brintos

brintos / linux-shallow public Read only

0
0
Text · 20.7 KiB · 357328d Raw
565 lines · plain
1====================2Credentials in Linux3====================4 5By: David Howells <dhowells@redhat.com>6 7.. contents:: :local:8 9Overview10========11 12There are several parts to the security check performed by Linux when one13object acts upon another:14 15 1. Objects.16 17     Objects are things in the system that may be acted upon directly by18     userspace programs.  Linux has a variety of actionable objects, including:19 20	- Tasks21	- Files/inodes22	- Sockets23	- Message queues24	- Shared memory segments25	- Semaphores26	- Keys27 28     As a part of the description of all these objects there is a set of29     credentials.  What's in the set depends on the type of object.30 31 2. Object ownership.32 33     Amongst the credentials of most objects, there will be a subset that34     indicates the ownership of that object.  This is used for resource35     accounting and limitation (disk quotas and task rlimits for example).36 37     In a standard UNIX filesystem, for instance, this will be defined by the38     UID marked on the inode.39 40 3. The objective context.41 42     Also amongst the credentials of those objects, there will be a subset that43     indicates the 'objective context' of that object.  This may or may not be44     the same set as in (2) - in standard UNIX files, for instance, this is the45     defined by the UID and the GID marked on the inode.46 47     The objective context is used as part of the security calculation that is48     carried out when an object is acted upon.49 50 4. Subjects.51 52     A subject is an object that is acting upon another object.53 54     Most of the objects in the system are inactive: they don't act on other55     objects within the system.  Processes/tasks are the obvious exception:56     they do stuff; they access and manipulate things.57 58     Objects other than tasks may under some circumstances also be subjects.59     For instance an open file may send SIGIO to a task using the UID and EUID60     given to it by a task that called ``fcntl(F_SETOWN)`` upon it.  In this case,61     the file struct will have a subjective context too.62 63 5. The subjective context.64 65     A subject has an additional interpretation of its credentials.  A subset66     of its credentials forms the 'subjective context'.  The subjective context67     is used as part of the security calculation that is carried out when a68     subject acts.69 70     A Linux task, for example, has the FSUID, FSGID and the supplementary71     group list for when it is acting upon a file - which are quite separate72     from the real UID and GID that normally form the objective context of the73     task.74 75 6. Actions.76 77     Linux has a number of actions available that a subject may perform upon an78     object.  The set of actions available depends on the nature of the subject79     and the object.80 81     Actions include reading, writing, creating and deleting files; forking or82     signalling and tracing tasks.83 84 7. Rules, access control lists and security calculations.85 86     When a subject acts upon an object, a security calculation is made.  This87     involves taking the subjective context, the objective context and the88     action, and searching one or more sets of rules to see whether the subject89     is granted or denied permission to act in the desired manner on the90     object, given those contexts.91 92     There are two main sources of rules:93 94     a. Discretionary access control (DAC):95 96	 Sometimes the object will include sets of rules as part of its97	 description.  This is an 'Access Control List' or 'ACL'.  A Linux98	 file may supply more than one ACL.99 100	 A traditional UNIX file, for example, includes a permissions mask that101	 is an abbreviated ACL with three fixed classes of subject ('user',102	 'group' and 'other'), each of which may be granted certain privileges103	 ('read', 'write' and 'execute' - whatever those map to for the object104	 in question).  UNIX file permissions do not allow the arbitrary105	 specification of subjects, however, and so are of limited use.106 107	 A Linux file might also sport a POSIX ACL.  This is a list of rules108	 that grants various permissions to arbitrary subjects.109 110     b. Mandatory access control (MAC):111 112	 The system as a whole may have one or more sets of rules that get113	 applied to all subjects and objects, regardless of their source.114	 SELinux and Smack are examples of this.115 116	 In the case of SELinux and Smack, each object is given a label as part117	 of its credentials.  When an action is requested, they take the118	 subject label, the object label and the action and look for a rule119	 that says that this action is either granted or denied.120 121 122Types of Credentials123====================124 125The Linux kernel supports the following types of credentials:126 127 1. Traditional UNIX credentials.128 129	- Real User ID130	- Real Group ID131 132     The UID and GID are carried by most, if not all, Linux objects, even if in133     some cases it has to be invented (FAT or CIFS files for example, which are134     derived from Windows).  These (mostly) define the objective context of135     that object, with tasks being slightly different in some cases.136 137	- Effective, Saved and FS User ID138	- Effective, Saved and FS Group ID139	- Supplementary groups140 141     These are additional credentials used by tasks only.  Usually, an142     EUID/EGID/GROUPS will be used as the subjective context, and real UID/GID143     will be used as the objective.  For tasks, it should be noted that this is144     not always true.145 146 2. Capabilities.147 148	- Set of permitted capabilities149	- Set of inheritable capabilities150	- Set of effective capabilities151	- Capability bounding set152 153     These are only carried by tasks.  They indicate superior capabilities154     granted piecemeal to a task that an ordinary task wouldn't otherwise have.155     These are manipulated implicitly by changes to the traditional UNIX156     credentials, but can also be manipulated directly by the ``capset()``157     system call.158 159     The permitted capabilities are those caps that the process might grant160     itself to its effective or permitted sets through ``capset()``.  This161     inheritable set might also be so constrained.162 163     The effective capabilities are the ones that a task is actually allowed to164     make use of itself.165 166     The inheritable capabilities are the ones that may get passed across167     ``execve()``.168 169     The bounding set limits the capabilities that may be inherited across170     ``execve()``, especially when a binary is executed that will execute as171     UID 0.172 173 3. Secure management flags (securebits).174 175     These are only carried by tasks.  These govern the way the above176     credentials are manipulated and inherited over certain operations such as177     execve().  They aren't used directly as objective or subjective178     credentials.179 180 4. Keys and keyrings.181 182     These are only carried by tasks.  They carry and cache security tokens183     that don't fit into the other standard UNIX credentials.  They are for184     making such things as network filesystem keys available to the file185     accesses performed by processes, without the necessity of ordinary186     programs having to know about security details involved.187 188     Keyrings are a special type of key.  They carry sets of other keys and can189     be searched for the desired key.  Each process may subscribe to a number190     of keyrings:191 192	Per-thread keying193	Per-process keyring194	Per-session keyring195 196     When a process accesses a key, if not already present, it will normally be197     cached on one of these keyrings for future accesses to find.198 199     For more information on using keys, see ``Documentation/security/keys/*``.200 201 5. LSM202 203     The Linux Security Module allows extra controls to be placed over the204     operations that a task may do.  Currently Linux supports several LSM205     options.206 207     Some work by labelling the objects in a system and then applying sets of208     rules (policies) that say what operations a task with one label may do to209     an object with another label.210 211 6. AF_KEY212 213     This is a socket-based approach to credential management for networking214     stacks [RFC 2367].  It isn't discussed by this document as it doesn't215     interact directly with task and file credentials; rather it keeps system216     level credentials.217 218 219When a file is opened, part of the opening task's subjective context is220recorded in the file struct created.  This allows operations using that file221struct to use those credentials instead of the subjective context of the task222that issued the operation.  An example of this would be a file opened on a223network filesystem where the credentials of the opened file should be presented224to the server, regardless of who is actually doing a read or a write upon it.225 226 227File Markings228=============229 230Files on disk or obtained over the network may have annotations that form the231objective security context of that file.  Depending on the type of filesystem,232this may include one or more of the following:233 234 * UNIX UID, GID, mode;235 * Windows user ID;236 * Access control list;237 * LSM security label;238 * UNIX exec privilege escalation bits (SUID/SGID);239 * File capabilities exec privilege escalation bits.240 241These are compared to the task's subjective security context, and certain242operations allowed or disallowed as a result.  In the case of execve(), the243privilege escalation bits come into play, and may allow the resulting process244extra privileges, based on the annotations on the executable file.245 246 247Task Credentials248================249 250In Linux, all of a task's credentials are held in (uid, gid) or through251(groups, keys, LSM security) a refcounted structure of type 'struct cred'.252Each task points to its credentials by a pointer called 'cred' in its253task_struct.254 255Once a set of credentials has been prepared and committed, it may not be256changed, barring the following exceptions:257 258 1. its reference count may be changed;259 260 2. the reference count on the group_info struct it points to may be changed;261 262 3. the reference count on the security data it points to may be changed;263 264 4. the reference count on any keyrings it points to may be changed;265 266 5. any keyrings it points to may be revoked, expired or have their security267    attributes changed; and268 269 6. the contents of any keyrings to which it points may be changed (the whole270    point of keyrings being a shared set of credentials, modifiable by anyone271    with appropriate access).272 273To alter anything in the cred struct, the copy-and-replace principle must be274adhered to.  First take a copy, then alter the copy and then use RCU to change275the task pointer to make it point to the new copy.  There are wrappers to aid276with this (see below).277 278A task may only alter its _own_ credentials; it is no longer permitted for a279task to alter another's credentials.  This means the ``capset()`` system call280is no longer permitted to take any PID other than the one of the current281process. Also ``keyctl_instantiate()`` and ``keyctl_negate()`` functions no282longer permit attachment to process-specific keyrings in the requesting283process as the instantiating process may need to create them.284 285 286Immutable Credentials287---------------------288 289Once a set of credentials has been made public (by calling ``commit_creds()``290for example), it must be considered immutable, barring two exceptions:291 292 1. The reference count may be altered.293 294 2. While the keyring subscriptions of a set of credentials may not be295    changed, the keyrings subscribed to may have their contents altered.296 297To catch accidental credential alteration at compile time, struct task_struct298has _const_ pointers to its credential sets, as does struct file.  Furthermore,299certain functions such as ``get_cred()`` and ``put_cred()`` operate on const300pointers, thus rendering casts unnecessary, but require to temporarily ditch301the const qualification to be able to alter the reference count.302 303 304Accessing Task Credentials305--------------------------306 307A task being able to alter only its own credentials permits the current process308to read or replace its own credentials without the need for any form of locking309-- which simplifies things greatly.  It can just call::310 311	const struct cred *current_cred()312 313to get a pointer to its credentials structure, and it doesn't have to release314it afterwards.315 316There are convenience wrappers for retrieving specific aspects of a task's317credentials (the value is simply returned in each case)::318 319	uid_t current_uid(void)		Current's real UID320	gid_t current_gid(void)		Current's real GID321	uid_t current_euid(void)	Current's effective UID322	gid_t current_egid(void)	Current's effective GID323	uid_t current_fsuid(void)	Current's file access UID324	gid_t current_fsgid(void)	Current's file access GID325	kernel_cap_t current_cap(void)	Current's effective capabilities326	struct user_struct *current_user(void)  Current's user account327 328There are also convenience wrappers for retrieving specific associated pairs of329a task's credentials::330 331	void current_uid_gid(uid_t *, gid_t *);332	void current_euid_egid(uid_t *, gid_t *);333	void current_fsuid_fsgid(uid_t *, gid_t *);334 335which return these pairs of values through their arguments after retrieving336them from the current task's credentials.337 338 339In addition, there is a function for obtaining a reference on the current340process's current set of credentials::341 342	const struct cred *get_current_cred(void);343 344and functions for getting references to one of the credentials that don't345actually live in struct cred::346 347	struct user_struct *get_current_user(void);348	struct group_info *get_current_groups(void);349 350which get references to the current process's user accounting structure and351supplementary groups list respectively.352 353Once a reference has been obtained, it must be released with ``put_cred()``,354``free_uid()`` or ``put_group_info()`` as appropriate.355 356 357Accessing Another Task's Credentials358------------------------------------359 360While a task may access its own credentials without the need for locking, the361same is not true of a task wanting to access another task's credentials.  It362must use the RCU read lock and ``rcu_dereference()``.363 364The ``rcu_dereference()`` is wrapped by::365 366	const struct cred *__task_cred(struct task_struct *task);367 368This should be used inside the RCU read lock, as in the following example::369 370	void foo(struct task_struct *t, struct foo_data *f)371	{372		const struct cred *tcred;373		...374		rcu_read_lock();375		tcred = __task_cred(t);376		f->uid = tcred->uid;377		f->gid = tcred->gid;378		f->groups = get_group_info(tcred->groups);379		rcu_read_unlock();380		...381	}382 383Should it be necessary to hold another task's credentials for a long period of384time, and possibly to sleep while doing so, then the caller should get a385reference on them using::386 387	const struct cred *get_task_cred(struct task_struct *task);388 389This does all the RCU magic inside of it.  The caller must call put_cred() on390the credentials so obtained when they're finished with.391 392.. note::393   The result of ``__task_cred()`` should not be passed directly to394   ``get_cred()`` as this may race with ``commit_cred()``.395 396There are a couple of convenience functions to access bits of another task's397credentials, hiding the RCU magic from the caller::398 399	uid_t task_uid(task)		Task's real UID400	uid_t task_euid(task)		Task's effective UID401 402If the caller is holding the RCU read lock at the time anyway, then::403 404	__task_cred(task)->uid405	__task_cred(task)->euid406 407should be used instead.  Similarly, if multiple aspects of a task's credentials408need to be accessed, RCU read lock should be used, ``__task_cred()`` called,409the result stored in a temporary pointer and then the credential aspects called410from that before dropping the lock.  This prevents the potentially expensive411RCU magic from being invoked multiple times.412 413Should some other single aspect of another task's credentials need to be414accessed, then this can be used::415 416	task_cred_xxx(task, member)417 418where 'member' is a non-pointer member of the cred struct.  For instance::419 420	uid_t task_cred_xxx(task, suid);421 422will retrieve 'struct cred::suid' from the task, doing the appropriate RCU423magic.  This may not be used for pointer members as what they point to may424disappear the moment the RCU read lock is dropped.425 426 427Altering Credentials428--------------------429 430As previously mentioned, a task may only alter its own credentials, and may not431alter those of another task.  This means that it doesn't need to use any432locking to alter its own credentials.433 434To alter the current process's credentials, a function should first prepare a435new set of credentials by calling::436 437	struct cred *prepare_creds(void);438 439this locks current->cred_replace_mutex and then allocates and constructs a440duplicate of the current process's credentials, returning with the mutex still441held if successful.  It returns NULL if not successful (out of memory).442 443The mutex prevents ``ptrace()`` from altering the ptrace state of a process444while security checks on credentials construction and changing is taking place445as the ptrace state may alter the outcome, particularly in the case of446``execve()``.447 448The new credentials set should be altered appropriately, and any security449checks and hooks done.  Both the current and the proposed sets of credentials450are available for this purpose as current_cred() will return the current set451still at this point.452 453When replacing the group list, the new list must be sorted before it454is added to the credential, as a binary search is used to test for455membership.  In practice, this means groups_sort() should be456called before set_groups() or set_current_groups().457groups_sort() must not be called on a ``struct group_list`` which458is shared as it may permute elements as part of the sorting process459even if the array is already sorted.460 461When the credential set is ready, it should be committed to the current process462by calling::463 464	int commit_creds(struct cred *new);465 466This will alter various aspects of the credentials and the process, giving the467LSM a chance to do likewise, then it will use ``rcu_assign_pointer()`` to468actually commit the new credentials to ``current->cred``, it will release469``current->cred_replace_mutex`` to allow ``ptrace()`` to take place, and it470will notify the scheduler and others of the changes.471 472This function is guaranteed to return 0, so that it can be tail-called at the473end of such functions as ``sys_setresuid()``.474 475Note that this function consumes the caller's reference to the new credentials.476The caller should _not_ call ``put_cred()`` on the new credentials afterwards.477 478Furthermore, once this function has been called on a new set of credentials,479those credentials may _not_ be changed further.480 481 482Should the security checks fail or some other error occur after483``prepare_creds()`` has been called, then the following function should be484invoked::485 486	void abort_creds(struct cred *new);487 488This releases the lock on ``current->cred_replace_mutex`` that489``prepare_creds()`` got and then releases the new credentials.490 491 492A typical credentials alteration function would look something like this::493 494	int alter_suid(uid_t suid)495	{496		struct cred *new;497		int ret;498 499		new = prepare_creds();500		if (!new)501			return -ENOMEM;502 503		new->suid = suid;504		ret = security_alter_suid(new);505		if (ret < 0) {506			abort_creds(new);507			return ret;508		}509 510		return commit_creds(new);511	}512 513 514Managing Credentials515--------------------516 517There are some functions to help manage credentials:518 519 - ``void put_cred(const struct cred *cred);``520 521     This releases a reference to the given set of credentials.  If the522     reference count reaches zero, the credentials will be scheduled for523     destruction by the RCU system.524 525 - ``const struct cred *get_cred(const struct cred *cred);``526 527     This gets a reference on a live set of credentials, returning a pointer to528     that set of credentials.529 530 - ``struct cred *get_new_cred(struct cred *cred);``531 532     This gets a reference on a set of credentials that is under construction533     and is thus still mutable, returning a pointer to that set of credentials.534 535 536Open File Credentials537=====================538 539When a new file is opened, a reference is obtained on the opening task's540credentials and this is attached to the file struct as ``f_cred`` in place of541``f_uid`` and ``f_gid``.  Code that used to access ``file->f_uid`` and542``file->f_gid`` should now access ``file->f_cred->fsuid`` and543``file->f_cred->fsgid``.544 545It is safe to access ``f_cred`` without the use of RCU or locking because the546pointer will not change over the lifetime of the file struct, and nor will the547contents of the cred struct pointed to, barring the exceptions listed above548(see the Task Credentials section).549 550To avoid "confused deputy" privilege escalation attacks, access control checks551during subsequent operations on an opened file should use these credentials552instead of "current"'s credentials, as the file may have been passed to a more553privileged process.554 555Overriding the VFS's Use of Credentials556=======================================557 558Under some circumstances it is desirable to override the credentials used by559the VFS, and that can be done by calling into such as ``vfs_mkdir()`` with a560different set of credentials.  This is done in the following places:561 562 * ``sys_faccessat()``.563 * ``do_coredump()``.564 * nfs4recover.c.565