msg.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983
  1. /*
  2. * linux/ipc/msg.c
  3. * Copyright (C) 1992 Krishna Balasubramanian
  4. *
  5. * Removed all the remaining kerneld mess
  6. * Catch the -EFAULT stuff properly
  7. * Use GFP_KERNEL for messages as in 1.2
  8. * Fixed up the unchecked user space derefs
  9. * Copyright (C) 1998 Alan Cox & Andi Kleen
  10. *
  11. * /proc/sysvipc/msg support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
  12. *
  13. * mostly rewritten, threaded and wake-one semantics added
  14. * MSGMAX limit removed, sysctl's added
  15. * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
  16. *
  17. * support for audit of ipc object properties and permission changes
  18. * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  19. *
  20. * namespaces support
  21. * OpenVZ, SWsoft Inc.
  22. * Pavel Emelianov <xemul@openvz.org>
  23. */
  24. #include <linux/capability.h>
  25. #include <linux/slab.h>
  26. #include <linux/msg.h>
  27. #include <linux/spinlock.h>
  28. #include <linux/init.h>
  29. #include <linux/mm.h>
  30. #include <linux/proc_fs.h>
  31. #include <linux/list.h>
  32. #include <linux/security.h>
  33. #include <linux/sched.h>
  34. #include <linux/syscalls.h>
  35. #include <linux/audit.h>
  36. #include <linux/seq_file.h>
  37. #include <linux/rwsem.h>
  38. #include <linux/nsproxy.h>
  39. #include <linux/ipc_namespace.h>
  40. #include <asm/current.h>
  41. #include <asm/uaccess.h>
  42. #include "util.h"
  43. /*
  44. * one msg_receiver structure for each sleeping receiver:
  45. */
  46. struct msg_receiver {
  47. struct list_head r_list;
  48. struct task_struct *r_tsk;
  49. int r_mode;
  50. long r_msgtype;
  51. long r_maxsize;
  52. struct msg_msg *volatile r_msg;
  53. };
  54. /* one msg_sender for each sleeping sender */
  55. struct msg_sender {
  56. struct list_head list;
  57. struct task_struct *tsk;
  58. };
  59. #define SEARCH_ANY 1
  60. #define SEARCH_EQUAL 2
  61. #define SEARCH_NOTEQUAL 3
  62. #define SEARCH_LESSEQUAL 4
  63. #define msg_ids(ns) ((ns)->ids[IPC_MSG_IDS])
  64. #define msg_unlock(msq) ipc_unlock(&(msq)->q_perm)
  65. static void freeque(struct ipc_namespace *, struct kern_ipc_perm *);
  66. static int newque(struct ipc_namespace *, struct ipc_params *);
  67. #ifdef CONFIG_PROC_FS
  68. static int sysvipc_msg_proc_show(struct seq_file *s, void *it);
  69. #endif
  70. /*
  71. * Scale msgmni with the available lowmem size: the memory dedicated to msg
  72. * queues should occupy at most 1/MSG_MEM_SCALE of lowmem.
  73. * Also take into account the number of nsproxies created so far.
  74. * This should be done staying within the (MSGMNI , IPCMNI/nr_ipc_ns) range.
  75. */
  76. void recompute_msgmni(struct ipc_namespace *ns)
  77. {
  78. struct sysinfo i;
  79. unsigned long allowed;
  80. int nb_ns;
  81. si_meminfo(&i);
  82. allowed = (((i.totalram - i.totalhigh) / MSG_MEM_SCALE) * i.mem_unit)
  83. / MSGMNB;
  84. nb_ns = atomic_read(&nr_ipc_ns);
  85. allowed /= nb_ns;
  86. if (allowed < MSGMNI) {
  87. ns->msg_ctlmni = MSGMNI;
  88. goto out_callback;
  89. }
  90. if (allowed > IPCMNI / nb_ns) {
  91. ns->msg_ctlmni = IPCMNI / nb_ns;
  92. goto out_callback;
  93. }
  94. ns->msg_ctlmni = allowed;
  95. out_callback:
  96. printk(KERN_INFO "msgmni has been set to %d for ipc namespace %p\n",
  97. ns->msg_ctlmni, ns);
  98. }
  99. void msg_init_ns(struct ipc_namespace *ns)
  100. {
  101. ns->msg_ctlmax = MSGMAX;
  102. ns->msg_ctlmnb = MSGMNB;
  103. recompute_msgmni(ns);
  104. atomic_set(&ns->msg_bytes, 0);
  105. atomic_set(&ns->msg_hdrs, 0);
  106. ipc_init_ids(&ns->ids[IPC_MSG_IDS]);
  107. }
  108. #ifdef CONFIG_IPC_NS
  109. void msg_exit_ns(struct ipc_namespace *ns)
  110. {
  111. free_ipcs(ns, &msg_ids(ns), freeque);
  112. }
  113. #endif
  114. void __init msg_init(void)
  115. {
  116. msg_init_ns(&init_ipc_ns);
  117. ipc_init_proc_interface("sysvipc/msg",
  118. " key msqid perms cbytes qnum lspid lrpid uid gid cuid cgid stime rtime ctime\n",
  119. IPC_MSG_IDS, sysvipc_msg_proc_show);
  120. }
  121. /*
  122. * This routine is called in the paths where the rw_mutex is held to protect
  123. * access to the idr tree.
  124. */
  125. static inline struct msg_queue *msg_lock_check_down(struct ipc_namespace *ns,
  126. int id)
  127. {
  128. struct kern_ipc_perm *ipcp = ipc_lock_check_down(&msg_ids(ns), id);
  129. if (IS_ERR(ipcp))
  130. return (struct msg_queue *)ipcp;
  131. return container_of(ipcp, struct msg_queue, q_perm);
  132. }
  133. /*
  134. * msg_lock_(check_) routines are called in the paths where the rw_mutex
  135. * is not held.
  136. */
  137. static inline struct msg_queue *msg_lock(struct ipc_namespace *ns, int id)
  138. {
  139. struct kern_ipc_perm *ipcp = ipc_lock(&msg_ids(ns), id);
  140. if (IS_ERR(ipcp))
  141. return (struct msg_queue *)ipcp;
  142. return container_of(ipcp, struct msg_queue, q_perm);
  143. }
  144. static inline struct msg_queue *msg_lock_check(struct ipc_namespace *ns,
  145. int id)
  146. {
  147. struct kern_ipc_perm *ipcp = ipc_lock_check(&msg_ids(ns), id);
  148. if (IS_ERR(ipcp))
  149. return (struct msg_queue *)ipcp;
  150. return container_of(ipcp, struct msg_queue, q_perm);
  151. }
  152. static inline void msg_rmid(struct ipc_namespace *ns, struct msg_queue *s)
  153. {
  154. ipc_rmid(&msg_ids(ns), &s->q_perm);
  155. }
  156. /**
  157. * newque - Create a new msg queue
  158. * @ns: namespace
  159. * @params: ptr to the structure that contains the key and msgflg
  160. *
  161. * Called with msg_ids.rw_mutex held (writer)
  162. */
  163. static int newque(struct ipc_namespace *ns, struct ipc_params *params)
  164. {
  165. struct msg_queue *msq;
  166. int id, retval;
  167. key_t key = params->key;
  168. int msgflg = params->flg;
  169. msq = ipc_rcu_alloc(sizeof(*msq));
  170. if (!msq)
  171. return -ENOMEM;
  172. msq->q_perm.mode = msgflg & S_IRWXUGO;
  173. msq->q_perm.key = key;
  174. msq->q_perm.security = NULL;
  175. retval = security_msg_queue_alloc(msq);
  176. if (retval) {
  177. ipc_rcu_putref(msq);
  178. return retval;
  179. }
  180. /*
  181. * ipc_addid() locks msq
  182. */
  183. id = ipc_addid(&msg_ids(ns), &msq->q_perm, ns->msg_ctlmni);
  184. if (id < 0) {
  185. security_msg_queue_free(msq);
  186. ipc_rcu_putref(msq);
  187. return id;
  188. }
  189. msq->q_stime = msq->q_rtime = 0;
  190. msq->q_ctime = get_seconds();
  191. msq->q_cbytes = msq->q_qnum = 0;
  192. msq->q_qbytes = ns->msg_ctlmnb;
  193. msq->q_lspid = msq->q_lrpid = 0;
  194. INIT_LIST_HEAD(&msq->q_messages);
  195. INIT_LIST_HEAD(&msq->q_receivers);
  196. INIT_LIST_HEAD(&msq->q_senders);
  197. msg_unlock(msq);
  198. return msq->q_perm.id;
  199. }
  200. static inline void ss_add(struct msg_queue *msq, struct msg_sender *mss)
  201. {
  202. mss->tsk = current;
  203. current->state = TASK_INTERRUPTIBLE;
  204. list_add_tail(&mss->list, &msq->q_senders);
  205. }
  206. static inline void ss_del(struct msg_sender *mss)
  207. {
  208. if (mss->list.next != NULL)
  209. list_del(&mss->list);
  210. }
  211. static void ss_wakeup(struct list_head *h, int kill)
  212. {
  213. struct list_head *tmp;
  214. tmp = h->next;
  215. while (tmp != h) {
  216. struct msg_sender *mss;
  217. mss = list_entry(tmp, struct msg_sender, list);
  218. tmp = tmp->next;
  219. if (kill)
  220. mss->list.next = NULL;
  221. wake_up_process(mss->tsk);
  222. }
  223. }
  224. static void expunge_all(struct msg_queue *msq, int res)
  225. {
  226. struct list_head *tmp;
  227. tmp = msq->q_receivers.next;
  228. while (tmp != &msq->q_receivers) {
  229. struct msg_receiver *msr;
  230. msr = list_entry(tmp, struct msg_receiver, r_list);
  231. tmp = tmp->next;
  232. msr->r_msg = NULL;
  233. wake_up_process(msr->r_tsk);
  234. smp_mb();
  235. msr->r_msg = ERR_PTR(res);
  236. }
  237. }
  238. /*
  239. * freeque() wakes up waiters on the sender and receiver waiting queue,
  240. * removes the message queue from message queue ID IDR, and cleans up all the
  241. * messages associated with this queue.
  242. *
  243. * msg_ids.rw_mutex (writer) and the spinlock for this message queue are held
  244. * before freeque() is called. msg_ids.rw_mutex remains locked on exit.
  245. */
  246. static void freeque(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  247. {
  248. struct list_head *tmp;
  249. struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
  250. expunge_all(msq, -EIDRM);
  251. ss_wakeup(&msq->q_senders, 1);
  252. msg_rmid(ns, msq);
  253. msg_unlock(msq);
  254. tmp = msq->q_messages.next;
  255. while (tmp != &msq->q_messages) {
  256. struct msg_msg *msg = list_entry(tmp, struct msg_msg, m_list);
  257. tmp = tmp->next;
  258. atomic_dec(&ns->msg_hdrs);
  259. free_msg(msg);
  260. }
  261. atomic_sub(msq->q_cbytes, &ns->msg_bytes);
  262. security_msg_queue_free(msq);
  263. ipc_rcu_putref(msq);
  264. }
  265. /*
  266. * Called with msg_ids.rw_mutex and ipcp locked.
  267. */
  268. static inline int msg_security(struct kern_ipc_perm *ipcp, int msgflg)
  269. {
  270. struct msg_queue *msq = container_of(ipcp, struct msg_queue, q_perm);
  271. return security_msg_queue_associate(msq, msgflg);
  272. }
  273. asmlinkage long sys_msgget(key_t key, int msgflg)
  274. {
  275. struct ipc_namespace *ns;
  276. struct ipc_ops msg_ops;
  277. struct ipc_params msg_params;
  278. ns = current->nsproxy->ipc_ns;
  279. msg_ops.getnew = newque;
  280. msg_ops.associate = msg_security;
  281. msg_ops.more_checks = NULL;
  282. msg_params.key = key;
  283. msg_params.flg = msgflg;
  284. return ipcget(ns, &msg_ids(ns), &msg_ops, &msg_params);
  285. }
  286. static inline unsigned long
  287. copy_msqid_to_user(void __user *buf, struct msqid64_ds *in, int version)
  288. {
  289. switch(version) {
  290. case IPC_64:
  291. return copy_to_user(buf, in, sizeof(*in));
  292. case IPC_OLD:
  293. {
  294. struct msqid_ds out;
  295. memset(&out, 0, sizeof(out));
  296. ipc64_perm_to_ipc_perm(&in->msg_perm, &out.msg_perm);
  297. out.msg_stime = in->msg_stime;
  298. out.msg_rtime = in->msg_rtime;
  299. out.msg_ctime = in->msg_ctime;
  300. if (in->msg_cbytes > USHRT_MAX)
  301. out.msg_cbytes = USHRT_MAX;
  302. else
  303. out.msg_cbytes = in->msg_cbytes;
  304. out.msg_lcbytes = in->msg_cbytes;
  305. if (in->msg_qnum > USHRT_MAX)
  306. out.msg_qnum = USHRT_MAX;
  307. else
  308. out.msg_qnum = in->msg_qnum;
  309. if (in->msg_qbytes > USHRT_MAX)
  310. out.msg_qbytes = USHRT_MAX;
  311. else
  312. out.msg_qbytes = in->msg_qbytes;
  313. out.msg_lqbytes = in->msg_qbytes;
  314. out.msg_lspid = in->msg_lspid;
  315. out.msg_lrpid = in->msg_lrpid;
  316. return copy_to_user(buf, &out, sizeof(out));
  317. }
  318. default:
  319. return -EINVAL;
  320. }
  321. }
  322. static inline unsigned long
  323. copy_msqid_from_user(struct msqid64_ds *out, void __user *buf, int version)
  324. {
  325. switch(version) {
  326. case IPC_64:
  327. if (copy_from_user(out, buf, sizeof(*out)))
  328. return -EFAULT;
  329. return 0;
  330. case IPC_OLD:
  331. {
  332. struct msqid_ds tbuf_old;
  333. if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
  334. return -EFAULT;
  335. out->msg_perm.uid = tbuf_old.msg_perm.uid;
  336. out->msg_perm.gid = tbuf_old.msg_perm.gid;
  337. out->msg_perm.mode = tbuf_old.msg_perm.mode;
  338. if (tbuf_old.msg_qbytes == 0)
  339. out->msg_qbytes = tbuf_old.msg_lqbytes;
  340. else
  341. out->msg_qbytes = tbuf_old.msg_qbytes;
  342. return 0;
  343. }
  344. default:
  345. return -EINVAL;
  346. }
  347. }
  348. /*
  349. * This function handles some msgctl commands which require the rw_mutex
  350. * to be held in write mode.
  351. * NOTE: no locks must be held, the rw_mutex is taken inside this function.
  352. */
  353. static int msgctl_down(struct ipc_namespace *ns, int msqid, int cmd,
  354. struct msqid_ds __user *buf, int version)
  355. {
  356. struct kern_ipc_perm *ipcp;
  357. struct msqid64_ds msqid64;
  358. struct msg_queue *msq;
  359. int err;
  360. if (cmd == IPC_SET) {
  361. if (copy_msqid_from_user(&msqid64, buf, version))
  362. return -EFAULT;
  363. }
  364. down_write(&msg_ids(ns).rw_mutex);
  365. msq = msg_lock_check_down(ns, msqid);
  366. if (IS_ERR(msq)) {
  367. err = PTR_ERR(msq);
  368. goto out_up;
  369. }
  370. ipcp = &msq->q_perm;
  371. err = audit_ipc_obj(ipcp);
  372. if (err)
  373. goto out_unlock;
  374. if (cmd == IPC_SET) {
  375. err = audit_ipc_set_perm(msqid64.msg_qbytes,
  376. msqid64.msg_perm.uid,
  377. msqid64.msg_perm.gid,
  378. msqid64.msg_perm.mode);
  379. if (err)
  380. goto out_unlock;
  381. }
  382. if (current->euid != ipcp->cuid &&
  383. current->euid != ipcp->uid &&
  384. !capable(CAP_SYS_ADMIN)) {
  385. /* We _could_ check for CAP_CHOWN above, but we don't */
  386. err = -EPERM;
  387. goto out_unlock;
  388. }
  389. err = security_msg_queue_msgctl(msq, cmd);
  390. if (err)
  391. goto out_unlock;
  392. switch (cmd) {
  393. case IPC_RMID:
  394. freeque(ns, ipcp);
  395. goto out_up;
  396. case IPC_SET:
  397. if (msqid64.msg_qbytes > ns->msg_ctlmnb &&
  398. !capable(CAP_SYS_RESOURCE)) {
  399. err = -EPERM;
  400. goto out_unlock;
  401. }
  402. msq->q_qbytes = msqid64.msg_qbytes;
  403. ipc_update_perm(&msqid64.msg_perm, ipcp);
  404. msq->q_ctime = get_seconds();
  405. /* sleeping receivers might be excluded by
  406. * stricter permissions.
  407. */
  408. expunge_all(msq, -EAGAIN);
  409. /* sleeping senders might be able to send
  410. * due to a larger queue size.
  411. */
  412. ss_wakeup(&msq->q_senders, 0);
  413. break;
  414. default:
  415. err = -EINVAL;
  416. }
  417. out_unlock:
  418. msg_unlock(msq);
  419. out_up:
  420. up_write(&msg_ids(ns).rw_mutex);
  421. return err;
  422. }
  423. asmlinkage long sys_msgctl(int msqid, int cmd, struct msqid_ds __user *buf)
  424. {
  425. struct msg_queue *msq;
  426. int err, version;
  427. struct ipc_namespace *ns;
  428. if (msqid < 0 || cmd < 0)
  429. return -EINVAL;
  430. version = ipc_parse_version(&cmd);
  431. ns = current->nsproxy->ipc_ns;
  432. switch (cmd) {
  433. case IPC_INFO:
  434. case MSG_INFO:
  435. {
  436. struct msginfo msginfo;
  437. int max_id;
  438. if (!buf)
  439. return -EFAULT;
  440. /*
  441. * We must not return kernel stack data.
  442. * due to padding, it's not enough
  443. * to set all member fields.
  444. */
  445. err = security_msg_queue_msgctl(NULL, cmd);
  446. if (err)
  447. return err;
  448. memset(&msginfo, 0, sizeof(msginfo));
  449. msginfo.msgmni = ns->msg_ctlmni;
  450. msginfo.msgmax = ns->msg_ctlmax;
  451. msginfo.msgmnb = ns->msg_ctlmnb;
  452. msginfo.msgssz = MSGSSZ;
  453. msginfo.msgseg = MSGSEG;
  454. down_read(&msg_ids(ns).rw_mutex);
  455. if (cmd == MSG_INFO) {
  456. msginfo.msgpool = msg_ids(ns).in_use;
  457. msginfo.msgmap = atomic_read(&ns->msg_hdrs);
  458. msginfo.msgtql = atomic_read(&ns->msg_bytes);
  459. } else {
  460. msginfo.msgmap = MSGMAP;
  461. msginfo.msgpool = MSGPOOL;
  462. msginfo.msgtql = MSGTQL;
  463. }
  464. max_id = ipc_get_maxid(&msg_ids(ns));
  465. up_read(&msg_ids(ns).rw_mutex);
  466. if (copy_to_user(buf, &msginfo, sizeof(struct msginfo)))
  467. return -EFAULT;
  468. return (max_id < 0) ? 0 : max_id;
  469. }
  470. case MSG_STAT: /* msqid is an index rather than a msg queue id */
  471. case IPC_STAT:
  472. {
  473. struct msqid64_ds tbuf;
  474. int success_return;
  475. if (!buf)
  476. return -EFAULT;
  477. if (cmd == MSG_STAT) {
  478. msq = msg_lock(ns, msqid);
  479. if (IS_ERR(msq))
  480. return PTR_ERR(msq);
  481. success_return = msq->q_perm.id;
  482. } else {
  483. msq = msg_lock_check(ns, msqid);
  484. if (IS_ERR(msq))
  485. return PTR_ERR(msq);
  486. success_return = 0;
  487. }
  488. err = -EACCES;
  489. if (ipcperms(&msq->q_perm, S_IRUGO))
  490. goto out_unlock;
  491. err = security_msg_queue_msgctl(msq, cmd);
  492. if (err)
  493. goto out_unlock;
  494. memset(&tbuf, 0, sizeof(tbuf));
  495. kernel_to_ipc64_perm(&msq->q_perm, &tbuf.msg_perm);
  496. tbuf.msg_stime = msq->q_stime;
  497. tbuf.msg_rtime = msq->q_rtime;
  498. tbuf.msg_ctime = msq->q_ctime;
  499. tbuf.msg_cbytes = msq->q_cbytes;
  500. tbuf.msg_qnum = msq->q_qnum;
  501. tbuf.msg_qbytes = msq->q_qbytes;
  502. tbuf.msg_lspid = msq->q_lspid;
  503. tbuf.msg_lrpid = msq->q_lrpid;
  504. msg_unlock(msq);
  505. if (copy_msqid_to_user(buf, &tbuf, version))
  506. return -EFAULT;
  507. return success_return;
  508. }
  509. case IPC_SET:
  510. case IPC_RMID:
  511. err = msgctl_down(ns, msqid, cmd, buf, version);
  512. return err;
  513. default:
  514. return -EINVAL;
  515. }
  516. out_unlock:
  517. msg_unlock(msq);
  518. return err;
  519. }
  520. static int testmsg(struct msg_msg *msg, long type, int mode)
  521. {
  522. switch(mode)
  523. {
  524. case SEARCH_ANY:
  525. return 1;
  526. case SEARCH_LESSEQUAL:
  527. if (msg->m_type <=type)
  528. return 1;
  529. break;
  530. case SEARCH_EQUAL:
  531. if (msg->m_type == type)
  532. return 1;
  533. break;
  534. case SEARCH_NOTEQUAL:
  535. if (msg->m_type != type)
  536. return 1;
  537. break;
  538. }
  539. return 0;
  540. }
  541. static inline int pipelined_send(struct msg_queue *msq, struct msg_msg *msg)
  542. {
  543. struct list_head *tmp;
  544. tmp = msq->q_receivers.next;
  545. while (tmp != &msq->q_receivers) {
  546. struct msg_receiver *msr;
  547. msr = list_entry(tmp, struct msg_receiver, r_list);
  548. tmp = tmp->next;
  549. if (testmsg(msg, msr->r_msgtype, msr->r_mode) &&
  550. !security_msg_queue_msgrcv(msq, msg, msr->r_tsk,
  551. msr->r_msgtype, msr->r_mode)) {
  552. list_del(&msr->r_list);
  553. if (msr->r_maxsize < msg->m_ts) {
  554. msr->r_msg = NULL;
  555. wake_up_process(msr->r_tsk);
  556. smp_mb();
  557. msr->r_msg = ERR_PTR(-E2BIG);
  558. } else {
  559. msr->r_msg = NULL;
  560. msq->q_lrpid = task_pid_vnr(msr->r_tsk);
  561. msq->q_rtime = get_seconds();
  562. wake_up_process(msr->r_tsk);
  563. smp_mb();
  564. msr->r_msg = msg;
  565. return 1;
  566. }
  567. }
  568. }
  569. return 0;
  570. }
  571. long do_msgsnd(int msqid, long mtype, void __user *mtext,
  572. size_t msgsz, int msgflg)
  573. {
  574. struct msg_queue *msq;
  575. struct msg_msg *msg;
  576. int err;
  577. struct ipc_namespace *ns;
  578. ns = current->nsproxy->ipc_ns;
  579. if (msgsz > ns->msg_ctlmax || (long) msgsz < 0 || msqid < 0)
  580. return -EINVAL;
  581. if (mtype < 1)
  582. return -EINVAL;
  583. msg = load_msg(mtext, msgsz);
  584. if (IS_ERR(msg))
  585. return PTR_ERR(msg);
  586. msg->m_type = mtype;
  587. msg->m_ts = msgsz;
  588. msq = msg_lock_check(ns, msqid);
  589. if (IS_ERR(msq)) {
  590. err = PTR_ERR(msq);
  591. goto out_free;
  592. }
  593. for (;;) {
  594. struct msg_sender s;
  595. err = -EACCES;
  596. if (ipcperms(&msq->q_perm, S_IWUGO))
  597. goto out_unlock_free;
  598. err = security_msg_queue_msgsnd(msq, msg, msgflg);
  599. if (err)
  600. goto out_unlock_free;
  601. if (msgsz + msq->q_cbytes <= msq->q_qbytes &&
  602. 1 + msq->q_qnum <= msq->q_qbytes) {
  603. break;
  604. }
  605. /* queue full, wait: */
  606. if (msgflg & IPC_NOWAIT) {
  607. err = -EAGAIN;
  608. goto out_unlock_free;
  609. }
  610. ss_add(msq, &s);
  611. ipc_rcu_getref(msq);
  612. msg_unlock(msq);
  613. schedule();
  614. ipc_lock_by_ptr(&msq->q_perm);
  615. ipc_rcu_putref(msq);
  616. if (msq->q_perm.deleted) {
  617. err = -EIDRM;
  618. goto out_unlock_free;
  619. }
  620. ss_del(&s);
  621. if (signal_pending(current)) {
  622. err = -ERESTARTNOHAND;
  623. goto out_unlock_free;
  624. }
  625. }
  626. msq->q_lspid = task_tgid_vnr(current);
  627. msq->q_stime = get_seconds();
  628. if (!pipelined_send(msq, msg)) {
  629. /* noone is waiting for this message, enqueue it */
  630. list_add_tail(&msg->m_list, &msq->q_messages);
  631. msq->q_cbytes += msgsz;
  632. msq->q_qnum++;
  633. atomic_add(msgsz, &ns->msg_bytes);
  634. atomic_inc(&ns->msg_hdrs);
  635. }
  636. err = 0;
  637. msg = NULL;
  638. out_unlock_free:
  639. msg_unlock(msq);
  640. out_free:
  641. if (msg != NULL)
  642. free_msg(msg);
  643. return err;
  644. }
  645. asmlinkage long
  646. sys_msgsnd(int msqid, struct msgbuf __user *msgp, size_t msgsz, int msgflg)
  647. {
  648. long mtype;
  649. if (get_user(mtype, &msgp->mtype))
  650. return -EFAULT;
  651. return do_msgsnd(msqid, mtype, msgp->mtext, msgsz, msgflg);
  652. }
  653. static inline int convert_mode(long *msgtyp, int msgflg)
  654. {
  655. /*
  656. * find message of correct type.
  657. * msgtyp = 0 => get first.
  658. * msgtyp > 0 => get first message of matching type.
  659. * msgtyp < 0 => get message with least type must be < abs(msgtype).
  660. */
  661. if (*msgtyp == 0)
  662. return SEARCH_ANY;
  663. if (*msgtyp < 0) {
  664. *msgtyp = -*msgtyp;
  665. return SEARCH_LESSEQUAL;
  666. }
  667. if (msgflg & MSG_EXCEPT)
  668. return SEARCH_NOTEQUAL;
  669. return SEARCH_EQUAL;
  670. }
  671. long do_msgrcv(int msqid, long *pmtype, void __user *mtext,
  672. size_t msgsz, long msgtyp, int msgflg)
  673. {
  674. struct msg_queue *msq;
  675. struct msg_msg *msg;
  676. int mode;
  677. struct ipc_namespace *ns;
  678. if (msqid < 0 || (long) msgsz < 0)
  679. return -EINVAL;
  680. mode = convert_mode(&msgtyp, msgflg);
  681. ns = current->nsproxy->ipc_ns;
  682. msq = msg_lock_check(ns, msqid);
  683. if (IS_ERR(msq))
  684. return PTR_ERR(msq);
  685. for (;;) {
  686. struct msg_receiver msr_d;
  687. struct list_head *tmp;
  688. msg = ERR_PTR(-EACCES);
  689. if (ipcperms(&msq->q_perm, S_IRUGO))
  690. goto out_unlock;
  691. msg = ERR_PTR(-EAGAIN);
  692. tmp = msq->q_messages.next;
  693. while (tmp != &msq->q_messages) {
  694. struct msg_msg *walk_msg;
  695. walk_msg = list_entry(tmp, struct msg_msg, m_list);
  696. if (testmsg(walk_msg, msgtyp, mode) &&
  697. !security_msg_queue_msgrcv(msq, walk_msg, current,
  698. msgtyp, mode)) {
  699. msg = walk_msg;
  700. if (mode == SEARCH_LESSEQUAL &&
  701. walk_msg->m_type != 1) {
  702. msg = walk_msg;
  703. msgtyp = walk_msg->m_type - 1;
  704. } else {
  705. msg = walk_msg;
  706. break;
  707. }
  708. }
  709. tmp = tmp->next;
  710. }
  711. if (!IS_ERR(msg)) {
  712. /*
  713. * Found a suitable message.
  714. * Unlink it from the queue.
  715. */
  716. if ((msgsz < msg->m_ts) && !(msgflg & MSG_NOERROR)) {
  717. msg = ERR_PTR(-E2BIG);
  718. goto out_unlock;
  719. }
  720. list_del(&msg->m_list);
  721. msq->q_qnum--;
  722. msq->q_rtime = get_seconds();
  723. msq->q_lrpid = task_tgid_vnr(current);
  724. msq->q_cbytes -= msg->m_ts;
  725. atomic_sub(msg->m_ts, &ns->msg_bytes);
  726. atomic_dec(&ns->msg_hdrs);
  727. ss_wakeup(&msq->q_senders, 0);
  728. msg_unlock(msq);
  729. break;
  730. }
  731. /* No message waiting. Wait for a message */
  732. if (msgflg & IPC_NOWAIT) {
  733. msg = ERR_PTR(-ENOMSG);
  734. goto out_unlock;
  735. }
  736. list_add_tail(&msr_d.r_list, &msq->q_receivers);
  737. msr_d.r_tsk = current;
  738. msr_d.r_msgtype = msgtyp;
  739. msr_d.r_mode = mode;
  740. if (msgflg & MSG_NOERROR)
  741. msr_d.r_maxsize = INT_MAX;
  742. else
  743. msr_d.r_maxsize = msgsz;
  744. msr_d.r_msg = ERR_PTR(-EAGAIN);
  745. current->state = TASK_INTERRUPTIBLE;
  746. msg_unlock(msq);
  747. schedule();
  748. /* Lockless receive, part 1:
  749. * Disable preemption. We don't hold a reference to the queue
  750. * and getting a reference would defeat the idea of a lockless
  751. * operation, thus the code relies on rcu to guarantee the
  752. * existance of msq:
  753. * Prior to destruction, expunge_all(-EIRDM) changes r_msg.
  754. * Thus if r_msg is -EAGAIN, then the queue not yet destroyed.
  755. * rcu_read_lock() prevents preemption between reading r_msg
  756. * and the spin_lock() inside ipc_lock_by_ptr().
  757. */
  758. rcu_read_lock();
  759. /* Lockless receive, part 2:
  760. * Wait until pipelined_send or expunge_all are outside of
  761. * wake_up_process(). There is a race with exit(), see
  762. * ipc/mqueue.c for the details.
  763. */
  764. msg = (struct msg_msg*)msr_d.r_msg;
  765. while (msg == NULL) {
  766. cpu_relax();
  767. msg = (struct msg_msg *)msr_d.r_msg;
  768. }
  769. /* Lockless receive, part 3:
  770. * If there is a message or an error then accept it without
  771. * locking.
  772. */
  773. if (msg != ERR_PTR(-EAGAIN)) {
  774. rcu_read_unlock();
  775. break;
  776. }
  777. /* Lockless receive, part 3:
  778. * Acquire the queue spinlock.
  779. */
  780. ipc_lock_by_ptr(&msq->q_perm);
  781. rcu_read_unlock();
  782. /* Lockless receive, part 4:
  783. * Repeat test after acquiring the spinlock.
  784. */
  785. msg = (struct msg_msg*)msr_d.r_msg;
  786. if (msg != ERR_PTR(-EAGAIN))
  787. goto out_unlock;
  788. list_del(&msr_d.r_list);
  789. if (signal_pending(current)) {
  790. msg = ERR_PTR(-ERESTARTNOHAND);
  791. out_unlock:
  792. msg_unlock(msq);
  793. break;
  794. }
  795. }
  796. if (IS_ERR(msg))
  797. return PTR_ERR(msg);
  798. msgsz = (msgsz > msg->m_ts) ? msg->m_ts : msgsz;
  799. *pmtype = msg->m_type;
  800. if (store_msg(mtext, msg, msgsz))
  801. msgsz = -EFAULT;
  802. free_msg(msg);
  803. return msgsz;
  804. }
  805. asmlinkage long sys_msgrcv(int msqid, struct msgbuf __user *msgp, size_t msgsz,
  806. long msgtyp, int msgflg)
  807. {
  808. long err, mtype;
  809. err = do_msgrcv(msqid, &mtype, msgp->mtext, msgsz, msgtyp, msgflg);
  810. if (err < 0)
  811. goto out;
  812. if (put_user(mtype, &msgp->mtype))
  813. err = -EFAULT;
  814. out:
  815. return err;
  816. }
  817. #ifdef CONFIG_PROC_FS
  818. static int sysvipc_msg_proc_show(struct seq_file *s, void *it)
  819. {
  820. struct msg_queue *msq = it;
  821. return seq_printf(s,
  822. "%10d %10d %4o %10lu %10lu %5u %5u %5u %5u %5u %5u %10lu %10lu %10lu\n",
  823. msq->q_perm.key,
  824. msq->q_perm.id,
  825. msq->q_perm.mode,
  826. msq->q_cbytes,
  827. msq->q_qnum,
  828. msq->q_lspid,
  829. msq->q_lrpid,
  830. msq->q_perm.uid,
  831. msq->q_perm.gid,
  832. msq->q_perm.cuid,
  833. msq->q_perm.cgid,
  834. msq->q_stime,
  835. msq->q_rtime,
  836. msq->q_ctime);
  837. }
  838. #endif