Index: head/sys/kern/sysv_msg.c =================================================================== --- head/sys/kern/sysv_msg.c (revision 100522) +++ head/sys/kern/sysv_msg.c (revision 100523) @@ -1,1238 +1,1137 @@ /* $FreeBSD$ */ /* * 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 "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 *); -#define MSG_DEBUG -#undef MSG_DEBUG_OK +#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 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) { - printf("msginfo.msgssz=%d (0x%x)\n", msginfo.msgssz, - 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) { - printf("msginfo.msgseg=%d\n", msginfo.msgseg); + 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; } } 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); 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 /* { u_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 >= sizeof(msgcalls)/sizeof(msgcalls[0])) return (EINVAL); mtx_lock(&Giant); error = (*msgcalls[uap->which])(td, &uap->a2); mtx_unlock(&Giant); 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; -#ifdef MSG_DEBUG_OK - printf("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr); -#endif + DPRINTF(("call to msgctl(%d, %d, 0x%x)\n", msqid, cmd, user_msqptr)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { -#ifdef MSG_DEBUG_OK - printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid, - msginfo.msgmni); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("no such msqid\n"); -#endif + DPRINTF(("no such msqid\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { -#ifdef MSG_DEBUG_OK - printf("wrong sequence number\n"); -#endif + 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 ((error = copyin(user_msqptr, &msqbuf, sizeof(msqbuf))) != 0) goto done2; if (msqbuf.msg_qbytes > msqptr->msg_qbytes) { error = suser(td); if (error) goto done2; } if (msqbuf.msg_qbytes > msginfo.msgmnb) { -#ifdef MSG_DEBUG_OK - printf("can't increase msg_qbytes beyond %d (truncating)\n", - msginfo.msgmnb); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("can't reduce msg_qbytes to 0\n"); -#endif + 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))) { -#ifdef MSG_DEBUG_OK - printf("requester doesn't have read access\n"); -#endif + DPRINTF(("requester doesn't have read access\n")); goto done2; } error = copyout(msqptr, user_msqptr, sizeof(struct msqid_ds)); break; default: -#ifdef MSG_DEBUG_OK - printf("invalid command %d\n", cmd); -#endif + DPRINTF(("invalid command %d\n", cmd)); error = EINVAL; goto done2; } if (error == 0) td->td_retval[0] = rval; done2: mtx_unlock(&Giant); 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; -#ifdef MSG_DEBUG_OK - printf("msgget(0x%x, 0%o)\n", key, msgflg); -#endif + DPRINTF(("msgget(0x%x, 0%o)\n", key, msgflg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); 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) { -#ifdef MSG_DEBUG_OK - printf("found public key\n"); -#endif + DPRINTF(("found public key\n")); if ((msgflg & IPC_CREAT) && (msgflg & IPC_EXCL)) { -#ifdef MSG_DEBUG_OK - printf("not exclusive\n"); -#endif + DPRINTF(("not exclusive\n")); error = EEXIST; goto done2; } if ((error = ipcperm(td, &msqptr->msg_perm, msgflg & 0700 ))) { -#ifdef MSG_DEBUG_OK - printf("requester doesn't have 0%o access\n", - msgflg & 0700); -#endif + DPRINTF(("requester doesn't have 0%o access\n", + msgflg & 0700)); goto done2; } goto found; } } -#ifdef MSG_DEBUG_OK - printf("need to allocate the msqid_ds\n"); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("no more msqid_ds's available\n"); -#endif + DPRINTF(("no more msqid_ds's available\n")); error = ENOSPC; goto done2; } -#ifdef MSG_DEBUG_OK - printf("msqid %d is available\n", msqid); -#endif + 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_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 { -#ifdef MSG_DEBUG_OK - printf("didn't find it and wasn't asked to create it\n"); -#endif + 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(&Giant); return (error); } #ifndef _SYS_SYSPROTO_H_ struct msgsnd_args { int msqid; 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; 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; -#ifdef MSG_DEBUG_OK - printf("call to msgsnd(%d, 0x%x, %d, %d)\n", msqid, user_msgp, msgsz, - msgflg); -#endif + 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(&Giant); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { -#ifdef MSG_DEBUG_OK - printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid, - msginfo.msgmni); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("no such message queue id\n"); -#endif + DPRINTF(("no such message queue id\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { -#ifdef MSG_DEBUG_OK - printf("wrong sequence number\n"); -#endif + DPRINTF(("wrong sequence number\n")); error = EINVAL; goto done2; } if ((error = ipcperm(td, &msqptr->msg_perm, IPC_W))) { -#ifdef MSG_DEBUG_OK - printf("requester doesn't have write access\n"); -#endif + DPRINTF(("requester doesn't have write access\n")); goto done2; } segs_needed = (msgsz + msginfo.msgssz - 1) / msginfo.msgssz; -#ifdef MSG_DEBUG_OK - printf("msgsz=%d, msgssz=%d, segs_needed=%d\n", msgsz, msginfo.msgssz, - segs_needed); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("msgsz > msqptr->msg_qbytes\n"); -#endif + DPRINTF(("msgsz > msqptr->msg_qbytes\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.mode & MSG_LOCKED) { -#ifdef MSG_DEBUG_OK - printf("msqid is locked\n"); -#endif + DPRINTF(("msqid is locked\n")); need_more_resources = 1; } if (msgsz + msqptr->msg_cbytes > msqptr->msg_qbytes) { -#ifdef MSG_DEBUG_OK - printf("msgsz + msg_cbytes > msg_qbytes\n"); -#endif + DPRINTF(("msgsz + msg_cbytes > msg_qbytes\n")); need_more_resources = 1; } if (segs_needed > nfree_msgmaps) { -#ifdef MSG_DEBUG_OK - printf("segs_needed > nfree_msgmaps\n"); -#endif + DPRINTF(("segs_needed > nfree_msgmaps\n")); need_more_resources = 1; } if (free_msghdrs == NULL) { -#ifdef MSG_DEBUG_OK - printf("no more msghdrs\n"); -#endif + DPRINTF(("no more msghdrs\n")); need_more_resources = 1; } if (need_more_resources) { int we_own_it; if ((msgflg & IPC_NOWAIT) != 0) { -#ifdef MSG_DEBUG_OK - printf("need more resources but caller doesn't want to wait\n"); -#endif + DPRINTF(("need more resources but caller " + "doesn't want to wait\n")); error = EAGAIN; goto done2; } if ((msqptr->msg_perm.mode & MSG_LOCKED) != 0) { -#ifdef MSG_DEBUG_OK - printf("we don't own the msqid_ds\n"); -#endif + DPRINTF(("we don't own the msqid_ds\n")); we_own_it = 0; } else { /* Force later arrivals to wait for our request */ -#ifdef MSG_DEBUG_OK - printf("we own the msqid_ds\n"); -#endif + DPRINTF(("we own the msqid_ds\n")); msqptr->msg_perm.mode |= MSG_LOCKED; we_own_it = 1; } -#ifdef MSG_DEBUG_OK - printf("goodnight\n"); -#endif + DPRINTF(("goodnight\n")); error = tsleep(msqptr, (PZERO - 4) | PCATCH, "msgwait", 0); -#ifdef MSG_DEBUG_OK - printf("good morning, error=%d\n", error); -#endif + DPRINTF(("good morning, error=%d\n", error)); if (we_own_it) msqptr->msg_perm.mode &= ~MSG_LOCKED; if (error != 0) { -#ifdef MSG_DEBUG_OK - printf("msgsnd: interrupted system call\n"); -#endif + DPRINTF(("msgsnd: interrupted system call\n")); error = EINTR; goto done2; } /* * Make sure that the msq queue still exists */ if (msqptr->msg_qbytes == 0) { -#ifdef MSG_DEBUG_OK - printf("msqid deleted\n"); -#endif + DPRINTF(("msqid deleted\n")); error = EIDRM; goto done2; } } else { -#ifdef MSG_DEBUG_OK - printf("got all the resources that we need\n"); -#endif + 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"); -#ifdef MSG_DEBUG_OK - printf("allocating segment %d to message\n", next); -#endif + 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 */ if ((error = copyin(user_msgp, &msghdr->msg_type, sizeof(msghdr->msg_type))) != 0) { -#ifdef MSG_DEBUG_OK - printf("error %d copying the message type\n", error); -#endif + DPRINTF(("error %d copying the message type\n", error)); msg_freehdr(msghdr); msqptr->msg_perm.mode &= ~MSG_LOCKED; wakeup(msqptr); goto done2; } user_msgp = (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); -#ifdef MSG_DEBUG_OK - printf("mtype (%d) < 1\n", msghdr->msg_type); -#endif + 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"); if ((error = copyin(user_msgp, &msgpool[next * msginfo.msgssz], tlen)) != 0) { -#ifdef MSG_DEBUG_OK - printf("error %d copying in message segment\n", error); -#endif + DPRINTF(("error %d copying in message segment\n", + error)); msg_freehdr(msghdr); msqptr->msg_perm.mode &= ~MSG_LOCKED; wakeup(msqptr); goto done2; } msgsz -= tlen; user_msgp = (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(&Giant); 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; -#ifdef MSG_DEBUG_OK - printf("call to msgrcv(%d, 0x%x, %d, %ld, %d)\n", msqid, user_msgp, - msgsz, msgtyp, msgflg); -#endif + 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); mtx_lock(&Giant); msqid = IPCID_TO_IX(msqid); if (msqid < 0 || msqid >= msginfo.msgmni) { -#ifdef MSG_DEBUG_OK - printf("msqid (%d) out of range (0<=msqid<%d)\n", msqid, - msginfo.msgmni); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("no such message queue id\n"); -#endif + DPRINTF(("no such message queue id\n")); error = EINVAL; goto done2; } if (msqptr->msg_perm.seq != IPCID_TO_SEQ(uap->msqid)) { -#ifdef MSG_DEBUG_OK - printf("wrong sequence number\n"); -#endif + DPRINTF(("wrong sequence number\n")); error = EINVAL; goto done2; } if ((error = ipcperm(td, &msqptr->msg_perm, IPC_R))) { -#ifdef MSG_DEBUG_OK - printf("requester doesn't have read access\n"); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("first message on the queue is too big (want %d, got %d)\n", - msgsz, msghdr->msg_ts); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("found message type %d, requested %d\n", - msghdr->msg_type, msgtyp); -#endif + DPRINTF(("found message type %d, " + "requested %d\n", + msghdr->msg_type, msgtyp)); if (msgsz < msghdr->msg_ts && (msgflg & MSG_NOERROR) == 0) { -#ifdef MSG_DEBUG_OK - printf("requested message on the queue is too big (want %d, got %d)\n", - msgsz, msghdr->msg_ts); -#endif + 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) { -#ifdef MSG_DEBUG_OK - printf("no appropriate message found (msgtyp=%d)\n", - msgtyp); -#endif + DPRINTF(("no appropriate message found (msgtyp=%d)\n", + msgtyp)); /* The SVID says to return ENOMSG. */ error = ENOMSG; goto done2; } /* * Wait for something to happen */ -#ifdef MSG_DEBUG_OK - printf("msgrcv: goodnight\n"); -#endif + DPRINTF(("msgrcv: goodnight\n")); error = tsleep(msqptr, (PZERO - 4) | PCATCH, "msgwait", 0); -#ifdef MSG_DEBUG_OK - printf("msgrcv: good morning (error=%d)\n", error); -#endif + DPRINTF(("msgrcv: good morning (error=%d)\n", error)); if (error != 0) { -#ifdef MSG_DEBUG_OK - printf("msgsnd: interrupted system call\n"); -#endif + 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)) { -#ifdef MSG_DEBUG_OK - printf("msqid deleted\n"); -#endif + 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). */ -#ifdef MSG_DEBUG_OK - printf("found a message, msgsz=%d, msg_ts=%d\n", msgsz, - msghdr->msg_ts); -#endif + 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. */ error = copyout(&(msghdr->msg_type), user_msgp, sizeof(msghdr->msg_type)); if (error != 0) { -#ifdef MSG_DEBUG_OK - printf("error (%d) copying out message type\n", error); -#endif + 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"); error = copyout(&msgpool[next * msginfo.msgssz], user_msgp, tlen); if (error != 0) { -#ifdef MSG_DEBUG_OK - printf("error (%d) copying out message segment\n", - error); -#endif + 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(&Giant); 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_RD, &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_RD, &msginfo.msgssz, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, msgseg, CTLFLAG_RD, &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 100522) +++ head/sys/kern/sysv_sem.c (revision 100523) @@ -1,1190 +1,1140 @@ /* $FreeBSD$ */ /* * Implementation of SVID semaphores * * Author: Daniel Boulet * * This software is provided ``AS IS'' without any warranties of any kind. */ #include "opt_sysvipc.h" #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(struct proc *p); static int sysctl_sema(SYSCTL_HANDLER_ARGS); #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, 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 int semtot = 0; static struct semid_ds *sema; /* semaphore id pool */ static struct sem *sem; /* semaphore pool */ static struct sem_undo *semu_list; /* list of active undo structures */ static int *semu; /* undo structure pool */ 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 { struct 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_RD, &seminfo.semmni, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmns, CTLFLAG_RD, &seminfo.semmns, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmnu, CTLFLAG_RD, &seminfo.semmnu, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semmsl, CTLFLAG_RW, &seminfo.semmsl, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semopm, CTLFLAG_RD, &seminfo.semopm, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semume, CTLFLAG_RD, &seminfo.semume, 0, ""); SYSCTL_INT(_kern_ipc, OID_AUTO, semusz, CTLFLAG_RD, &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) { register 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); if (sem == NULL) panic("sem is NULL"); sema = malloc(sizeof(struct semid_ds) * seminfo.semmni, M_SEM, M_WAITOK); if (sema == NULL) panic("sema is NULL"); semu = malloc(seminfo.semmnu * seminfo.semusz, M_SEM, M_WAITOK); if (semu == NULL) panic("semu is NULL"); for (i = 0; i < seminfo.semmni; i++) { sema[i].sem_base = 0; sema[i].sem_perm.mode = 0; } for (i = 0; i < seminfo.semmnu; i++) { register struct sem_undo *suptr = SEMU(i); suptr->un_proc = NULL; } semu_list = NULL; at_exit(semexit_myhook); } static int semunload(void) { if (semtot != 0) return (EBUSY); free(sem, M_SEM); free(sema, M_SEM); free(semu, M_SEM); rm_at_exit(semexit_myhook); 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 /* { u_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 >= sizeof(semcalls)/sizeof(semcalls[0])) return (EINVAL); mtx_lock(&Giant); error = (*semcalls[uap->which])(td, &uap->a2); mtx_unlock(&Giant); 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; { register int i; register struct sem_undo *suptr; register struct sem_undo **supptr; int attempt; /* * 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) { suptr->un_next = semu_list; semu_list = suptr; 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; supptr = &semu_list; while ((suptr = *supptr) != NULL) { if (suptr->un_cnt == 0) { suptr->un_proc = NULL; *supptr = suptr->un_next; did_something = 1; } else supptr = &(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) register struct thread *td; struct sem_undo **supptr; int semid, semnum; int adjval; { struct proc *p = td->td_proc; register struct sem_undo *suptr; register struct undo *sunptr; int i; /* Look for and remember the sem_undo if the caller doesn't provide it */ suptr = *supptr; if (suptr == NULL) { for (suptr = semu_list; suptr != NULL; suptr = suptr->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; { register struct sem_undo *suptr; for (suptr = semu_list; suptr != NULL; suptr = suptr->un_next) { register struct undo *sunptr = &suptr->un_ent[0]; register 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++; } } } /* * 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; register struct __semctl_args *uap; { int semid = uap->semid; int semnum = uap->semnum; int cmd = uap->cmd; union semun *arg = uap->arg; union semun real_arg; struct ucred *cred = td->td_ucred; int i, rval, error; struct semid_ds sbuf; register struct semid_ds *semaptr; u_short usval; -#ifdef SEM_DEBUG - printf("call to semctl(%d, %d, %d, 0x%x)\n", semid, semnum, cmd, arg); -#endif + DPRINTF(("call to semctl(%d, %d, %d, 0x%x)\n", + semid, semnum, cmd, arg)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); switch(cmd) { case SEM_STAT: if (semid < 0 || semid >= seminfo.semmni) UGAR(EINVAL); semaptr = &sema[semid]; if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 ) UGAR(EINVAL); if ((error = ipcperm(td, &semaptr->sem_perm, IPC_R))) UGAR(error); if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) UGAR(error); 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; goto done2; } semid = IPCID_TO_IX(semid); if (semid < 0 || semid >= seminfo.semmni) { error = EINVAL; goto done2; } semaptr = &sema[semid]; if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0 || semaptr->sem_perm.seq != IPCID_TO_SEQ(uap->semid)) { error = EINVAL; goto done2; } error = 0; rval = 0; switch (cmd) { case IPC_RMID: 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_clear(semid, -1); wakeup(semaptr); break; case IPC_SET: if ((error = ipcperm(td, &semaptr->sem_perm, IPC_M))) goto done2; if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; if ((error = copyin(real_arg.buf, &sbuf, sizeof(sbuf))) != 0) { 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 = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; error = copyout(semaptr, real_arg.buf, sizeof(struct semid_ds)); break; case GETNCNT: 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: 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: 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 = ipcperm(td, &semaptr->sem_perm, IPC_R))) goto done2; if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; for (i = 0; i < semaptr->sem_nsems; i++) { error = copyout(&semaptr->sem_base[i].semval, &real_arg.array[i], sizeof(real_arg.array[0])); if (error != 0) break; } break; case GETZCNT: 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 = ipcperm(td, &semaptr->sem_perm, IPC_W))) goto done2; if (semnum < 0 || semnum >= semaptr->sem_nsems) { error = EINVAL; goto done2; } if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) 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_clear(semid, semnum); wakeup(semaptr); break; case SETALL: if ((error = ipcperm(td, &semaptr->sem_perm, IPC_W))) goto done2; if ((error = copyin(arg, &real_arg, sizeof(real_arg))) != 0) goto done2; for (i = 0; i < semaptr->sem_nsems; i++) { error = copyin(&real_arg.array[i], &usval, sizeof(real_arg.array[0])); if (error != 0) break; if (usval > seminfo.semvmx) { error = ERANGE; break; } semaptr->sem_base[i].semval = usval; } semundo_clear(semid, -1); wakeup(semaptr); break; default: error = EINVAL; break; } if (error == 0) td->td_retval[0] = rval; done2: mtx_unlock(&Giant); 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; register 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; -#ifdef SEM_DEBUG - printf("semget(0x%x, %d, 0%o)\n", key, nsems, semflg); -#endif + 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) { -#ifdef SEM_DEBUG - printf("found public key\n"); -#endif + 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) { -#ifdef SEM_DEBUG - printf("too small\n"); -#endif + DPRINTF(("too small\n")); error = EINVAL; goto done2; } if ((semflg & IPC_CREAT) && (semflg & IPC_EXCL)) { -#ifdef SEM_DEBUG - printf("not exclusive\n"); -#endif + DPRINTF(("not exclusive\n")); error = EEXIST; goto done2; } goto found; } } -#ifdef SEM_DEBUG - printf("need to allocate the semid_ds\n"); -#endif + DPRINTF(("need to allocate the semid_ds\n")); if (key == IPC_PRIVATE || (semflg & IPC_CREAT)) { if (nsems <= 0 || nsems > seminfo.semmsl) { -#ifdef SEM_DEBUG - printf("nsems out of range (0<%d<=%d)\n", nsems, - seminfo.semmsl); -#endif + DPRINTF(("nsems out of range (0<%d<=%d)\n", nsems, + seminfo.semmsl)); error = EINVAL; goto done2; } if (nsems > seminfo.semmns - semtot) { -#ifdef SEM_DEBUG - printf("not enough semaphores left (need %d, got %d)\n", - nsems, seminfo.semmns - semtot); -#endif + 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) { -#ifdef SEM_DEBUG - printf("no more semid_ds's available\n"); -#endif + DPRINTF(("no more semid_ds's available\n")); error = ENOSPC; goto done2; } -#ifdef SEM_DEBUG - printf("semid %d is available\n", semid); -#endif + 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); -#ifdef SEM_DEBUG - printf("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, - &sem[semtot]); -#endif + DPRINTF(("sembase = 0x%x, next = 0x%x\n", sema[semid].sem_base, + &sem[semtot])); } else { -#ifdef SEM_DEBUG - printf("didn't find it and wasn't asked to create it\n"); -#endif + 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; u_int nsops; }; #endif /* * MPSAFE */ int semop(td, uap) struct thread *td; register struct semop_args *uap; { int semid = uap->semid; u_int nsops = uap->nsops; struct sembuf *sops = NULL; register struct semid_ds *semaptr; register struct sembuf *sopptr = 0; register struct sem *semptr = 0; struct sem_undo *suptr; int i, j, error; int do_wakeup, do_undos; -#ifdef SEM_DEBUG - printf("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops); -#endif + DPRINTF(("call to semop(%d, 0x%x, %u)\n", semid, sops, nsops)); if (!jail_sysvipc_allowed && jailed(td->td_ucred)) return (ENOSYS); mtx_lock(&Giant); semid = IPCID_TO_IX(semid); /* Convert back to zero origin */ if (semid < 0 || semid >= seminfo.semmni) { error = EINVAL; goto done2; } semaptr = &sema[semid]; 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; } if (nsops > seminfo.semopm) { -#ifdef SEM_DEBUG - printf("too many sops (max=%d, nsops=%d)\n", seminfo.semopm, - nsops); -#endif + DPRINTF(("too many sops (max=%d, nsops=%d)\n", seminfo.semopm, + nsops)); error = E2BIG; goto done2; } /* Allocate memory for sem_ops */ sops = malloc(nsops * sizeof(sops[0]), M_SEM, M_WAITOK); if (!sops) panic("Failed to allocate %d sem_ops", nsops); if ((error = copyin(uap->sops, sops, nsops * sizeof(sops[0]))) != 0) { -#ifdef SEM_DEBUG - printf("error = %d from copyin(%08x, %08x, %d)\n", error, - uap->sops, sops, nsops * sizeof(sops[0])); -#endif + DPRINTF(("error = %d from copyin(%08x, %08x, %d)\n", error, + uap->sops, sops, nsops * sizeof(sops[0]))); 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))) { -#ifdef SEM_DEBUG - printf("error = %d from ipaccess\n", error); -#endif + 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]; -#ifdef SEM_DEBUG - printf("semop: semaptr=%x, sem_base=%x, semptr=%x, sem[%d]=%d : op=%d, flag=%s\n", + 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"); -#endif + (sopptr->sem_flg & IPC_NOWAIT) ? + "nowait" : "wait")); if (sopptr->sem_op < 0) { if (semptr->semval + sopptr->sem_op < 0) { -#ifdef SEM_DEBUG - printf("semop: can't do it now\n"); -#endif + 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) { -#ifdef SEM_DEBUG - printf("semop: not zero now\n"); -#endif + 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 */ -#ifdef SEM_DEBUG - printf("semop: rollback 0 through %d\n", i-1); -#endif + 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++; -#ifdef SEM_DEBUG - printf("semop: good night!\n"); -#endif + DPRINTF(("semop: good night!\n")); error = tsleep(semaptr, (PZERO - 4) | PCATCH, "semwait", 0); -#ifdef SEM_DEBUG - printf("semop: good morning (error=%d)!\n", error); -#endif + DPRINTF(("semop: good morning (error=%d)!\n", error)); if (error != 0) { error = EINTR; goto done2; } -#ifdef SEM_DEBUG - printf("semop: good morning!\n"); -#endif + 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) { 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 = i - 1; j >= 0; j--) { if ((sops[j].sem_flg & SEM_UNDO) == 0) continue; adjval = sops[j].sem_op; if (adjval == 0) continue; if (semundo_adjust(td, &suptr, semid, sops[j].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; -#ifdef SEM_DEBUG - printf("error = %d from semundo_adjust\n", error); -#endif + DPRINTF(("error = %d from semundo_adjust\n", error)); goto done2; } /* loop through the sops */ } /* 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) { -#ifdef SEM_DEBUG - printf("semop: doing wakeup\n"); -#endif + DPRINTF(("semop: doing wakeup\n")); wakeup(semaptr); -#ifdef SEM_DEBUG - printf("semop: back from wakeup\n"); -#endif + DPRINTF(("semop: back from wakeup\n")); } -#ifdef SEM_DEBUG - printf("semop: done\n"); -#endif + DPRINTF(("semop: done\n")); td->td_retval[0] = 0; done2: if (sops) free(sops, M_SEM); mtx_unlock(&Giant); return (error); } /* * Go through the undo structures for this process and apply the adjustments to * semaphores. */ static void semexit_myhook(p) struct proc *p; { register struct sem_undo *suptr; register struct sem_undo **supptr; /* * Go through the chain of undo vectors looking for one * associated with this process. */ for (supptr = &semu_list; (suptr = *supptr) != NULL; supptr = &suptr->un_next) { if (suptr->un_proc == p) break; } if (suptr == NULL) return; -#ifdef SEM_DEBUG - printf("proc @%08x has undo structure with %d entries\n", p, - suptr->un_cnt); -#endif + 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; semaptr = &sema[semid]; if ((semaptr->sem_perm.mode & SEM_ALLOC) == 0) panic("semexit - semid not allocated"); if (semnum >= semaptr->sem_nsems) panic("semexit - semnum out of range"); -#ifdef SEM_DEBUG - printf("semexit: %08x id=%d num=%d(adj=%d) ; sem=%d\n", + 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); -#endif + 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); -#ifdef SEM_DEBUG - printf("semexit: back from wakeup\n"); -#endif + DPRINTF(("semexit: back from wakeup\n")); } } /* * Deallocate the undo vector. */ -#ifdef SEM_DEBUG - printf("removing vector\n"); -#endif + DPRINTF(("removing vector\n")); suptr->un_proc = NULL; *supptr = suptr->un_next; } static int sysctl_sema(SYSCTL_HANDLER_ARGS) { return (SYSCTL_OUT(req, sema, sizeof(struct semid_ds) * seminfo.semmni)); }