xref: /titanic_52/usr/src/contrib/ast/src/lib/libast/vmalloc/vmmopen.c (revision 906afcb89d0412cc073b95c2d701a804a8cdb62c)
1 /***********************************************************************
2 *                                                                      *
3 *               This software is part of the ast package               *
4 *          Copyright (c) 1985-2012 AT&T Intellectual Property          *
5 *                      and is licensed under the                       *
6 *                 Eclipse Public License, Version 1.0                  *
7 *                    by AT&T Intellectual Property                     *
8 *                                                                      *
9 *                A copy of the License is available at                 *
10 *          http://www.eclipse.org/org/documents/epl-v10.html           *
11 *         (with md5 checksum b35adb5213ca9657e911e9befb180842)         *
12 *                                                                      *
13 *              Information and Software Systems Research               *
14 *                            AT&T Research                             *
15 *                           Florham Park NJ                            *
16 *                                                                      *
17 *                 Glenn Fowler <gsf@research.att.com>                  *
18 *                  David Korn <dgk@research.att.com>                   *
19 *                   Phong Vo <kpv@research.att.com>                    *
20 *                                                                      *
21 ***********************************************************************/
22 #if defined(_UWIN) && defined(_BLD_ast)
23 
24 void _STUB_vmmapopen(){}
25 
26 #else
27 
28 #include	"vmhdr.h"
29 #include	<sys/types.h>
30 #include	<string.h>
31 #if _hdr_unistd
32 #include	<unistd.h>
33 #endif
34 
35 #undef	ALIGN	/* some sys/param.h define this */
36 
37 #include	<sys/mman.h>	/* mmap() headers	*/
38 #include	<sys/file.h>
39 #include	<sys/stat.h>
40 #include	<fcntl.h>
41 
42 #include	<sys/shm.h>	/* shm headers		*/
43 #include	<sys/ipc.h>
44 
45 #undef	ALIGN
46 #define ALIGN	sizeof(struct _align_s)
47 
48 /* Create a region to allocate based on mmap() or shmget().
49 ** Both ways provide for share memory allocation.
50 ** mmap() also allows for allocating persistent data.
51 **
52 ** Written by Kiem-Phong Vo (kpv@research.att.com)
53 */
54 
55 #define MM_INIT		001	/* initialization mode	*/
56 
57 #define MM_RELEASE	010	/* release share mem	*/
58 #define MM_CLEANUP	020	/* clean up resources	*/
59 
60 /* magic word signaling region is being initialized */
61 #define MM_LETMEDOIT	((unsigned int)(('N'<<24) | ('B'<<16) | ('&'<<8) | ('I')) )
62 
63 /* magic word signaling file/segment is ready */
64 #define	MM_MAGIC	((unsigned int)(('P'<<24) | ('&'<<16) | ('N'<<8) | ('8')) )
65 
66 /* default mimimum region size */
67 #define MM_MINSIZE	(64*_Vmpagesize)
68 
69 /* macros to get the data section and size */
70 #define MMHEAD(file)	ROUND(sizeof(Mmvm_t)+strlen(file), ALIGN)
71 #define MMDATA(mmvm)	((Vmuchar_t*)(mmvm)->base + MMHEAD(mmvm->file))
72 #define MMSIZE(mmvm)	((mmvm)->size - MMHEAD(mmvm->file))
73 
74 #ifdef S_IRUSR
75 #define FILE_MODE	(S_IRUSR|S_IWUSR|S_IRGRP|S_IROTH)
76 #else
77 #define FILE_MODE	0644
78 #endif
79 
80 /* to store key/value pairs for application */
81 typedef struct _mmuser_s	Mmuser_t;
82 struct _mmuser_s
83 {	Mmuser_t*	next;	/* link list		*/
84 	int		key;	/* identifying key	*/
85 	Void_t*		val;	/* associated value	*/
86 };
87 
88 typedef struct _mmvm_s
89 {	unsigned int	magic;	/* magic bytes		*/
90 	Void_t*		base;	/* address to map to	*/
91 	ssize_t		size;	/* total data size	*/
92 	ssize_t		busy;	/* amount in use	*/
93 	Mmuser_t*	user;	/* stored (key,val)'s	*/
94 	int		proj;	/* project number	*/
95 	char		file[1];/* file name		*/
96 } Mmvm_t;
97 
98 typedef struct _mmdisc_s
99 {	Vmdisc_t	disc;	/* Vmalloc discipline	*/
100 	int		flag;	/* various modes	*/
101 	Mmvm_t*		mmvm;	/* shared memory data	*/
102 	ssize_t		size;	/* desired file size	*/
103 	int		shmid;	/* ID of the shared mem	*/
104 	int		proj;	/* shm project ID 	*/
105 	char		file[1];/* backing store/ftok()	*/
106 } Mmdisc_t;
107 
108 #if DEBUG
109 #include	<stdio.h>
110 #include	<string.h>
111 int _vmmdump(Vmalloc_t* vm, int fd)
112 {
113 	char		mesg[1024];
114 	Mmdisc_t	*mmdc = (Mmdisc_t*)vm->disc;
115 
116 	fd = fd < 0 ? 2 : fd;
117 	sprintf(mesg, "File: %s\n", mmdc->file ); write(fd, mesg, strlen(mesg));
118 	sprintf(mesg, "Project: %10d\n", mmdc->proj); write(fd, mesg, strlen(mesg));
119 	sprintf(mesg, "Memory:  %#010lx\n", mmdc->mmvm); write(fd, mesg, strlen(mesg));
120 	sprintf(mesg, "Size:    %10d\n", mmdc->size); write(fd, mesg, strlen(mesg));
121 	sprintf(mesg, "Shmid:   %10d\n", mmdc->shmid); write(fd, mesg, strlen(mesg));
122 
123 	sprintf(mesg, "File header:\n"); write(fd, mesg, strlen(mesg));
124 	sprintf(mesg, "Magic:   %10d\n", mmdc->mmvm->magic); write(fd, mesg, strlen(mesg));
125 	sprintf(mesg, "Base:    %#010lx\n", mmdc->mmvm->base); write(fd, mesg, strlen(mesg));
126 	sprintf(mesg, "Size:    %10d\n", mmdc->mmvm->size); write(fd, mesg, strlen(mesg));
127 	sprintf(mesg, "Busy:    %10d\n", mmdc->mmvm->busy); write(fd, mesg, strlen(mesg));
128 	return 0;
129 }
130 #endif /*DEBUG*/
131 
132 /* fix the mapped address for a region */
133 static Mmvm_t* mmfix(Mmvm_t* mmvm, Mmdisc_t* mmdc, int fd)
134 {
135 	Void_t	*base = mmvm->base;
136 	ssize_t	size = mmvm->size;
137 
138 	if(base != (Void_t*)mmvm) /* mmvm is not right yet */
139 	{	/**/ASSERT(!base || (base && (VLONG(base)%_Vmpagesize) == 0) );
140 		if(mmdc->proj < 0)
141 		{	munmap((Void_t*)mmvm, size);
142 			mmvm = (Mmvm_t*)mmap(base, size, (PROT_READ|PROT_WRITE),
143 					     (MAP_FIXED|MAP_SHARED), fd, (off_t)0 );
144 		}
145 		else
146 		{	shmdt((Void_t*)mmvm);
147 			mmvm = (Mmvm_t*)shmat(mmdc->shmid, base, 0);
148 		}
149 		if(!mmvm || mmvm == (Mmvm_t*)(-1) )
150 			mmvm = NIL(Mmvm_t*);
151 	}
152 
153 	return mmvm;
154 }
155 
156 /* initialize region data */
157 static int mminit(Mmdisc_t* mmdc)
158 {
159 	Void_t		*base;
160 	int		try, k;
161 	int		fd = -1;
162 	key_t		key = -1;
163 	ssize_t		extent, size = 0;
164 	Mmvm_t		*mmvm = NIL(Mmvm_t*);
165 	int		rv = -1;
166 
167 	if(mmdc->mmvm) /* already done this */
168 		return 0;
169 
170 	/* fixed size region so make it reasonably large */
171 	if((size = mmdc->size) < MM_MINSIZE )
172 		size =  MM_MINSIZE;
173 	size += MMHEAD(mmdc->file) + ALIGN;
174 	size  = ROUND(size, _Vmpagesize);
175 
176 	/* this op can happen simultaneously in different processes */
177 	if((fd = open(mmdc->file, O_RDWR|O_CREAT, FILE_MODE)) < 0)
178 		return -1;
179 
180 	/* get/create the initial segment of data */
181 	if(mmdc->proj < 0 ) /* proj < 0 means doing mmap() */
182 	{	/* Note that the location being written to is always zero! */
183 		if((extent = (ssize_t)lseek(fd, (off_t)0, SEEK_END)) < 0)
184 			goto done;
185 		if(extent < size) /* make the file size large enough */
186 			if(lseek(fd, (off_t)size, 0) != (off_t)size || write(fd, "", 1) != 1 )
187 				goto done;
188 
189 		/* map the file into memory */
190 		mmvm = (Mmvm_t*)mmap(NIL(Void_t*), size, (PROT_READ|PROT_WRITE),
191 		 		     MAP_SHARED, fd, (off_t)0 );
192 	}
193 	else
194 	{	/* make the key and get/create an id for the share mem segment */
195 		if((key = ftok(mmdc->file, mmdc->proj)) < 0 )
196 			goto done;
197 		if((mmdc->shmid = shmget(key, size, IPC_CREAT|FILE_MODE)) < 0 )
198 			goto done;
199 
200 		/* map the data segment into memory */
201 		mmvm = (Mmvm_t*)shmat(mmdc->shmid, NIL(Void_t*), 0);
202 	}
203 
204 	if(!mmvm || mmvm == (Mmvm_t*)(-1) ) /* initial mapping failed */
205 		goto done;
206 
207 	/* all processes compete for the chore to initialize data */
208 	if(asocasint(&mmvm->magic, 0, MM_LETMEDOIT) == 0 ) /* lucky winner: us! */
209 	{	if(!(base = vmmaddress(size)) ) /* get a suitable base for the map */
210 			base = (Void_t*)mmvm;
211 		mmdc->flag |= MM_INIT;
212 		mmvm->base = base;
213 		mmvm->size = size;
214 		mmvm->busy = 0;
215 		mmvm->proj = mmdc->proj;
216 		strcpy(mmvm->file, mmdc->file);
217 		if(mmdc->proj < 0 ) /* flush to file */
218 			msync((Void_t*)mmvm, MMHEAD(mmvm->file), MS_SYNC);
219 
220 		if(mmvm->base != (Void_t*)mmvm) /* not yet at the right address */
221 			if(!(mmvm = mmfix(mmvm, mmdc, fd)) )
222 				goto done;
223 		rv = 0; /* success, return this value to indicate a new map */
224 	}
225 	else /* wait for someone else to finish initialization */
226 	{	/**/ASSERT(!(mmdc->flag&MM_INIT));
227 		if(mmvm->magic != MM_LETMEDOIT && mmvm->magic != MM_MAGIC)
228 			goto done;
229 
230 		for(try = 0, k = 0;; ASOLOOP(k) ) /* waiting */
231 		{	if(asocasint(&mmvm->magic, MM_MAGIC, MM_MAGIC) == MM_MAGIC )
232 				break;
233 			else if((try += 1) <= 0 ) /* too many tries */
234 				goto done;
235 		}
236 
237 		/* mapped the wrong memory */
238 		if(mmvm->proj != mmdc->proj || strcmp(mmvm->file, mmdc->file) != 0 )
239 			goto done;
240 
241 		if(mmvm->base != (Void_t*)mmvm) /* not yet at the right address */
242 			if(!(mmvm = mmfix(mmvm, mmdc, fd)) )
243 				goto done;
244 		rv = 1; /* success, return this value to indicate a finished map */
245 	}
246 
247 done:	(void)close(fd);
248 
249 	if(rv >= 0 ) /* successful construction of region */
250 	{	/**/ASSERT(mmvm && mmvm != (Mmvm_t*)(-1));
251 		mmdc->mmvm = mmvm;
252 	}
253 	else if(mmvm && mmvm != (Mmvm_t*)(-1)) /* error, remove map */
254 	{	if(mmdc->proj < 0)
255 			(void)munmap((Void_t*)mmvm, size);
256 		else	(void)shmdt((Void_t*)mmvm);
257 	}
258 
259 	return rv;
260 }
261 
262 #if __STD_C /* end a file mapping */
263 static int mmend(Mmdisc_t* mmdc)
264 #else
265 static int mmend(mmdc)
266 Mmdisc_t*	mmdc;
267 #endif
268 {
269 	Mmvm_t		*mmvm;
270 	struct shmid_ds	shmds;
271 
272 	if(!(mmvm = mmdc->mmvm) )
273 		return 0;
274 
275 	if(mmdc->proj < 0 )
276 	{	(void)msync(mmvm->base, mmvm->size, MS_ASYNC);
277 		if(mmdc->flag&MM_RELEASE)
278 		{	if(mmvm->base )
279 				(void)munmap(mmvm->base, mmvm->size);
280 		}
281 		if(mmdc->flag&MM_CLEANUP)
282 			(void)unlink(mmdc->file);
283 	}
284 	else
285 	{	if(mmdc->flag&MM_RELEASE)
286 		{	if(mmvm->base )
287 				(void)shmdt(mmvm->base);
288 		}
289 		if(mmdc->flag&MM_CLEANUP)
290 		{	if(mmdc->shmid >= 0 )
291 				(void)shmctl(mmdc->shmid, IPC_RMID, &shmds);
292 		}
293 	}
294 
295 	mmdc->mmvm = NIL(Mmvm_t*);
296 	return 0;
297 }
298 
299 #if __STD_C
300 static Void_t* mmgetmem(Vmalloc_t* vm, Void_t* caddr,
301 			size_t csize, size_t nsize, Vmdisc_t* disc)
302 #else
303 static Void_t* mmgetmem(vm, caddr, csize, nsize, disc)
304 Vmalloc_t*	vm;
305 Void_t*		caddr;
306 size_t		csize;
307 size_t		nsize;
308 Vmdisc_t*	disc;
309 #endif
310 {
311 	Mmvm_t		*mmvm;
312 	Mmdisc_t	*mmdc = (Mmdisc_t*)disc;
313 
314 	if(!(mmvm = mmdc->mmvm) ) /* bad data */
315 		return NIL(Void_t*);
316 
317 	/* this region allows only a single busy block! */
318 	if(caddr) /* resizing/freeing an existing block */
319 	{	if(caddr == MMDATA(mmvm) && nsize <= MMSIZE(mmvm) )
320 		{	mmvm->busy = nsize;
321 			return MMDATA(mmvm);
322 		}
323 		else	return NIL(Void_t*);
324 	}
325 	else /* requesting a new block */
326 	{	if(mmvm->busy == 0 )
327 		{	mmvm->busy = nsize;
328 			return MMDATA(mmvm);
329 		}
330 		else	return NIL(Void_t*);
331 	}
332 }
333 
334 #if __STD_C
335 static int mmexcept(Vmalloc_t* vm, int type, Void_t* data, Vmdisc_t* disc)
336 #else
337 static int mmexcept(vm, type, data, disc)
338 Vmalloc_t*	vm;
339 int		type;
340 Void_t*		data;
341 Vmdisc_t*	disc;
342 #endif
343 {
344 	int		rv;
345 	Void_t		*base;
346 	Mmdisc_t	*mmdc = (Mmdisc_t*)disc;
347 
348 	if(type == VM_OPEN)
349 	{	if(data) /* VM_OPEN event at start of vmopen() */
350 		{	if((rv = mminit(mmdc)) < 0 ) /* initialization failed */
351 				return -1;
352 			else if(rv == 0) /* just started a new map */
353 			{	/**/ASSERT(mmdc->flag&MM_INIT);
354 				/**/ASSERT(mmdc->mmvm->magic == MM_LETMEDOIT);
355 				return 0;
356 			}
357 			else /* an existing map was reconstructed */
358 			{	/**/ASSERT(!(mmdc->flag&MM_INIT));
359 				/**/ASSERT(mmdc->mmvm->magic == MM_MAGIC);
360 				*((Void_t**)data) = MMDATA(mmdc->mmvm);
361 				return 1;
362 			}
363 		}
364 		else	return 0;
365 	}
366 	else if(type == VM_ENDOPEN) /* at end of vmopen() */
367 	{	if(mmdc->flag&MM_INIT) /* this is the initializing process! */
368 		{	/**/ASSERT(mmdc->mmvm->magic == MM_LETMEDOIT);
369 			asocasint(&mmdc->mmvm->magic, MM_LETMEDOIT, MM_MAGIC);
370 
371 			if(mmdc->proj < 0) /* sync data to file now */
372 				msync((Void_t*)mmdc->mmvm, MMHEAD(mmdc->file), MS_SYNC);
373 		} /**/ASSERT(mmdc->mmvm->magic == MM_MAGIC);
374 		return 0;
375 	}
376 	else if(type == VM_CLOSE)
377 		return 1; /* tell vmclose not to free memory segments */
378 	else if(type == VM_ENDCLOSE) /* this is the final closing event */
379 	{	(void)mmend(mmdc);
380 		(void)vmfree(Vmheap, mmdc);
381 		return 0; /* all done */
382 	}
383 	else	return 0;
384 }
385 
386 #if __STD_C
387 Vmalloc_t* vmmopen(char* file, int proj, ssize_t size )
388 #else
389 Vmalloc_t* vmmopen(file, proj, size )
390 char*		file;	/* file for key or data backing */
391 int		proj;	/* project ID, < 0 doing mmap	*/
392 ssize_t		size;	/* desired size for mem segment	*/
393 #endif
394 {
395 	Vmalloc_t	*vm;
396 	Mmdisc_t	*mmdc;
397 
398 	GETPAGESIZE(_Vmpagesize);
399 
400 	if(!file || !file[0] )
401 		return NIL(Vmalloc_t*);
402 
403 	/* create discipline structure for getting memory from mmap */
404 	if(!(mmdc = vmalloc(Vmheap, sizeof(Mmdisc_t)+strlen(file))) )
405 		return NIL(Vmalloc_t*);
406 	memset(mmdc, 0, sizeof(Mmdisc_t));
407 	mmdc->disc.memoryf = mmgetmem;
408 	mmdc->disc.exceptf = mmexcept;
409 	mmdc->disc.round   = _Vmpagesize; /* round request to this size */
410 	mmdc->mmvm = NIL(Mmvm_t*);
411 	mmdc->size = size;
412 	mmdc->shmid = -1;
413 	mmdc->flag = 0;
414 	mmdc->proj = proj;
415 	strcpy(mmdc->file, file);
416 
417 	/* now open the Vmalloc_t handle to return to application */
418 	if(!(vm = vmopen(&mmdc->disc, Vmbest, VM_SHARE)) )
419 	{	(void)mmend(mmdc);
420 		(void)vmfree(Vmheap, mmdc);
421 		return NIL(Vmalloc_t*);
422 	}
423 	else
424 	{	/**/ASSERT(mmdc->mmvm && mmdc->mmvm->magic == MM_MAGIC);
425 		return vm;
426 	}
427 }
428 
429 /* to store (key,value) data in the map */
430 #if __STD_C
431 Void_t* vmmvalue(Vmalloc_t* vm, int key, Void_t* val, int oper)
432 #else
433 Void_t* vmmvalue(vm, key, val, oper)
434 Vmalloc_t*	vm;	/* a region based on vmmapopen	*/
435 int		key;	/* key of data to be set	*/
436 Void_t*		val;	/* data to be set		*/
437 int		oper;	/* operation type		*/
438 #endif
439 {
440 	Mmuser_t	*u;
441 	Mmdisc_t	*mmdc = (Mmdisc_t*)vm->disc;
442 	Mmvm_t		*mmvm = mmdc->mmvm;
443 
444 	/* check to see if operation is well-defined */
445 	if(oper != VM_MMGET && oper != VM_MMSET && oper != VM_MMADD)
446 		return NIL(Void_t*);
447 
448 	SETLOCK(vm, 0);
449 
450 	/* find the key */
451 	for(u = mmvm->user; u; u = u->next)
452 		if(u->key == key)
453 			break;
454 
455 	if(!u && (oper == VM_MMSET || oper == VM_MMADD) )
456 	{	if((u = KPVALLOC(vm, sizeof(Mmuser_t), vm->meth.allocf)) )
457 		{	u->val  = NIL(Void_t*);
458 			u->key  = key;
459 			u->next = mmvm->user;
460 			mmvm->user = u;
461 		}
462 	}
463 
464 	if(u) /* update data and set value to return */
465 	{	if(oper == VM_MMSET)
466 			u->val = val;
467 		else if(oper == VM_MMADD)
468 			u->val = (Void_t*)((long)(u->val) + (long)(val));
469 		val = u->val;
470 	}
471 	else	val = NIL(Void_t*);
472 
473 	CLRLOCK(vm, 0);
474 
475 	return val;
476 }
477 
478 void vmmrelease(Vmalloc_t* vm, int type)
479 {
480 	Mmdisc_t	*mmdc = (Mmdisc_t*)vm->disc;
481 
482 	mmdc->flag |= MM_RELEASE;
483 	if(type > 0)
484 		mmdc->flag |= MM_CLEANUP;
485 }
486 
487 /* suggest an address usable for mapping memory */
488 Void_t* vmmaddress(size_t size)
489 {
490 #if !defined(_map_min) || !defined(_map_max) || !defined(_map_dir)
491 	return NIL(Void_t*);
492 #else
493 	Void_t			*avail;
494 	static Vmuchar_t	*min = (Vmuchar_t*)_map_min;
495 	static Vmuchar_t	*max = (Vmuchar_t*)_map_max;
496 
497 	GETPAGESIZE(_Vmpagesize);
498 	size = ROUND(size, _Vmpagesize);
499 
500 	if(_map_dir == 0 || (min+size) > max)
501 		avail = NIL(Void_t*);
502 	else if(_map_dir > 0)
503 	{	avail = (Void_t*)min;
504 		min += size;
505 	}
506 	else
507 	{	max -= size;
508 		avail = (Void_t*)max;
509 	}
510 
511 	return avail;
512 #endif
513 }
514 
515 #endif
516