1.. SPDX-License-Identifier: GPL-2.0
2
3====================================================================
4Miscellaneous Device control operations for the autofs kernel module
5====================================================================
6
7The problem
8===========
9
10There is a problem with active restarts in autofs (that is to say
11restarting autofs when there are busy mounts).
12
13During normal operation autofs uses a file descriptor opened on the
14directory that is being managed in order to be able to issue control
15operations. Using a file descriptor gives ioctl operations access to
16autofs specific information stored in the super block. The operations
17are things such as setting an autofs mount catatonic, setting the
18expire timeout and requesting expire checks. As is explained below,
19certain types of autofs triggered mounts can end up covering an autofs
20mount itself which prevents us being able to use open(2) to obtain a
21file descriptor for these operations if we don't already have one open.
22
23Currently autofs uses "umount -l" (lazy umount) to clear active mounts
24at restart. While using lazy umount works for most cases, anything that
25needs to walk back up the mount tree to construct a path, such as
26getcwd(2) and the proc file system /proc/<pid>/cwd, no longer works
27because the point from which the path is constructed has been detached
28from the mount tree.
29
30The actual problem with autofs is that it can't reconnect to existing
31mounts. Immediately one thinks of just adding the ability to remount
32autofs file systems would solve it, but alas, that can't work. This is
33because autofs direct mounts and the implementation of "on demand mount
34and expire" of nested mount trees have the file system mounted directly
35on top of the mount trigger directory dentry.
36
37For example, there are two types of automount maps, direct (in the kernel
38module source you will see a third type called an offset, which is just
39a direct mount in disguise) and indirect.
40
41Here is a master map with direct and indirect map entries::
42
43    /-      /etc/auto.direct
44    /test   /etc/auto.indirect
45
46and the corresponding map files::
47
48    /etc/auto.direct:
49
50    /automount/dparse/g6  budgie:/autofs/export1
51    /automount/dparse/g1  shark:/autofs/export1
52    and so on.
53
54/etc/auto.indirect::
55
56    g1    shark:/autofs/export1
57    g6    budgie:/autofs/export1
58    and so on.
59
60For the above indirect map an autofs file system is mounted on /test and
61mounts are triggered for each sub-directory key by the inode lookup
62operation. So we see a mount of shark:/autofs/export1 on /test/g1, for
63example.
64
65The way that direct mounts are handled is by making an autofs mount on
66each full path, such as /automount/dparse/g1, and using it as a mount
67trigger. So when we walk on the path we mount shark:/autofs/export1 "on
68top of this mount point". Since these are always directories we can
69use the follow_link inode operation to trigger the mount.
70
71But, each entry in direct and indirect maps can have offsets (making
72them multi-mount map entries).
73
74For example, an indirect mount map entry could also be::
75
76    g1  \
77    /        shark:/autofs/export5/testing/test \
78    /s1      shark:/autofs/export/testing/test/s1 \
79    /s2      shark:/autofs/export5/testing/test/s2 \
80    /s1/ss1  shark:/autofs/export1 \
81    /s2/ss2  shark:/autofs/export2
82
83and a similarly a direct mount map entry could also be::
84
85    /automount/dparse/g1 \
86	/       shark:/autofs/export5/testing/test \
87	/s1     shark:/autofs/export/testing/test/s1 \
88	/s2     shark:/autofs/export5/testing/test/s2 \
89	/s1/ss1 shark:/autofs/export2 \
90	/s2/ss2 shark:/autofs/export2
91
92One of the issues with version 4 of autofs was that, when mounting an
93entry with a large number of offsets, possibly with nesting, we needed
94to mount and umount all of the offsets as a single unit. Not really a
95problem, except for people with a large number of offsets in map entries.
96This mechanism is used for the well known "hosts" map and we have seen
97cases (in 2.4) where the available number of mounts are exhausted or
98where the number of privileged ports available is exhausted.
99
100In version 5 we mount only as we go down the tree of offsets and
101similarly for expiring them which resolves the above problem. There is
102somewhat more detail to the implementation but it isn't needed for the
103sake of the problem explanation. The one important detail is that these
104offsets are implemented using the same mechanism as the direct mounts
105above and so the mount points can be covered by a mount.
106
107The current autofs implementation uses an ioctl file descriptor opened
108on the mount point for control operations. The references held by the
109descriptor are accounted for in checks made to determine if a mount is
110in use and is also used to access autofs file system information held
111in the mount super block. So the use of a file handle needs to be
112retained.
113
114
115The Solution
116============
117
118To be able to restart autofs leaving existing direct, indirect and
119offset mounts in place we need to be able to obtain a file handle
120for these potentially covered autofs mount points. Rather than just
121implement an isolated operation it was decided to re-implement the
122existing ioctl interface and add new operations to provide this
123functionality.
124
125In addition, to be able to reconstruct a mount tree that has busy mounts,
126the uid and gid of the last user that triggered the mount needs to be
127available because these can be used as macro substitution variables in
128autofs maps. They are recorded at mount request time and an operation
129has been added to retrieve them.
130
131Since we're re-implementing the control interface, a couple of other
132problems with the existing interface have been addressed. First, when
133a mount or expire operation completes a status is returned to the
134kernel by either a "send ready" or a "send fail" operation. The
135"send fail" operation of the ioctl interface could only ever send
136ENOENT so the re-implementation allows user space to send an actual
137status. Another expensive operation in user space, for those using
138very large maps, is discovering if a mount is present. Usually this
139involves scanning /proc/mounts and since it needs to be done quite
140often it can introduce significant overhead when there are many entries
141in the mount table. An operation to lookup the mount status of a mount
142point dentry (covered or not) has also been added.
143
144Current kernel development policy recommends avoiding the use of the
145ioctl mechanism in favor of systems such as Netlink. An implementation
146using this system was attempted to evaluate its suitability and it was
147found to be inadequate, in this case. The Generic Netlink system was
148used for this as raw Netlink would lead to a significant increase in
149complexity. There's no question that the Generic Netlink system is an
150elegant solution for common case ioctl functions but it's not a complete
151replacement probably because its primary purpose in life is to be a
152message bus implementation rather than specifically an ioctl replacement.
153While it would be possible to work around this there is one concern
154that lead to the decision to not use it. This is that the autofs
155expire in the daemon has become far to complex because umount
156candidates are enumerated, almost for no other reason than to "count"
157the number of times to call the expire ioctl. This involves scanning
158the mount table which has proved to be a big overhead for users with
159large maps. The best way to improve this is try and get back to the
160way the expire was done long ago. That is, when an expire request is
161issued for a mount (file handle) we should continually call back to
162the daemon until we can't umount any more mounts, then return the
163appropriate status to the daemon. At the moment we just expire one
164mount at a time. A Generic Netlink implementation would exclude this
165possibility for future development due to the requirements of the
166message bus architecture.
167
168
169autofs Miscellaneous Device mount control interface
170====================================================
171
172The control interface is opening a device node, typically /dev/autofs.
173
174All the ioctls use a common structure to pass the needed parameter
175information and return operation results::
176
177    struct autofs_dev_ioctl {
178	    __u32 ver_major;
179	    __u32 ver_minor;
180	    __u32 size;             /* total size of data passed in
181				    * including this struct */
182	    __s32 ioctlfd;          /* automount command fd */
183
184	    /* Command parameters */
185	    union {
186		    struct args_protover		protover;
187		    struct args_protosubver		protosubver;
188		    struct args_openmount		openmount;
189		    struct args_ready		ready;
190		    struct args_fail		fail;
191		    struct args_setpipefd		setpipefd;
192		    struct args_timeout		timeout;
193		    struct args_requester		requester;
194		    struct args_expire		expire;
195		    struct args_askumount		askumount;
196		    struct args_ismountpoint	ismountpoint;
197	    };
198
199	    char path[0];
200    };
201
202The ioctlfd field is a mount point file descriptor of an autofs mount
203point. It is returned by the open call and is used by all calls except
204the check for whether a given path is a mount point, where it may
205optionally be used to check a specific mount corresponding to a given
206mount point file descriptor, and when requesting the uid and gid of the
207last successful mount on a directory within the autofs file system.
208
209The union is used to communicate parameters and results of calls made
210as described below.
211
212The path field is used to pass a path where it is needed and the size field
213is used account for the increased structure length when translating the
214structure sent from user space.
215
216This structure can be initialized before setting specific fields by using
217the void function call init_autofs_dev_ioctl(``struct autofs_dev_ioctl *``).
218
219All of the ioctls perform a copy of this structure from user space to
220kernel space and return -EINVAL if the size parameter is smaller than
221the structure size itself, -ENOMEM if the kernel memory allocation fails
222or -EFAULT if the copy itself fails. Other checks include a version check
223of the compiled in user space version against the module version and a
224mismatch results in a -EINVAL return. If the size field is greater than
225the structure size then a path is assumed to be present and is checked to
226ensure it begins with a "/" and is NULL terminated, otherwise -EINVAL is
227returned. Following these checks, for all ioctl commands except
228AUTOFS_DEV_IOCTL_VERSION_CMD, AUTOFS_DEV_IOCTL_OPENMOUNT_CMD and
229AUTOFS_DEV_IOCTL_CLOSEMOUNT_CMD the ioctlfd is validated and if it is
230not a valid descriptor or doesn't correspond to an autofs mount point
231an error of -EBADF, -ENOTTY or -EINVAL (not an autofs descriptor) is
232returned.
233
234
235The ioctls
236==========
237
238An example of an implementation which uses this interface can be seen
239in autofs version 5.0.4 and later in file lib/dev-ioctl-lib.c of the
240distribution tar available for download from kernel.org in directory
241/pub/linux/daemons/autofs/v5.
242
243The device node ioctl operations implemented by this interface are:
244
245
246AUTOFS_DEV_IOCTL_VERSION
247------------------------
248
249Get the major and minor version of the autofs device ioctl kernel module
250implementation. It requires an initialized struct autofs_dev_ioctl as an
251input parameter and sets the version information in the passed in structure.
252It returns 0 on success or the error -EINVAL if a version mismatch is
253detected.
254
255
256AUTOFS_DEV_IOCTL_PROTOVER_CMD and AUTOFS_DEV_IOCTL_PROTOSUBVER_CMD
257------------------------------------------------------------------
258
259Get the major and minor version of the autofs protocol version understood
260by loaded module. This call requires an initialized struct autofs_dev_ioctl
261with the ioctlfd field set to a valid autofs mount point descriptor
262and sets the requested version number in version field of struct args_protover
263or sub_version field of struct args_protosubver. These commands return
2640 on success or one of the negative error codes if validation fails.
265
266
267AUTOFS_DEV_IOCTL_OPENMOUNT and AUTOFS_DEV_IOCTL_CLOSEMOUNT
268----------------------------------------------------------
269
270Obtain and release a file descriptor for an autofs managed mount point
271path. The open call requires an initialized struct autofs_dev_ioctl with
272the path field set and the size field adjusted appropriately as well
273as the devid field of struct args_openmount set to the device number of
274the autofs mount. The device number can be obtained from the mount options
275shown in /proc/mounts. The close call requires an initialized struct
276autofs_dev_ioct with the ioctlfd field set to the descriptor obtained
277from the open call. The release of the file descriptor can also be done
278with close(2) so any open descriptors will also be closed at process exit.
279The close call is included in the implemented operations largely for
280completeness and to provide for a consistent user space implementation.
281
282
283AUTOFS_DEV_IOCTL_READY_CMD and AUTOFS_DEV_IOCTL_FAIL_CMD
284--------------------------------------------------------
285
286Return mount and expire result status from user space to the kernel.
287Both of these calls require an initialized struct autofs_dev_ioctl
288with the ioctlfd field set to the descriptor obtained from the open
289call and the token field of struct args_ready or struct args_fail set
290to the wait queue token number, received by user space in the foregoing
291mount or expire request. The status field of struct args_fail is set to
292the errno of the operation. It is set to 0 on success.
293
294
295AUTOFS_DEV_IOCTL_SETPIPEFD_CMD
296------------------------------
297
298Set the pipe file descriptor used for kernel communication to the daemon.
299Normally this is set at mount time using an option but when reconnecting
300to a existing mount we need to use this to tell the autofs mount about
301the new kernel pipe descriptor. In order to protect mounts against
302incorrectly setting the pipe descriptor we also require that the autofs
303mount be catatonic (see next call).
304
305The call requires an initialized struct autofs_dev_ioctl with the
306ioctlfd field set to the descriptor obtained from the open call and
307the pipefd field of struct args_setpipefd set to descriptor of the pipe.
308On success the call also sets the process group id used to identify the
309controlling process (eg. the owning automount(8) daemon) to the process
310group of the caller.
311
312
313AUTOFS_DEV_IOCTL_CATATONIC_CMD
314------------------------------
315
316Make the autofs mount point catatonic. The autofs mount will no longer
317issue mount requests, the kernel communication pipe descriptor is released
318and any remaining waits in the queue released.
319
320The call requires an initialized struct autofs_dev_ioctl with the
321ioctlfd field set to the descriptor obtained from the open call.
322
323
324AUTOFS_DEV_IOCTL_TIMEOUT_CMD
325----------------------------
326
327Set the expire timeout for mounts within an autofs mount point.
328
329The call requires an initialized struct autofs_dev_ioctl with the
330ioctlfd field set to the descriptor obtained from the open call.
331
332
333AUTOFS_DEV_IOCTL_REQUESTER_CMD
334------------------------------
335
336Return the uid and gid of the last process to successfully trigger a the
337mount on the given path dentry.
338
339The call requires an initialized struct autofs_dev_ioctl with the path
340field set to the mount point in question and the size field adjusted
341appropriately. Upon return the uid field of struct args_requester contains
342the uid and gid field the gid.
343
344When reconstructing an autofs mount tree with active mounts we need to
345re-connect to mounts that may have used the original process uid and
346gid (or string variations of them) for mount lookups within the map entry.
347This call provides the ability to obtain this uid and gid so they may be
348used by user space for the mount map lookups.
349
350
351AUTOFS_DEV_IOCTL_EXPIRE_CMD
352---------------------------
353
354Issue an expire request to the kernel for an autofs mount. Typically
355this ioctl is called until no further expire candidates are found.
356
357The call requires an initialized struct autofs_dev_ioctl with the
358ioctlfd field set to the descriptor obtained from the open call. In
359addition an immediate expire that's independent of the mount timeout,
360and a forced expire that's independent of whether the mount is busy,
361can be requested by setting the how field of struct args_expire to
362AUTOFS_EXP_IMMEDIATE or AUTOFS_EXP_FORCED, respectively . If no
363expire candidates can be found the ioctl returns -1 with errno set to
364EAGAIN.
365
366This call causes the kernel module to check the mount corresponding
367to the given ioctlfd for mounts that can be expired, issues an expire
368request back to the daemon and waits for completion.
369
370AUTOFS_DEV_IOCTL_ASKUMOUNT_CMD
371------------------------------
372
373Checks if an autofs mount point is in use.
374
375The call requires an initialized struct autofs_dev_ioctl with the
376ioctlfd field set to the descriptor obtained from the open call and
377it returns the result in the may_umount field of struct args_askumount,
3781 for busy and 0 otherwise.
379
380
381AUTOFS_DEV_IOCTL_ISMOUNTPOINT_CMD
382---------------------------------
383
384Check if the given path is a mountpoint.
385
386The call requires an initialized struct autofs_dev_ioctl. There are two
387possible variations. Both use the path field set to the path of the mount
388point to check and the size field adjusted appropriately. One uses the
389ioctlfd field to identify a specific mount point to check while the other
390variation uses the path and optionally in.type field of struct args_ismountpoint
391set to an autofs mount type. The call returns 1 if this is a mount point
392and sets out.devid field to the device number of the mount and out.magic
393field to the relevant super block magic number (described below) or 0 if
394it isn't a mountpoint. In both cases the device number (as returned
395by new_encode_dev()) is returned in out.devid field.
396
397If supplied with a file descriptor we're looking for a specific mount,
398not necessarily at the top of the mounted stack. In this case the path
399the descriptor corresponds to is considered a mountpoint if it is itself
400a mountpoint or contains a mount, such as a multi-mount without a root
401mount. In this case we return 1 if the descriptor corresponds to a mount
402point and also returns the super magic of the covering mount if there
403is one or 0 if it isn't a mountpoint.
404
405If a path is supplied (and the ioctlfd field is set to -1) then the path
406is looked up and is checked to see if it is the root of a mount. If a
407type is also given we are looking for a particular autofs mount and if
408a match isn't found a fail is returned. If the located path is the
409root of a mount 1 is returned along with the super magic of the mount
410or 0 otherwise.
411