Message ID | 5154AAB4.2000701@redhat.com |
---|---|
State | Not Applicable, archived |
Headers | show |
On Thu 28-03-13 15:40:20, Eric Sandeen wrote: > This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a > userspace application to request 32-bit rather than 64-bit > hashes from readdir on an indexed / dx / htree directory. > > Gluster had been relying on the top bits of the d_off being > free; there are some reports that filling all 64 bits breaks > Samba as well. The infrastructure to return 32-bit hashes > already exists; NFS can turn it on, and it's turned on for > 32-bit processes as well. So it's just a matter of flipping > on the f_mode flag before readdir starts. > > Care needs to be taken that we don't change the FMODE flag > after readdir has been started, so we make sure that > filp->private_data has not yet been set before we set the flag. > (Thanks Zach!). > > Pre-submission-fixes-by: Zach Brown <zab@redhat.com> > Signed-off-by: Eric Sandeen <sandeen@redhat.com> > --- > > V2: > fix "readir" typo > rename goto target to *_out like others > remove parameter; we can't really ever turn this back off once it's used. > closing and reopening is the only way to get back to 64 bit hashes. Looks good. Just one nit below: > diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h > + case EXT3_IOC_32BITHASH: { > + int err = 0; > + > + /* Serialize with readdir */ > + if ((err = mutex_lock_killable(&inode->i_mutex))) > + return err; > + > + /* protect f_mode */ > + spin_lock(&filp->f_lock); > + > + /* Only valid for htree directories */ > + if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) { Won't it be better to return ENOTDIR in !S_ISDIR case? > + err = -EINVAL; > + goto hash32bits_out; > + } > + > + /* Have we already started readdir on this dx dir? */ > + if (filp->private_data) { > + err = -EINVAL; And here maybe EBUSY, but in this case I'm really undecided. > + goto hash32bits_out; > + } > + > + filp->f_mode |= FMODE_32BITHASH; > +hash32bits_out: > + spin_unlock(&filp->f_lock); > + mutex_unlock(&inode->i_mutex); > + return err; > + } > case FITRIM: { > > struct super_block *sb = inode->i_sb; Honza
On 3/29/13 6:43 AM, Jan Kara wrote: > On Thu 28-03-13 15:40:20, Eric Sandeen wrote: >> This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a >> userspace application to request 32-bit rather than 64-bit >> hashes from readdir on an indexed / dx / htree directory. >> >> Gluster had been relying on the top bits of the d_off being >> free; there are some reports that filling all 64 bits breaks >> Samba as well. The infrastructure to return 32-bit hashes >> already exists; NFS can turn it on, and it's turned on for >> 32-bit processes as well. So it's just a matter of flipping >> on the f_mode flag before readdir starts. >> >> Care needs to be taken that we don't change the FMODE flag >> after readdir has been started, so we make sure that >> filp->private_data has not yet been set before we set the flag. >> (Thanks Zach!). >> >> Pre-submission-fixes-by: Zach Brown <zab@redhat.com> >> Signed-off-by: Eric Sandeen <sandeen@redhat.com> >> --- >> >> V2: >> fix "readir" typo >> rename goto target to *_out like others >> remove parameter; we can't really ever turn this back off once it's used. >> closing and reopening is the only way to get back to 64 bit hashes. > Looks good. Just one nit below: Oh, of course, duh! I'll send V3, thanks. -Eric >> diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h >> + case EXT3_IOC_32BITHASH: { >> + int err = 0; >> + >> + /* Serialize with readdir */ >> + if ((err = mutex_lock_killable(&inode->i_mutex))) >> + return err; >> + >> + /* protect f_mode */ >> + spin_lock(&filp->f_lock); >> + >> + /* Only valid for htree directories */ >> + if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) { > Won't it be better to return ENOTDIR in !S_ISDIR case? > >> + err = -EINVAL; >> + goto hash32bits_out; >> + } >> + >> + /* Have we already started readdir on this dx dir? */ >> + if (filp->private_data) { >> + err = -EINVAL; > And here maybe EBUSY, but in this case I'm really undecided. > >> + goto hash32bits_out; >> + } >> + >> + filp->f_mode |= FMODE_32BITHASH; >> +hash32bits_out: >> + spin_unlock(&filp->f_lock); >> + mutex_unlock(&inode->i_mutex); >> + return err; >> + } >> case FITRIM: { >> >> struct super_block *sb = inode->i_sb; > > Honza > -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/ext3/dir.c b/fs/ext3/dir.c index 87eccbb..83df29f 100644 --- a/fs/ext3/dir.c +++ b/fs/ext3/dir.c @@ -47,7 +47,7 @@ static unsigned char get_dtype(struct super_block *sb, int filetype) * * Return 1 if it is a dx dir, 0 if not */ -static int is_dx_dir(struct inode *inode) +int is_dx_dir(struct inode *inode) { struct super_block *sb = inode->i_sb; diff --git a/fs/ext3/ext3.h b/fs/ext3/ext3.h index e85ff15..f3018f4 100644 --- a/fs/ext3/ext3.h +++ b/fs/ext3/ext3.h @@ -220,6 +220,7 @@ struct ext3_new_group_data { #endif #define EXT3_IOC_GETRSVSZ _IOR('f', 5, long) #define EXT3_IOC_SETRSVSZ _IOW('f', 6, long) +#define EXT3_IOC_32BITHASH _IO('f', 13) /* * ioctl commands in 32 bit emulation @@ -1010,6 +1011,7 @@ extern void ext3_rsv_window_add(struct super_block *sb, struct ext3_reserve_wind extern int ext3_trim_fs(struct super_block *sb, struct fstrim_range *range); /* dir.c */ +extern int is_dx_dir(struct inode *inode); extern int ext3_check_dir_entry(const char *, struct inode *, struct ext3_dir_entry_2 *, struct buffer_head *, unsigned long); diff --git a/fs/ext3/ioctl.c b/fs/ext3/ioctl.c index 4d96e9a..3811664 100644 --- a/fs/ext3/ioctl.c +++ b/fs/ext3/ioctl.c @@ -251,6 +251,34 @@ group_add_out: mnt_drop_write_file(filp); return err; } + case EXT3_IOC_32BITHASH: { + int err = 0; + + /* Serialize with readdir */ + if ((err = mutex_lock_killable(&inode->i_mutex))) + return err; + + /* protect f_mode */ + spin_lock(&filp->f_lock); + + /* Only valid for htree directories */ + if (!S_ISDIR(inode->i_mode) || !is_dx_dir(inode)) { + err = -EINVAL; + goto hash32bits_out; + } + + /* Have we already started readdir on this dx dir? */ + if (filp->private_data) { + err = -EINVAL; + goto hash32bits_out; + } + + filp->f_mode |= FMODE_32BITHASH; +hash32bits_out: + spin_unlock(&filp->f_lock); + mutex_unlock(&inode->i_mutex); + return err; + } case FITRIM: { struct super_block *sb = inode->i_sb;
This adds a new ioctl, EXT3_IOC_32BITHASH, which allows a userspace application to request 32-bit rather than 64-bit hashes from readdir on an indexed / dx / htree directory. Gluster had been relying on the top bits of the d_off being free; there are some reports that filling all 64 bits breaks Samba as well. The infrastructure to return 32-bit hashes already exists; NFS can turn it on, and it's turned on for 32-bit processes as well. So it's just a matter of flipping on the f_mode flag before readdir starts. Care needs to be taken that we don't change the FMODE flag after readdir has been started, so we make sure that filp->private_data has not yet been set before we set the flag. (Thanks Zach!). Pre-submission-fixes-by: Zach Brown <zab@redhat.com> Signed-off-by: Eric Sandeen <sandeen@redhat.com> --- V2: fix "readir" typo rename goto target to *_out like others remove parameter; we can't really ever turn this back off once it's used. closing and reopening is the only way to get back to 64 bit hashes. -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html