Quoted 6 lines
I guess I fail to see what kind of guarantees from the fs you're looking for? When you speak about journal, I guess you're looking for some kind of ordering but it's hard to be sure about exact requirements. So can you please spell those out? Once we know requirements, we can speak about the API.
Quoted 7 lines
The requirement is quite easy to explain with an example, less easy to generalize. I illustrated the requirement with the example:
O_TMPFILE;write();DATA_BARRIER;linkat()The requirement of the
DATA_BARRIERoperation is that if the effects oflinkat()are observed after crash, then the effects ofwrite()must also be observed after crash.
OK, understood and I agree this makes sense to ask for... And you're not the only one. Application programmers often *assume* that if they do write(file_new), rename(file_new, file), they will either see old or new file contents after a crash. They are wrong but ext4 has heuristics to detect such behavior and at least initiate writeback on rename to make time window when data loss can happen smaller. Because there was lot of controversy around this when users were transitioning from ext3 (where this worked) to ext4. If there was API with decent performance to achieve this, I guess at least some applications would start using it.
Quoted 5 lines
At this point one may say that fdatasync(2) meets the requirements of a
DATA_BARRIERoperation, so I need to introduce a non-requirement. The non-requirement is thatDATA_BARRIERoperation does not need to be aDATA_INTEGRITYoperation, meaning that after crash, there is no requirement to be able to find or access the written data.
Well, I guess your other real requirement is good performance ;) How that's achieved is internal to the filesystem.
Quoted 3 lines
Now it should be clear to filesystem developers why the
DATA_BARRIERrequirement is easier to meet with far less performance penalty than fdatasync(2) on xfs and ext4.
Well, it won't be free either but yes, you can save a transaction commit / disk cache flush compared to fdatasync(2).
Quoted 5 lines
Here is a clumsy attempt to generalize the guaranty of such operation, without referring to internal fs implementation details:
"Make sure that all file data hits persistent storage before the next
file metadata change hits persistent storage".
Quoted 8 lines
Does that make sense to you?
Is the requirement clear?How does this sound for an API:
sync_file_range(fd, SYNC_FILE_RANGE_WRITE_DATA_BARRIER, 0, 0);which individual filesystems could support or EOPNOTSUPP via the ->fsync()
file_operation?
Yes, this sounds workable to me but there are also other options - e.g. somehow pass the information to linkat() / rename() / whatever that the ordering is required. This might be easier for some filesystems to implement. So it might be good to make a short session at LSF/MM about this so that developers of other filesystems can say what interface they would be able to implement...
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 119 lines · Jan Kara
Quoted 117 lines · Amir Goldstein
Quoted 89 lines · Jan Kara
Quoted 55 lines · Amir Goldstein
Quoted 54 lines · Jan Kara
Quoted 30 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. Is is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}. Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
sync_file_range(2)man page describes this flag combintaion as "write-for-data-integrity operation", although some may argue against this definition.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.One intended use case for this API is creating a dependency between a new file's content and its link into the namepsace without requiring journal commit and flushing of disk volatile caches:
fd = open(".", O_TMPFILE | O_RDWR); write(fd, newdata, count); sync_file_range(fd, SYNC_FILE_RANGE_WRITE_AND_WAIT, 0, 0); linkat(AT_EMPTY_PATH, fd, AT_FDCWD, newfile);Quoted 6 lines
For many local filesystems, ext4 and xfs included, the sequence above will guaranty that after crash, either 'newfile' exists with 'newdata' content or 'newfile' does not exists. For some applications, this guaranty is strong enough and the effects of
sync_file_range(2), even withWB_SYNC_ALL, are far less intrusive to other writers in the system than the effects of fdatasync(2).I agree that this paragraph is true but I don't want any userspace program rely on this. We've been through this when ext3 got converted to ext4 and it has caused a lot of complaints. Ext3 had provided rather strong data vs metadata ordering due to the way journalling was implemented. Then ext4 came, implemented delayed allocation and somewhat changed how journalling works and suddently userspace programmers were caught by surprise their code working by luck stopped working. And they were complaining that when their code worked without fsync(2) before, it should work after it as well. So it took a lot of explaining that their applications are broken and Ted has also implemented some workarounds to make at least the most common mistakes silently fixed by the kernel most of the time.
So I'm against providing 90% data integrity guarantees because there'll be many people who'll think they can get away with it and then complain when they won't once our implementation changes.
Quoted 7 lines
Quoted 3 lines
Rather I'd modify the manpage to not say that
SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTERis a write-for-data-integrity.OK. I do agree that manpage is misleading and that the language describing this flag combination should be toned down. I do not object to adding more and more disclaimers about this API not being a data integrity API.
I don't think we need more disclaimers, I'd just reword that "write-for-data-integrity" bit.
Quoted 13 lines
But the requirement I have is a real life workload and fdatasync(2) is not at all a viable option when application is creating many files per second.
I need to export the functionality of data writeback to userspace. Objecting to expose this functionality via the interface that has been documented to expose it for years and used to expose it in the past (until the Fixes commit) is a bit weird IMO, even though I do completely relate to your concern.
I don't mind removing the "intended use case" part of commit message if that helps reducing the chance that people misunderstand the guaranties of the API.
Quoted 5 lines
Another thing I could do is introduce a new flag for
sync_file_range()that will really mean what I want it to mean (all data will be written back and all related inode metadata changes will be committed to journal before the next inode metadata change is committed). For the sake of argument let's call itSYNC_FILE_RANGE_DATA_DEPENDENCY.So there are two parts to your requirements:
Quoted 8 lines
1) Data writeback for all file pages has been submitted (+ completed).
2) What happens with related metadata.
sync_file_range()is fine for 1) although I agree thatWB_SYNC_NONEmode can still end up skipping some pages due to unrelated writeback / lock contention so havingsync_file_range()mode doingWB_SYNC_ALLwriteback makes some sense to me.OK. does it make enough sense for you to ACK my patch if the commit message was toned down to:
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. It is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
Quoted 8 lines
Quoted 6 lines
Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the range sync request.Fixes 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE")Signed-off-by Amir Goldstein <[email protected]>Yes, I'd be fine with that.
Great.
I'll send V2 with your ACK.
I guess since Andrew merged the Fixes commit, I'll point to patch at him as well.
Quoted 59 lines
Quoted 14 lines
Quoted 7 lines
What happens with 2) is pretty much up to the filesystem. You are speaking about the journal but that is very much a filesystem specific detail. What if the filesystem doesn't have journal at all? E.g. ext2? We cannot really expose a generic API that silently fails to work based on fs internal details.
Quoted 6 lines
I guess I fail to see what kind of guarantees from the fs you're looking for? When you speak about journal, I guess you're looking for some kind of ordering but it's hard to be sure about exact requirements. So can you please spell those out? Once we know requirements, we can speak about the API.
Quoted 7 lines
The requirement is quite easy to explain with an example, less easy to generalize. I illustrated the requirement with the example:
O_TMPFILE;write();DATA_BARRIER;linkat()The requirement of the
DATA_BARRIERoperation is that if the effects oflinkat()are observed after crash, then the effects ofwrite()must also be observed after crash.OK, understood and I agree this makes sense to ask for... And you're not the only one. Application programmers often *assume* that if they do
write(file_new),rename(file_new, file), they will either see old or new file contents after a crash. They are wrong but ext4 has heuristics to detect such behavior and at least initiate writeback on rename to make time window when data loss can happen smaller. Because there was lot of controversy around this when users were transitioning from ext3 (where this worked) to ext4. If there was API with decent performance to achieve this, I guess at least some applications would start using it.Quoted 5 lines
At this point one may say that fdatasync(2) meets the requirements of a
DATA_BARRIERoperation, so I need to introduce a non-requirement. The non-requirement is thatDATA_BARRIERoperation does not need to be aDATA_INTEGRITYoperation, meaning that after crash, there is no requirement to be able to find or access the written data.Well, I guess your other real requirement is good performance ;) How that's achieved is internal to the filesystem.
Quoted 3 lines
Now it should be clear to filesystem developers why the
DATA_BARRIERrequirement is easier to meet with far less performance penalty than fdatasync(2) on xfs and ext4.Well, it won't be free either but yes, you can save a transaction commit / disk cache flush compared to fdatasync(2).
Quoted 11 lines
Here is a clumsy attempt to generalize the guaranty of such operation, without referring to internal fs implementation details:
"Make sure that all file data hits persistent storage before the next
file metadata change hits persistent storage".Does that make sense to you?
Is the requirement clear?How does this sound for an API:
sync_file_range(fd, SYNC_FILE_RANGE_WRITE_DATA_BARRIER, 0, 0);
Quoted 7 lines
Quoted 2 lines
which individual filesystems could support or EOPNOTSUPP via the ->fsync()
file_operation?Yes, this sounds workable to me but there are also other options - e.g. somehow pass the information to
linkat()/rename()/ whatever that the ordering is required. This might be easier for some filesystems to implement.
I like this idea. rename() and linkat() are definitely the most probable metadata operation that needs this barrier. But I wouldn't want to limit this capability to rename() and linkat() alone.
Overlayfs for example, could use a similar data write barrier before remove xattr (when "hydrating" a metacopy upper).
Quoted 3 lines
So it might be good to make a short session at LSF/MM about this so that developers of other filesystems can say what interface they would be able to implement...
Yeh, its on my short list for things to discuss.
[CC: Anna & Josef]
Thanks,
Amir.
PATCH v4 1/1 fs/sync.c: sync_file_range(2) may use WB_SYNC_ALL writeback
In reply to Amir GoldsteinCommit 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback") claims that sync_file_range(2) syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) to WB_SYNC_NONE.
This claim is only partially true. It is true for users that use the flag SYNC_FILE_RANGE_WRITE by itself, as does PostgreSQL, the user that was the reason for changing to WB_SYNC_NONE writeback.
However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
Re-brand that flag combination as SYNC_FILE_RANGE_WRITE_AND_WAIT and use the helper filemap_write_and_wait_range(), that uses WB_SYNC_ALL writeback, to perform the full range sync request.
---
Andrew,
V2 of this patch is on your mmtotm queue.
However, I had already sent out V3 with a braino fix and Dave Chinner just added more review comments which I had addressed in this version.
Thanks,
Amir.
Changes since v3:
- Remove unneeded change to VALID_FLAGS (Dave)
- Call file_fdatawait_range() before writeback (Dave)
Changes since v2:
- Return after filemap_write_and_wait_range()
Changes since v1:
- Remove non-guaranties of the API from commit message
- Added ACK by Jan
fs/sync.c | 20 +++++++++++++++----- include/uapi/linux/fs.h | 3 +++ 2 files changed, 18 insertions(+), 5 deletions(-)
fs/sync.c +15 -5
@@ -235,9 +235,9 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd) }/* -* sys_sync_file_range() permits finely controlled syncing over a segment of +* ksys_sync_file_range() permits finely controlled syncing over a segment of * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is -* zero then sys_sync_file_range() will operate from offset out to EOF. +* zero then ksys_sync_file_range() will operate from offset out to EOF. * * The flag bits are: *
@@ -254,7 +254,7 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* Useful combinations of the flag bits are: * * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages -* in the range which were dirty on entry to sys_sync_file_range() are placed +* in the range which were dirty on entry to ksys_sync_file_range() are placed * under writeout. This is a start-write-for-data-integrity operation. * * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
@@ -266,10 +266,13 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait * for that operation to complete and to return the result. * -* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: +* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER +* (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): * a traditional sync() operation. This is a write-for-data-integrity operation * which will ensure that all pages in the range which were dirty on entry to -* sys_sync_file_range() are committed to disk. +* ksys_sync_file_range() are written to disk. It should be noted that disk +* caches are not flushed by this call, so there are no guarantees here that the +* data will be available on disk after a crash. * * * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
@@ -344,6 +347,13 @@ int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, goto out_put; } +if ((flags &SYNC_FILE_RANGE_WRITE_AND_WAIT ) == +SYNC_FILE_RANGE_WRITE_AND_WAIT ) { +/* Unlike SYNC_FILE_RANGE_WRITE alone, uses WB_SYNC_ALL */ + ret =filemap_write_and_wait_range (mapping, offset, endbyte); + goto out_put; + } +if (flags &SYNC_FILE_RANGE_WRITE ) { ret =__filemap_fdatawrite_range (mapping, offset, endbyte,WB_SYNC_NONE );
include/uapi/linux/fs.h +3 -0
@@ -320,6 +320,9 @@ struct fscrypt_key {#define SYNC_FILE_RANGE_WAIT_BEFORE 1#define SYNC_FILE_RANGE_WRITE 2#define SYNC_FILE_RANGE_WAIT_AFTER 4 +#define SYNC_FILE_RANGE_WRITE_AND_WAIT (SYNC_FILE_RANGE_WRITE | \ + SYNC_FILE_RANGE_WAIT_BEFORE | \ + SYNC_FILE_RANGE_WAIT_AFTER)/* * Flags for preadv2/pwritev2:
Signature · 1 line
2.17.1
Quoted 93 lines · Amir Goldstein
Commit
23d0127096cb("fs/sync.c: makesync_file_range(2)useWB_SYNC_NONEwriteback") claims thatsync_file_range(2)syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) toWB_SYNC_NONE.This claim is only partially true. It is true for users that use the flag
SYNC_FILE_RANGE_WRITEby itself, as does PostgreSQL, the user that was the reason for changing toWB_SYNC_NONEwriteback.However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.Re-brand that flag combination as
SYNC_FILE_RANGE_WRITE_AND_WAITand use the helperfilemap_write_and_wait_range(), that usesWB_SYNC_ALLwriteback, to perform the full range sync request.Fixes 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE")Signed-off-by Amir Goldstein <[email protected]>Acked-by Jan Kara <[email protected]>Cc Dave Chinner <[email protected]>Cc Al Viro <[email protected]>---
Andrew,
V2 of this patch is on your mmtotm queue.
However, I had already sent out V3 with a braino fix and Dave Chinner just added more review comments which I had addressed in this version.Thanks,
Amir.Changes since v3:
- Remove unneeded change toVALID_FLAGS(Dave)
- Callfile_fdatawait_range()before writeback (Dave)Changes since v2:
- Return afterfilemap_write_and_wait_range()Changes since v1:
- Remove non-guaranties of the API from commit message
- Added ACK by Janfs/sync.c | 20 +++++++++++++++----- include/uapi/linux/fs.h | 3 +++ 2 files changed, 18 insertions(+), 5 deletions(-)fs/sync.c +8
-5@@ -235,9 +235,9 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd) }/* -* sys_sync_file_range() permits finely controlled syncing over a segment of +* ksys_sync_file_range() permits finely controlled syncing over a segment of * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is -* zero then sys_sync_file_range() will operate from offset out to EOF. +* zero then ksys_sync_file_range() will operate from offset out to EOF. * * The flag bits are: * @@ -254,7 +254,7 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* Useful combinations of the flag bits are: * * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages -* in the range which were dirty on entry to sys_sync_file_range() are placed +* in the range which were dirty on entry to ksys_sync_file_range() are placed * under writeout. This is a start-write-for-data-integrity operation. * * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which @@ -266,10 +266,13 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait * for that operation to complete and to return the result. * -* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: +* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER +* (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): * a traditional sync() operation. This is a write-for-data-integrity operation * which will ensure that all pages in the range which were dirty on entry to -* sys_sync_file_range() are committed to disk. +* ksys_sync_file_range() are written to disk. It should be noted that disk +* caches are not flushed by this call, so there are no guarantees here that the +* data will be available on disk after a crash. * * * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
fs/sync.c int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, · line 355
Quoted 9 lines
fs/sync.c +6
-0@@ -344,6 +347,13 @@ int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, goto out_put; } +if ((flags &SYNC_FILE_RANGE_WRITE_AND_WAIT ) == +SYNC_FILE_RANGE_WRITE_AND_WAIT ) { +/* Unlike SYNC_FILE_RANGE_WRITE alone, uses WB_SYNC_ALL */ + ret =filemap_write_and_wait_range (mapping, offset, endbyte); + goto out_put; + }
Clunky, now that I look at it in context.
+ int sync_mode = WB_SYNC_NONE;
+
+ if ((flags & SYNC_FILE_RANGE_WRITE_AND_WAIT) ==
+ SYNC_FILE_RANGE_WRITE_AND_WAIT)
+ sync_mode = WB_SYNC_ALL;
.....
if (flags & SYNC_FILE_RANGE_WRITE) {
ret = __filemap_fdatawrite_range(mapping, offset, endbyte,- WB_SYNC_NONE);
+ sync_mode);
Cheers,
Dave.
Signature · 2 lines
Dave Chinner [email protected]
PATCH v5 1/1 fs/sync.c: sync_file_range(2) may use WB_SYNC_ALL writeback
Commit 23d0127096cb ("fs/sync.c: make sync_file_range(2) use WB_SYNC_NONE writeback") claims that sync_file_range(2) syscall was "created for userspace to be able to issue background writeout and so waiting for in-flight IO is undesirable there" and changes the writeback (back) to WB_SYNC_NONE.
This claim is only partially true. It is true for users that use the flag SYNC_FILE_RANGE_WRITE by itself, as does PostgreSQL, the user that was the reason for changing to WB_SYNC_NONE writeback.
However, that claim is not true for users that use that flag combination SYNC_FILE_RANGE_{WAIT_BEFORE|WRITE|_WAIT_AFTER}.
Those users explicitly requested to wait for in-flight IO as well as to writeback of dirty pages.
Re-brand that flag combination as SYNC_FILE_RANGE_WRITE_AND_WAIT and use WB_SYNC_ALL writeback to perform the full range sync request.
---
Andrew,
One more version addressing another comment by Dave Chinner.
Thanks,
Amir.
Changes since v4:
- Don't use filemap_write_and_wait_range() helper (Dave)
Changes since v3:
- Remove unneeded change to VALID_FLAGS (Dave)
- Call file_fdatawait_range() before writeback (Dave)
Changes since v2:
- Return after filemap_write_and_wait_range()
Changes since v1:
- Remove non-guaranties of the API from commit message
- Added ACK by Jan
fs/sync.c | 21 +++++++++++++++------ include/uapi/linux/fs.h | 3 +++ 2 files changed, 18 insertions(+), 6 deletions(-)
fs/sync.c +15 -6
@@ -235,9 +235,9 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd) }/* -* sys_sync_file_range() permits finely controlled syncing over a segment of +* ksys_sync_file_range() permits finely controlled syncing over a segment of * a file in the range offset .. (offset+nbytes-1) inclusive. If nbytes is -* zero then sys_sync_file_range() will operate from offset out to EOF. +* zero then ksys_sync_file_range() will operate from offset out to EOF. * * The flag bits are: *
@@ -254,7 +254,7 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* Useful combinations of the flag bits are: * * SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE: ensures that all pages -* in the range which were dirty on entry to sys_sync_file_range() are placed +* in the range which were dirty on entry to ksys_sync_file_range() are placed * under writeout. This is a start-write-for-data-integrity operation. * * SYNC_FILE_RANGE_WRITE: start writeout of all dirty pages in the range which
@@ -266,10 +266,13 @@ SYSCALL_DEFINE1(fdatasync, unsigned int, fd)* earlier SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE operation to wait * for that operation to complete and to return the result. * -* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER: +* SYNC_FILE_RANGE_WAIT_BEFORE|SYNC_FILE_RANGE_WRITE|SYNC_FILE_RANGE_WAIT_AFTER +* (a.k.a. SYNC_FILE_RANGE_WRITE_AND_WAIT): * a traditional sync() operation. This is a write-for-data-integrity operation * which will ensure that all pages in the range which were dirty on entry to -* sys_sync_file_range() are committed to disk. +* ksys_sync_file_range() are written to disk. It should be noted that disk +* caches are not flushed by this call, so there are no guarantees here that the +* data will be available on disk after a crash. * * * SYNC_FILE_RANGE_WAIT_BEFORE and SYNC_FILE_RANGE_WAIT_AFTER will detect any
@@ -345,8 +348,14 @@ int ksys_sync_file_range(int fd, loff_t offset, loff_t nbytes, }if (flags &SYNC_FILE_RANGE_WRITE ) { +int sync_mode =WB_SYNC_NONE ; + +if ((flags &SYNC_FILE_RANGE_WRITE_AND_WAIT ) == +SYNC_FILE_RANGE_WRITE_AND_WAIT ) + sync_mode =WB_SYNC_ALL ; + ret =__filemap_fdatawrite_range (mapping, offset, endbyte, -WB_SYNC_NONE ); + sync_mode);if (ret <0 ) goto out_put; }
include/uapi/linux/fs.h +3 -0
@@ -320,6 +320,9 @@ struct fscrypt_key {#define SYNC_FILE_RANGE_WAIT_BEFORE 1#define SYNC_FILE_RANGE_WRITE 2#define SYNC_FILE_RANGE_WAIT_AFTER 4 +#define SYNC_FILE_RANGE_WRITE_AND_WAIT (SYNC_FILE_RANGE_WRITE | \ + SYNC_FILE_RANGE_WAIT_BEFORE | \ + SYNC_FILE_RANGE_WAIT_AFTER)/* * Flags for preadv2/pwritev2:
Signature · 1 line
2.17.1