2# CDDL HEADER START
4# The contents of this file are subject to the terms of the
5# Common Development and Distribution License (the "License").
6# You may not use this file except in compliance with the License.
8# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9# or http://www.opensolaris.org/os/licensing.
10# See the License for the specific language governing permissions
11# and limitations under the License.
13# When distributing Covered Code, include this CDDL HEADER in each
14# file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15# If applicable, add the following below this CDDL HEADER, with the
16# fields enclosed by brackets "" replaced with your own identifying
17# information: Portions Copyright [yyyy] [name of copyright owner]
19# CDDL HEADER END
22# Copyright 2006 Sun Microsystems, Inc. All rights reserved.
23# Use is subject to license terms.
26# Process Model Unification
28The Solaris Process Model Unification project:
29 PSARC/2002/117 Solaris Process Model Unification
30 4470917 Solaris Process Model Unification
31folded libthread into libc and has led to some fundamental changes
32in the rules by which code in libc must be developed and maintained.
34All code in libc must be both MT-Safe and Fork-Safe
35and where possible (almost everywhere), Async-Signal-Safe.
37To this end, the following rules should be followed:
39Almost all internal libc locks (mutexes and read-write locks)
40should be acquired and released via these interfaces:
42 mutex_t some_lock = DEFAULTMUTEX;
45 ... do something critical ...
48 rwlock_t some_rw_lock = DEFAULTRWLOCK;
51 ... multiple threads can do something ...
55 ... only one thread can do something ...
58The above l* versions of the mutex and rwlock interfaces do more
59than the ordinary interfaces: They define critical regions in
60which the calling thread cannot be suspended (making the region
61fork-safe) and in which the calling thread has all signals deferred
62(making the region async-signal-safe).
64However, certain rules apply to the code within these critical regions:
66 - The code must be of guaranteed short duration; no calls
67 to interfaces that might block indefinitely are allowed.
68 This means no calls into stdio or syslog() and no calls
69 to cond_wait() unless there is a guarantee of an almost-
70 immediate call to cond_signal() or cond_broadcast()
71 from elsewhere.
73 - The code cannot call any non-l* synchronization
74 primitives (mutex_lock(), _private_mutex_lock(),
75 rw_wrlock(), rw_rdlock(), sema_wait(), etc.)
77 - The code cannot call any functions outside of libc,
78 including application callbacks and functions from
79 dlopen()ed objects, such as those in the I18N code.
81 - Because malloc(), calloc(), realloc(), and free()
82 are designed to be interposed upon, they fall into
83 the previous case of prohibition. None of these can
84 be called by a thread while in a critical region.
86There is a private memory allocator for use internally to libc.
87It cannot be interposed upon and it is safe to use while in
88a critical region (or for that matter while not in a critical
89region; it is async-signal-safe and fork-safe):
91 void *lmalloc(size_t);
92 void lfree(void *, size_t);
94 void *libc_malloc(size_t);
95 void *libc_realloc(void *, size_t);
96 char *libc_strdup(const char *);
97 void libc_free(void *);
99lmalloc() and lfree() are the basic interfaces. The libc_*()
100variants are built on top of lmalloc()/lfree() but they have
101the same interface signatures as the corresponding functions
102without the 'libc_' prefix. lmalloc() and libc_malloc()
103return zeroed memory blocks. Note that lmalloc()/lfree()
104require the caller to remember the size parameter passed
105to lmalloc() and to pass the same value to lfree().
107Memory allocated by lmalloc() can only be freed by lfree().
108Memory allocated by libc_malloc(), libc_realloc(), or libc_strdup()
109can only be freed by libc_free(). Never pass such allocated
110memory out of libc if the caller of libc is expected to free it.
112lmalloc()/lfree() is a small and simple power of two allocator.
113Do not use it as a general-purpose allocator. Be kind to it.
115There is a special mutual exclusion interface that exists for
116cases, like code in the I18N interfaces, where mutual exclusion
117is required but the above rules cannot be followed:
119 int fork_lock_enter(const char *);
120 void fork_lock_exit(void);
122fork_lock_enter() does triple-duty. Not only does it serialize
123calls to fork() and forkall(), but it also serializes calls to
124thr_suspend() (fork() and forkall() also suspend other threads),
125and furthermore it serializes I18N calls to functions in other
126dlopen()ed L10N objects that might be calling malloc()/free().
127Use it in general like this:
129 (void) fork_lock_enter(NULL);
130 ... serialized; do something that might call malloc ...
133The 'const char *' argument to fork_lock_enter() should always
134be NULL except for two special cases:
135 - When called from fork() or forkall()
136 - When called from pthread_atfork()
137This enforces the prohibition against calling fork() or pthread_atfork()
138from a pthread_atfork()-registered fork handler function while a fork()
139prologue or epilogue is in progress. If _THREAD_ERROR_DETECTION is set
140to 1 or 2 in the environment, such cases will draw a nasty message and
141will dump core if _THREAD_ERROR_DETECTION=2. fork_lock_enter() returns
142non-zero only if it is called from a fork handler. This is of interest
143only to callers that have to do something about this condition; the
144return value should be ignored in all other cases (fork_lock_enter()
145never actually fails).
147It is an error to call fork_lock_enter() while in a critical region
148(that is, while holding any internal libc lock).
150On return from fork_lock_enter(), no internal libc locks are held
151but a flag has been set to cause other callers of fork_lock_enter()
152to delay (via _cond_wait()) until fork_lock_exit() is called.
154These are the rules to follow for memory allocation:
156 - If a function acquires an internal libc lock or is called while
157 an internal libc lock is held:
159 * The malloc family cannot be used.
161 * lmalloc or libc_malloc should be used. The memory must
162 be released by lfree or libc_free, respectively.
164 * lfree takes an argument to tell the size of the releasing
165 memory. If the function does not know the size at the
166 releasing point, libc_malloc and libc_free should be used.
168 * As the memory allocated by lmalloc or libc_malloc needs
169 to be released by lfree or libc_free and these are internal
170 to libc, they cannot be used to allocate memory that might
171 be released by application code outside libc.
173 * If the memory allocation by malloc() cannot be avoided and
174 the scalability of the function does not matter much, the
175 function can be serialized with fork_lock_enter() instead
176 of lmutex_lock().
178 * If the memory allocation by malloc() cannot be avoided and
179 the scalability of the function does matter, another
180 implementation of the function will be necessary.
182In a DEBUG build of libc:
183 make THREAD_DEBUG=-DTHREAD_DEBUG install
184many of these rules are enforced by ASSERT() statements scattered about
185in the libc sources. This is the default mode for building libc when
186a DEBUG nightly build is performed.
188# libaio/librt Implementation In libc
190The putback of the project:
191 6416832 libaio and librt can and should be folded into libc
192introduced several libc-private locking interfaces:
193 void sig_mutex_lock(mutex_t *);
194 void sig_mutex_unlock(mutex_t *);
195 int sig_mutex_trylock(mutex_t *);
196 int sig_cond_wait(cond_t *, mutex_t *);
197 int sig_cond_reltimedwait(cond_t *, mutex_t *, const timespec_t *);
198which are declared in both "thr_uberdata.h" and "mtlib.h".
200They are used in specialized code in libc, like the asynchronous i/o code.
201Unlike the lmutex_lock() and lmutex_unlock() interfaces described above,
202these interfaces do not define critical regions, but signals are
203deferred while locks acquired by these functions are held, making
204their use be async-signal safe. Calls to malloc(), calloc(), realloc(),
205and free() are permissible while holding such locks.
207These interfaces were brought over from code in the former libaio
208and librt and are necessary because, where they are used, the code
209must execute potentially long-term waits and must be cancelable.
210sig_cond_wait() and sig_cond_reltimedwait() are cancellation points.
212These interfaces are available for other uses inside libc, as
213the need arises. (There is no need if the code does not perform
214long-term waits.) Just follow a few rules to be self-consistent:
215 - Don't mix calls to mutex_[un]lock(), lmutex_[un]lock() and
216 sig_mutex_[un]lock() on the same mutex.
217 - Don't call cond_wait() with a mutex acquired by sig_mutex_lock();
218 call sig_cond_wait() or sig_cond_reltimedwait().
219 - Use pthread_cleanup_push() and pthread_cleanup_pop() to make
220 your code cancellation-safe.
221 - The sig_*() interfaces are not in themselves fork-safe.
222 You have to employ other logic to make your code fork-safe.
223 See the tail of postfork1_child() for examples.
225# Removal Of Synonym Symbols
227The following project:
228 PSARC 2008/309 expunge synonyms.h
229 6700179 expunge synonyms.h
231Removed the historical "synonym" symbols from the C Library.
233Historically, for every public function symbol in the C library a second,
234private, symbol of the same value was defined to be used internally by libc
235(generally, one or the other will be a weak symbol, precisely which is
238These synonym symbols existed such that an application which provided
239otherwise conflicting symbols could interpose on the version in libc without
240compromising libc itself, that is if libc defines fopen() which needs open() it
241would call _open() and an application defined open() would not cause fopen()
242to break. This was made transparent to code within libc via a header,
243synonyms.h, which would #define open _open, for all such symbols.
245Since ON now uses direct bindings extensively all symbols not explicitly
246marked "NODIRECT" are directly bound within libc anyway, and none of this is
247actually necessary. Thus synonyms.h was removed, and no new synonym symbols
248need be added. However, unfortunately, certain of the private symbols were
249inadvertently exposed to applications, and several are known to have been
250used, thus these existing synonyms must continue to exist to maintain
251compatibility. A preloadable library, /lib/c_synonyms.so.1 is provided which
252also provides the historical names with underscore prefixes to allow any other
253incorrect application to continue to function.
255It should never be necessary to add additional synonym symbols to libc nor to
256add underscore prefixed aliases to c_synonyms.so.1.
258# libc Internals Scoped Protected
260The integration of the fix for:
261 6689238 libc needs global protection against ld.so.1
263Scopes all function symbols within libc protected, excepting those that need
264to be accepting of interposition and to have a consistent version called both
265within and without libc (basically, the malloc() family).
267This causes references by libc to itself to be permanently and unavoidably
268bound, and thus to never enter ld.so (and potentially from there audit
269libraries or other such support libraries). This maintains an otherwise
270complicated to verify invariant: within critical sections (with any internal
271lock held, etc) execution can never leave the context of libc. Previously
272this was done with a selection of known-to-be-problematic functions having
273weak synonyms scoped private, but this was both difficult to verify, difficult
274to remember, and thus always at least somewhat incomplete.
276In summary, any new function symbol in libc must be scoped protected unless it
277is one of a very small group of functions that must allow interposed versions
278to be bound to from the C library itself -- it is grossly unlikely that more
279of these will occur.