kernel - Fix missing chain wait
[dragonfly.git] / sys / kern / sysv_shm.c
1 /*
2  * Copyright (c) 1994 Adam Glass and Charles Hannum.  All rights reserved.
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. All advertising materials mentioning features or use of this software
13  *    must display the following acknowledgement:
14  *      This product includes software developed by Adam Glass and Charles
15  *      Hannum.
16  * 4. The names of the authors may not be used to endorse or promote products
17  *    derived from this software without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR
20  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
21  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
22  * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
24  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
25  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
26  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
27  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
28  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
29  */
30
31 #include "opt_compat.h"
32 #include "opt_sysvipc.h"
33
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/sysproto.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
39 #include <sys/shm.h>
40 #include <sys/proc.h>
41 #include <sys/malloc.h>
42 #include <sys/mman.h>
43 #include <sys/stat.h>
44 #include <sys/sysent.h>
45 #include <sys/jail.h>
46
47 #include <sys/mplock2.h>
48
49 #include <vm/vm.h>
50 #include <vm/vm_param.h>
51 #include <sys/lock.h>
52 #include <vm/pmap.h>
53 #include <vm/vm_object.h>
54 #include <vm/vm_map.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
57
58 static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments");
59
60 struct oshmctl_args;
61 static int sys_oshmctl (struct proc *p, struct oshmctl_args *uap);
62
63 static int shmget_allocate_segment (struct proc *p, struct shmget_args *uap, int mode);
64 static int shmget_existing (struct proc *p, struct shmget_args *uap, int mode, int segnum);
65
66 /* XXX casting to (sy_call_t *) is bogus, as usual. */
67 static sy_call_t *shmcalls[] = {
68         (sy_call_t *)sys_shmat, (sy_call_t *)sys_oshmctl,
69         (sy_call_t *)sys_shmdt, (sy_call_t *)sys_shmget,
70         (sy_call_t *)sys_shmctl
71 };
72
73 #define SHMSEG_FREE             0x0200
74 #define SHMSEG_REMOVED          0x0400
75 #define SHMSEG_ALLOCATED        0x0800
76 #define SHMSEG_WANTED           0x1000
77
78 static int shm_last_free, shm_committed, shmalloced;
79 int shm_nused;
80 static struct shmid_ds  *shmsegs;
81
82 struct shm_handle {
83         /* vm_offset_t kva; */
84         vm_object_t shm_object;
85 };
86
87 struct shmmap_state {
88         vm_offset_t va;
89         int shmid;
90 };
91
92 static void shm_deallocate_segment (struct shmid_ds *);
93 static int shm_find_segment_by_key (key_t);
94 static struct shmid_ds *shm_find_segment_by_shmid (int);
95 static int shm_delete_mapping (struct vmspace *vm, struct shmmap_state *);
96 static void shmrealloc (void);
97 static void shminit (void *);
98
99 /*
100  * Tuneable values
101  */
102 #ifndef SHMMIN
103 #define SHMMIN  1
104 #endif
105 #ifndef SHMMNI
106 #define SHMMNI  512
107 #endif
108 #ifndef SHMSEG
109 #define SHMSEG  1024
110 #endif
111
112 struct  shminfo shminfo = {
113         0,
114         SHMMIN,
115         SHMMNI,
116         SHMSEG,
117         0
118 };
119
120 static int shm_use_phys;
121
122 TUNABLE_LONG("kern.ipc.shmmin", &shminfo.shmmin);
123 TUNABLE_LONG("kern.ipc.shmmni", &shminfo.shmmni);
124 TUNABLE_LONG("kern.ipc.shmseg", &shminfo.shmseg);
125 TUNABLE_LONG("kern.ipc.shmmaxpgs", &shminfo.shmall);
126 TUNABLE_INT("kern.ipc.shm_use_phys", &shm_use_phys);
127
128 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0,
129     "Max shared memory segment size");
130 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0,
131     "Min shared memory segment size");
132 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RD, &shminfo.shmmni, 0,
133     "Max number of shared memory identifiers");
134 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RW, &shminfo.shmseg, 0,
135     "Max shared memory segments per process");
136 SYSCTL_LONG(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0,
137     "Max pages of shared memory");
138 SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0,
139     "Use phys pager allocation instead of swap pager allocation");
140
141 static int
142 shm_find_segment_by_key(key_t key)
143 {
144         int i;
145
146         for (i = 0; i < shmalloced; i++) {
147                 if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) &&
148                     shmsegs[i].shm_perm.key == key)
149                         return i;
150         }
151         return -1;
152 }
153
154 static struct shmid_ds *
155 shm_find_segment_by_shmid(int shmid)
156 {
157         int segnum;
158         struct shmid_ds *shmseg;
159
160         segnum = IPCID_TO_IX(shmid);
161         if (segnum < 0 || segnum >= shmalloced)
162                 return NULL;
163         shmseg = &shmsegs[segnum];
164         if ((shmseg->shm_perm.mode & (SHMSEG_ALLOCATED | SHMSEG_REMOVED))
165             != SHMSEG_ALLOCATED ||
166             shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) {
167                 return NULL;
168         }
169         return shmseg;
170 }
171
172 static void
173 shm_deallocate_segment(struct shmid_ds *shmseg)
174 {
175         struct shm_handle *shm_handle;
176         size_t size;
177
178         shm_handle = shmseg->shm_internal;
179         vm_object_deallocate(shm_handle->shm_object);
180         kfree((caddr_t)shm_handle, M_SHM);
181         shmseg->shm_internal = NULL;
182         size = round_page(shmseg->shm_segsz);
183         shm_committed -= btoc(size);
184         shm_nused--;
185         shmseg->shm_perm.mode = SHMSEG_FREE;
186 }
187
188 static int
189 shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s)
190 {
191         struct shmid_ds *shmseg;
192         int segnum, result;
193         size_t size;
194
195         segnum = IPCID_TO_IX(shmmap_s->shmid);
196         shmseg = &shmsegs[segnum];
197         size = round_page(shmseg->shm_segsz);
198         result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size);
199         if (result != KERN_SUCCESS)
200                 return EINVAL;
201         shmmap_s->shmid = -1;
202         shmseg->shm_dtime = time_second;
203         if ((--shmseg->shm_nattch <= 0) &&
204             (shmseg->shm_perm.mode & SHMSEG_REMOVED)) {
205                 shm_deallocate_segment(shmseg);
206                 shm_last_free = segnum;
207         }
208         return 0;
209 }
210
211 /*
212  * MPALMOSTSAFE
213  */
214 int
215 sys_shmdt(struct shmdt_args *uap)
216 {
217         struct thread *td = curthread;
218         struct proc *p = td->td_proc;
219         struct shmmap_state *shmmap_s;
220         long i;
221         int error;
222
223         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
224                 return (ENOSYS);
225
226         get_mplock();
227         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
228         if (shmmap_s == NULL) {
229                 error = EINVAL;
230                 goto done;
231         }
232         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
233                 if (shmmap_s->shmid != -1 &&
234                     shmmap_s->va == (vm_offset_t)uap->shmaddr)
235                         break;
236         }
237         if (i == shminfo.shmseg)
238                 error = EINVAL;
239         else
240                 error = shm_delete_mapping(p->p_vmspace, shmmap_s);
241 done:
242         rel_mplock();
243         return (error);
244 }
245
246 /*
247  * MPALMOSTSAFE
248  */
249 int
250 sys_shmat(struct shmat_args *uap)
251 {
252         struct thread *td = curthread;
253         struct proc *p = td->td_proc;
254         int error, flags;
255         long i;
256         struct shmid_ds *shmseg;
257         struct shmmap_state *shmmap_s = NULL;
258         struct shm_handle *shm_handle;
259         vm_offset_t attach_va;
260         vm_prot_t prot;
261         vm_size_t size;
262         int rv;
263
264         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
265                 return (ENOSYS);
266
267         get_mplock();
268 again:
269         shmmap_s = (struct shmmap_state *)p->p_vmspace->vm_shm;
270         if (shmmap_s == NULL) {
271                 size = shminfo.shmseg * sizeof(struct shmmap_state);
272                 shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
273                 for (i = 0; i < shminfo.shmseg; i++)
274                         shmmap_s[i].shmid = -1;
275                 if (p->p_vmspace->vm_shm != NULL) {
276                         kfree(shmmap_s, M_SHM);
277                         goto again;
278                 }
279                 p->p_vmspace->vm_shm = (caddr_t)shmmap_s;
280         }
281         shmseg = shm_find_segment_by_shmid(uap->shmid);
282         if (shmseg == NULL) {
283                 error = EINVAL;
284                 goto done;
285         }
286         error = ipcperm(p, &shmseg->shm_perm,
287                         (uap->shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W);
288         if (error)
289                 goto done;
290         for (i = 0; i < shminfo.shmseg; i++) {
291                 if (shmmap_s->shmid == -1)
292                         break;
293                 shmmap_s++;
294         }
295         if (i >= shminfo.shmseg) {
296                 error = EMFILE;
297                 goto done;
298         }
299         size = round_page(shmseg->shm_segsz);
300 #ifdef VM_PROT_READ_IS_EXEC
301         prot = VM_PROT_READ | VM_PROT_EXECUTE;
302 #else
303         prot = VM_PROT_READ;
304 #endif
305         if ((uap->shmflg & SHM_RDONLY) == 0)
306                 prot |= VM_PROT_WRITE;
307         flags = MAP_ANON | MAP_SHARED;
308         if (uap->shmaddr) {
309                 flags |= MAP_FIXED;
310                 if (uap->shmflg & SHM_RND) {
311                         attach_va = (vm_offset_t)uap->shmaddr & ~(SHMLBA-1);
312                 } else if (((vm_offset_t)uap->shmaddr & (SHMLBA-1)) == 0) {
313                         attach_va = (vm_offset_t)uap->shmaddr;
314                 } else {
315                         error = EINVAL;
316                         goto done;
317                 }
318         } else {
319                 /*
320                  * This is just a hint to vm_map_find() about where to put it.
321                  */
322                 attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz);
323         }
324
325         shm_handle = shmseg->shm_internal;
326         vm_object_hold(shm_handle->shm_object);
327         vm_object_chain_wait(shm_handle->shm_object);
328         vm_object_reference_locked(shm_handle->shm_object);
329         rv = vm_map_find(&p->p_vmspace->vm_map, 
330                          shm_handle->shm_object, 0,
331                          &attach_va,
332                          size, PAGE_SIZE,
333                          ((flags & MAP_FIXED) ? 0 : 1), 
334                          VM_MAPTYPE_NORMAL,
335                          prot, prot,
336                          0);
337         vm_object_drop(shm_handle->shm_object);
338         if (rv != KERN_SUCCESS) {
339                 vm_object_deallocate(shm_handle->shm_object);
340                 error = ENOMEM;
341                 goto done;
342         }
343         vm_map_inherit(&p->p_vmspace->vm_map,
344                        attach_va, attach_va + size, VM_INHERIT_SHARE);
345
346         KKASSERT(shmmap_s->shmid == -1);
347         shmmap_s->va = attach_va;
348         shmmap_s->shmid = uap->shmid;
349         shmseg->shm_lpid = p->p_pid;
350         shmseg->shm_atime = time_second;
351         shmseg->shm_nattch++;
352         uap->sysmsg_resultp = (void *)attach_va;
353         error = 0;
354 done:
355         rel_mplock();
356         return error;
357 }
358
359 struct oshmid_ds {
360         struct  ipc_perm shm_perm;      /* operation perms */
361         int     shm_segsz;              /* size of segment (bytes) */
362         ushort  shm_cpid;               /* pid, creator */
363         ushort  shm_lpid;               /* pid, last operation */
364         short   shm_nattch;             /* no. of current attaches */
365         time_t  shm_atime;              /* last attach time */
366         time_t  shm_dtime;              /* last detach time */
367         time_t  shm_ctime;              /* last change time */
368         void    *shm_handle;            /* internal handle for shm segment */
369 };
370
371 struct oshmctl_args {
372         struct sysmsg sysmsg;
373         int shmid;
374         int cmd;
375         struct oshmid_ds *ubuf;
376 };
377
378 /*
379  * MPALMOSTSAFE
380  */
381 static int
382 sys_oshmctl(struct proc *p, struct oshmctl_args *uap)
383 {
384 #ifdef COMPAT_43
385         struct thread *td = curthread;
386         struct shmid_ds *shmseg;
387         struct oshmid_ds outbuf;
388         int error;
389
390         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
391                 return (ENOSYS);
392
393         get_mplock();
394         shmseg = shm_find_segment_by_shmid(uap->shmid);
395         if (shmseg == NULL) {
396                 error = EINVAL;
397                 goto done;
398         }
399
400         switch (uap->cmd) {
401         case IPC_STAT:
402                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
403                 if (error)
404                         break;
405                 outbuf.shm_perm = shmseg->shm_perm;
406                 outbuf.shm_segsz = shmseg->shm_segsz;
407                 outbuf.shm_cpid = shmseg->shm_cpid;
408                 outbuf.shm_lpid = shmseg->shm_lpid;
409                 outbuf.shm_nattch = shmseg->shm_nattch;
410                 outbuf.shm_atime = shmseg->shm_atime;
411                 outbuf.shm_dtime = shmseg->shm_dtime;
412                 outbuf.shm_ctime = shmseg->shm_ctime;
413                 outbuf.shm_handle = shmseg->shm_internal;
414                 error = copyout((caddr_t)&outbuf, uap->ubuf, sizeof(outbuf));
415                 break;
416         default:
417                 /* XXX casting to (sy_call_t *) is bogus, as usual. */
418                 error = sys_shmctl((struct shmctl_args *)uap);
419         }
420 done:
421         rel_mplock();
422         return error;
423 #else
424         return EINVAL;
425 #endif
426 }
427
428 /*
429  * MPALMOSTSAFE
430  */
431 int
432 sys_shmctl(struct shmctl_args *uap)
433 {
434         struct thread *td = curthread;
435         struct proc *p = td->td_proc;
436         int error;
437         struct shmid_ds inbuf;
438         struct shmid_ds *shmseg;
439
440         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
441                 return (ENOSYS);
442
443         get_mplock();
444         shmseg = shm_find_segment_by_shmid(uap->shmid);
445         if (shmseg == NULL) {
446                 error = EINVAL;
447                 goto done;
448         }
449
450         switch (uap->cmd) {
451         case IPC_STAT:
452                 error = ipcperm(p, &shmseg->shm_perm, IPC_R);
453                 if (error == 0)
454                         error = copyout(shmseg, uap->buf, sizeof(inbuf));
455                 break;
456         case IPC_SET:
457                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
458                 if (error == 0)
459                         error = copyin(uap->buf, &inbuf, sizeof(inbuf));
460                 if (error == 0) {
461                         shmseg->shm_perm.uid = inbuf.shm_perm.uid;
462                         shmseg->shm_perm.gid = inbuf.shm_perm.gid;
463                         shmseg->shm_perm.mode =
464                             (shmseg->shm_perm.mode & ~ACCESSPERMS) |
465                             (inbuf.shm_perm.mode & ACCESSPERMS);
466                         shmseg->shm_ctime = time_second;
467                 }
468                 break;
469         case IPC_RMID:
470                 error = ipcperm(p, &shmseg->shm_perm, IPC_M);
471                 if (error == 0) {
472                         shmseg->shm_perm.key = IPC_PRIVATE;
473                         shmseg->shm_perm.mode |= SHMSEG_REMOVED;
474                         if (shmseg->shm_nattch <= 0) {
475                                 shm_deallocate_segment(shmseg);
476                                 shm_last_free = IPCID_TO_IX(uap->shmid);
477                         }
478                 }
479                 break;
480 #if 0
481         case SHM_LOCK:
482         case SHM_UNLOCK:
483 #endif
484         default:
485                 error = EINVAL;
486                 break;
487         }
488 done:
489         rel_mplock();
490         return error;
491 }
492
493 static int
494 shmget_existing(struct proc *p, struct shmget_args *uap, int mode, int segnum)
495 {
496         struct shmid_ds *shmseg;
497         int error;
498
499         shmseg = &shmsegs[segnum];
500         if (shmseg->shm_perm.mode & SHMSEG_REMOVED) {
501                 /*
502                  * This segment is in the process of being allocated.  Wait
503                  * until it's done, and look the key up again (in case the
504                  * allocation failed or it was freed).
505                  */
506                 shmseg->shm_perm.mode |= SHMSEG_WANTED;
507                 error = tsleep((caddr_t)shmseg, PCATCH, "shmget", 0);
508                 if (error)
509                         return error;
510                 return EAGAIN;
511         }
512         if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL))
513                 return EEXIST;
514         error = ipcperm(p, &shmseg->shm_perm, mode);
515         if (error)
516                 return error;
517         if (uap->size && uap->size > shmseg->shm_segsz)
518                 return EINVAL;
519         uap->sysmsg_result = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
520         return 0;
521 }
522
523 static int
524 shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode)
525 {
526         int i, segnum, shmid;
527         size_t size;
528         struct ucred *cred = p->p_ucred;
529         struct shmid_ds *shmseg;
530         struct shm_handle *shm_handle;
531
532         if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax)
533                 return EINVAL;
534         if (shm_nused >= shminfo.shmmni) /* any shmids left? */
535                 return ENOSPC;
536         size = round_page(uap->size);
537         if (shm_committed + btoc(size) > shminfo.shmall)
538                 return ENOMEM;
539         if (shm_last_free < 0) {
540                 shmrealloc();   /* maybe expand the shmsegs[] array */
541                 for (i = 0; i < shmalloced; i++) {
542                         if (shmsegs[i].shm_perm.mode & SHMSEG_FREE)
543                                 break;
544                 }
545                 if (i == shmalloced)
546                         return ENOSPC;
547                 segnum = i;
548         } else  {
549                 segnum = shm_last_free;
550                 shm_last_free = -1;
551         }
552         shmseg = &shmsegs[segnum];
553         /*
554          * In case we sleep in malloc(), mark the segment present but deleted
555          * so that noone else tries to create the same key.
556          */
557         shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED;
558         shmseg->shm_perm.key = uap->key;
559         shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff;
560         shm_handle = kmalloc(sizeof(struct shm_handle), M_SHM, M_WAITOK);
561         shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm);
562         
563         /*
564          * We make sure that we have allocated a pager before we need
565          * to.
566          */
567         if (shm_use_phys) {
568                 shm_handle->shm_object =
569                    phys_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
570         } else {
571                 shm_handle->shm_object =
572                    swap_pager_alloc(NULL, size, VM_PROT_DEFAULT, 0);
573         }
574         vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING);
575         vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT);
576
577         shmseg->shm_internal = shm_handle;
578         shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid;
579         shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid;
580         shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) |
581             (mode & ACCESSPERMS) | SHMSEG_ALLOCATED;
582         shmseg->shm_segsz = uap->size;
583         shmseg->shm_cpid = p->p_pid;
584         shmseg->shm_lpid = shmseg->shm_nattch = 0;
585         shmseg->shm_atime = shmseg->shm_dtime = 0;
586         shmseg->shm_ctime = time_second;
587         shm_committed += btoc(size);
588         shm_nused++;
589
590         /*
591          * If a physical mapping is desired and we have a ton of free pages
592          * we pre-allocate the pages here in order to avoid on-the-fly
593          * allocation later.  This has a big effect on database warm-up
594          * times since DFly supports concurrent page faults coming from the
595          * same VM object for pages which already exist.
596          *
597          * This can hang the kernel for a while so only do it if shm_use_phys
598          * is set to 2 or higher.
599          */
600         if (shm_use_phys > 1) {
601                 vm_pindex_t pi, pmax;
602                 vm_page_t m;
603
604                 pmax = round_page(shmseg->shm_segsz) >> PAGE_SHIFT;
605                 vm_object_hold(shm_handle->shm_object);
606                 if (pmax > vmstats.v_free_count)
607                         pmax = vmstats.v_free_count;
608                 for (pi = 0; pi < pmax; ++pi) {
609                         m = vm_page_grab(shm_handle->shm_object, pi,
610                                          VM_ALLOC_SYSTEM | VM_ALLOC_NULL_OK |
611                                          VM_ALLOC_ZERO);
612                         if (m == NULL)
613                                 break;
614                         vm_pager_get_page(shm_handle->shm_object, &m, 1);
615                         vm_page_activate(m);
616                         vm_page_wakeup(m);
617                         lwkt_yield();
618                 }
619                 vm_object_drop(shm_handle->shm_object);
620         }
621
622         if (shmseg->shm_perm.mode & SHMSEG_WANTED) {
623                 /*
624                  * Somebody else wanted this key while we were asleep.  Wake
625                  * them up now.
626                  */
627                 shmseg->shm_perm.mode &= ~SHMSEG_WANTED;
628                 wakeup((caddr_t)shmseg);
629         }
630         uap->sysmsg_result = shmid;
631         return 0;
632 }
633
634 /*
635  * MPALMOSTSAFE
636  */
637 int
638 sys_shmget(struct shmget_args *uap)
639 {
640         struct thread *td = curthread;
641         struct proc *p = td->td_proc;
642         int segnum, mode, error;
643
644         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
645                 return (ENOSYS);
646
647         mode = uap->shmflg & ACCESSPERMS;
648         get_mplock();
649
650         if (uap->key != IPC_PRIVATE) {
651         again:
652                 segnum = shm_find_segment_by_key(uap->key);
653                 if (segnum >= 0) {
654                         error = shmget_existing(p, uap, mode, segnum);
655                         if (error == EAGAIN)
656                                 goto again;
657                         goto done;
658                 }
659                 if ((uap->shmflg & IPC_CREAT) == 0) {
660                         error = ENOENT;
661                         goto done;
662                 }
663         }
664         error = shmget_allocate_segment(p, uap, mode);
665 done:
666         rel_mplock();
667         return (error);
668 }
669
670 /*
671  * shmsys_args(int which, int a2, ...) (VARARGS)
672  *
673  * MPALMOSTSAFE
674  */
675 int
676 sys_shmsys(struct shmsys_args *uap)
677 {
678         struct thread *td = curthread;
679         unsigned int which = (unsigned int)uap->which;
680         int error;
681
682         if (!jail_sysvipc_allowed && td->td_ucred->cr_prison != NULL)
683                 return (ENOSYS);
684
685         if (which >= NELEM(shmcalls))
686                 return EINVAL;
687         get_mplock();
688         bcopy(&uap->a2, &uap->which,
689                 sizeof(struct shmsys_args) - offsetof(struct shmsys_args, a2));
690         error = ((*shmcalls[which])(uap));
691         rel_mplock();
692
693         return(error);
694 }
695
696 void
697 shmfork(struct proc *p1, struct proc *p2)
698 {
699         struct shmmap_state *shmmap_s;
700         size_t size;
701         int i;
702
703         get_mplock();
704         size = shminfo.shmseg * sizeof(struct shmmap_state);
705         shmmap_s = kmalloc(size, M_SHM, M_WAITOK);
706         bcopy((caddr_t)p1->p_vmspace->vm_shm, (caddr_t)shmmap_s, size);
707         p2->p_vmspace->vm_shm = (caddr_t)shmmap_s;
708         for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) {
709                 if (shmmap_s->shmid != -1)
710                         shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++;
711         }
712         rel_mplock();
713 }
714
715 void
716 shmexit(struct vmspace *vm)
717 {
718         struct shmmap_state *base, *shm;
719         int i;
720
721         if ((base = (struct shmmap_state *)vm->vm_shm) != NULL) {
722                 vm->vm_shm = NULL;
723                 get_mplock();
724                 for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) {
725                         if (shm->shmid != -1)
726                                 shm_delete_mapping(vm, shm);
727                 }
728                 kfree(base, M_SHM);
729                 rel_mplock();
730         }
731 }
732
733 static void
734 shmrealloc(void)
735 {
736         int i;
737         struct shmid_ds *newsegs;
738
739         if (shmalloced >= shminfo.shmmni)
740                 return;
741
742         newsegs = kmalloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK);
743         for (i = 0; i < shmalloced; i++)
744                 bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0]));
745         for (; i < shminfo.shmmni; i++) {
746                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
747                 shmsegs[i].shm_perm.seq = 0;
748         }
749         kfree(shmsegs, M_SHM);
750         shmsegs = newsegs;
751         shmalloced = shminfo.shmmni;
752 }
753
754 static void
755 shminit(void *dummy)
756 {
757         int i;
758
759         /*
760          * If not overridden by a tunable set the maximum shm to
761          * 2/3 of main memory.
762          */
763         if (shminfo.shmall == 0)
764                 shminfo.shmall = (size_t)vmstats.v_page_count * 2 / 3;
765
766         shminfo.shmmax = shminfo.shmall * PAGE_SIZE;
767         shmalloced = shminfo.shmmni;
768         shmsegs = kmalloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK);
769         for (i = 0; i < shmalloced; i++) {
770                 shmsegs[i].shm_perm.mode = SHMSEG_FREE;
771                 shmsegs[i].shm_perm.seq = 0;
772         }
773         shm_last_free = 0;
774         shm_nused = 0;
775         shm_committed = 0;
776 }
777 SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL);