| Message ID | 20240814090231.963520-1-raven@themaw.net (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] autofs: add per dentry expire timeout | expand |
On Wed, Aug 14, 2024 at 05:02:31PM GMT, Ian Kent wrote: > Add ability to set per-dentry mount expire timeout to autofs. > > There are two fairly well known automounter map formats, the autofs > format and the amd format (more or less System V and Berkley). > > Some time ago Linux autofs added an amd map format parser that > implemented a fair amount of the amd functionality. This was done > within the autofs infrastructure and some functionality wasn't > implemented because it either didn't make sense or required extra > kernel changes. The idea was to restrict changes to be within the > existing autofs functionality as much as possible and leave changes > with a wider scope to be considered later. > > One of these changes is implementing the amd options: > 1) "unmount", expire this mount according to a timeout (same as the > current autofs default). > 2) "nounmount", don't expire this mount (same as setting the autofs > timeout to 0 except only for this specific mount) . > 3) "utimeout=<seconds>", expire this mount using the specified > timeout (again same as setting the autofs timeout but only for > this mount). > > To implement these options per-dentry expire timeouts need to be > implemented for autofs indirect mounts. This is because all map keys > (mounts) for autofs indirect mounts use an expire timeout stored in > the autofs mount super block info. structure and all indirect mounts > use the same expire timeout. > > Now I have a request to add the "nounmount" option so I need to add > the per-dentry expire handling to the kernel implementation to do this. > > The implementation uses the trailing path component to identify the > mount (and is also used as the autofs map key) which is passed in the > autofs_dev_ioctl structure path field. The expire timeout is passed > in autofs_dev_ioctl timeout field (well, of the timeout union). > > If the passed in timeout is equal to -1 the per-dentry timeout and > flag are cleared providing for the "unmount" option. If the timeout > is greater than or equal to 0 the timeout is set to the value and the > flag is also set. If the dentry timeout is 0 the dentry will not expire > by timeout which enables the implementation of the "nounmount" option > for the specific mount. When the dentry timeout is greater than zero it > allows for the implementation of the "utimeout=<seconds>" option. > > Signed-off-by: Ian Kent <raven@themaw.net> > --- > fs/autofs/autofs_i.h | 4 ++ > fs/autofs/dev-ioctl.c | 97 ++++++++++++++++++++++++++++++++++-- > fs/autofs/expire.c | 7 ++- > fs/autofs/inode.c | 2 + > include/uapi/linux/auto_fs.h | 2 +- > 5 files changed, 104 insertions(+), 8 deletions(-) > > diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h > index 8c1d587b3eef..77c7991d89aa 100644 > --- a/fs/autofs/autofs_i.h > +++ b/fs/autofs/autofs_i.h > @@ -62,6 +62,7 @@ struct autofs_info { > struct list_head expiring; > > struct autofs_sb_info *sbi; > + unsigned long exp_timeout; > unsigned long last_used; > int count; > > @@ -81,6 +82,9 @@ struct autofs_info { > */ > #define AUTOFS_INF_PENDING (1<<2) /* dentry pending mount */ > > +#define AUTOFS_INF_EXPIRE_SET (1<<3) /* per-dentry expire timeout set for > + this mount point. > + */ > struct autofs_wait_queue { > wait_queue_head_t queue; > struct autofs_wait_queue *next; > diff --git a/fs/autofs/dev-ioctl.c b/fs/autofs/dev-ioctl.c > index 5bf781ea6d67..f011e026358e 100644 > --- a/fs/autofs/dev-ioctl.c > +++ b/fs/autofs/dev-ioctl.c > @@ -128,7 +128,13 @@ static int validate_dev_ioctl(int cmd, struct autofs_dev_ioctl *param) > goto out; > } > > + /* Setting the per-dentry expire timeout requires a trailing > + * path component, ie. no '/', so invert the logic of the > + * check_name() return for AUTOFS_DEV_IOCTL_TIMEOUT_CMD. > + */ > err = check_name(param->path); > + if (cmd == AUTOFS_DEV_IOCTL_TIMEOUT_CMD) > + err = err ? 0 : -EINVAL; > if (err) { > pr_warn("invalid path supplied for cmd(0x%08x)\n", > cmd); > @@ -396,16 +402,97 @@ static int autofs_dev_ioctl_catatonic(struct file *fp, > return 0; > } > > -/* Set the autofs mount timeout */ > +/* > + * Set the autofs mount expire timeout. > + * > + * There are two places an expire timeout can be set, in the autofs > + * super block info. (this is all that's needed for direct and offset > + * mounts because there's a distinct mount corresponding to each of > + * these) and per-dentry within within the dentry info. If a per-dentry > + * timeout is set it will override the expire timeout set in the parent > + * autofs super block info. > + * > + * If setting the autofs super block expire timeout the autofs_dev_ioctl > + * size field will be equal to the autofs_dev_ioctl structure size. If > + * setting the per-dentry expire timeout the mount point name is passed > + * in the autofs_dev_ioctl path field and the size field updated to > + * reflect this. > + * > + * Setting the autofs mount expire timeout sets the timeout in the super > + * block info. struct. Setting the per-dentry timeout does a little more. > + * If the timeout is equal to -1 the per-dentry timeout (and flag) is > + * cleared which reverts to using the super block timeout, otherwise if > + * timeout is 0 the timeout is set to this value and the flag is left > + * set which disables expiration for the mount point, lastly the flag > + * and the timeout are set enabling the dentry to use this timeout. > + */ > static int autofs_dev_ioctl_timeout(struct file *fp, > struct autofs_sb_info *sbi, > struct autofs_dev_ioctl *param) > { > - unsigned long timeout; > + unsigned long timeout = param->timeout.timeout; > + > + /* If setting the expire timeout for an individual indirect > + * mount point dentry the mount trailing component path is > + * placed in param->path and param->size adjusted to account > + * for it otherwise param->size it is set to the structure > + * size. > + */ > + if (param->size == AUTOFS_DEV_IOCTL_SIZE) { > + param->timeout.timeout = sbi->exp_timeout / HZ; > + sbi->exp_timeout = timeout * HZ; > + } else { > + struct dentry *base = fp->f_path.dentry; > + struct inode *inode = base->d_inode; > + int path_len = param->size - AUTOFS_DEV_IOCTL_SIZE - 1; > + struct dentry *dentry; > + struct autofs_info *ino; > + > + if (!autofs_type_indirect(sbi->type)) > + return -EINVAL; > + > + /* An expire timeout greater than the superblock timeout > + * could be a problem at shutdown but the super block > + * timeout itself can change so all we can really do is > + * warn the user. > + */ > + if (timeout >= sbi->exp_timeout) > + pr_warn("per-mount expire timeout is greater than " > + "the parent autofs mount timeout which could " > + "prevent shutdown\n"); Wouldn't it be possible to just record the lowest known per-dentry timeout in idk sbi->exp_lower_bound and reject sbi->exp_timeout changes that go below that?
On 14/8/24 22:39, Christian Brauner wrote: > On Wed, Aug 14, 2024 at 05:02:31PM GMT, Ian Kent wrote: >> Add ability to set per-dentry mount expire timeout to autofs. >> >> There are two fairly well known automounter map formats, the autofs >> format and the amd format (more or less System V and Berkley). >> >> Some time ago Linux autofs added an amd map format parser that >> implemented a fair amount of the amd functionality. This was done >> within the autofs infrastructure and some functionality wasn't >> implemented because it either didn't make sense or required extra >> kernel changes. The idea was to restrict changes to be within the >> existing autofs functionality as much as possible and leave changes >> with a wider scope to be considered later. >> >> One of these changes is implementing the amd options: >> 1) "unmount", expire this mount according to a timeout (same as the >> current autofs default). >> 2) "nounmount", don't expire this mount (same as setting the autofs >> timeout to 0 except only for this specific mount) . >> 3) "utimeout=<seconds>", expire this mount using the specified >> timeout (again same as setting the autofs timeout but only for >> this mount). >> >> To implement these options per-dentry expire timeouts need to be >> implemented for autofs indirect mounts. This is because all map keys >> (mounts) for autofs indirect mounts use an expire timeout stored in >> the autofs mount super block info. structure and all indirect mounts >> use the same expire timeout. >> >> Now I have a request to add the "nounmount" option so I need to add >> the per-dentry expire handling to the kernel implementation to do this. >> >> The implementation uses the trailing path component to identify the >> mount (and is also used as the autofs map key) which is passed in the >> autofs_dev_ioctl structure path field. The expire timeout is passed >> in autofs_dev_ioctl timeout field (well, of the timeout union). >> >> If the passed in timeout is equal to -1 the per-dentry timeout and >> flag are cleared providing for the "unmount" option. If the timeout >> is greater than or equal to 0 the timeout is set to the value and the >> flag is also set. If the dentry timeout is 0 the dentry will not expire >> by timeout which enables the implementation of the "nounmount" option >> for the specific mount. When the dentry timeout is greater than zero it >> allows for the implementation of the "utimeout=<seconds>" option. >> >> Signed-off-by: Ian Kent <raven@themaw.net> >> --- >> fs/autofs/autofs_i.h | 4 ++ >> fs/autofs/dev-ioctl.c | 97 ++++++++++++++++++++++++++++++++++-- >> fs/autofs/expire.c | 7 ++- >> fs/autofs/inode.c | 2 + >> include/uapi/linux/auto_fs.h | 2 +- >> 5 files changed, 104 insertions(+), 8 deletions(-) >> >> diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h >> index 8c1d587b3eef..77c7991d89aa 100644 >> --- a/fs/autofs/autofs_i.h >> +++ b/fs/autofs/autofs_i.h >> @@ -62,6 +62,7 @@ struct autofs_info { >> struct list_head expiring; >> >> struct autofs_sb_info *sbi; >> + unsigned long exp_timeout; >> unsigned long last_used; >> int count; >> >> @@ -81,6 +82,9 @@ struct autofs_info { >> */ >> #define AUTOFS_INF_PENDING (1<<2) /* dentry pending mount */ >> >> +#define AUTOFS_INF_EXPIRE_SET (1<<3) /* per-dentry expire timeout set for >> + this mount point. >> + */ >> struct autofs_wait_queue { >> wait_queue_head_t queue; >> struct autofs_wait_queue *next; >> diff --git a/fs/autofs/dev-ioctl.c b/fs/autofs/dev-ioctl.c >> index 5bf781ea6d67..f011e026358e 100644 >> --- a/fs/autofs/dev-ioctl.c >> +++ b/fs/autofs/dev-ioctl.c >> @@ -128,7 +128,13 @@ static int validate_dev_ioctl(int cmd, struct autofs_dev_ioctl *param) >> goto out; >> } >> >> + /* Setting the per-dentry expire timeout requires a trailing >> + * path component, ie. no '/', so invert the logic of the >> + * check_name() return for AUTOFS_DEV_IOCTL_TIMEOUT_CMD. >> + */ >> err = check_name(param->path); >> + if (cmd == AUTOFS_DEV_IOCTL_TIMEOUT_CMD) >> + err = err ? 0 : -EINVAL; >> if (err) { >> pr_warn("invalid path supplied for cmd(0x%08x)\n", >> cmd); >> @@ -396,16 +402,97 @@ static int autofs_dev_ioctl_catatonic(struct file *fp, >> return 0; >> } >> >> -/* Set the autofs mount timeout */ >> +/* >> + * Set the autofs mount expire timeout. >> + * >> + * There are two places an expire timeout can be set, in the autofs >> + * super block info. (this is all that's needed for direct and offset >> + * mounts because there's a distinct mount corresponding to each of >> + * these) and per-dentry within within the dentry info. If a per-dentry >> + * timeout is set it will override the expire timeout set in the parent >> + * autofs super block info. >> + * >> + * If setting the autofs super block expire timeout the autofs_dev_ioctl >> + * size field will be equal to the autofs_dev_ioctl structure size. If >> + * setting the per-dentry expire timeout the mount point name is passed >> + * in the autofs_dev_ioctl path field and the size field updated to >> + * reflect this. >> + * >> + * Setting the autofs mount expire timeout sets the timeout in the super >> + * block info. struct. Setting the per-dentry timeout does a little more. >> + * If the timeout is equal to -1 the per-dentry timeout (and flag) is >> + * cleared which reverts to using the super block timeout, otherwise if >> + * timeout is 0 the timeout is set to this value and the flag is left >> + * set which disables expiration for the mount point, lastly the flag >> + * and the timeout are set enabling the dentry to use this timeout. >> + */ >> static int autofs_dev_ioctl_timeout(struct file *fp, >> struct autofs_sb_info *sbi, >> struct autofs_dev_ioctl *param) >> { >> - unsigned long timeout; >> + unsigned long timeout = param->timeout.timeout; >> + >> + /* If setting the expire timeout for an individual indirect >> + * mount point dentry the mount trailing component path is >> + * placed in param->path and param->size adjusted to account >> + * for it otherwise param->size it is set to the structure >> + * size. >> + */ >> + if (param->size == AUTOFS_DEV_IOCTL_SIZE) { >> + param->timeout.timeout = sbi->exp_timeout / HZ; >> + sbi->exp_timeout = timeout * HZ; >> + } else { >> + struct dentry *base = fp->f_path.dentry; >> + struct inode *inode = base->d_inode; >> + int path_len = param->size - AUTOFS_DEV_IOCTL_SIZE - 1; >> + struct dentry *dentry; >> + struct autofs_info *ino; >> + >> + if (!autofs_type_indirect(sbi->type)) >> + return -EINVAL; >> + >> + /* An expire timeout greater than the superblock timeout >> + * could be a problem at shutdown but the super block >> + * timeout itself can change so all we can really do is >> + * warn the user. >> + */ >> + if (timeout >= sbi->exp_timeout) >> + pr_warn("per-mount expire timeout is greater than " >> + "the parent autofs mount timeout which could " >> + "prevent shutdown\n"); > Wouldn't it be possible to just record the lowest known per-dentry > timeout in idk sbi->exp_lower_bound and reject sbi->exp_timeout changes > that go below that? Not sure I understand what your saying here. The (amd) auto-mounted mounts are each meant to be able to expire independently, according to the timeout set for each of them rather than use the global timeout set in the autofs (parent) mount. But your comment is useful because I do use a polling mechanism in user space to check for expiration and the frequency of checking becomes a problem when I introduce this. Setting a lower bound may make the frequency to short (for very large directories) and I think there will be difficulties working out when the frequency needs to be reduced after changes. But these are problems that I think can be left to user space, not sure yet. For now it will be adequate to go with the default frequency based on the parent mount as it is now (ie. a check frequency of one quarter of the expire global timeout). Ian
On 15/8/24 11:10, Ian Kent wrote: > On 14/8/24 22:39, Christian Brauner wrote: >> On Wed, Aug 14, 2024 at 05:02:31PM GMT, Ian Kent wrote: >>> Add ability to set per-dentry mount expire timeout to autofs. >>> >>> There are two fairly well known automounter map formats, the autofs >>> format and the amd format (more or less System V and Berkley). >>> >>> Some time ago Linux autofs added an amd map format parser that >>> implemented a fair amount of the amd functionality. This was done >>> within the autofs infrastructure and some functionality wasn't >>> implemented because it either didn't make sense or required extra >>> kernel changes. The idea was to restrict changes to be within the >>> existing autofs functionality as much as possible and leave changes >>> with a wider scope to be considered later. >>> >>> One of these changes is implementing the amd options: >>> 1) "unmount", expire this mount according to a timeout (same as the >>> current autofs default). >>> 2) "nounmount", don't expire this mount (same as setting the autofs >>> timeout to 0 except only for this specific mount) . >>> 3) "utimeout=<seconds>", expire this mount using the specified >>> timeout (again same as setting the autofs timeout but only for >>> this mount). >>> >>> To implement these options per-dentry expire timeouts need to be >>> implemented for autofs indirect mounts. This is because all map keys >>> (mounts) for autofs indirect mounts use an expire timeout stored in >>> the autofs mount super block info. structure and all indirect mounts >>> use the same expire timeout. >>> >>> Now I have a request to add the "nounmount" option so I need to add >>> the per-dentry expire handling to the kernel implementation to do this. >>> >>> The implementation uses the trailing path component to identify the >>> mount (and is also used as the autofs map key) which is passed in the >>> autofs_dev_ioctl structure path field. The expire timeout is passed >>> in autofs_dev_ioctl timeout field (well, of the timeout union). >>> >>> If the passed in timeout is equal to -1 the per-dentry timeout and >>> flag are cleared providing for the "unmount" option. If the timeout >>> is greater than or equal to 0 the timeout is set to the value and the >>> flag is also set. If the dentry timeout is 0 the dentry will not expire >>> by timeout which enables the implementation of the "nounmount" option >>> for the specific mount. When the dentry timeout is greater than zero it >>> allows for the implementation of the "utimeout=<seconds>" option. >>> >>> Signed-off-by: Ian Kent <raven@themaw.net> >>> --- >>> fs/autofs/autofs_i.h | 4 ++ >>> fs/autofs/dev-ioctl.c | 97 >>> ++++++++++++++++++++++++++++++++++-- >>> fs/autofs/expire.c | 7 ++- >>> fs/autofs/inode.c | 2 + >>> include/uapi/linux/auto_fs.h | 2 +- >>> 5 files changed, 104 insertions(+), 8 deletions(-) >>> >>> diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h >>> index 8c1d587b3eef..77c7991d89aa 100644 >>> --- a/fs/autofs/autofs_i.h >>> +++ b/fs/autofs/autofs_i.h >>> @@ -62,6 +62,7 @@ struct autofs_info { >>> struct list_head expiring; >>> struct autofs_sb_info *sbi; >>> + unsigned long exp_timeout; >>> unsigned long last_used; >>> int count; >>> @@ -81,6 +82,9 @@ struct autofs_info { >>> */ >>> #define AUTOFS_INF_PENDING (1<<2) /* dentry pending mount */ >>> +#define AUTOFS_INF_EXPIRE_SET (1<<3) /* per-dentry expire >>> timeout set for >>> + this mount point. >>> + */ >>> struct autofs_wait_queue { >>> wait_queue_head_t queue; >>> struct autofs_wait_queue *next; >>> diff --git a/fs/autofs/dev-ioctl.c b/fs/autofs/dev-ioctl.c >>> index 5bf781ea6d67..f011e026358e 100644 >>> --- a/fs/autofs/dev-ioctl.c >>> +++ b/fs/autofs/dev-ioctl.c >>> @@ -128,7 +128,13 @@ static int validate_dev_ioctl(int cmd, struct >>> autofs_dev_ioctl *param) >>> goto out; >>> } >>> + /* Setting the per-dentry expire timeout requires a trailing >>> + * path component, ie. no '/', so invert the logic of the >>> + * check_name() return for AUTOFS_DEV_IOCTL_TIMEOUT_CMD. >>> + */ >>> err = check_name(param->path); >>> + if (cmd == AUTOFS_DEV_IOCTL_TIMEOUT_CMD) >>> + err = err ? 0 : -EINVAL; >>> if (err) { >>> pr_warn("invalid path supplied for cmd(0x%08x)\n", >>> cmd); >>> @@ -396,16 +402,97 @@ static int autofs_dev_ioctl_catatonic(struct >>> file *fp, >>> return 0; >>> } >>> -/* Set the autofs mount timeout */ >>> +/* >>> + * Set the autofs mount expire timeout. >>> + * >>> + * There are two places an expire timeout can be set, in the autofs >>> + * super block info. (this is all that's needed for direct and offset >>> + * mounts because there's a distinct mount corresponding to each of >>> + * these) and per-dentry within within the dentry info. If a >>> per-dentry >>> + * timeout is set it will override the expire timeout set in the >>> parent >>> + * autofs super block info. >>> + * >>> + * If setting the autofs super block expire timeout the >>> autofs_dev_ioctl >>> + * size field will be equal to the autofs_dev_ioctl structure size. If >>> + * setting the per-dentry expire timeout the mount point name is >>> passed >>> + * in the autofs_dev_ioctl path field and the size field updated to >>> + * reflect this. >>> + * >>> + * Setting the autofs mount expire timeout sets the timeout in the >>> super >>> + * block info. struct. Setting the per-dentry timeout does a little >>> more. >>> + * If the timeout is equal to -1 the per-dentry timeout (and flag) is >>> + * cleared which reverts to using the super block timeout, >>> otherwise if >>> + * timeout is 0 the timeout is set to this value and the flag is left >>> + * set which disables expiration for the mount point, lastly the flag >>> + * and the timeout are set enabling the dentry to use this timeout. >>> + */ >>> static int autofs_dev_ioctl_timeout(struct file *fp, >>> struct autofs_sb_info *sbi, >>> struct autofs_dev_ioctl *param) >>> { >>> - unsigned long timeout; >>> + unsigned long timeout = param->timeout.timeout; >>> + >>> + /* If setting the expire timeout for an individual indirect >>> + * mount point dentry the mount trailing component path is >>> + * placed in param->path and param->size adjusted to account >>> + * for it otherwise param->size it is set to the structure >>> + * size. >>> + */ >>> + if (param->size == AUTOFS_DEV_IOCTL_SIZE) { >>> + param->timeout.timeout = sbi->exp_timeout / HZ; >>> + sbi->exp_timeout = timeout * HZ; >>> + } else { >>> + struct dentry *base = fp->f_path.dentry; >>> + struct inode *inode = base->d_inode; >>> + int path_len = param->size - AUTOFS_DEV_IOCTL_SIZE - 1; >>> + struct dentry *dentry; >>> + struct autofs_info *ino; >>> + >>> + if (!autofs_type_indirect(sbi->type)) >>> + return -EINVAL; >>> + >>> + /* An expire timeout greater than the superblock timeout >>> + * could be a problem at shutdown but the super block >>> + * timeout itself can change so all we can really do is >>> + * warn the user. >>> + */ >>> + if (timeout >= sbi->exp_timeout) >>> + pr_warn("per-mount expire timeout is greater than " >>> + "the parent autofs mount timeout which could " >>> + "prevent shutdown\n"); >> Wouldn't it be possible to just record the lowest known per-dentry >> timeout in idk sbi->exp_lower_bound and reject sbi->exp_timeout changes >> that go below that? > > Not sure I understand what your saying here. Having re-read what you said I think I understand what your saying now. I was mislead by the lower bound wording, the warning talks about a per-dentry timeout setting that's longer than the timeout set for this indirect autofs mount. TBH I'm not even sure this warning is worthwhile since most user space builds should just leave in-use mounts mounted at exit and automount should just re-instate them at startup and continue on as though nothing had happened. But I thought it worth a warning so that users know the timeouts they are using are longer than the parent (global for these indirect mounts) timeout. I can certainly get rid of it if you prefer. Ian > > > The (amd) auto-mounted mounts are each meant to be able to expire > independently, > > according to the timeout set for each of them rather than use the > global timeout set > > in the autofs (parent) mount. > > > But your comment is useful because I do use a polling mechanism in > user space to check > > for expiration and the frequency of checking becomes a problem when I > introduce this. > > Setting a lower bound may make the frequency to short (for very large > directories) and > > I think there will be difficulties working out when the frequency > needs to be reduced > > after changes. But these are problems that I think can be left to user > space, not sure > > yet. For now it will be adequate to go with the default frequency > based on the parent > > mount as it is now (ie. a check frequency of one quarter of the expire > global timeout). > > > Ian > >
On Wed, 14 Aug 2024 17:02:31 +0800, Ian Kent wrote: > Add ability to set per-dentry mount expire timeout to autofs. > > There are two fairly well known automounter map formats, the autofs > format and the amd format (more or less System V and Berkley). > > Some time ago Linux autofs added an amd map format parser that > implemented a fair amount of the amd functionality. This was done > within the autofs infrastructure and some functionality wasn't > implemented because it either didn't make sense or required extra > kernel changes. The idea was to restrict changes to be within the > existing autofs functionality as much as possible and leave changes > with a wider scope to be considered later. > > [...] Applied to the vfs.misc branch of the vfs/vfs.git tree. Patches in the vfs.misc branch should appear in linux-next soon. Please report any outstanding bugs that were missed during review in a new review to the original patch series allowing us to drop it. It's encouraged to provide Acked-bys and Reviewed-bys even though the patch has now been applied. If possible patch trailers will be updated. Note that commit hashes shown below are subject to change due to rebase, trailer updates or similar. If in doubt, please check the listed branch. tree: https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git branch: vfs.misc [1/1] autofs: add per dentry expire timeout https://git.kernel.org/vfs/vfs/c/70089655316e
diff --git a/fs/autofs/autofs_i.h b/fs/autofs/autofs_i.h index 8c1d587b3eef..77c7991d89aa 100644 --- a/fs/autofs/autofs_i.h +++ b/fs/autofs/autofs_i.h @@ -62,6 +62,7 @@ struct autofs_info { struct list_head expiring; struct autofs_sb_info *sbi; + unsigned long exp_timeout; unsigned long last_used; int count; @@ -81,6 +82,9 @@ struct autofs_info { */ #define AUTOFS_INF_PENDING (1<<2) /* dentry pending mount */ +#define AUTOFS_INF_EXPIRE_SET (1<<3) /* per-dentry expire timeout set for + this mount point. + */ struct autofs_wait_queue { wait_queue_head_t queue; struct autofs_wait_queue *next; diff --git a/fs/autofs/dev-ioctl.c b/fs/autofs/dev-ioctl.c index 5bf781ea6d67..f011e026358e 100644 --- a/fs/autofs/dev-ioctl.c +++ b/fs/autofs/dev-ioctl.c @@ -128,7 +128,13 @@ static int validate_dev_ioctl(int cmd, struct autofs_dev_ioctl *param) goto out; } + /* Setting the per-dentry expire timeout requires a trailing + * path component, ie. no '/', so invert the logic of the + * check_name() return for AUTOFS_DEV_IOCTL_TIMEOUT_CMD. + */ err = check_name(param->path); + if (cmd == AUTOFS_DEV_IOCTL_TIMEOUT_CMD) + err = err ? 0 : -EINVAL; if (err) { pr_warn("invalid path supplied for cmd(0x%08x)\n", cmd); @@ -396,16 +402,97 @@ static int autofs_dev_ioctl_catatonic(struct file *fp, return 0; } -/* Set the autofs mount timeout */ +/* + * Set the autofs mount expire timeout. + * + * There are two places an expire timeout can be set, in the autofs + * super block info. (this is all that's needed for direct and offset + * mounts because there's a distinct mount corresponding to each of + * these) and per-dentry within within the dentry info. If a per-dentry + * timeout is set it will override the expire timeout set in the parent + * autofs super block info. + * + * If setting the autofs super block expire timeout the autofs_dev_ioctl + * size field will be equal to the autofs_dev_ioctl structure size. If + * setting the per-dentry expire timeout the mount point name is passed + * in the autofs_dev_ioctl path field and the size field updated to + * reflect this. + * + * Setting the autofs mount expire timeout sets the timeout in the super + * block info. struct. Setting the per-dentry timeout does a little more. + * If the timeout is equal to -1 the per-dentry timeout (and flag) is + * cleared which reverts to using the super block timeout, otherwise if + * timeout is 0 the timeout is set to this value and the flag is left + * set which disables expiration for the mount point, lastly the flag + * and the timeout are set enabling the dentry to use this timeout. + */ static int autofs_dev_ioctl_timeout(struct file *fp, struct autofs_sb_info *sbi, struct autofs_dev_ioctl *param) { - unsigned long timeout; + unsigned long timeout = param->timeout.timeout; + + /* If setting the expire timeout for an individual indirect + * mount point dentry the mount trailing component path is + * placed in param->path and param->size adjusted to account + * for it otherwise param->size it is set to the structure + * size. + */ + if (param->size == AUTOFS_DEV_IOCTL_SIZE) { + param->timeout.timeout = sbi->exp_timeout / HZ; + sbi->exp_timeout = timeout * HZ; + } else { + struct dentry *base = fp->f_path.dentry; + struct inode *inode = base->d_inode; + int path_len = param->size - AUTOFS_DEV_IOCTL_SIZE - 1; + struct dentry *dentry; + struct autofs_info *ino; + + if (!autofs_type_indirect(sbi->type)) + return -EINVAL; + + /* An expire timeout greater than the superblock timeout + * could be a problem at shutdown but the super block + * timeout itself can change so all we can really do is + * warn the user. + */ + if (timeout >= sbi->exp_timeout) + pr_warn("per-mount expire timeout is greater than " + "the parent autofs mount timeout which could " + "prevent shutdown\n"); + + inode_lock_shared(inode); + dentry = try_lookup_one_len(param->path, base, path_len); + inode_unlock_shared(inode); + if (IS_ERR_OR_NULL(dentry)) + return dentry ? PTR_ERR(dentry) : -ENOENT; + ino = autofs_dentry_ino(dentry); + if (!ino) { + dput(dentry); + return -ENOENT; + } + + if (ino->exp_timeout && ino->flags & AUTOFS_INF_EXPIRE_SET) + param->timeout.timeout = ino->exp_timeout / HZ; + else + param->timeout.timeout = sbi->exp_timeout / HZ; + + if (timeout == -1) { + /* Revert to using the super block timeout */ + ino->flags &= ~AUTOFS_INF_EXPIRE_SET; + ino->exp_timeout = 0; + } else { + /* Set the dentry expire flag and timeout. + * + * If timeout is 0 it will prevent the expire + * of this particular automount. + */ + ino->flags |= AUTOFS_INF_EXPIRE_SET; + ino->exp_timeout = timeout * HZ; + } + dput(dentry); + } - timeout = param->timeout.timeout; - param->timeout.timeout = sbi->exp_timeout / HZ; - sbi->exp_timeout = timeout * HZ; return 0; } diff --git a/fs/autofs/expire.c b/fs/autofs/expire.c index 39d8c84c16f4..5c2d459e1e48 100644 --- a/fs/autofs/expire.c +++ b/fs/autofs/expire.c @@ -429,8 +429,6 @@ static struct dentry *autofs_expire_indirect(struct super_block *sb, if (!root) return NULL; - timeout = sbi->exp_timeout; - dentry = NULL; while ((dentry = get_next_positive_subdir(dentry, root))) { spin_lock(&sbi->fs_lock); @@ -441,6 +439,11 @@ static struct dentry *autofs_expire_indirect(struct super_block *sb, } spin_unlock(&sbi->fs_lock); + if (ino->flags & AUTOFS_INF_EXPIRE_SET) + timeout = ino->exp_timeout; + else + timeout = sbi->exp_timeout; + expired = should_expire(dentry, mnt, timeout, how); if (!expired) continue; diff --git a/fs/autofs/inode.c b/fs/autofs/inode.c index cf792d4de4f1..068f273757bf 100644 --- a/fs/autofs/inode.c +++ b/fs/autofs/inode.c @@ -19,6 +19,7 @@ struct autofs_info *autofs_new_ino(struct autofs_sb_info *sbi) INIT_LIST_HEAD(&ino->expiring); ino->last_used = jiffies; ino->sbi = sbi; + ino->exp_timeout = -1; ino->count = 1; } return ino; @@ -28,6 +29,7 @@ void autofs_clean_ino(struct autofs_info *ino) { ino->uid = GLOBAL_ROOT_UID; ino->gid = GLOBAL_ROOT_GID; + ino->exp_timeout = -1; ino->last_used = jiffies; } diff --git a/include/uapi/linux/auto_fs.h b/include/uapi/linux/auto_fs.h index 1f7925afad2d..8081df849743 100644 --- a/include/uapi/linux/auto_fs.h +++ b/include/uapi/linux/auto_fs.h @@ -23,7 +23,7 @@ #define AUTOFS_MIN_PROTO_VERSION 3 #define AUTOFS_MAX_PROTO_VERSION 5 -#define AUTOFS_PROTO_SUBVERSION 5 +#define AUTOFS_PROTO_SUBVERSION 6 /* * The wait_queue_token (autofs_wqt_t) is part of a structure which is passed
Add ability to set per-dentry mount expire timeout to autofs. There are two fairly well known automounter map formats, the autofs format and the amd format (more or less System V and Berkley). Some time ago Linux autofs added an amd map format parser that implemented a fair amount of the amd functionality. This was done within the autofs infrastructure and some functionality wasn't implemented because it either didn't make sense or required extra kernel changes. The idea was to restrict changes to be within the existing autofs functionality as much as possible and leave changes with a wider scope to be considered later. One of these changes is implementing the amd options: 1) "unmount", expire this mount according to a timeout (same as the current autofs default). 2) "nounmount", don't expire this mount (same as setting the autofs timeout to 0 except only for this specific mount) . 3) "utimeout=<seconds>", expire this mount using the specified timeout (again same as setting the autofs timeout but only for this mount). To implement these options per-dentry expire timeouts need to be implemented for autofs indirect mounts. This is because all map keys (mounts) for autofs indirect mounts use an expire timeout stored in the autofs mount super block info. structure and all indirect mounts use the same expire timeout. Now I have a request to add the "nounmount" option so I need to add the per-dentry expire handling to the kernel implementation to do this. The implementation uses the trailing path component to identify the mount (and is also used as the autofs map key) which is passed in the autofs_dev_ioctl structure path field. The expire timeout is passed in autofs_dev_ioctl timeout field (well, of the timeout union). If the passed in timeout is equal to -1 the per-dentry timeout and flag are cleared providing for the "unmount" option. If the timeout is greater than or equal to 0 the timeout is set to the value and the flag is also set. If the dentry timeout is 0 the dentry will not expire by timeout which enables the implementation of the "nounmount" option for the specific mount. When the dentry timeout is greater than zero it allows for the implementation of the "utimeout=<seconds>" option. Signed-off-by: Ian Kent <raven@themaw.net> --- fs/autofs/autofs_i.h | 4 ++ fs/autofs/dev-ioctl.c | 97 ++++++++++++++++++++++++++++++++++-- fs/autofs/expire.c | 7 ++- fs/autofs/inode.c | 2 + include/uapi/linux/auto_fs.h | 2 +- 5 files changed, 104 insertions(+), 8 deletions(-)