vfs: fix copy_file_range() averts filesystem freeze protection
Subsystems (5)
PATCH v2 1/1 vfs: fix copy_file_range() averts filesystem freeze protection
Commit 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies") removed fallback to generic_copy_file_range() for cross-fs cases inside vfs_copy_file_range().
To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to generic_copy_file_range() was added in nfsd and ksmbd code, but that call is missing sb_start_write(), fsnotify hooks and more.
Ideally, nfsd and ksmbd would pass a flag to vfs_copy_file_range() that will take care of the fallback, but that code would be subtle and we got vfs_copy_file_range() logic wrong too many times already.
Instead, add a flag to explicitly request vfs_copy_file_range() to perform only generic_copy_file_range() and let nfsd and ksmbd use this flag only in the fallback path.
This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code paths to reduce the risk of further regressions.
---
Hi Linus,
I've tried Al, but he seems to be AFK, so since you ended up applying the regressing commit, I might as well send you the fix as well.
I intentionally chose a fix "for dummies", because I'd like to end this copy_file_range() regression streak.
I ran the copy_range fstests group on ext4/xfs/overlay to verify no regressions in local fs and nfsv3/nfsv4 to test server-side-copy.
I also patched copy_file_range() locally to test the "dumb" fallback code on local fs.
Namje tested ksmbd.
Please apply.
Thanks,
Amir.
Changes since v1:
- Added Tested-by's
fs/ksmbd/vfs.c | 6 +++--- fs/nfsd/vfs.c | 4 ++-- fs/read_write.c | 19 +++++++++++++++---- include/linux/fs.h | 8 ++++++++ 4 files changed, 28 insertions(+), 9 deletions(-)
fs/ksmbd/vfs.c +3 -3
@@ -1794,9 +1794,9 @@ int ksmbd_vfs_copy_file_ranges(struct ksmbd_work *work, ret =vfs_copy_file_range (src_fp->filp, src_off, dst_fp->filp, dst_off, len,0 );if (ret == -EOPNOTSUPP || ret == -EXDEV ) - ret =generic_copy_file_range (src_fp->filp, src_off, - dst_fp->filp, dst_off, - len,0 ); + ret =vfs_copy_file_range (src_fp->filp, src_off, + dst_fp->filp, dst_off, len, +COPY_FILE_SPLICE );if (ret <0 )return ret;
fs/nfsd/vfs.c +2 -2
@@ -596,8 +596,8 @@ ssize_t nfsd_copy_file_range(struct file *src, u64 src_pos, struct file *dst, ret =vfs_copy_file_range (src, src_pos, dst, dst_pos, count,0 );if (ret == -EOPNOTSUPP || ret == -EXDEV ) - ret =generic_copy_file_range (src, src_pos, dst, dst_pos, - count,0 ); + ret =vfs_copy_file_range (src, src_pos, dst, dst_pos, count, +COPY_FILE_SPLICE );return ret; }
fs/read_write.c +15 -4
@@ -1388,6 +1388,8 @@ ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in, struct file *file_out, loff_t pos_out, size_t len, unsigned int flags) { + lockdep_assert(sb_write_started(file_inode(file_out)->i_sb)); +return do_splice_direct (file_in, &pos_in, file_out, &pos_out, len >MAX_RW_COUNT ?MAX_RW_COUNT : len,0 ); }
@@ -1424,7 +1426,9 @@ static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,* and several different sets of file_operations, but they all end up * using the same ->copy_file_range() function pointer. */ -if (file_out->f_op->copy_file_range) { +if (flags ©_FILE_SPLICE ) { +/* cross sb splice is allowed */ + }else if (file_out->f_op->copy_file_range) {if (file_in->f_op->copy_file_range != file_out->f_op->copy_file_range)return -EXDEV ;
@@ -1474,8 +1478,9 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, size_t len, unsigned int flags) { ssize_t ret; +bool splice = flags ©_FILE_SPLICE ; -if (flags !=0 ) +if (flags & ~COPY_FILE_SPLICE )return -EINVAL ; ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,
@@ -1501,14 +1506,14 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,* same sb using clone, but for filesystems where both clone and copy * are supported (e.g. nfs,cifs), we only call the copy method. */ -if (file_out->f_op->copy_file_range) { +if (!splice && file_out->f_op->copy_file_range) { ret = file_out->f_op->copy_file_range (file_in, pos_in, file_out, pos_out, len, flags); goto done; } -if (file_in->f_op->remap_file_range && +if (!splice && file_in->f_op->remap_file_range &&file_inode (file_in)->i_sb ==file_inode (file_out)->i_sb) { ret = file_in->f_op->remap_file_range(file_in, pos_in, file_out, pos_out,
@@ -1528,6 +1533,8 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,* consistent story about which filesystems support copy_file_range() * and which filesystems do not, that will allow userspace tools to * make consistent desicions w.r.t using copy_file_range(). +* +* We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE. */ ret =generic_copy_file_range (file_in, pos_in, file_out, pos_out, len, flags);
@@ -1582,6 +1589,10 @@ SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, pos_out = f_out.file->f_pos; } + ret = -EINVAL ; +if (flags !=0 ) + goto out; + ret =vfs_copy_file_range (f_in.file, pos_in, f_out.file, pos_out, len, flags);if (ret >0 ) {
include/linux/fs.h +8 -0
@@ -2089,6 +2089,14 @@ struct dir_context {*/ #define REMAP_FILE_ADVISORY (REMAP_FILE_CAN_SHORTEN) +/* +* These flags control the behavior of vfs_copy_file_range(). +* They are not available to the user via syscall. +* +* COPY_FILE_SPLICE: call splice direct instead of fs clone/copy ops +*/ +#define COPY_FILE_SPLICE (1 << 0) +struct iov_iter ;struct io_uring_cmd ;
Signature · 1 line
2.25.1
Quoted 37 lines · Amir Goldstein
Commit
868f9f2f8e00("vfs: fixcopy_file_range()regression in cross-fs copies") removed fallback togeneric_copy_file_range()for cross-fs cases insidevfs_copy_file_range().To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to
generic_copy_file_range()was added in nfsd and ksmbd code, but that call is missingsb_start_write(), fsnotify hooks and more.Ideally, nfsd and ksmbd would pass a flag to
vfs_copy_file_range()that will take care of the fallback, but that code would be subtle and we gotvfs_copy_file_range()logic wrong too many times already.Instead, add a flag to explicitly request
vfs_copy_file_range()to perform onlygeneric_copy_file_range()and let nfsd and ksmbd use this flag only in the fallback path.This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code paths to reduce the risk of further regressions.
Fixes 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies")Tested-by Namjae Jeon <[email protected]>Tested-by Luis Henriques <[email protected]>Signed-off-by Amir Goldstein <[email protected]>---
Hi Linus,
I've tried Al, but he seems to be AFK, so since you ended up applying the regressing commit, I might as well send you the fix as well.
I intentionally chose a fix "for dummies", because I'd like to end this
copy_file_range()regression streak.I ran the
copy_rangefstests group on ext4/xfs/overlay to verify no regressions in local fs and nfsv3/nfsv4 to test server-side-copy.
Quoted 7 lines
I also patched
copy_file_range()locally to test the "dumb" fallback code on local fs.Namje tested ksmbd.
Please apply.
Ping.
Happy Thanksgiving!
Amir.
Quoted 136 lines
Changes since v1:
- Added Tested-by'sfs/ksmbd/vfs.c | 6 +++--- fs/nfsd/vfs.c | 4 ++-- fs/read_write.c | 19 +++++++++++++++---- include/linux/fs.h | 8 ++++++++ 4 files changed, 28 insertions(+), 9 deletions(-)fs/ksmbd/vfs.c +3
-3@@ -1794,9 +1794,9 @@ int ksmbd_vfs_copy_file_ranges(struct ksmbd_work *work, ret =vfs_copy_file_range (src_fp->filp, src_off, dst_fp->filp, dst_off, len,0 );if (ret == -EOPNOTSUPP || ret == -EXDEV ) - ret =generic_copy_file_range (src_fp->filp, src_off, - dst_fp->filp, dst_off, - len,0 ); + ret =vfs_copy_file_range (src_fp->filp, src_off, + dst_fp->filp, dst_off, len, +COPY_FILE_SPLICE );if (ret <0 )return ret;fs/nfsd/vfs.c +2
-2@@ -596,8 +596,8 @@ ssize_t nfsd_copy_file_range(struct file *src, u64 src_pos, struct file *dst, ret =vfs_copy_file_range (src, src_pos, dst, dst_pos, count,0 );if (ret == -EOPNOTSUPP || ret == -EXDEV ) - ret =generic_copy_file_range (src, src_pos, dst, dst_pos, - count,0 ); + ret =vfs_copy_file_range (src, src_pos, dst, dst_pos, count, +COPY_FILE_SPLICE );return ret; }fs/read_write.c +15
-4@@ -1388,6 +1388,8 @@ ssize_t generic_copy_file_range(struct file *file_in, loff_t pos_in, struct file *file_out, loff_t pos_out, size_t len, unsigned int flags) { + lockdep_assert(sb_write_started(file_inode(file_out)->i_sb)); +return do_splice_direct (file_in, &pos_in, file_out, &pos_out, len >MAX_RW_COUNT ?MAX_RW_COUNT : len,0 ); }@@ -1424,7 +1426,9 @@ static int generic_copy_file_checks(struct file *file_in, loff_t pos_in,* and several different sets of file_operations, but they all end up * using the same ->copy_file_range() function pointer. */ -if (file_out->f_op->copy_file_range) { +if (flags ©_FILE_SPLICE ) { +/* cross sb splice is allowed */ + }else if (file_out->f_op->copy_file_range) {if (file_in->f_op->copy_file_range != file_out->f_op->copy_file_range)return -EXDEV ;@@ -1474,8 +1478,9 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in, size_t len, unsigned int flags) { ssize_t ret; +bool splice = flags ©_FILE_SPLICE ; -if (flags !=0 ) +if (flags & ~COPY_FILE_SPLICE )return -EINVAL ; ret = generic_copy_file_checks(file_in, pos_in, file_out, pos_out, &len,@@ -1501,14 +1506,14 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,* same sb using clone, but for filesystems where both clone and copy * are supported (e.g. nfs,cifs), we only call the copy method. */ -if (file_out->f_op->copy_file_range) { +if (!splice && file_out->f_op->copy_file_range) { ret = file_out->f_op->copy_file_range (file_in, pos_in, file_out, pos_out, len, flags); goto done; } -if (file_in->f_op->remap_file_range && +if (!splice && file_in->f_op->remap_file_range &&file_inode (file_in)->i_sb ==file_inode (file_out)->i_sb) { ret = file_in->f_op->remap_file_range(file_in, pos_in, file_out, pos_out,@@ -1528,6 +1533,8 @@ ssize_t vfs_copy_file_range(struct file *file_in, loff_t pos_in,* consistent story about which filesystems support copy_file_range() * and which filesystems do not, that will allow userspace tools to * make consistent desicions w.r.t using copy_file_range(). +* +* We also get here if caller (e.g. nfsd) requested COPY_FILE_SPLICE. */ ret =generic_copy_file_range (file_in, pos_in, file_out, pos_out, len, flags);@@ -1582,6 +1589,10 @@ SYSCALL_DEFINE6(copy_file_range, int, fd_in, loff_t __user *, off_in, pos_out = f_out.file->f_pos; } + ret = -EINVAL ; +if (flags !=0 ) + goto out; + ret =vfs_copy_file_range (f_in.file, pos_in, f_out.file, pos_out, len, flags);if (ret >0 ) {include/linux/fs.h +8
-1@@ -2089,6 +2089,14 @@ struct dir_context {*/ #define REMAP_FILE_ADVISORY (REMAP_FILE_CAN_SHORTEN) +/* +* These flags control the behavior of vfs_copy_file_range(). +* They are not available to the user via syscall. +* +* COPY_FILE_SPLICE: call splice direct instead of fs clone/copy ops +*/ +#define COPY_FILE_SPLICE (1 << 0) +struct iov_iter ;struct io_uring_cmd ; --2.25.1
Quoted 16 lines · Amir Goldstein
Commit
868f9f2f8e00("vfs: fixcopy_file_range()regression in cross-fs copies") removed fallback togeneric_copy_file_range()for cross-fs cases insidevfs_copy_file_range().To preserve behavior of nfsd and ksmbd server-side-copy, the fallback to
generic_copy_file_range()was added in nfsd and ksmbd code, but that call is missingsb_start_write(), fsnotify hooks and more.Ideally, nfsd and ksmbd would pass a flag to
vfs_copy_file_range()that will take care of the fallback, but that code would be subtle and we gotvfs_copy_file_range()logic wrong too many times already.Instead, add a flag to explicitly request
vfs_copy_file_range()to perform onlygeneric_copy_file_range()and let nfsd and ksmbd use this flag only in the fallback path.
Quoted 7 lines
This choise keeps the logic changes to minimum in the non-nfsd/ksmbd code paths to reduce the risk of further regressions.
Fixes 868f9f2f8e00 ("vfs: fix copy_file_range() regression in cross-fs copies")Tested-by Namjae Jeon <[email protected]>Tested-by Luis Henriques <[email protected]>Signed-off-by Amir Goldstein <[email protected]>
Applied...
Ok, this is finally in my tree now. Thanks,
Linus