ipc,msg: make msgctl_nolock lockless
[linux-3.10.git] / ipc / shm.c
1 /*
2  * linux/ipc/shm.c
3  * Copyright (C) 1992, 1993 Krishna Balasubramanian
4  *       Many improvements/fixes by Bruno Haible.
5  * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
6  * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
7  *
8  * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
9  * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
10  * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
11  * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
12  * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
13  * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
14  * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
15  *
16  * support for audit of ipc object properties and permission changes
17  * Dustin Kirkland <dustin.kirkland@us.ibm.com>
18  *
19  * namespaces support
20  * OpenVZ, SWsoft Inc.
21  * Pavel Emelianov <xemul@openvz.org>
22  */
23
24 #include <linux/slab.h>
25 #include <linux/mm.h>
26 #include <linux/hugetlb.h>
27 #include <linux/shm.h>
28 #include <linux/init.h>
29 #include <linux/file.h>
30 #include <linux/mman.h>
31 #include <linux/shmem_fs.h>
32 #include <linux/security.h>
33 #include <linux/syscalls.h>
34 #include <linux/audit.h>
35 #include <linux/capability.h>
36 #include <linux/ptrace.h>
37 #include <linux/seq_file.h>
38 #include <linux/rwsem.h>
39 #include <linux/nsproxy.h>
40 #include <linux/mount.h>
41 #include <linux/ipc_namespace.h>
42
43 #include <asm/uaccess.h>
44
45 #include "util.h"
46
47 struct shm_file_data {
48         int id;
49         struct ipc_namespace *ns;
50         struct file *file;
51         const struct vm_operations_struct *vm_ops;
52 };
53
54 #define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
55
56 static const struct file_operations shm_file_operations;
57 static const struct vm_operations_struct shm_vm_ops;
58
59 #define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
60
61 #define shm_unlock(shp)                 \
62         ipc_unlock(&(shp)->shm_perm)
63
64 static int newseg(struct ipc_namespace *, struct ipc_params *);
65 static void shm_open(struct vm_area_struct *vma);
66 static void shm_close(struct vm_area_struct *vma);
67 static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
68 #ifdef CONFIG_PROC_FS
69 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
70 #endif
71
72 void shm_init_ns(struct ipc_namespace *ns)
73 {
74         ns->shm_ctlmax = SHMMAX;
75         ns->shm_ctlall = SHMALL;
76         ns->shm_ctlmni = SHMMNI;
77         ns->shm_rmid_forced = 0;
78         ns->shm_tot = 0;
79         ipc_init_ids(&shm_ids(ns));
80 }
81
82 /*
83  * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
84  * Only shm_ids.rw_mutex remains locked on exit.
85  */
86 static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
87 {
88         struct shmid_kernel *shp;
89         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
90
91         if (shp->shm_nattch){
92                 shp->shm_perm.mode |= SHM_DEST;
93                 /* Do not find it any more */
94                 shp->shm_perm.key = IPC_PRIVATE;
95                 shm_unlock(shp);
96         } else
97                 shm_destroy(ns, shp);
98 }
99
100 #ifdef CONFIG_IPC_NS
101 void shm_exit_ns(struct ipc_namespace *ns)
102 {
103         free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
104         idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
105 }
106 #endif
107
108 static int __init ipc_ns_init(void)
109 {
110         shm_init_ns(&init_ipc_ns);
111         return 0;
112 }
113
114 pure_initcall(ipc_ns_init);
115
116 void __init shm_init (void)
117 {
118         ipc_init_proc_interface("sysvipc/shm",
119 #if BITS_PER_LONG <= 32
120                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
121 #else
122                                 "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
123 #endif
124                                 IPC_SHM_IDS, sysvipc_shm_proc_show);
125 }
126
127 /*
128  * shm_lock_(check_) routines are called in the paths where the rw_mutex
129  * is not necessarily held.
130  */
131 static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
132 {
133         struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
134
135         if (IS_ERR(ipcp))
136                 return (struct shmid_kernel *)ipcp;
137
138         return container_of(ipcp, struct shmid_kernel, shm_perm);
139 }
140
141 static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
142 {
143         rcu_read_lock();
144         ipc_lock_object(&ipcp->shm_perm);
145 }
146
147 static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
148                                                 int id)
149 {
150         struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
151
152         if (IS_ERR(ipcp))
153                 return (struct shmid_kernel *)ipcp;
154
155         return container_of(ipcp, struct shmid_kernel, shm_perm);
156 }
157
158 static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
159 {
160         ipc_rmid(&shm_ids(ns), &s->shm_perm);
161 }
162
163
164 /* This is called by fork, once for every shm attach. */
165 static void shm_open(struct vm_area_struct *vma)
166 {
167         struct file *file = vma->vm_file;
168         struct shm_file_data *sfd = shm_file_data(file);
169         struct shmid_kernel *shp;
170
171         shp = shm_lock(sfd->ns, sfd->id);
172         BUG_ON(IS_ERR(shp));
173         shp->shm_atim = get_seconds();
174         shp->shm_lprid = task_tgid_vnr(current);
175         shp->shm_nattch++;
176         shm_unlock(shp);
177 }
178
179 /*
180  * shm_destroy - free the struct shmid_kernel
181  *
182  * @ns: namespace
183  * @shp: struct to free
184  *
185  * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
186  * but returns with shp unlocked and freed.
187  */
188 static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
189 {
190         ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
191         shm_rmid(ns, shp);
192         shm_unlock(shp);
193         if (!is_file_hugepages(shp->shm_file))
194                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
195         else if (shp->mlock_user)
196                 user_shm_unlock(file_inode(shp->shm_file)->i_size,
197                                                 shp->mlock_user);
198         fput (shp->shm_file);
199         security_shm_free(shp);
200         ipc_rcu_putref(shp);
201 }
202
203 /*
204  * shm_may_destroy - identifies whether shm segment should be destroyed now
205  *
206  * Returns true if and only if there are no active users of the segment and
207  * one of the following is true:
208  *
209  * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
210  *
211  * 2) sysctl kernel.shm_rmid_forced is set to 1.
212  */
213 static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
214 {
215         return (shp->shm_nattch == 0) &&
216                (ns->shm_rmid_forced ||
217                 (shp->shm_perm.mode & SHM_DEST));
218 }
219
220 /*
221  * remove the attach descriptor vma.
222  * free memory for segment if it is marked destroyed.
223  * The descriptor has already been removed from the current->mm->mmap list
224  * and will later be kfree()d.
225  */
226 static void shm_close(struct vm_area_struct *vma)
227 {
228         struct file * file = vma->vm_file;
229         struct shm_file_data *sfd = shm_file_data(file);
230         struct shmid_kernel *shp;
231         struct ipc_namespace *ns = sfd->ns;
232
233         down_write(&shm_ids(ns).rw_mutex);
234         /* remove from the list of attaches of the shm segment */
235         shp = shm_lock(ns, sfd->id);
236         BUG_ON(IS_ERR(shp));
237         shp->shm_lprid = task_tgid_vnr(current);
238         shp->shm_dtim = get_seconds();
239         shp->shm_nattch--;
240         if (shm_may_destroy(ns, shp))
241                 shm_destroy(ns, shp);
242         else
243                 shm_unlock(shp);
244         up_write(&shm_ids(ns).rw_mutex);
245 }
246
247 /* Called with ns->shm_ids(ns).rw_mutex locked */
248 static int shm_try_destroy_current(int id, void *p, void *data)
249 {
250         struct ipc_namespace *ns = data;
251         struct kern_ipc_perm *ipcp = p;
252         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
253
254         if (shp->shm_creator != current)
255                 return 0;
256
257         /*
258          * Mark it as orphaned to destroy the segment when
259          * kernel.shm_rmid_forced is changed.
260          * It is noop if the following shm_may_destroy() returns true.
261          */
262         shp->shm_creator = NULL;
263
264         /*
265          * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
266          * is not set, it shouldn't be deleted here.
267          */
268         if (!ns->shm_rmid_forced)
269                 return 0;
270
271         if (shm_may_destroy(ns, shp)) {
272                 shm_lock_by_ptr(shp);
273                 shm_destroy(ns, shp);
274         }
275         return 0;
276 }
277
278 /* Called with ns->shm_ids(ns).rw_mutex locked */
279 static int shm_try_destroy_orphaned(int id, void *p, void *data)
280 {
281         struct ipc_namespace *ns = data;
282         struct kern_ipc_perm *ipcp = p;
283         struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
284
285         /*
286          * We want to destroy segments without users and with already
287          * exit'ed originating process.
288          *
289          * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
290          */
291         if (shp->shm_creator != NULL)
292                 return 0;
293
294         if (shm_may_destroy(ns, shp)) {
295                 shm_lock_by_ptr(shp);
296                 shm_destroy(ns, shp);
297         }
298         return 0;
299 }
300
301 void shm_destroy_orphaned(struct ipc_namespace *ns)
302 {
303         down_write(&shm_ids(ns).rw_mutex);
304         if (shm_ids(ns).in_use)
305                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
306         up_write(&shm_ids(ns).rw_mutex);
307 }
308
309
310 void exit_shm(struct task_struct *task)
311 {
312         struct ipc_namespace *ns = task->nsproxy->ipc_ns;
313
314         if (shm_ids(ns).in_use == 0)
315                 return;
316
317         /* Destroy all already created segments, but not mapped yet */
318         down_write(&shm_ids(ns).rw_mutex);
319         if (shm_ids(ns).in_use)
320                 idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
321         up_write(&shm_ids(ns).rw_mutex);
322 }
323
324 static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
325 {
326         struct file *file = vma->vm_file;
327         struct shm_file_data *sfd = shm_file_data(file);
328
329         return sfd->vm_ops->fault(vma, vmf);
330 }
331
332 #ifdef CONFIG_NUMA
333 static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
334 {
335         struct file *file = vma->vm_file;
336         struct shm_file_data *sfd = shm_file_data(file);
337         int err = 0;
338         if (sfd->vm_ops->set_policy)
339                 err = sfd->vm_ops->set_policy(vma, new);
340         return err;
341 }
342
343 static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
344                                         unsigned long addr)
345 {
346         struct file *file = vma->vm_file;
347         struct shm_file_data *sfd = shm_file_data(file);
348         struct mempolicy *pol = NULL;
349
350         if (sfd->vm_ops->get_policy)
351                 pol = sfd->vm_ops->get_policy(vma, addr);
352         else if (vma->vm_policy)
353                 pol = vma->vm_policy;
354
355         return pol;
356 }
357 #endif
358
359 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
360 {
361         struct shm_file_data *sfd = shm_file_data(file);
362         int ret;
363
364         ret = sfd->file->f_op->mmap(sfd->file, vma);
365         if (ret != 0)
366                 return ret;
367         sfd->vm_ops = vma->vm_ops;
368 #ifdef CONFIG_MMU
369         BUG_ON(!sfd->vm_ops->fault);
370 #endif
371         vma->vm_ops = &shm_vm_ops;
372         shm_open(vma);
373
374         return ret;
375 }
376
377 static int shm_release(struct inode *ino, struct file *file)
378 {
379         struct shm_file_data *sfd = shm_file_data(file);
380
381         put_ipc_ns(sfd->ns);
382         shm_file_data(file) = NULL;
383         kfree(sfd);
384         return 0;
385 }
386
387 static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
388 {
389         struct shm_file_data *sfd = shm_file_data(file);
390
391         if (!sfd->file->f_op->fsync)
392                 return -EINVAL;
393         return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
394 }
395
396 static long shm_fallocate(struct file *file, int mode, loff_t offset,
397                           loff_t len)
398 {
399         struct shm_file_data *sfd = shm_file_data(file);
400
401         if (!sfd->file->f_op->fallocate)
402                 return -EOPNOTSUPP;
403         return sfd->file->f_op->fallocate(file, mode, offset, len);
404 }
405
406 static unsigned long shm_get_unmapped_area(struct file *file,
407         unsigned long addr, unsigned long len, unsigned long pgoff,
408         unsigned long flags)
409 {
410         struct shm_file_data *sfd = shm_file_data(file);
411         return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
412                                                 pgoff, flags);
413 }
414
415 static const struct file_operations shm_file_operations = {
416         .mmap           = shm_mmap,
417         .fsync          = shm_fsync,
418         .release        = shm_release,
419 #ifndef CONFIG_MMU
420         .get_unmapped_area      = shm_get_unmapped_area,
421 #endif
422         .llseek         = noop_llseek,
423         .fallocate      = shm_fallocate,
424 };
425
426 static const struct file_operations shm_file_operations_huge = {
427         .mmap           = shm_mmap,
428         .fsync          = shm_fsync,
429         .release        = shm_release,
430         .get_unmapped_area      = shm_get_unmapped_area,
431         .llseek         = noop_llseek,
432         .fallocate      = shm_fallocate,
433 };
434
435 int is_file_shm_hugepages(struct file *file)
436 {
437         return file->f_op == &shm_file_operations_huge;
438 }
439
440 static const struct vm_operations_struct shm_vm_ops = {
441         .open   = shm_open,     /* callback for a new vm-area open */
442         .close  = shm_close,    /* callback for when the vm-area is released */
443         .fault  = shm_fault,
444 #if defined(CONFIG_NUMA)
445         .set_policy = shm_set_policy,
446         .get_policy = shm_get_policy,
447 #endif
448 };
449
450 /**
451  * newseg - Create a new shared memory segment
452  * @ns: namespace
453  * @params: ptr to the structure that contains key, size and shmflg
454  *
455  * Called with shm_ids.rw_mutex held as a writer.
456  */
457
458 static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
459 {
460         key_t key = params->key;
461         int shmflg = params->flg;
462         size_t size = params->u.size;
463         int error;
464         struct shmid_kernel *shp;
465         size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
466         struct file * file;
467         char name[13];
468         int id;
469         vm_flags_t acctflag = 0;
470
471         if (size < SHMMIN || size > ns->shm_ctlmax)
472                 return -EINVAL;
473
474         if (ns->shm_tot + numpages > ns->shm_ctlall)
475                 return -ENOSPC;
476
477         shp = ipc_rcu_alloc(sizeof(*shp));
478         if (!shp)
479                 return -ENOMEM;
480
481         shp->shm_perm.key = key;
482         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
483         shp->mlock_user = NULL;
484
485         shp->shm_perm.security = NULL;
486         error = security_shm_alloc(shp);
487         if (error) {
488                 ipc_rcu_putref(shp);
489                 return error;
490         }
491
492         sprintf (name, "SYSV%08x", key);
493         if (shmflg & SHM_HUGETLB) {
494                 struct hstate *hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT)
495                                                 & SHM_HUGE_MASK);
496                 size_t hugesize;
497
498                 if (!hs) {
499                         error = -EINVAL;
500                         goto no_file;
501                 }
502                 hugesize = ALIGN(size, huge_page_size(hs));
503
504                 /* hugetlb_file_setup applies strict accounting */
505                 if (shmflg & SHM_NORESERVE)
506                         acctflag = VM_NORESERVE;
507                 file = hugetlb_file_setup(name, hugesize, acctflag,
508                                   &shp->mlock_user, HUGETLB_SHMFS_INODE,
509                                 (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
510         } else {
511                 /*
512                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
513                  * if it's asked for.
514                  */
515                 if  ((shmflg & SHM_NORESERVE) &&
516                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
517                         acctflag = VM_NORESERVE;
518                 file = shmem_file_setup(name, size, acctflag);
519         }
520         error = PTR_ERR(file);
521         if (IS_ERR(file))
522                 goto no_file;
523
524         id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
525         if (id < 0) {
526                 error = id;
527                 goto no_id;
528         }
529
530         shp->shm_cprid = task_tgid_vnr(current);
531         shp->shm_lprid = 0;
532         shp->shm_atim = shp->shm_dtim = 0;
533         shp->shm_ctim = get_seconds();
534         shp->shm_segsz = size;
535         shp->shm_nattch = 0;
536         shp->shm_file = file;
537         shp->shm_creator = current;
538
539         /*
540          * shmid gets reported as "inode#" in /proc/pid/maps.
541          * proc-ps tools use this. Changing this will break them.
542          */
543         file_inode(file)->i_ino = shp->shm_perm.id;
544
545         ns->shm_tot += numpages;
546         error = shp->shm_perm.id;
547
548         ipc_unlock_object(&shp->shm_perm);
549         rcu_read_unlock();
550         return error;
551
552 no_id:
553         if (is_file_hugepages(file) && shp->mlock_user)
554                 user_shm_unlock(size, shp->mlock_user);
555         fput(file);
556 no_file:
557         security_shm_free(shp);
558         ipc_rcu_putref(shp);
559         return error;
560 }
561
562 /*
563  * Called with shm_ids.rw_mutex and ipcp locked.
564  */
565 static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
566 {
567         struct shmid_kernel *shp;
568
569         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
570         return security_shm_associate(shp, shmflg);
571 }
572
573 /*
574  * Called with shm_ids.rw_mutex and ipcp locked.
575  */
576 static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
577                                 struct ipc_params *params)
578 {
579         struct shmid_kernel *shp;
580
581         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
582         if (shp->shm_segsz < params->u.size)
583                 return -EINVAL;
584
585         return 0;
586 }
587
588 SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
589 {
590         struct ipc_namespace *ns;
591         struct ipc_ops shm_ops;
592         struct ipc_params shm_params;
593
594         ns = current->nsproxy->ipc_ns;
595
596         shm_ops.getnew = newseg;
597         shm_ops.associate = shm_security;
598         shm_ops.more_checks = shm_more_checks;
599
600         shm_params.key = key;
601         shm_params.flg = shmflg;
602         shm_params.u.size = size;
603
604         return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
605 }
606
607 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
608 {
609         switch(version) {
610         case IPC_64:
611                 return copy_to_user(buf, in, sizeof(*in));
612         case IPC_OLD:
613             {
614                 struct shmid_ds out;
615
616                 memset(&out, 0, sizeof(out));
617                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
618                 out.shm_segsz   = in->shm_segsz;
619                 out.shm_atime   = in->shm_atime;
620                 out.shm_dtime   = in->shm_dtime;
621                 out.shm_ctime   = in->shm_ctime;
622                 out.shm_cpid    = in->shm_cpid;
623                 out.shm_lpid    = in->shm_lpid;
624                 out.shm_nattch  = in->shm_nattch;
625
626                 return copy_to_user(buf, &out, sizeof(out));
627             }
628         default:
629                 return -EINVAL;
630         }
631 }
632
633 static inline unsigned long
634 copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
635 {
636         switch(version) {
637         case IPC_64:
638                 if (copy_from_user(out, buf, sizeof(*out)))
639                         return -EFAULT;
640                 return 0;
641         case IPC_OLD:
642             {
643                 struct shmid_ds tbuf_old;
644
645                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
646                         return -EFAULT;
647
648                 out->shm_perm.uid       = tbuf_old.shm_perm.uid;
649                 out->shm_perm.gid       = tbuf_old.shm_perm.gid;
650                 out->shm_perm.mode      = tbuf_old.shm_perm.mode;
651
652                 return 0;
653             }
654         default:
655                 return -EINVAL;
656         }
657 }
658
659 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
660 {
661         switch(version) {
662         case IPC_64:
663                 return copy_to_user(buf, in, sizeof(*in));
664         case IPC_OLD:
665             {
666                 struct shminfo out;
667
668                 if(in->shmmax > INT_MAX)
669                         out.shmmax = INT_MAX;
670                 else
671                         out.shmmax = (int)in->shmmax;
672
673                 out.shmmin      = in->shmmin;
674                 out.shmmni      = in->shmmni;
675                 out.shmseg      = in->shmseg;
676                 out.shmall      = in->shmall; 
677
678                 return copy_to_user(buf, &out, sizeof(out));
679             }
680         default:
681                 return -EINVAL;
682         }
683 }
684
685 /*
686  * Calculate and add used RSS and swap pages of a shm.
687  * Called with shm_ids.rw_mutex held as a reader
688  */
689 static void shm_add_rss_swap(struct shmid_kernel *shp,
690         unsigned long *rss_add, unsigned long *swp_add)
691 {
692         struct inode *inode;
693
694         inode = file_inode(shp->shm_file);
695
696         if (is_file_hugepages(shp->shm_file)) {
697                 struct address_space *mapping = inode->i_mapping;
698                 struct hstate *h = hstate_file(shp->shm_file);
699                 *rss_add += pages_per_huge_page(h) * mapping->nrpages;
700         } else {
701 #ifdef CONFIG_SHMEM
702                 struct shmem_inode_info *info = SHMEM_I(inode);
703                 spin_lock(&info->lock);
704                 *rss_add += inode->i_mapping->nrpages;
705                 *swp_add += info->swapped;
706                 spin_unlock(&info->lock);
707 #else
708                 *rss_add += inode->i_mapping->nrpages;
709 #endif
710         }
711 }
712
713 /*
714  * Called with shm_ids.rw_mutex held as a reader
715  */
716 static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
717                 unsigned long *swp)
718 {
719         int next_id;
720         int total, in_use;
721
722         *rss = 0;
723         *swp = 0;
724
725         in_use = shm_ids(ns).in_use;
726
727         for (total = 0, next_id = 0; total < in_use; next_id++) {
728                 struct kern_ipc_perm *ipc;
729                 struct shmid_kernel *shp;
730
731                 ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
732                 if (ipc == NULL)
733                         continue;
734                 shp = container_of(ipc, struct shmid_kernel, shm_perm);
735
736                 shm_add_rss_swap(shp, rss, swp);
737
738                 total++;
739         }
740 }
741
742 /*
743  * This function handles some shmctl commands which require the rw_mutex
744  * to be held in write mode.
745  * NOTE: no locks must be held, the rw_mutex is taken inside this function.
746  */
747 static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
748                        struct shmid_ds __user *buf, int version)
749 {
750         struct kern_ipc_perm *ipcp;
751         struct shmid64_ds shmid64;
752         struct shmid_kernel *shp;
753         int err;
754
755         if (cmd == IPC_SET) {
756                 if (copy_shmid_from_user(&shmid64, buf, version))
757                         return -EFAULT;
758         }
759
760         down_write(&shm_ids(ns).rw_mutex);
761         rcu_read_lock();
762
763         ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
764                                &shmid64.shm_perm, 0);
765         if (IS_ERR(ipcp)) {
766                 err = PTR_ERR(ipcp);
767                 /* the ipc lock is not held upon failure */
768                 goto out_unlock1;
769         }
770
771         shp = container_of(ipcp, struct shmid_kernel, shm_perm);
772
773         err = security_shm_shmctl(shp, cmd);
774         if (err)
775                 goto out_unlock0;
776
777         switch (cmd) {
778         case IPC_RMID:
779                 /* do_shm_rmid unlocks the ipc object and rcu */
780                 do_shm_rmid(ns, ipcp);
781                 goto out_up;
782         case IPC_SET:
783                 err = ipc_update_perm(&shmid64.shm_perm, ipcp);
784                 if (err)
785                         goto out_unlock0;
786                 shp->shm_ctim = get_seconds();
787                 break;
788         default:
789                 err = -EINVAL;
790         }
791
792 out_unlock0:
793         ipc_unlock_object(&shp->shm_perm);
794 out_unlock1:
795         rcu_read_unlock();
796 out_up:
797         up_write(&shm_ids(ns).rw_mutex);
798         return err;
799 }
800
801 SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
802 {
803         struct shmid_kernel *shp;
804         int err, version;
805         struct ipc_namespace *ns;
806
807         if (cmd < 0 || shmid < 0) {
808                 err = -EINVAL;
809                 goto out;
810         }
811
812         version = ipc_parse_version(&cmd);
813         ns = current->nsproxy->ipc_ns;
814
815         switch (cmd) { /* replace with proc interface ? */
816         case IPC_INFO:
817         {
818                 struct shminfo64 shminfo;
819
820                 err = security_shm_shmctl(NULL, cmd);
821                 if (err)
822                         return err;
823
824                 memset(&shminfo, 0, sizeof(shminfo));
825                 shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
826                 shminfo.shmmax = ns->shm_ctlmax;
827                 shminfo.shmall = ns->shm_ctlall;
828
829                 shminfo.shmmin = SHMMIN;
830                 if(copy_shminfo_to_user (buf, &shminfo, version))
831                         return -EFAULT;
832
833                 down_read(&shm_ids(ns).rw_mutex);
834                 err = ipc_get_maxid(&shm_ids(ns));
835                 up_read(&shm_ids(ns).rw_mutex);
836
837                 if(err<0)
838                         err = 0;
839                 goto out;
840         }
841         case SHM_INFO:
842         {
843                 struct shm_info shm_info;
844
845                 err = security_shm_shmctl(NULL, cmd);
846                 if (err)
847                         return err;
848
849                 memset(&shm_info, 0, sizeof(shm_info));
850                 down_read(&shm_ids(ns).rw_mutex);
851                 shm_info.used_ids = shm_ids(ns).in_use;
852                 shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
853                 shm_info.shm_tot = ns->shm_tot;
854                 shm_info.swap_attempts = 0;
855                 shm_info.swap_successes = 0;
856                 err = ipc_get_maxid(&shm_ids(ns));
857                 up_read(&shm_ids(ns).rw_mutex);
858                 if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
859                         err = -EFAULT;
860                         goto out;
861                 }
862
863                 err = err < 0 ? 0 : err;
864                 goto out;
865         }
866         case SHM_STAT:
867         case IPC_STAT:
868         {
869                 struct shmid64_ds tbuf;
870                 int result;
871
872                 if (cmd == SHM_STAT) {
873                         shp = shm_lock(ns, shmid);
874                         if (IS_ERR(shp)) {
875                                 err = PTR_ERR(shp);
876                                 goto out;
877                         }
878                         result = shp->shm_perm.id;
879                 } else {
880                         shp = shm_lock_check(ns, shmid);
881                         if (IS_ERR(shp)) {
882                                 err = PTR_ERR(shp);
883                                 goto out;
884                         }
885                         result = 0;
886                 }
887                 err = -EACCES;
888                 if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
889                         goto out_unlock;
890                 err = security_shm_shmctl(shp, cmd);
891                 if (err)
892                         goto out_unlock;
893                 memset(&tbuf, 0, sizeof(tbuf));
894                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
895                 tbuf.shm_segsz  = shp->shm_segsz;
896                 tbuf.shm_atime  = shp->shm_atim;
897                 tbuf.shm_dtime  = shp->shm_dtim;
898                 tbuf.shm_ctime  = shp->shm_ctim;
899                 tbuf.shm_cpid   = shp->shm_cprid;
900                 tbuf.shm_lpid   = shp->shm_lprid;
901                 tbuf.shm_nattch = shp->shm_nattch;
902                 shm_unlock(shp);
903                 if(copy_shmid_to_user (buf, &tbuf, version))
904                         err = -EFAULT;
905                 else
906                         err = result;
907                 goto out;
908         }
909         case SHM_LOCK:
910         case SHM_UNLOCK:
911         {
912                 struct file *shm_file;
913
914                 shp = shm_lock_check(ns, shmid);
915                 if (IS_ERR(shp)) {
916                         err = PTR_ERR(shp);
917                         goto out;
918                 }
919
920                 audit_ipc_obj(&(shp->shm_perm));
921
922                 if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
923                         kuid_t euid = current_euid();
924                         err = -EPERM;
925                         if (!uid_eq(euid, shp->shm_perm.uid) &&
926                             !uid_eq(euid, shp->shm_perm.cuid))
927                                 goto out_unlock;
928                         if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
929                                 goto out_unlock;
930                 }
931
932                 err = security_shm_shmctl(shp, cmd);
933                 if (err)
934                         goto out_unlock;
935
936                 shm_file = shp->shm_file;
937                 if (is_file_hugepages(shm_file))
938                         goto out_unlock;
939
940                 if (cmd == SHM_LOCK) {
941                         struct user_struct *user = current_user();
942                         err = shmem_lock(shm_file, 1, user);
943                         if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
944                                 shp->shm_perm.mode |= SHM_LOCKED;
945                                 shp->mlock_user = user;
946                         }
947                         goto out_unlock;
948                 }
949
950                 /* SHM_UNLOCK */
951                 if (!(shp->shm_perm.mode & SHM_LOCKED))
952                         goto out_unlock;
953                 shmem_lock(shm_file, 0, shp->mlock_user);
954                 shp->shm_perm.mode &= ~SHM_LOCKED;
955                 shp->mlock_user = NULL;
956                 get_file(shm_file);
957                 shm_unlock(shp);
958                 shmem_unlock_mapping(shm_file->f_mapping);
959                 fput(shm_file);
960                 goto out;
961         }
962         case IPC_RMID:
963         case IPC_SET:
964                 err = shmctl_down(ns, shmid, cmd, buf, version);
965                 return err;
966         default:
967                 return -EINVAL;
968         }
969
970 out_unlock:
971         shm_unlock(shp);
972 out:
973         return err;
974 }
975
976 /*
977  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
978  *
979  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
980  * "raddr" thing points to kernel space, and there has to be a wrapper around
981  * this.
982  */
983 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
984               unsigned long shmlba)
985 {
986         struct shmid_kernel *shp;
987         unsigned long addr;
988         unsigned long size;
989         struct file * file;
990         int    err;
991         unsigned long flags;
992         unsigned long prot;
993         int acc_mode;
994         struct ipc_namespace *ns;
995         struct shm_file_data *sfd;
996         struct path path;
997         fmode_t f_mode;
998         unsigned long populate = 0;
999
1000         err = -EINVAL;
1001         if (shmid < 0)
1002                 goto out;
1003         else if ((addr = (ulong)shmaddr)) {
1004                 if (addr & (shmlba - 1)) {
1005                         if (shmflg & SHM_RND)
1006                                 addr &= ~(shmlba - 1);     /* round down */
1007                         else
1008 #ifndef __ARCH_FORCE_SHMLBA
1009                                 if (addr & ~PAGE_MASK)
1010 #endif
1011                                         goto out;
1012                 }
1013                 flags = MAP_SHARED | MAP_FIXED;
1014         } else {
1015                 if ((shmflg & SHM_REMAP))
1016                         goto out;
1017
1018                 flags = MAP_SHARED;
1019         }
1020
1021         if (shmflg & SHM_RDONLY) {
1022                 prot = PROT_READ;
1023                 acc_mode = S_IRUGO;
1024                 f_mode = FMODE_READ;
1025         } else {
1026                 prot = PROT_READ | PROT_WRITE;
1027                 acc_mode = S_IRUGO | S_IWUGO;
1028                 f_mode = FMODE_READ | FMODE_WRITE;
1029         }
1030         if (shmflg & SHM_EXEC) {
1031                 prot |= PROT_EXEC;
1032                 acc_mode |= S_IXUGO;
1033         }
1034
1035         /*
1036          * We cannot rely on the fs check since SYSV IPC does have an
1037          * additional creator id...
1038          */
1039         ns = current->nsproxy->ipc_ns;
1040         shp = shm_lock_check(ns, shmid);
1041         if (IS_ERR(shp)) {
1042                 err = PTR_ERR(shp);
1043                 goto out;
1044         }
1045
1046         err = -EACCES;
1047         if (ipcperms(ns, &shp->shm_perm, acc_mode))
1048                 goto out_unlock;
1049
1050         err = security_shm_shmat(shp, shmaddr, shmflg);
1051         if (err)
1052                 goto out_unlock;
1053
1054         path = shp->shm_file->f_path;
1055         path_get(&path);
1056         shp->shm_nattch++;
1057         size = i_size_read(path.dentry->d_inode);
1058         shm_unlock(shp);
1059
1060         err = -ENOMEM;
1061         sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1062         if (!sfd)
1063                 goto out_put_dentry;
1064
1065         file = alloc_file(&path, f_mode,
1066                           is_file_hugepages(shp->shm_file) ?
1067                                 &shm_file_operations_huge :
1068                                 &shm_file_operations);
1069         err = PTR_ERR(file);
1070         if (IS_ERR(file))
1071                 goto out_free;
1072
1073         file->private_data = sfd;
1074         file->f_mapping = shp->shm_file->f_mapping;
1075         sfd->id = shp->shm_perm.id;
1076         sfd->ns = get_ipc_ns(ns);
1077         sfd->file = shp->shm_file;
1078         sfd->vm_ops = NULL;
1079
1080         err = security_mmap_file(file, prot, flags);
1081         if (err)
1082                 goto out_fput;
1083
1084         down_write(&current->mm->mmap_sem);
1085         if (addr && !(shmflg & SHM_REMAP)) {
1086                 err = -EINVAL;
1087                 if (find_vma_intersection(current->mm, addr, addr + size))
1088                         goto invalid;
1089                 /*
1090                  * If shm segment goes below stack, make sure there is some
1091                  * space left for the stack to grow (at least 4 pages).
1092                  */
1093                 if (addr < current->mm->start_stack &&
1094                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1095                         goto invalid;
1096         }
1097                 
1098         addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1099         *raddr = addr;
1100         err = 0;
1101         if (IS_ERR_VALUE(addr))
1102                 err = (long)addr;
1103 invalid:
1104         up_write(&current->mm->mmap_sem);
1105         if (populate)
1106                 mm_populate(addr, populate);
1107
1108 out_fput:
1109         fput(file);
1110
1111 out_nattch:
1112         down_write(&shm_ids(ns).rw_mutex);
1113         shp = shm_lock(ns, shmid);
1114         BUG_ON(IS_ERR(shp));
1115         shp->shm_nattch--;
1116         if (shm_may_destroy(ns, shp))
1117                 shm_destroy(ns, shp);
1118         else
1119                 shm_unlock(shp);
1120         up_write(&shm_ids(ns).rw_mutex);
1121
1122 out:
1123         return err;
1124
1125 out_unlock:
1126         shm_unlock(shp);
1127         goto out;
1128
1129 out_free:
1130         kfree(sfd);
1131 out_put_dentry:
1132         path_put(&path);
1133         goto out_nattch;
1134 }
1135
1136 SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1137 {
1138         unsigned long ret;
1139         long err;
1140
1141         err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1142         if (err)
1143                 return err;
1144         force_successful_syscall_return();
1145         return (long)ret;
1146 }
1147
1148 /*
1149  * detach and kill segment if marked destroyed.
1150  * The work is done in shm_close.
1151  */
1152 SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1153 {
1154         struct mm_struct *mm = current->mm;
1155         struct vm_area_struct *vma;
1156         unsigned long addr = (unsigned long)shmaddr;
1157         int retval = -EINVAL;
1158 #ifdef CONFIG_MMU
1159         loff_t size = 0;
1160         struct vm_area_struct *next;
1161 #endif
1162
1163         if (addr & ~PAGE_MASK)
1164                 return retval;
1165
1166         down_write(&mm->mmap_sem);
1167
1168         /*
1169          * This function tries to be smart and unmap shm segments that
1170          * were modified by partial mlock or munmap calls:
1171          * - It first determines the size of the shm segment that should be
1172          *   unmapped: It searches for a vma that is backed by shm and that
1173          *   started at address shmaddr. It records it's size and then unmaps
1174          *   it.
1175          * - Then it unmaps all shm vmas that started at shmaddr and that
1176          *   are within the initially determined size.
1177          * Errors from do_munmap are ignored: the function only fails if
1178          * it's called with invalid parameters or if it's called to unmap
1179          * a part of a vma. Both calls in this function are for full vmas,
1180          * the parameters are directly copied from the vma itself and always
1181          * valid - therefore do_munmap cannot fail. (famous last words?)
1182          */
1183         /*
1184          * If it had been mremap()'d, the starting address would not
1185          * match the usual checks anyway. So assume all vma's are
1186          * above the starting address given.
1187          */
1188         vma = find_vma(mm, addr);
1189
1190 #ifdef CONFIG_MMU
1191         while (vma) {
1192                 next = vma->vm_next;
1193
1194                 /*
1195                  * Check if the starting address would match, i.e. it's
1196                  * a fragment created by mprotect() and/or munmap(), or it
1197                  * otherwise it starts at this address with no hassles.
1198                  */
1199                 if ((vma->vm_ops == &shm_vm_ops) &&
1200                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1201
1202
1203                         size = file_inode(vma->vm_file)->i_size;
1204                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1205                         /*
1206                          * We discovered the size of the shm segment, so
1207                          * break out of here and fall through to the next
1208                          * loop that uses the size information to stop
1209                          * searching for matching vma's.
1210                          */
1211                         retval = 0;
1212                         vma = next;
1213                         break;
1214                 }
1215                 vma = next;
1216         }
1217
1218         /*
1219          * We need look no further than the maximum address a fragment
1220          * could possibly have landed at. Also cast things to loff_t to
1221          * prevent overflows and make comparisons vs. equal-width types.
1222          */
1223         size = PAGE_ALIGN(size);
1224         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1225                 next = vma->vm_next;
1226
1227                 /* finding a matching vma now does not alter retval */
1228                 if ((vma->vm_ops == &shm_vm_ops) &&
1229                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1230
1231                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1232                 vma = next;
1233         }
1234
1235 #else /* CONFIG_MMU */
1236         /* under NOMMU conditions, the exact address to be destroyed must be
1237          * given */
1238         retval = -EINVAL;
1239         if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1240                 do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1241                 retval = 0;
1242         }
1243
1244 #endif
1245
1246         up_write(&mm->mmap_sem);
1247         return retval;
1248 }
1249
1250 #ifdef CONFIG_PROC_FS
1251 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1252 {
1253         struct user_namespace *user_ns = seq_user_ns(s);
1254         struct shmid_kernel *shp = it;
1255         unsigned long rss = 0, swp = 0;
1256
1257         shm_add_rss_swap(shp, &rss, &swp);
1258
1259 #if BITS_PER_LONG <= 32
1260 #define SIZE_SPEC "%10lu"
1261 #else
1262 #define SIZE_SPEC "%21lu"
1263 #endif
1264
1265         return seq_printf(s,
1266                           "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1267                           "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1268                           SIZE_SPEC " " SIZE_SPEC "\n",
1269                           shp->shm_perm.key,
1270                           shp->shm_perm.id,
1271                           shp->shm_perm.mode,
1272                           shp->shm_segsz,
1273                           shp->shm_cprid,
1274                           shp->shm_lprid,
1275                           shp->shm_nattch,
1276                           from_kuid_munged(user_ns, shp->shm_perm.uid),
1277                           from_kgid_munged(user_ns, shp->shm_perm.gid),
1278                           from_kuid_munged(user_ns, shp->shm_perm.cuid),
1279                           from_kgid_munged(user_ns, shp->shm_perm.cgid),
1280                           shp->shm_atim,
1281                           shp->shm_dtim,
1282                           shp->shm_ctim,
1283                           rss * PAGE_SIZE,
1284                           swp * PAGE_SIZE);
1285 }
1286 #endif