Index: head/sys/kern/sysv_msg.c =================================================================== --- head/sys/kern/sysv_msg.c (revision 122200) +++ head/sys/kern/sysv_msg.c (revision 122201) @@ -1,1151 +1,1151 @@ /* * Implementation of SVID messages * * Author: Daniel Boulet * * Copyright 1993 Daniel Boulet and RTMX Inc. * * This system call was implemented by Daniel Boulet under contract from RTMX. * * Redistribution and use in source forms, with and without modification, * are permitted provided that this entire comment appears intact. * * Redistribution in binary form may occur without any restrictions. * Obviously, it would be nice if you gave credit where credit is due * but requiring it would be too onerous. * * This software is provided ``AS IS'' without any warranties of any kind. */ #include __FBSDID("$FreeBSD$"); #include "opt_sysvipc.h" #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_MSG, "msg", "SVID compatible message queues"); static void msginit(void); static int msgunload(void); static int sysvmsg_modload(struct module *, int, void *); #ifdef MSG_DEBUG #define DPRINTF(a) printf a #else #define DPRINTF(a) #endif static void msg_freehdr(struct msg *msghdr); /* XXX casting to (sy_call_t *) is bogus, as usual. */ static sy_call_t *msgcalls[] = { (sy_call_t *)msgctl, (sy_call_t *)msgget, (sy_call_t *)msgsnd, (sy_call_t *)msgrcv }; struct msg { struct msg *msg_next; /* next msg in the chain */ long msg_type; /* type of this message */ /* >0 -> type of this message */ /* 0 -> free header */ u_short msg_ts; /* size of this message */ short msg_spot; /* location of start of msg in buffer */ }; #ifndef MSGSSZ #define MSGSSZ 8 /* Each segment must be 2^N long */ #endif #ifndef MSGSEG #define MSGSEG 2048 /* must be less than 32767 */ #endif #define MSGMAX (MSGSSZ*MSGSEG) #ifndef MSGMNB #define MSGMNB 2048 /* max # of bytes in a queue */ #endif #ifndef MSGMNI #define MSGMNI 40 #endif #ifndef MSGTQL #define MSGTQL 40 #endif /* * Based on the configuration parameters described in an SVR2 (yes, two) * config(1m) man page. * * Each message is broken up and stored in segments that are msgssz bytes * long. For efficiency reasons, this should be a power of two. Also, * it doesn't make sense if it is less than 8 or greater than about 256. * Consequently, msginit in kern/sysv_msg.c checks that msgssz is a power of * two between 8 and 1024 inclusive (and panic's if it isn't). */ struct msginfo msginfo = { MSGMAX, /* max chars in a message */ MSGMNI, /* # of message queue identifiers */ MSGMNB, /* max chars in a queue */ MSGTQL, /* max messages in system */ MSGSSZ, /* size of a message segment */ /* (must be small power of 2 greater than 4) */ MSGSEG /* number of message segments */ }; /* * macros to convert between msqid_ds's and msqid's. * (specific to this implementation) */ #define MSQID(ix,ds) ((ix) & 0xffff | (((ds).msg_perm.seq << 16) & 0xffff0000)) #define MSQID_IX(id) ((id) & 0xffff) #define MSQID_SEQ(id) (((id) >> 16) & 0xffff) /* * The rest of this file is specific to this particular implementation. */ struct msgmap { short next; /* next segment in buffer */ /* -1 -> available */ /* 0..(MSGSEG-1) -> index of next segment */ }; #define MSG_LOCKED 01000 /* Is this msqid_ds locked? */ static int nfree_msgmaps; /* # of free map entries */ static short free_msgmaps; /* head of linked list of free map entries */ static struct msg *free_msghdrs;/* list of free msg headers */ static char *msgpool; /* MSGMAX byte long msg buffer pool */ static struct msgmap *msgmaps; /* MSGSEG msgmap structures */ static struct msg *msghdrs; /* MSGTQL msg headers */ static struct msqid_ds *msqids; /* MSGMNI msqid_ds struct's */ static struct mtx msq_mtx; /* global mutex for message queues. */ static void msginit() { register int i; TUNABLE_INT_FETCH("kern.ipc.msgseg", &msginfo.msgseg); TUNABLE_INT_FETCH("kern.ipc.msgssz", &msginfo.msgssz); msginfo.msgmax = msginfo.msgseg * msginfo.msgssz; TUNABLE_INT_FETCH("kern.ipc.msgmni", &msginfo.msgmni); msgpool = malloc(msginfo.msgmax, M_MSG, M_WAITOK); if (msgpool == NULL) panic("msgpool is NULL"); msgmaps = malloc(sizeof(struct msgmap) * msginfo.msgseg, M_MSG, M_WAITOK); if (msgmaps == NULL) panic("msgmaps is NULL"); msghdrs = malloc(sizeof(struct msg) * msginfo.msgtql, M_MSG, M_WAITOK); if (msghdrs == NULL) panic("msghdrs is NULL"); msqids = malloc(sizeof(struct msqid_ds) * msginfo.msgmni, M_MSG, M_WAITOK); if (msqids == NULL) panic("msqids is NULL"); /* * msginfo.msgssz should be a power of two for efficiency reasons. * It is also pretty silly if msginfo.msgssz is less than 8 * or greater than about 256 so ... */ i = 8; while (i < 1024 && i != msginfo.msgssz) i <<= 1; if (i != msginfo.msgssz) { DPRINTF(("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, msginfo.msgssz)); panic("msginfo.msgssz not a small power of 2"); } if (msginfo.msgseg > 32767) { DPRINTF(("msginfo.msgseg=%d\n", msginfo.msgseg)); panic("msginfo.msgseg > 32767"); } if (msgmaps == NULL) panic("msgmaps is NULL"); for (i = 0; i < msginfo.msgseg; i++) { if (i > 0) msgmaps[i-1].next = i; msgmaps[i].next = -1; /* implies entry is available */ } free_msgmaps = 0; nfree_msgmaps = msginfo.msgseg; if (msghdrs == NULL) panic("msghdrs is NULL"); for (i = 0; i < msginfo.msgtql; i++) { msghdrs[i].msg_type = 0; if (i > 0) msghdrs[i-1].msg_next = &msghdrs[i]; msghdrs[i].msg_next = NULL; } free_msghdrs = &msghdrs[0]; if (msqids == NULL) panic("msqids is NULL"); for (i = 0; i < msginfo.msgmni; i++) { msqids[i].msg_qbytes = 0; /* implies entry is available */ msqids[i].msg_perm.seq = 0; /* reset to a known value */ msqids[i].msg_perm.mode = 0; } mtx_init(&msq_mtx, "msq", NULL, MTX_DEF); } static int msgunload() { struct msqid_ds *msqptr; int msqid; for (msqid = 0; msqid < msginfo.msgmni; msqid++) { /* * Look for an unallocated and unlocked msqid_ds. * msqid_ds's can be locked by msgsnd or msgrcv while * they are copying the message in/out. We can't * re-use the entry until they release it. */ msqptr = &msqids[msqid]; if (msqptr->msg_qbytes != 0 || (msqptr->msg_perm.mode & MSG_LOCKED) != 0) break; } if (msqid != msginfo.msgmni) return (EBUSY); free(msgpool, M_MSG); free(msgmaps, M_MSG); free(msghdrs, M_MSG); free(msqids, M_MSG); mtx_destroy(&msq_mtx); return (0); } static int sysvmsg_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: msginit(); break; case MOD_UNLOAD: error = msgunload(); break; case MOD_SHUTDOWN: break; default: error = EINVAL; break; } return (error); } static moduledata_t sysvmsg_mod = { "sysvmsg", &sysvmsg_modload, NULL }; SYSCALL_MODULE_HELPER(msgsys); SYSCALL_MODULE_HELPER(msgctl); SYSCALL_MODULE_HELPER(msgget); SYSCALL_MODULE_HELPER(msgsnd); SYSCALL_MODULE_HELPER(msgrcv); DECLARE_MODULE(sysvmsg, sysvmsg_mod, SI_SUB_SYSV_MSG, SI_ORDER_FIRST); MODULE_VERSION(sysvmsg, 1); /* * Entry point for all MSG calls * * MPSAFE */ int msgsys(td, uap) struct thread *td; /* XXX actually varargs. */ struct msgsys_args /* { int which; int a2; int a3; int a4; int a5; int a6; } */ *uap; { int error; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); if (uap->which < 0 || uap->which >= sizeof(msgcalls)/sizeof(msgcalls[0])) return (EINVAL); error = (*msgcalls[uap->which])(td, &uap->a2); return (error); } static void msg_freehdr(msghdr) struct msg *msghdr; { while (msghdr->msg_ts > 0) { short next; if (msghdr->msg_spot < 0 || msghdr->msg_spot >= msginfo.msgseg) panic("msghdr->msg_spot out of range"); next = msgmaps[msghdr->msg_spot].next; msgmaps[msghdr->msg_spot].next = free_msgmaps; free_msgmaps = msghdr->msg_spot; nfree_msgmaps++; msghdr->msg_spot = next; if (msghdr->msg_ts >= msginfo.msgssz) msghdr->msg_ts -= msginfo.msgssz; else msghdr->msg_ts = 0; } if (msghdr->msg_spot != -1) panic("msghdr->msg_spot != -1"); msghdr->msg_next = free_msghdrs; free_msghdrs = msghdr; } #ifndef _SYS_SYSPROTO_H_ struct msgctl_args { int msqid; int cmd; struct msqid_ds *buf; }; #endif /* * MPSAFE */ int msgctl(td, uap) struct thread *td; register struct msgctl_args *uap; { int msqid = uap->msqid; int cmd = uap->cmd; struct msqid_ds *user_msqptr = uap->buf; int rval, error; struct msqid_ds msqbuf; register struct msqid_ds *msqptr; DPRINTF(("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, msginfo.msgmni)); return (EINVAL); } if (cmd == IPC_SET && (error = copyin(user_msqptr, &msqbuf, sizeof(msqbuf))) != 0) return (error); msqptr = &msqids[msqid]; mtx_lock(&msq_mtx); if (msqptr->msg_qbytes == 0) { DPRINTF(("no such msqid\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { DPRINTF(("wrong sequence number\n")); error = EINVAL; goto done2; } error = 0; rval = 0; switch (cmd) { case IPC_RMID: { struct msg *msghdr; if ((error = ipcperm(td, &msqptr->msg_perm, IPC_M))) goto done2; /* Free the message headers */ msghdr = msqptr->msg_first; while (msghdr != NULL) { struct msg *msghdr_tmp; /* Free the segments of each message */ msqptr->msg_cbytes -= msghdr->msg_ts; msqptr->msg_qnum--; msghdr_tmp = msghdr; msghdr = msghdr->msg_next; msg_freehdr(msghdr_tmp); } if (msqptr->msg_cbytes != 0) panic("msg_cbytes is screwed up"); if (msqptr->msg_qnum != 0) panic("msg_qnum is screwed up"); msqptr->msg_qbytes = 0; /* Mark it as free */ wakeup(msqptr); } break; case IPC_SET: if ((error = ipcperm(td, &msqptr->msg_perm, IPC_M))) goto done2; if (msqbuf.msg_qbytes > msqptr->msg_qbytes) { error = suser(td); if (error) goto done2; } if (msqbuf.msg_qbytes > msginfo.msgmnb) { DPRINTF(("can't increase msg_qbytes beyond %d" "(truncating)\n", msginfo.msgmnb)); msqbuf.msg_qbytes = msginfo.msgmnb; /* silently restrict qbytes to system limit */ } if (msqbuf.msg_qbytes == 0) { DPRINTF(("can't reduce msg_qbytes to 0\n")); error = EINVAL; /* non-standard errno! */ goto done2; } msqptr->msg_perm.uid = msqbuf.msg_perm.uid; /* change the owner */ msqptr->msg_perm.gid = msqbuf.msg_perm.gid; /* change the owner */ msqptr->msg_perm.mode = (msqptr->msg_perm.mode & ~0777) | (msqbuf.msg_perm.mode & 0777); msqptr->msg_qbytes = msqbuf.msg_qbytes; msqptr->msg_ctime = time_second; break; case IPC_STAT: if ((error = ipcperm(td, &msqptr->msg_perm, IPC_R))) { DPRINTF(("requester doesn't have read access\n")); goto done2; } break; default: DPRINTF(("invalid command %d\n", cmd)); error = EINVAL; goto done2; } if (error == 0) td->td_retval[0] = rval; done2: mtx_unlock(&msq_mtx); if (cmd == IPC_STAT && error == 0) error = copyout(msqptr, user_msqptr, sizeof(struct msqid_ds)); return(error); } #ifndef _SYS_SYSPROTO_H_ struct msgget_args { key_t key; int msgflg; }; #endif /* * MPSAFE */ int msgget(td, uap) struct thread *td; register struct msgget_args *uap; { int msqid, error = 0; int key = uap->key; int msgflg = uap->msgflg; struct ucred *cred = td->td_ucred; register struct msqid_ds *msqptr = NULL; DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&msq_mtx); if (key != IPC_PRIVATE) { for (msqid = 0; msqid < msginfo.msgmni; msqid++) { msqptr = &msqids[msqid]; if (msqptr->msg_qbytes != 0 && msqptr->msg_perm.key == key) break; } if (msqid < msginfo.msgmni) { DPRINTF(("found public key\n")); if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { DPRINTF(("not exclusive\n")); error = EEXIST; goto done2; } - if ((error = ipcperm(td, &msqptr->msg_perm, msgflg & 0700 ))) { + if ((error = ipcperm(td, &msqptr->msg_perm, msgflg & 0700))) { DPRINTF(("requester doesn't have 0%o access\n", msgflg & 0700)); goto done2; } goto found; } } DPRINTF(("need to allocate the msqid_ds\n")); if (key == IPC_PRIVATE || (msgflg & IPC_CREAT)) { for (msqid = 0; msqid < msginfo.msgmni; msqid++) { /* * Look for an unallocated and unlocked msqid_ds. * msqid_ds's can be locked by msgsnd or msgrcv while * they are copying the message in/out. We can't * re-use the entry until they release it. */ msqptr = &msqids[msqid]; if (msqptr->msg_qbytes == 0 && (msqptr->msg_perm.mode & MSG_LOCKED) == 0) break; } if (msqid == msginfo.msgmni) { DPRINTF(("no more msqid_ds's available\n")); error = ENOSPC; goto done2; } DPRINTF(("msqid %d is available\n", msqid)); msqptr->msg_perm.key = key; msqptr->msg_perm.cuid = cred->cr_uid; msqptr->msg_perm.uid = cred->cr_uid; msqptr->msg_perm.cgid = cred->cr_gid; msqptr->msg_perm.gid = cred->cr_gid; msqptr->msg_perm.mode = (msgflg & 0777); /* Make sure that the returned msqid is unique */ msqptr->msg_perm.seq = (msqptr->msg_perm.seq + 1) & 0x7fff; msqptr->msg_first = NULL; msqptr->msg_last = NULL; msqptr->msg_cbytes = 0; msqptr->msg_qnum = 0; msqptr->msg_qbytes = msginfo.msgmnb; msqptr->msg_lspid = 0; msqptr->msg_lrpid = 0; msqptr->msg_stime = 0; msqptr->msg_rtime = 0; msqptr->msg_ctime = time_second; } else { DPRINTF(("didn't find it and wasn't asked to create it\n")); error = ENOENT; goto done2; } found: /* Construct the unique msqid */ td->td_retval[0] = IXSEQ_TO_IPCID(msqid, msqptr->msg_perm); done2: mtx_unlock(&msq_mtx); return (error); } #ifndef _SYS_SYSPROTO_H_ struct msgsnd_args { int msqid; const void *msgp; size_t msgsz; int msgflg; }; #endif /* * MPSAFE */ int msgsnd(td, uap) struct thread *td; register struct msgsnd_args *uap; { int msqid = uap->msqid; const void *user_msgp = uap->msgp; size_t msgsz = uap->msgsz; int msgflg = uap->msgflg; int segs_needed, error = 0; register struct msqid_ds *msqptr; register struct msg *msghdr; short next; DPRINTF(("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz, msgflg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&msq_mtx); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, msginfo.msgmni)); error = EINVAL; goto done2; } msqptr = &msqids[msqid]; if (msqptr->msg_qbytes == 0) { DPRINTF(("no such message queue id\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { DPRINTF(("wrong sequence number\n")); error = EINVAL; goto done2; } if ((error = ipcperm(td, &msqptr->msg_perm, IPC_W))) { DPRINTF(("requester doesn't have write access\n")); goto done2; } segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; DPRINTF(("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz, segs_needed)); for (;;) { int need_more_resources = 0; /* * check msgsz * (inside this loop in case msg_qbytes changes while we sleep) */ if (msgsz > msqptr->msg_qbytes) { DPRINTF(("msgsz > msqptr->msg_qbytes\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.mode & MSG_LOCKED) { DPRINTF(("msqid is locked\n")); need_more_resources = 1; } if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes) { DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); need_more_resources = 1; } if (segs_needed > nfree_msgmaps) { DPRINTF(("segs_needed > nfree_msgmaps\n")); need_more_resources = 1; } if (free_msghdrs == NULL) { DPRINTF(("no more msghdrs\n")); need_more_resources = 1; } if (need_more_resources) { int we_own_it; if ((msgflg & IPC_NOWAIT) != 0) { DPRINTF(("need more resources but caller " "doesn't want to wait\n")); error = EAGAIN; goto done2; } if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { DPRINTF(("we don't own the msqid_ds\n")); we_own_it = 0; } else { /* Force later arrivals to wait for our request */ DPRINTF(("we own the msqid_ds\n")); msqptr->msg_perm.mode |= MSG_LOCKED; we_own_it = 1; } DPRINTF(("goodnight\n")); error = msleep(msqptr, &msq_mtx, (PZERO - 4) | PCATCH, "msgwait", 0); DPRINTF(("good morning, error=%d\n", error)); if (we_own_it) msqptr->msg_perm.mode &= ~MSG_LOCKED; if (error != 0) { DPRINTF(("msgsnd: interrupted system call\n")); error = EINTR; goto done2; } /* * Make sure that the msq queue still exists */ if (msqptr->msg_qbytes == 0) { DPRINTF(("msqid deleted\n")); error = EIDRM; goto done2; } } else { DPRINTF(("got all the resources that we need\n")); break; } } /* * We have the resources that we need. * Make sure! */ if (msqptr->msg_perm.mode & MSG_LOCKED) panic("msg_perm.mode & MSG_LOCKED"); if (segs_needed > nfree_msgmaps) panic("segs_needed > nfree_msgmaps"); if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes) panic("msgsz + msg_cbytes > msg_qbytes"); if (free_msghdrs == NULL) panic("no more msghdrs"); /* * Re-lock the msqid_ds in case we page-fault when copying in the * message */ if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) panic("msqid_ds is already locked"); msqptr->msg_perm.mode |= MSG_LOCKED; /* * Allocate a message header */ msghdr = free_msghdrs; free_msghdrs = msghdr->msg_next; msghdr->msg_spot = -1; msghdr->msg_ts = msgsz; /* * Allocate space for the message */ while (segs_needed > 0) { if (nfree_msgmaps <= 0) panic("not enough msgmaps"); if (free_msgmaps == -1) panic("nil free_msgmaps"); next = free_msgmaps; if (next <= -1) panic("next too low #1"); if (next >= msginfo.msgseg) panic("next out of range #1"); DPRINTF(("allocating segment %d to message\n", next)); free_msgmaps = msgmaps[next].next; nfree_msgmaps--; msgmaps[next].next = msghdr->msg_spot; msghdr->msg_spot = next; segs_needed--; } /* * Copy in the message type */ mtx_unlock(&msq_mtx); if ((error = copyin(user_msgp, &msghdr->msg_type, sizeof(msghdr->msg_type))) != 0) { mtx_lock(&msq_mtx); DPRINTF(("error %d copying the message type\n", error)); msg_freehdr(msghdr); msqptr->msg_perm.mode &= ~MSG_LOCKED; wakeup(msqptr); goto done2; } mtx_lock(&msq_mtx); user_msgp = (const char *)user_msgp + sizeof(msghdr->msg_type); /* * Validate the message type */ if (msghdr->msg_type < 1) { msg_freehdr(msghdr); msqptr->msg_perm.mode &= ~MSG_LOCKED; wakeup(msqptr); DPRINTF(("mtype (%d) < 1\n", msghdr->msg_type)); error = EINVAL; goto done2; } /* * Copy in the message body */ next = msghdr->msg_spot; while (msgsz > 0) { size_t tlen; if (msgsz > msginfo.msgssz) tlen = msginfo.msgssz; else tlen = msgsz; if (next <= -1) panic("next too low #2"); if (next >= msginfo.msgseg) panic("next out of range #2"); mtx_unlock(&msq_mtx); if ((error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen)) != 0) { mtx_lock(&msq_mtx); DPRINTF(("error %d copying in message segment\n", error)); msg_freehdr(msghdr); msqptr->msg_perm.mode &= ~MSG_LOCKED; wakeup(msqptr); goto done2; } mtx_lock(&msq_mtx); msgsz -= tlen; user_msgp = (const char *)user_msgp + tlen; next = msgmaps[next].next; } if (next != -1) panic("didn't use all the msg segments"); /* * We've got the message. Unlock the msqid_ds. */ msqptr->msg_perm.mode &= ~MSG_LOCKED; /* * Make sure that the msqid_ds is still allocated. */ if (msqptr->msg_qbytes == 0) { msg_freehdr(msghdr); wakeup(msqptr); error = EIDRM; goto done2; } /* * Put the message into the queue */ if (msqptr->msg_first == NULL) { msqptr->msg_first = msghdr; msqptr->msg_last = msghdr; } else { msqptr->msg_last->msg_next = msghdr; msqptr->msg_last = msghdr; } msqptr->msg_last->msg_next = NULL; msqptr->msg_cbytes += msghdr->msg_ts; msqptr->msg_qnum++; msqptr->msg_lspid = td->td_proc->p_pid; msqptr->msg_stime = time_second; wakeup(msqptr); td->td_retval[0] = 0; done2: mtx_unlock(&msq_mtx); return (error); } #ifndef _SYS_SYSPROTO_H_ struct msgrcv_args { int msqid; void *msgp; size_t msgsz; long msgtyp; int msgflg; }; #endif /* * MPSAFE */ int msgrcv(td, uap) struct thread *td; register struct msgrcv_args *uap; { int msqid = uap->msqid; void *user_msgp = uap->msgp; size_t msgsz = uap->msgsz; long msgtyp = uap->msgtyp; int msgflg = uap->msgflg; size_t len; register struct msqid_ds *msqptr; register struct msg *msghdr; int error = 0; short next; DPRINTF(("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp, msgsz, msgtyp, msgflg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { DPRINTF(("msqid (%d) out of range (0<=msqid<%d)\n", msqid, msginfo.msgmni)); return (EINVAL); } msqptr = &msqids[msqid]; mtx_lock(&msq_mtx); if (msqptr->msg_qbytes == 0) { DPRINTF(("no such message queue id\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { DPRINTF(("wrong sequence number\n")); error = EINVAL; goto done2; } if ((error = ipcperm(td, &msqptr->msg_perm, IPC_R))) { DPRINTF(("requester doesn't have read access\n")); goto done2; } msghdr = NULL; while (msghdr == NULL) { if (msgtyp == 0) { msghdr = msqptr->msg_first; if (msghdr != NULL) { if (msgsz < msghdr->msg_ts && (msgflg & MSG_NOERROR) == 0) { DPRINTF(("first message on the queue " "is too big (want %d, got %d)\n", msgsz, msghdr->msg_ts)); error = E2BIG; goto done2; } if (msqptr->msg_first == msqptr->msg_last) { msqptr->msg_first = NULL; msqptr->msg_last = NULL; } else { msqptr->msg_first = msghdr->msg_next; if (msqptr->msg_first == NULL) panic("msg_first/last screwed up #1"); } } } else { struct msg *previous; struct msg **prev; previous = NULL; prev = &(msqptr->msg_first); while ((msghdr = *prev) != NULL) { /* * Is this message's type an exact match or is * this message's type less than or equal to * the absolute value of a negative msgtyp? * Note that the second half of this test can * NEVER be true if msgtyp is positive since * msg_type is always positive! */ if (msgtyp == msghdr->msg_type || msghdr->msg_type <= -msgtyp) { DPRINTF(("found message type %d, " "requested %d\n", msghdr->msg_type, msgtyp)); if (msgsz < msghdr->msg_ts && (msgflg & MSG_NOERROR) == 0) { DPRINTF(("requested message " "on the queue is too big " "(want %d, got %d)\n", msgsz, msghdr->msg_ts)); error = E2BIG; goto done2; } *prev = msghdr->msg_next; if (msghdr == msqptr->msg_last) { if (previous == NULL) { if (prev != &msqptr->msg_first) panic("msg_first/last screwed up #2"); msqptr->msg_first = NULL; msqptr->msg_last = NULL; } else { if (prev == &msqptr->msg_first) panic("msg_first/last screwed up #3"); msqptr->msg_last = previous; } } break; } previous = msghdr; prev = &(msghdr->msg_next); } } /* * We've either extracted the msghdr for the appropriate * message or there isn't one. * If there is one then bail out of this loop. */ if (msghdr != NULL) break; /* * Hmph! No message found. Does the user want to wait? */ if ((msgflg & IPC_NOWAIT) != 0) { DPRINTF(("no appropriate message found (msgtyp=%d)\n", msgtyp)); /* The SVID says to return ENOMSG. */ error = ENOMSG; goto done2; } /* * Wait for something to happen */ DPRINTF(("msgrcv: goodnight\n")); error = msleep(msqptr, &msq_mtx, (PZERO - 4) | PCATCH, "msgwait", 0); DPRINTF(("msgrcv: good morning (error=%d)\n", error)); if (error != 0) { DPRINTF(("msgsnd: interrupted system call\n")); error = EINTR; goto done2; } /* * Make sure that the msq queue still exists */ if (msqptr->msg_qbytes == 0 || msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { DPRINTF(("msqid deleted\n")); error = EIDRM; goto done2; } } /* * Return the message to the user. * * First, do the bookkeeping (before we risk being interrupted). */ msqptr->msg_cbytes -= msghdr->msg_ts; msqptr->msg_qnum--; msqptr->msg_lrpid = td->td_proc->p_pid; msqptr->msg_rtime = time_second; /* * Make msgsz the actual amount that we'll be returning. * Note that this effectively truncates the message if it is too long * (since msgsz is never increased). */ DPRINTF(("found a message, msgsz=%d, msg_ts=%d\n", msgsz, msghdr->msg_ts)); if (msgsz > msghdr->msg_ts) msgsz = msghdr->msg_ts; /* * Return the type to the user. */ mtx_unlock(&msq_mtx); error = copyout(&(msghdr->msg_type), user_msgp, sizeof(msghdr->msg_type)); mtx_lock(&msq_mtx); if (error != 0) { DPRINTF(("error (%d) copying out message type\n", error)); msg_freehdr(msghdr); wakeup(msqptr); goto done2; } user_msgp = (char *)user_msgp + sizeof(msghdr->msg_type); /* * Return the segments to the user */ next = msghdr->msg_spot; for (len = 0; len < msgsz; len += msginfo.msgssz) { size_t tlen; if (msgsz - len > msginfo.msgssz) tlen = msginfo.msgssz; else tlen = msgsz - len; if (next <= -1) panic("next too low #3"); if (next >= msginfo.msgseg) panic("next out of range #3"); mtx_unlock(&msq_mtx); error = copyout(&msgpool[next * msginfo.msgssz], user_msgp, tlen); mtx_lock(&msq_mtx); if (error != 0) { DPRINTF(("error (%d) copying out message segment\n", error)); msg_freehdr(msghdr); wakeup(msqptr); goto done2; } user_msgp = (char *)user_msgp + tlen; next = msgmaps[next].next; } /* * Done, return the actual number of bytes copied out. */ msg_freehdr(msghdr); wakeup(msqptr); td->td_retval[0] = msgsz; done2: mtx_unlock(&msq_mtx); return (error); } static int sysctl_msqids(SYSCTL_HANDLER_ARGS) { return (SYSCTL_OUT(req, msqids, sizeof(struct msqid_ds) * msginfo.msgmni)); } SYSCTL_DECL(_kern_ipc); SYSCTL_INT(_kern_ipc, OID_AUTO, msgmax, CTLFLAG_RD, &msginfo.msgmax, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgmni, CTLFLAG_RDTUN, &msginfo.msgmni, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgmnb, CTLFLAG_RD, &msginfo.msgmnb, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgtql, CTLFLAG_RD, &msginfo.msgtql, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgssz, CTLFLAG_RDTUN, &msginfo.msgssz, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RDTUN, &msginfo.msgseg, 0, ""); SYSCTL_PROC(_kern_ipc, OID_AUTO, msqids, CTLFLAG_RD, NULL, 0, sysctl_msqids, "", "Message queue IDs"); Index: head/sys/kern/sysv_sem.c =================================================================== --- head/sys/kern/sysv_sem.c (revision 122200) +++ head/sys/kern/sysv_sem.c (revision 122201) @@ -1,1230 +1,1230 @@ /* * Implementation of SVID semaphores * * Author: Daniel Boulet * * This software is provided ``AS IS'' without any warranties of any kind. */ #include __FBSDID("$FreeBSD$"); #include "opt_sysvipc.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_SEM, "sem", "SVID compatible semaphores"); #ifdef SEM_DEBUG #define DPRINTF(a) printf a #else #define DPRINTF(a) #endif static void seminit(void); static int sysvsem_modload(struct module *, int, void *); static int semunload(void); static void semexit_myhook(void *arg, struct proc *p); static int sysctl_sema(SYSCTL_HANDLER_ARGS); static int semvalid(int semid, struct semid_ds *semaptr); #ifndef _SYS_SYSPROTO_H_ struct __semctl_args; int __semctl(struct thread *td, struct __semctl_args *uap); struct semget_args; int semget(struct thread *td, struct semget_args *uap); struct semop_args; int semop(struct thread *td, struct semop_args *uap); #endif static struct sem_undo *semu_alloc(struct thread *td); -static int semundo_adjust(struct thread *td, struct sem_undo **supptr, +static int semundo_adjust(struct thread *td, struct sem_undo **supptr, int semid, int semnum, int adjval); static void semundo_clear(int semid, int semnum); /* XXX casting to (sy_call_t *) is bogus, as usual. */ static sy_call_t *semcalls[] = { (sy_call_t *)__semctl, (sy_call_t *)semget, (sy_call_t *)semop }; static struct mtx sem_mtx; /* semaphore global lock */ static int semtot = 0; static struct semid_ds *sema; /* semaphore id pool */ static struct mtx *sema_mtx; /* semaphore id pool mutexes*/ static struct sem *sem; /* semaphore pool */ SLIST_HEAD(, sem_undo) semu_list; /* list of active undo structures */ static int *semu; /* undo structure pool */ static eventhandler_tag semexit_tag; #define SEMUNDO_MTX sem_mtx #define SEMUNDO_LOCK() mtx_lock(&SEMUNDO_MTX); #define SEMUNDO_UNLOCK() mtx_unlock(&SEMUNDO_MTX); #define SEMUNDO_LOCKASSERT(how) mtx_assert(&SEMUNDO_MTX, (how)); struct sem { u_short semval; /* semaphore value */ pid_t sempid; /* pid of last operation */ u_short semncnt; /* # awaiting semval > cval */ u_short semzcnt; /* # awaiting semval = 0 */ }; /* * Undo structure (one per process) */ struct sem_undo { SLIST_ENTRY(sem_undo) un_next; /* ptr to next active undo structure */ struct proc *un_proc; /* owner of this structure */ short un_cnt; /* # of active entries */ struct undo { short un_adjval; /* adjust on exit values */ short un_num; /* semaphore # */ int un_id; /* semid */ } un_ent[1]; /* undo entries */ }; /* * Configuration parameters */ #ifndef SEMMNI #define SEMMNI 10 /* # of semaphore identifiers */ #endif #ifndef SEMMNS #define SEMMNS 60 /* # of semaphores in system */ #endif #ifndef SEMUME #define SEMUME 10 /* max # of undo entries per process */ #endif #ifndef SEMMNU #define SEMMNU 30 /* # of undo structures in system */ #endif /* shouldn't need tuning */ #ifndef SEMMAP #define SEMMAP 30 /* # of entries in semaphore map */ #endif #ifndef SEMMSL #define SEMMSL SEMMNS /* max # of semaphores per id */ #endif #ifndef SEMOPM #define SEMOPM 100 /* max # of operations per semop call */ #endif #define SEMVMX 32767 /* semaphore maximum value */ #define SEMAEM 16384 /* adjust on exit max value */ /* * Due to the way semaphore memory is allocated, we have to ensure that * SEMUSZ is properly aligned. */ #define SEM_ALIGN(bytes) (((bytes) + (sizeof(long) - 1)) & ~(sizeof(long) - 1)) /* actual size of an undo structure */ #define SEMUSZ SEM_ALIGN(offsetof(struct sem_undo, un_ent[SEMUME])) /* * Macro to find a particular sem_undo vector */ #define SEMU(ix) \ ((struct sem_undo *)(((intptr_t)semu)+ix * seminfo.semusz)) /* * semaphore info struct */ struct seminfo seminfo = { SEMMAP, /* # of entries in semaphore map */ SEMMNI, /* # of semaphore identifiers */ SEMMNS, /* # of semaphores in system */ SEMMNU, /* # of undo structures in system */ SEMMSL, /* max # of semaphores per id */ SEMOPM, /* max # of operations per semop call */ SEMUME, /* max # of undo entries per process */ SEMUSZ, /* size in bytes of undo structure */ SEMVMX, /* semaphore maximum value */ SEMAEM /* adjust on exit max value */ }; SYSCTL_DECL(_kern_ipc); SYSCTL_INT(_kern_ipc, OID_AUTO, semmap, CTLFLAG_RW, &seminfo.semmap, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmni, CTLFLAG_RDTUN, &seminfo.semmni, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RDTUN, &seminfo.semmns, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RDTUN, &seminfo.semmnu, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RDTUN, &seminfo.semopm, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RDTUN, &seminfo.semume, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RDTUN, &seminfo.semusz, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semvmx, CTLFLAG_RW, &seminfo.semvmx, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semaem, CTLFLAG_RW, &seminfo.semaem, 0, ""); SYSCTL_PROC(_kern_ipc, OID_AUTO, sema, CTLFLAG_RD, NULL, 0, sysctl_sema, "", ""); static void seminit(void) { int i; TUNABLE_INT_FETCH("kern.ipc.semmap", &seminfo.semmap); TUNABLE_INT_FETCH("kern.ipc.semmni", &seminfo.semmni); TUNABLE_INT_FETCH("kern.ipc.semmns", &seminfo.semmns); TUNABLE_INT_FETCH("kern.ipc.semmnu", &seminfo.semmnu); TUNABLE_INT_FETCH("kern.ipc.semmsl", &seminfo.semmsl); TUNABLE_INT_FETCH("kern.ipc.semopm", &seminfo.semopm); TUNABLE_INT_FETCH("kern.ipc.semume", &seminfo.semume); TUNABLE_INT_FETCH("kern.ipc.semusz", &seminfo.semusz); TUNABLE_INT_FETCH("kern.ipc.semvmx", &seminfo.semvmx); TUNABLE_INT_FETCH("kern.ipc.semaem", &seminfo.semaem); sem = malloc(sizeof(struct sem) * seminfo.semmns, M_SEM, M_WAITOK); sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK); sema_mtx = malloc(sizeof(struct mtx) * seminfo.semmni, M_SEM, M_WAITOK | M_ZERO); semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK); for (i = 0; i < seminfo.semmni; i++) { sema[i].sem_base = 0; sema[i].sem_perm.mode = 0; } for (i = 0; i < seminfo.semmni; i++) mtx_init(&sema_mtx[i], "semid", NULL, MTX_DEF); for (i = 0; i < seminfo.semmnu; i++) { struct sem_undo *suptr = SEMU(i); suptr->un_proc = NULL; } SLIST_INIT(&semu_list); mtx_init(&sem_mtx, "sem", NULL, MTX_DEF); semexit_tag = EVENTHANDLER_REGISTER(process_exit, semexit_myhook, NULL, EVENTHANDLER_PRI_ANY); } static int semunload(void) { int i; if (semtot != 0) return (EBUSY); EVENTHANDLER_DEREGISTER(process_exit, semexit_tag); free(sem, M_SEM); free(sema, M_SEM); free(semu, M_SEM); for (i = 0; i < seminfo.semmni; i++) mtx_destroy(&sema_mtx[i]); mtx_destroy(&sem_mtx); return (0); } static int sysvsem_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: seminit(); break; case MOD_UNLOAD: error = semunload(); break; case MOD_SHUTDOWN: break; default: error = EINVAL; break; } return (error); } static moduledata_t sysvsem_mod = { "sysvsem", &sysvsem_modload, NULL }; SYSCALL_MODULE_HELPER(semsys); SYSCALL_MODULE_HELPER(__semctl); SYSCALL_MODULE_HELPER(semget); SYSCALL_MODULE_HELPER(semop); DECLARE_MODULE(sysvsem, sysvsem_mod, SI_SUB_SYSV_SEM, SI_ORDER_FIRST); MODULE_VERSION(sysvsem, 1); /* * Entry point for all SEM calls * * MPSAFE */ int semsys(td, uap) struct thread *td; /* XXX actually varargs. */ struct semsys_args /* { int which; int a2; int a3; int a4; int a5; } */ *uap; { int error; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); if (uap->which < 0 || uap->which >= sizeof(semcalls)/sizeof(semcalls[0])) return (EINVAL); error = (*semcalls[uap->which])(td, &uap->a2); return (error); } /* * Allocate a new sem_undo structure for a process * (returns ptr to structure or NULL if no more room) */ static struct sem_undo * semu_alloc(td) struct thread *td; { int i; struct sem_undo *suptr; struct sem_undo **supptr; int attempt; SEMUNDO_LOCKASSERT(MA_OWNED); /* * Try twice to allocate something. * (we'll purge any empty structures after the first pass so * two passes are always enough) */ for (attempt = 0; attempt < 2; attempt++) { /* * Look for a free structure. * Fill it in and return it if we find one. */ for (i = 0; i < seminfo.semmnu; i++) { suptr = SEMU(i); if (suptr->un_proc == NULL) { SLIST_INSERT_HEAD(&semu_list, suptr, un_next); suptr->un_cnt = 0; suptr->un_proc = td->td_proc; return(suptr); } } /* * We didn't find a free one, if this is the first attempt * then try to free some structures. */ if (attempt == 0) { /* All the structures are in use - try to free some */ int did_something = 0; SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) { if (suptr->un_cnt == 0) { suptr->un_proc = NULL; did_something = 1; *supptr = SLIST_NEXT(suptr, un_next); } } /* If we didn't free anything then just give-up */ if (!did_something) return(NULL); } else { /* * The second pass failed even though we freed * something after the first pass! * This is IMPOSSIBLE! */ panic("semu_alloc - second attempt failed"); } } return (NULL); } /* * Adjust a particular entry for a particular proc */ static int semundo_adjust(td, supptr, semid, semnum, adjval) struct thread *td; struct sem_undo **supptr; int semid, semnum; int adjval; { struct proc *p = td->td_proc; struct sem_undo *suptr; struct undo *sunptr; int i; SEMUNDO_LOCKASSERT(MA_OWNED); /* Look for and remember the sem_undo if the caller doesn't provide it */ suptr = *supptr; if (suptr == NULL) { SLIST_FOREACH(suptr, &semu_list, un_next) { if (suptr->un_proc == p) { *supptr = suptr; break; } } if (suptr == NULL) { if (adjval == 0) return(0); suptr = semu_alloc(td); if (suptr == NULL) return(ENOSPC); *supptr = suptr; } } /* * Look for the requested entry and adjust it (delete if adjval becomes * 0). */ sunptr = &suptr->un_ent[0]; for (i = 0; i < suptr->un_cnt; i++, sunptr++) { if (sunptr->un_id != semid || sunptr->un_num != semnum) continue; if (adjval != 0) { adjval += sunptr->un_adjval; if (adjval > seminfo.semaem || adjval < -seminfo.semaem) return (ERANGE); } sunptr->un_adjval = adjval; if (sunptr->un_adjval == 0) { suptr->un_cnt--; if (i < suptr->un_cnt) suptr->un_ent[i] = suptr->un_ent[suptr->un_cnt]; } return(0); } /* Didn't find the right entry - create it */ if (adjval == 0) return(0); if (adjval > seminfo.semaem || adjval < -seminfo.semaem) return (ERANGE); if (suptr->un_cnt != seminfo.semume) { sunptr = &suptr->un_ent[suptr->un_cnt]; suptr->un_cnt++; sunptr->un_adjval = adjval; sunptr->un_id = semid; sunptr->un_num = semnum; } else return(EINVAL); return(0); } static void semundo_clear(semid, semnum) int semid, semnum; { struct sem_undo *suptr; SEMUNDO_LOCKASSERT(MA_OWNED); SLIST_FOREACH(suptr, &semu_list, un_next) { struct undo *sunptr = &suptr->un_ent[0]; int i = 0; while (i < suptr->un_cnt) { if (sunptr->un_id == semid) { if (semnum == -1 || sunptr->un_num == semnum) { suptr->un_cnt--; if (i < suptr->un_cnt) { suptr->un_ent[i] = suptr->un_ent[suptr->un_cnt]; continue; } } if (semnum != -1) break; } i++, sunptr++; } } } static int semvalid(semid, semaptr) int semid; struct semid_ds *semaptr; { return ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || semaptr->sem_perm.seq != IPCID_TO_SEQ(semid) ? EINVAL : 0); } /* * Note that the user-mode half of this passes a union, not a pointer */ #ifndef _SYS_SYSPROTO_H_ struct __semctl_args { int semid; int semnum; int cmd; union semun *arg; }; #endif /* * MPSAFE */ int __semctl(td, uap) struct thread *td; struct __semctl_args *uap; { int semid = uap->semid; int semnum = uap->semnum; int cmd = uap->cmd; u_short *array; union semun *arg = uap->arg; union semun real_arg; struct ucred *cred = td->td_ucred; int i, rval, error; struct semid_ds sbuf; struct semid_ds *semaptr; struct mtx *sema_mtxp; u_short usval, count; DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); array = NULL; switch(cmd) { case SEM_STAT: if (semid < 0 || semid >= seminfo.semmni) return (EINVAL); if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) return (error); semaptr = &sema[semid]; sema_mtxp = &sema_mtx[semid]; mtx_lock(sema_mtxp); - if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ) { + if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) { error = EINVAL; goto done2; } if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; mtx_unlock(sema_mtxp); error = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds)); rval = IXSEQ_TO_IPCID(semid,semaptr->sem_perm); if (error == 0) td->td_retval[0] = rval; return (error); } semid = IPCID_TO_IX(semid); if (semid < 0 || semid >= seminfo.semmni) return (EINVAL); semaptr = &sema[semid]; sema_mtxp = &sema_mtx[semid]; error = 0; rval = 0; switch (cmd) { case IPC_RMID: mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M))) goto done2; semaptr->sem_perm.cuid = cred->cr_uid; semaptr->sem_perm.uid = cred->cr_uid; semtot -= semaptr->sem_nsems; for (i = semaptr->sem_base - sem; i < semtot; i++) sem[i] = sem[i + semaptr->sem_nsems]; for (i = 0; i < seminfo.semmni; i++) { if ((sema[i].sem_perm.mode & SEM_ALLOC) && sema[i].sem_base > semaptr->sem_base) sema[i].sem_base -= semaptr->sem_nsems; } semaptr->sem_perm.mode = 0; SEMUNDO_LOCK(); semundo_clear(semid, -1); SEMUNDO_UNLOCK(); wakeup(semaptr); break; case IPC_SET: if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0) goto done2; mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M))) goto done2; semaptr->sem_perm.uid = sbuf.sem_perm.uid; semaptr->sem_perm.gid = sbuf.sem_perm.gid; semaptr->sem_perm.mode = (semaptr->sem_perm.mode & ~0777) | (sbuf.sem_perm.mode & 0777); semaptr->sem_ctime = time_second; break; case IPC_STAT: if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; sbuf = *semaptr; mtx_unlock(sema_mtxp); error = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds)); break; case GETNCNT: mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } rval = semaptr->sem_base[semnum].semncnt; break; case GETPID: mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } rval = semaptr->sem_base[semnum].sempid; break; case GETVAL: mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } rval = semaptr->sem_base[semnum].semval; break; case GETALL: if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; array = malloc(sizeof(*array) * semaptr->sem_nsems, M_TEMP, M_WAITOK); mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; for (i = 0; i < semaptr->sem_nsems; i++) array[i] = semaptr->sem_base[i].semval; mtx_unlock(sema_mtxp); error = copyout(array, real_arg.array, i * sizeof(real_arg.array[0])); break; case GETZCNT: mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } rval = semaptr->sem_base[semnum].semzcnt; break; case SETVAL: if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } if (real_arg.val < 0 || real_arg.val > seminfo.semvmx) { error = ERANGE; goto done2; } semaptr->sem_base[semnum].semval = real_arg.val; SEMUNDO_LOCK(); semundo_clear(semid, semnum); SEMUNDO_UNLOCK(); wakeup(semaptr); break; case SETALL: mtx_lock(sema_mtxp); raced: if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; count = semaptr->sem_nsems; mtx_unlock(sema_mtxp); if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; array = malloc(sizeof(*array) * count, M_TEMP, M_WAITOK); copyin(real_arg.array, array, count * sizeof(*array)); if (error) break; mtx_lock(sema_mtxp); if ((error = semvalid(uap->semid, semaptr)) != 0) goto done2; /* we could have raced? */ if (count != semaptr->sem_nsems) { free(array, M_TEMP); array = NULL; goto raced; } if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W))) goto done2; for (i = 0; i < semaptr->sem_nsems; i++) { usval = array[i]; if (usval > seminfo.semvmx) { error = ERANGE; break; } semaptr->sem_base[i].semval = usval; } SEMUNDO_LOCK(); semundo_clear(semid, -1); SEMUNDO_UNLOCK(); wakeup(semaptr); break; default: error = EINVAL; break; } if (error == 0) td->td_retval[0] = rval; done2: if (mtx_owned(sema_mtxp)) mtx_unlock(sema_mtxp); if (array != NULL) free(array, M_TEMP); return(error); } #ifndef _SYS_SYSPROTO_H_ struct semget_args { key_t key; int nsems; int semflg; }; #endif /* * MPSAFE */ int semget(td, uap) struct thread *td; struct semget_args *uap; { int semid, error = 0; int key = uap->key; int nsems = uap->nsems; int semflg = uap->semflg; struct ucred *cred = td->td_ucred; DPRINTF(("semget(0x%x, %d, 0%o)\n", key, nsems, semflg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); if (key != IPC_PRIVATE) { for (semid = 0; semid < seminfo.semmni; semid++) { if ((sema[semid].sem_perm.mode & SEM_ALLOC) && sema[semid].sem_perm.key == key) break; } if (semid < seminfo.semmni) { DPRINTF(("found public key\n")); if ((error = ipcperm(td, &sema[semid].sem_perm, semflg & 0700))) { goto done2; } if (nsems > 0 && sema[semid].sem_nsems < nsems) { DPRINTF(("too small\n")); error = EINVAL; goto done2; } if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { DPRINTF(("not exclusive\n")); error = EEXIST; goto done2; } goto found; } } DPRINTF(("need to allocate the semid_ds\n")); if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { if (nsems <= 0 || nsems > seminfo.semmsl) { DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems, seminfo.semmsl)); error = EINVAL; goto done2; } if (nsems > seminfo.semmns - semtot) { DPRINTF(( "not enough semaphores left (need %d, got %d)\n", nsems, seminfo.semmns - semtot)); error = ENOSPC; goto done2; } for (semid = 0; semid < seminfo.semmni; semid++) { if ((sema[semid].sem_perm.mode & SEM_ALLOC) == 0) break; } if (semid == seminfo.semmni) { DPRINTF(("no more semid_ds's available\n")); error = ENOSPC; goto done2; } DPRINTF(("semid %d is available\n", semid)); sema[semid].sem_perm.key = key; sema[semid].sem_perm.cuid = cred->cr_uid; sema[semid].sem_perm.uid = cred->cr_uid; sema[semid].sem_perm.cgid = cred->cr_gid; sema[semid].sem_perm.gid = cred->cr_gid; sema[semid].sem_perm.mode = (semflg & 0777) | SEM_ALLOC; sema[semid].sem_perm.seq = (sema[semid].sem_perm.seq + 1) & 0x7fff; sema[semid].sem_nsems = nsems; sema[semid].sem_otime = 0; sema[semid].sem_ctime = time_second; sema[semid].sem_base = &sem[semtot]; semtot += nsems; bzero(sema[semid].sem_base, sizeof(sema[semid].sem_base[0])*nsems); DPRINTF(("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, &sem[semtot])); } else { DPRINTF(("didn't find it and wasn't asked to create it\n")); error = ENOENT; goto done2; } found: td->td_retval[0] = IXSEQ_TO_IPCID(semid, sema[semid].sem_perm); done2: mtx_unlock(&Giant); return (error); } #ifndef _SYS_SYSPROTO_H_ struct semop_args { int semid; struct sembuf *sops; size_t nsops; }; #endif /* * MPSAFE */ int semop(td, uap) struct thread *td; struct semop_args *uap; { int semid = uap->semid; size_t nsops = uap->nsops; struct sembuf *sops; struct semid_ds *semaptr; struct sembuf *sopptr = 0; struct sem *semptr = 0; struct sem_undo *suptr; struct mtx *sema_mtxp; size_t i, j, k; int error; int do_wakeup, do_undos; DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ if (semid < 0 || semid >= seminfo.semmni) return (EINVAL); /* Allocate memory for sem_ops */ if (nsops > seminfo.semopm) { DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm, nsops)); return (E2BIG); } sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK); if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) { DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error, uap->sops, sops, nsops * sizeof(sops[0]))); free(sops, M_SEM); return (error); } semaptr = &sema[semid]; sema_mtxp = &sema_mtx[semid]; mtx_lock(sema_mtxp); if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) { error = EINVAL; goto done2; } if (semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { error = EINVAL; goto done2; } /* * Initial pass thru sops to see what permissions are needed. * Also perform any checks that don't need repeating on each * attempt to satisfy the request vector. */ j = 0; /* permission needed */ do_undos = 0; for (i = 0; i < nsops; i++) { sopptr = &sops[i]; if (sopptr->sem_num >= semaptr->sem_nsems) { error = EFBIG; goto done2; } if (sopptr->sem_flg & SEM_UNDO && sopptr->sem_op != 0) do_undos = 1; j |= (sopptr->sem_op == 0) ? SEM_R : SEM_A; } if ((error = ipcperm(td, &semaptr->sem_perm, j))) { DPRINTF(("error = %d from ipaccess\n", error)); goto done2; } /* * Loop trying to satisfy the vector of requests. * If we reach a point where we must wait, any requests already * performed are rolled back and we go to sleep until some other * process wakes us up. At this point, we start all over again. * * This ensures that from the perspective of other tasks, a set * of requests is atomic (never partially satisfied). */ for (;;) { do_wakeup = 0; error = 0; /* error return if necessary */ for (i = 0; i < nsops; i++) { sopptr = &sops[i]; semptr = &semaptr->sem_base[sopptr->sem_num]; DPRINTF(( "semop: semaptr=%x, sem_base=%x, " "semptr=%x, sem[%d]=%d : op=%d, flag=%s\n", semaptr, semaptr->sem_base, semptr, sopptr->sem_num, semptr->semval, sopptr->sem_op, (sopptr->sem_flg & IPC_NOWAIT) ? "nowait" : "wait")); if (sopptr->sem_op < 0) { if (semptr->semval + sopptr->sem_op < 0) { DPRINTF(("semop: can't do it now\n")); break; } else { semptr->semval += sopptr->sem_op; if (semptr->semval == 0 && semptr->semzcnt > 0) do_wakeup = 1; } } else if (sopptr->sem_op == 0) { if (semptr->semval != 0) { DPRINTF(("semop: not zero now\n")); break; } } else if (semptr->semval + sopptr->sem_op > seminfo.semvmx) { error = ERANGE; break; } else { if (semptr->semncnt > 0) do_wakeup = 1; semptr->semval += sopptr->sem_op; } } /* * Did we get through the entire vector? */ if (i >= nsops) goto done; /* * No ... rollback anything that we've already done */ DPRINTF(("semop: rollback 0 through %d\n", i-1)); for (j = 0; j < i; j++) semaptr->sem_base[sops[j].sem_num].semval -= sops[j].sem_op; /* If we detected an error, return it */ if (error != 0) goto done2; /* * If the request that we couldn't satisfy has the * NOWAIT flag set then return with EAGAIN. */ if (sopptr->sem_flg & IPC_NOWAIT) { error = EAGAIN; goto done2; } if (sopptr->sem_op == 0) semptr->semzcnt++; else semptr->semncnt++; DPRINTF(("semop: good night!\n")); error = msleep(semaptr, sema_mtxp, (PZERO - 4) | PCATCH, "semwait", 0); DPRINTF(("semop: good morning (error=%d)!\n", error)); if (error != 0) { error = EINTR; goto done2; } DPRINTF(("semop: good morning!\n")); /* * Make sure that the semaphore still exists */ if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { error = EIDRM; goto done2; } /* * The semaphore is still alive. Readjust the count of * waiting processes. */ if (sopptr->sem_op == 0) semptr->semzcnt--; else semptr->semncnt--; } done: /* * Process any SEM_UNDO requests. */ if (do_undos) { SEMUNDO_LOCK(); suptr = NULL; for (i = 0; i < nsops; i++) { /* * We only need to deal with SEM_UNDO's for non-zero * op's. */ int adjval; if ((sops[i].sem_flg & SEM_UNDO) == 0) continue; adjval = sops[i].sem_op; if (adjval == 0) continue; error = semundo_adjust(td, &suptr, semid, sops[i].sem_num, -adjval); if (error == 0) continue; /* * Oh-Oh! We ran out of either sem_undo's or undo's. * Rollback the adjustments to this point and then * rollback the semaphore ups and down so we can return * with an error with all structures restored. We * rollback the undo's in the exact reverse order that * we applied them. This guarantees that we won't run * out of space as we roll things back out. */ for (j = 0; j < i; j++) { k = i - j - 1; if ((sops[k].sem_flg & SEM_UNDO) == 0) continue; adjval = sops[k].sem_op; if (adjval == 0) continue; if (semundo_adjust(td, &suptr, semid, sops[k].sem_num, adjval) != 0) panic("semop - can't undo undos"); } for (j = 0; j < nsops; j++) semaptr->sem_base[sops[j].sem_num].semval -= sops[j].sem_op; DPRINTF(("error = %d from semundo_adjust\n", error)); SEMUNDO_UNLOCK(); goto done2; } /* loop through the sops */ SEMUNDO_UNLOCK(); } /* if (do_undos) */ /* We're definitely done - set the sempid's and time */ for (i = 0; i < nsops; i++) { sopptr = &sops[i]; semptr = &semaptr->sem_base[sopptr->sem_num]; semptr->sempid = td->td_proc->p_pid; } semaptr->sem_otime = time_second; /* * Do a wakeup if any semaphore was up'd whilst something was * sleeping on it. */ if (do_wakeup) { DPRINTF(("semop: doing wakeup\n")); wakeup(semaptr); DPRINTF(("semop: back from wakeup\n")); } DPRINTF(("semop: done\n")); td->td_retval[0] = 0; done2: mtx_unlock(sema_mtxp); free(sops, M_SEM); return (error); } /* * Go through the undo structures for this process and apply the adjustments to * semaphores. */ static void semexit_myhook(arg, p) void *arg; struct proc *p; { struct sem_undo *suptr; struct sem_undo **supptr; /* * Go through the chain of undo vectors looking for one * associated with this process. */ SEMUNDO_LOCK(); SLIST_FOREACH_PREVPTR(suptr, supptr, &semu_list, un_next) { if (suptr->un_proc == p) break; } SEMUNDO_UNLOCK(); if (suptr == NULL) return; DPRINTF(("proc @%08x has undo structure with %d entries\n", p, suptr->un_cnt)); /* * If there are any active undo elements then process them. */ if (suptr->un_cnt > 0) { int ix; for (ix = 0; ix < suptr->un_cnt; ix++) { int semid = suptr->un_ent[ix].un_id; int semnum = suptr->un_ent[ix].un_num; int adjval = suptr->un_ent[ix].un_adjval; struct semid_ds *semaptr; struct mtx *sema_mtxp; semaptr = &sema[semid]; sema_mtxp = &sema_mtx[semid]; mtx_lock(sema_mtxp); SEMUNDO_LOCK(); if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) panic("semexit - semid not allocated"); if (semnum >= semaptr->sem_nsems) panic("semexit - semnum out of range"); DPRINTF(( "semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n", suptr->un_proc, suptr->un_ent[ix].un_id, suptr->un_ent[ix].un_num, suptr->un_ent[ix].un_adjval, semaptr->sem_base[semnum].semval)); if (adjval < 0) { if (semaptr->sem_base[semnum].semval < -adjval) semaptr->sem_base[semnum].semval = 0; else semaptr->sem_base[semnum].semval += adjval; } else semaptr->sem_base[semnum].semval += adjval; wakeup(semaptr); DPRINTF(("semexit: back from wakeup\n")); mtx_unlock(sema_mtxp); SEMUNDO_UNLOCK(); } } /* * Deallocate the undo vector. */ DPRINTF(("removing vector\n")); suptr->un_proc = NULL; *supptr = SLIST_NEXT(suptr, un_next); } static int sysctl_sema(SYSCTL_HANDLER_ARGS) { return (SYSCTL_OUT(req, sema, sizeof(struct semid_ds) * seminfo.semmni)); } Index: head/sys/kern/sysv_shm.c =================================================================== --- head/sys/kern/sysv_shm.c (revision 122200) +++ head/sys/kern/sysv_shm.c (revision 122201) @@ -1,953 +1,953 @@ /* $NetBSD: sysv_shm.c,v 1.23 1994/07/04 23:25:12 glass Exp $ */ /* * Copyright (c) 1994 Adam Glass and Charles Hannum. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by Adam Glass and Charles * Hannum. * 4. The names of the authors may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_compat.h" #include "opt_sysvipc.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_SHM, "shm", "SVID compatible shared memory segments"); struct oshmctl_args; static int oshmctl(struct thread *td, struct oshmctl_args *uap); static int shmget_allocate_segment(struct thread *td, struct shmget_args *uap, int mode); static int shmget_existing(struct thread *td, struct shmget_args *uap, int mode, int segnum); /* XXX casting to (sy_call_t *) is bogus, as usual. */ static sy_call_t *shmcalls[] = { (sy_call_t *)shmat, (sy_call_t *)oshmctl, (sy_call_t *)shmdt, (sy_call_t *)shmget, (sy_call_t *)shmctl }; #define SHMSEG_FREE 0x0200 #define SHMSEG_REMOVED 0x0400 #define SHMSEG_ALLOCATED 0x0800 #define SHMSEG_WANTED 0x1000 static int shm_last_free, shm_nused, shm_committed, shmalloced; static struct shmid_ds *shmsegs; struct shm_handle { /* vm_offset_t kva; */ vm_object_t shm_object; }; struct shmmap_state { vm_offset_t va; int shmid; }; static void shm_deallocate_segment(struct shmid_ds *); static int shm_find_segment_by_key(key_t); static struct shmid_ds *shm_find_segment_by_shmid(int); static struct shmid_ds *shm_find_segment_by_shmidx(int); static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *); static void shmrealloc(void); static void shminit(void); static int sysvshm_modload(struct module *, int, void *); static int shmunload(void); static void shmexit_myhook(struct vmspace *vm); static void shmfork_myhook(struct proc *p1, struct proc *p2); static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS); /* * Tuneable values. */ #ifndef SHMMAXPGS #define SHMMAXPGS 8192 /* Note: sysv shared memory is swap backed. */ #endif #ifndef SHMMAX #define SHMMAX (SHMMAXPGS*PAGE_SIZE) #endif #ifndef SHMMIN #define SHMMIN 1 #endif #ifndef SHMMNI #define SHMMNI 192 #endif #ifndef SHMSEG #define SHMSEG 128 #endif #ifndef SHMALL #define SHMALL (SHMMAXPGS) #endif struct shminfo shminfo = { SHMMAX, SHMMIN, SHMMNI, SHMSEG, SHMALL }; static int shm_use_phys; static int shm_allow_removed; SYSCTL_DECL(_kern_ipc); SYSCTL_INT(_kern_ipc, OID_AUTO, shmmax, CTLFLAG_RW, &shminfo.shmmax, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shmmin, CTLFLAG_RW, &shminfo.shmmin, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shmmni, CTLFLAG_RDTUN, &shminfo.shmmni, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shmseg, CTLFLAG_RDTUN, &shminfo.shmseg, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shmall, CTLFLAG_RW, &shminfo.shmall, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shm_use_phys, CTLFLAG_RW, &shm_use_phys, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, shm_allow_removed, CTLFLAG_RW, &shm_allow_removed, 0, ""); SYSCTL_PROC(_kern_ipc, OID_AUTO, shmsegs, CTLFLAG_RD, NULL, 0, sysctl_shmsegs, "", ""); static int shm_find_segment_by_key(key) key_t key; { int i; for (i = 0; i < shmalloced; i++) if ((shmsegs[i].shm_perm.mode & SHMSEG_ALLOCATED) && shmsegs[i].shm_perm.key == key) return (i); return (-1); } static struct shmid_ds * shm_find_segment_by_shmid(int shmid) { int segnum; struct shmid_ds *shmseg; segnum = IPCID_TO_IX(shmid); if (segnum < 0 || segnum >= shmalloced) return (NULL); shmseg = &shmsegs[segnum]; if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 || (!shm_allow_removed && (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0) || shmseg->shm_perm.seq != IPCID_TO_SEQ(shmid)) return (NULL); return (shmseg); } static struct shmid_ds * shm_find_segment_by_shmidx(int segnum) { struct shmid_ds *shmseg; if (segnum < 0 || segnum >= shmalloced) return (NULL); shmseg = &shmsegs[segnum]; if ((shmseg->shm_perm.mode & SHMSEG_ALLOCATED) == 0 || (!shm_allow_removed && (shmseg->shm_perm.mode & SHMSEG_REMOVED) != 0)) return (NULL); return (shmseg); } static void shm_deallocate_segment(shmseg) struct shmid_ds *shmseg; { struct shm_handle *shm_handle; size_t size; GIANT_REQUIRED; shm_handle = shmseg->shm_internal; vm_object_deallocate(shm_handle->shm_object); free(shm_handle, M_SHM); shmseg->shm_internal = NULL; size = round_page(shmseg->shm_segsz); shm_committed -= btoc(size); shm_nused--; shmseg->shm_perm.mode = SHMSEG_FREE; } static int shm_delete_mapping(struct vmspace *vm, struct shmmap_state *shmmap_s) { struct shmid_ds *shmseg; int segnum, result; size_t size; GIANT_REQUIRED; segnum = IPCID_TO_IX(shmmap_s->shmid); shmseg = &shmsegs[segnum]; size = round_page(shmseg->shm_segsz); result = vm_map_remove(&vm->vm_map, shmmap_s->va, shmmap_s->va + size); if (result != KERN_SUCCESS) return (EINVAL); shmmap_s->shmid = -1; shmseg->shm_dtime = time_second; if ((--shmseg->shm_nattch <= 0) && (shmseg->shm_perm.mode & SHMSEG_REMOVED)) { shm_deallocate_segment(shmseg); shm_last_free = segnum; } return (0); } #ifndef _SYS_SYSPROTO_H_ struct shmdt_args { const void *shmaddr; }; #endif /* * MPSAFE */ int shmdt(td, uap) struct thread *td; struct shmdt_args *uap; { struct proc *p = td->td_proc; struct shmmap_state *shmmap_s; int i; int error = 0; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); shmmap_s = p->p_vmspace->vm_shm; if (shmmap_s == NULL) { error = EINVAL; goto done2; } for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) { if (shmmap_s->shmid != -1 && shmmap_s->va == (vm_offset_t)uap->shmaddr) { break; } } if (i == shminfo.shmseg) { error = EINVAL; goto done2; } error = shm_delete_mapping(p->p_vmspace, shmmap_s); done2: mtx_unlock(&Giant); return (error); } #ifndef _SYS_SYSPROTO_H_ struct shmat_args { int shmid; const void *shmaddr; int shmflg; }; #endif /* * MPSAFE */ int kern_shmat(td, shmid, shmaddr, shmflg) struct thread *td; int shmid; const void *shmaddr; int shmflg; { struct proc *p = td->td_proc; int i, flags; struct shmid_ds *shmseg; struct shmmap_state *shmmap_s = NULL; struct shm_handle *shm_handle; vm_offset_t attach_va; vm_prot_t prot; vm_size_t size; int rv; int error = 0; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); shmmap_s = p->p_vmspace->vm_shm; if (shmmap_s == NULL) { size = shminfo.shmseg * sizeof(struct shmmap_state); shmmap_s = malloc(size, M_SHM, M_WAITOK); for (i = 0; i < shminfo.shmseg; i++) shmmap_s[i].shmid = -1; p->p_vmspace->vm_shm = shmmap_s; } shmseg = shm_find_segment_by_shmid(shmid); if (shmseg == NULL) { error = EINVAL; goto done2; } error = ipcperm(td, &shmseg->shm_perm, (shmflg & SHM_RDONLY) ? IPC_R : IPC_R|IPC_W); if (error) goto done2; for (i = 0; i < shminfo.shmseg; i++) { if (shmmap_s->shmid == -1) break; shmmap_s++; } if (i >= shminfo.shmseg) { error = EMFILE; goto done2; } size = round_page(shmseg->shm_segsz); #ifdef VM_PROT_READ_IS_EXEC prot = VM_PROT_READ | VM_PROT_EXECUTE; #else prot = VM_PROT_READ; #endif if ((shmflg & SHM_RDONLY) == 0) prot |= VM_PROT_WRITE; flags = MAP_ANON | MAP_SHARED; if (shmaddr) { flags |= MAP_FIXED; if (shmflg & SHM_RND) { attach_va = (vm_offset_t)shmaddr & ~(SHMLBA-1); } else if (((vm_offset_t)shmaddr & (SHMLBA-1)) == 0) { attach_va = (vm_offset_t)shmaddr; } else { error = EINVAL; goto done2; } } else { /* * This is just a hint to vm_map_find() about where to * put it. */ attach_va = round_page((vm_offset_t)p->p_vmspace->vm_taddr + maxtsiz + maxdsiz); } shm_handle = shmseg->shm_internal; vm_object_reference(shm_handle->shm_object); rv = vm_map_find(&p->p_vmspace->vm_map, shm_handle->shm_object, 0, &attach_va, size, (flags & MAP_FIXED)?0:1, prot, prot, 0); if (rv != KERN_SUCCESS) { error = ENOMEM; goto done2; } vm_map_inherit(&p->p_vmspace->vm_map, attach_va, attach_va + size, VM_INHERIT_SHARE); shmmap_s->va = attach_va; shmmap_s->shmid = shmid; shmseg->shm_lpid = p->p_pid; shmseg->shm_atime = time_second; shmseg->shm_nattch++; td->td_retval[0] = attach_va; done2: mtx_unlock(&Giant); return (error); } -int +int shmat(td, uap) struct thread *td; struct shmat_args *uap; { return kern_shmat(td, uap->shmid, uap->shmaddr, uap->shmflg); } struct oshmid_ds { struct ipc_perm shm_perm; /* operation perms */ int shm_segsz; /* size of segment (bytes) */ u_short shm_cpid; /* pid, creator */ u_short shm_lpid; /* pid, last operation */ short shm_nattch; /* no. of current attaches */ time_t shm_atime; /* last attach time */ time_t shm_dtime; /* last detach time */ time_t shm_ctime; /* last change time */ void *shm_handle; /* internal handle for shm segment */ }; struct oshmctl_args { int shmid; int cmd; struct oshmid_ds *ubuf; }; /* * MPSAFE */ static int oshmctl(td, uap) struct thread *td; struct oshmctl_args *uap; { #ifdef COMPAT_43 int error = 0; struct shmid_ds *shmseg; struct oshmid_ds outbuf; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); shmseg = shm_find_segment_by_shmid(uap->shmid); if (shmseg == NULL) { error = EINVAL; goto done2; } switch (uap->cmd) { case IPC_STAT: error = ipcperm(td, &shmseg->shm_perm, IPC_R); if (error) goto done2; outbuf.shm_perm = shmseg->shm_perm; outbuf.shm_segsz = shmseg->shm_segsz; outbuf.shm_cpid = shmseg->shm_cpid; outbuf.shm_lpid = shmseg->shm_lpid; outbuf.shm_nattch = shmseg->shm_nattch; outbuf.shm_atime = shmseg->shm_atime; outbuf.shm_dtime = shmseg->shm_dtime; outbuf.shm_ctime = shmseg->shm_ctime; outbuf.shm_handle = shmseg->shm_internal; error = copyout(&outbuf, uap->ubuf, sizeof(outbuf)); if (error) goto done2; break; default: /* XXX casting to (sy_call_t *) is bogus, as usual. */ error = ((sy_call_t *)shmctl)(td, uap); break; } done2: mtx_unlock(&Giant); return (error); #else return (EINVAL); #endif } #ifndef _SYS_SYSPROTO_H_ struct shmctl_args { int shmid; int cmd; struct shmid_ds *buf; }; #endif /* * MPSAFE */ int kern_shmctl(td, shmid, cmd, buf, bufsz) struct thread *td; int shmid; int cmd; void *buf; size_t *bufsz; { int error = 0; struct shmid_ds *shmseg; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); switch (cmd) { case IPC_INFO: memcpy(buf, &shminfo, sizeof(shminfo)); if (bufsz) *bufsz = sizeof(shminfo); td->td_retval[0] = shmalloced; goto done2; case SHM_INFO: { struct shm_info shm_info; shm_info.used_ids = shm_nused; shm_info.shm_rss = 0; /*XXX where to get from ? */ shm_info.shm_tot = 0; /*XXX where to get from ? */ shm_info.shm_swp = 0; /*XXX where to get from ? */ shm_info.swap_attempts = 0; /*XXX where to get from ? */ shm_info.swap_successes = 0; /*XXX where to get from ? */ memcpy(buf, &shm_info, sizeof(shm_info)); if (bufsz) *bufsz = sizeof(shm_info); td->td_retval[0] = shmalloced; goto done2; } } if (cmd == SHM_STAT) shmseg = shm_find_segment_by_shmidx(shmid); else shmseg = shm_find_segment_by_shmid(shmid); if (shmseg == NULL) { error = EINVAL; goto done2; } switch (cmd) { case SHM_STAT: case IPC_STAT: error = ipcperm(td, &shmseg->shm_perm, IPC_R); if (error) goto done2; memcpy(buf, shmseg, sizeof(struct shmid_ds)); if (bufsz) *bufsz = sizeof(struct shmid_ds); if (cmd == SHM_STAT) td->td_retval[0] = IXSEQ_TO_IPCID(shmid, shmseg->shm_perm); break; case IPC_SET: { struct shmid_ds *shmid; shmid = (struct shmid_ds *)buf; error = ipcperm(td, &shmseg->shm_perm, IPC_M); if (error) goto done2; shmseg->shm_perm.uid = shmid->shm_perm.uid; shmseg->shm_perm.gid = shmid->shm_perm.gid; shmseg->shm_perm.mode = (shmseg->shm_perm.mode & ~ACCESSPERMS) | (shmid->shm_perm.mode & ACCESSPERMS); shmseg->shm_ctime = time_second; break; } case IPC_RMID: error = ipcperm(td, &shmseg->shm_perm, IPC_M); if (error) goto done2; shmseg->shm_perm.key = IPC_PRIVATE; shmseg->shm_perm.mode |= SHMSEG_REMOVED; if (shmseg->shm_nattch <= 0) { shm_deallocate_segment(shmseg); shm_last_free = IPCID_TO_IX(shmid); } break; #if 0 case SHM_LOCK: case SHM_UNLOCK: #endif default: error = EINVAL; break; } done2: mtx_unlock(&Giant); return (error); } int shmctl(td, uap) struct thread *td; struct shmctl_args *uap; { int error = 0; struct shmid_ds buf; size_t bufsz; /* IPC_SET needs to copyin the buffer before calling kern_shmctl */ if (uap->cmd == IPC_SET) { if ((error = copyin(uap->buf, &buf, sizeof(struct shmid_ds)))) goto done; } error = kern_shmctl(td, uap->shmid, uap->cmd, (void *)&buf, &bufsz); if (error) goto done; /* Cases in which we need to copyout */ switch (uap->cmd) { case IPC_INFO: case SHM_INFO: case SHM_STAT: case IPC_STAT: error = copyout(&buf, uap->buf, bufsz); break; } done: if (error) { /* Invalidate the return value */ td->td_retval[0] = -1; } return (error); } #ifndef _SYS_SYSPROTO_H_ struct shmget_args { key_t key; size_t size; int shmflg; }; #endif static int shmget_existing(td, uap, mode, segnum) struct thread *td; struct shmget_args *uap; int mode; int segnum; { struct shmid_ds *shmseg; int error; shmseg = &shmsegs[segnum]; if (shmseg->shm_perm.mode & SHMSEG_REMOVED) { /* * This segment is in the process of being allocated. Wait * until it's done, and look the key up again (in case the * allocation failed or it was freed). */ shmseg->shm_perm.mode |= SHMSEG_WANTED; error = tsleep(shmseg, PLOCK | PCATCH, "shmget", 0); if (error) return (error); return (EAGAIN); } if ((uap->shmflg & (IPC_CREAT | IPC_EXCL)) == (IPC_CREAT | IPC_EXCL)) return (EEXIST); error = ipcperm(td, &shmseg->shm_perm, mode); if (error) return (error); if (uap->size && uap->size > shmseg->shm_segsz) return (EINVAL); td->td_retval[0] = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); return (0); } static int shmget_allocate_segment(td, uap, mode) struct thread *td; struct shmget_args *uap; int mode; { int i, segnum, shmid, size; struct ucred *cred = td->td_ucred; struct shmid_ds *shmseg; struct shm_handle *shm_handle; GIANT_REQUIRED; if (uap->size < shminfo.shmmin || uap->size > shminfo.shmmax) return (EINVAL); if (shm_nused >= shminfo.shmmni) /* Any shmids left? */ return (ENOSPC); size = round_page(uap->size); if (shm_committed + btoc(size) > shminfo.shmall) return (ENOMEM); if (shm_last_free < 0) { shmrealloc(); /* Maybe expand the shmsegs[] array. */ for (i = 0; i < shmalloced; i++) if (shmsegs[i].shm_perm.mode & SHMSEG_FREE) break; if (i == shmalloced) return (ENOSPC); segnum = i; } else { segnum = shm_last_free; shm_last_free = -1; } shmseg = &shmsegs[segnum]; /* * In case we sleep in malloc(), mark the segment present but deleted * so that noone else tries to create the same key. */ shmseg->shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED; shmseg->shm_perm.key = uap->key; shmseg->shm_perm.seq = (shmseg->shm_perm.seq + 1) & 0x7fff; shm_handle = (struct shm_handle *) malloc(sizeof(struct shm_handle), M_SHM, M_WAITOK); shmid = IXSEQ_TO_IPCID(segnum, shmseg->shm_perm); /* * We make sure that we have allocated a pager before we need * to. */ if (shm_use_phys) { shm_handle->shm_object = vm_pager_allocate(OBJT_PHYS, 0, size, VM_PROT_DEFAULT, 0); } else { shm_handle->shm_object = vm_pager_allocate(OBJT_SWAP, 0, size, VM_PROT_DEFAULT, 0); } VM_OBJECT_LOCK(shm_handle->shm_object); vm_object_clear_flag(shm_handle->shm_object, OBJ_ONEMAPPING); vm_object_set_flag(shm_handle->shm_object, OBJ_NOSPLIT); VM_OBJECT_UNLOCK(shm_handle->shm_object); shmseg->shm_internal = shm_handle; shmseg->shm_perm.cuid = shmseg->shm_perm.uid = cred->cr_uid; shmseg->shm_perm.cgid = shmseg->shm_perm.gid = cred->cr_gid; shmseg->shm_perm.mode = (shmseg->shm_perm.mode & SHMSEG_WANTED) | (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; shmseg->shm_segsz = uap->size; shmseg->shm_cpid = td->td_proc->p_pid; shmseg->shm_lpid = shmseg->shm_nattch = 0; shmseg->shm_atime = shmseg->shm_dtime = 0; shmseg->shm_ctime = time_second; shm_committed += btoc(size); shm_nused++; if (shmseg->shm_perm.mode & SHMSEG_WANTED) { /* * Somebody else wanted this key while we were asleep. Wake * them up now. */ shmseg->shm_perm.mode &= ~SHMSEG_WANTED; wakeup(shmseg); } td->td_retval[0] = shmid; return (0); } /* * MPSAFE */ int shmget(td, uap) struct thread *td; struct shmget_args *uap; { int segnum, mode; int error; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); mode = uap->shmflg & ACCESSPERMS; if (uap->key != IPC_PRIVATE) { again: segnum = shm_find_segment_by_key(uap->key); if (segnum >= 0) { error = shmget_existing(td, uap, mode, segnum); if (error == EAGAIN) goto again; goto done2; } if ((uap->shmflg & IPC_CREAT) == 0) { error = ENOENT; goto done2; } } error = shmget_allocate_segment(td, uap, mode); done2: mtx_unlock(&Giant); return (error); } /* * MPSAFE */ int shmsys(td, uap) struct thread *td; /* XXX actually varargs. */ struct shmsys_args /* { int which; int a2; int a3; int a4; } */ *uap; { int error; if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); if (uap->which < 0 || uap->which >= sizeof(shmcalls)/sizeof(shmcalls[0])) return (EINVAL); mtx_lock(&Giant); error = (*shmcalls[uap->which])(td, &uap->a2); mtx_unlock(&Giant); return (error); } static void shmfork_myhook(p1, p2) struct proc *p1, *p2; { struct shmmap_state *shmmap_s; size_t size; int i; size = shminfo.shmseg * sizeof(struct shmmap_state); shmmap_s = malloc(size, M_SHM, M_WAITOK); bcopy(p1->p_vmspace->vm_shm, shmmap_s, size); p2->p_vmspace->vm_shm = shmmap_s; for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) if (shmmap_s->shmid != -1) shmsegs[IPCID_TO_IX(shmmap_s->shmid)].shm_nattch++; } static void shmexit_myhook(struct vmspace *vm) { struct shmmap_state *base, *shm; int i; GIANT_REQUIRED; if ((base = vm->vm_shm) != NULL) { vm->vm_shm = NULL; for (i = 0, shm = base; i < shminfo.shmseg; i++, shm++) { if (shm->shmid != -1) shm_delete_mapping(vm, shm); } free(base, M_SHM); } } static void shmrealloc(void) { int i; struct shmid_ds *newsegs; if (shmalloced >= shminfo.shmmni) return; newsegs = malloc(shminfo.shmmni * sizeof(*newsegs), M_SHM, M_WAITOK); if (newsegs == NULL) return; for (i = 0; i < shmalloced; i++) bcopy(&shmsegs[i], &newsegs[i], sizeof(newsegs[0])); for (; i < shminfo.shmmni; i++) { shmsegs[i].shm_perm.mode = SHMSEG_FREE; shmsegs[i].shm_perm.seq = 0; } free(shmsegs, M_SHM); shmsegs = newsegs; shmalloced = shminfo.shmmni; } static void shminit() { int i; TUNABLE_INT_FETCH("kern.ipc.shmmaxpgs", &shminfo.shmall); for (i = PAGE_SIZE; i > 0; i--) { shminfo.shmmax = shminfo.shmall * PAGE_SIZE; if (shminfo.shmmax >= shminfo.shmall) break; } TUNABLE_INT_FETCH("kern.ipc.shmmin", &shminfo.shmmin); TUNABLE_INT_FETCH("kern.ipc.shmmni", &shminfo.shmmni); TUNABLE_INT_FETCH("kern.ipc.shmseg", &shminfo.shmseg); TUNABLE_INT_FETCH("kern.ipc.shm_use_phys", &shm_use_phys); shmalloced = shminfo.shmmni; shmsegs = malloc(shmalloced * sizeof(shmsegs[0]), M_SHM, M_WAITOK); if (shmsegs == NULL) panic("cannot allocate initial memory for sysvshm"); for (i = 0; i < shmalloced; i++) { shmsegs[i].shm_perm.mode = SHMSEG_FREE; shmsegs[i].shm_perm.seq = 0; } shm_last_free = 0; shm_nused = 0; shm_committed = 0; shmexit_hook = &shmexit_myhook; shmfork_hook = &shmfork_myhook; } static int shmunload() { if (shm_nused > 0) return (EBUSY); free(shmsegs, M_SHM); shmexit_hook = NULL; shmfork_hook = NULL; return (0); } static int sysctl_shmsegs(SYSCTL_HANDLER_ARGS) { return (SYSCTL_OUT(req, shmsegs, shmalloced * sizeof(shmsegs[0]))); } static int sysvshm_modload(struct module *module, int cmd, void *arg) { int error = 0; switch (cmd) { case MOD_LOAD: shminit(); break; case MOD_UNLOAD: error = shmunload(); break; case MOD_SHUTDOWN: break; default: error = EINVAL; break; } return (error); } static moduledata_t sysvshm_mod = { "sysvshm", &sysvshm_modload, NULL }; SYSCALL_MODULE_HELPER(shmsys); SYSCALL_MODULE_HELPER(shmat); SYSCALL_MODULE_HELPER(shmctl); SYSCALL_MODULE_HELPER(shmdt); SYSCALL_MODULE_HELPER(shmget); DECLARE_MODULE(sysvshm, sysvshm_mod, SI_SUB_SYSV_SHM, SI_ORDER_FIRST); MODULE_VERSION(sysvshm, 1);