X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..d9a64523371fa019c4575bb400cbbc3a50ac9903:/bsd/kern/sysv_shm.c?ds=sidebyside diff --git a/bsd/kern/sysv_shm.c b/bsd/kern/sysv_shm.c index dfe108c62..2fb45c996 100644 --- a/bsd/kern/sysv_shm.c +++ b/bsd/kern/sysv_shm.c @@ -83,7 +83,7 @@ #include #endif -#include +#include #include #include @@ -93,8 +93,10 @@ #include #include +#include #include +#include /* Uncomment this line to see MAC debugging output. */ /* #define MAC_DEBUG */ @@ -105,10 +107,7 @@ #endif #if SYSV_SHM -static void shminit(void *); -#if 0 -SYSINIT(sysv_shm, SI_SUB_SYSV_SHM, SI_ORDER_FIRST, shminit, NULL) -#endif 0 +static int shminit(void); static lck_grp_t *sysv_shm_subsys_lck_grp; static lck_grp_attr_t *sysv_shm_subsys_lck_grp_attr; @@ -121,8 +120,8 @@ static lck_mtx_t sysv_shm_subsys_mutex; static int oshmctl(void *p, void *uap, void *retval); static int shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, int * retval); static int shmget_existing(struct shmget_args *uap, int mode, int segnum, int * retval); -static void shmid_ds_64to32(struct user_shmid_ds *in, struct shmid_ds *out); -static void shmid_ds_32to64(struct shmid_ds *in, struct user_shmid_ds *out); +static void shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out); +static void shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out); /* XXX casting to (sy_call_t *) is bogus, as usual. */ static sy_call_t *shmcalls[] = { @@ -140,8 +139,17 @@ static int shm_last_free, shm_nused, shm_committed; struct shmid_kernel *shmsegs; /* 64 bit version */ static int shm_inited = 0; +/* + * Since anonymous memory chunks are limited to ANON_MAX_SIZE bytes, + * we have to keep a list of chunks when we want to handle a shared memory + * segment bigger than ANON_MAX_SIZE. + * Each chunk points to a VM named entry of up to ANON_MAX_SIZE bytes + * of anonymous memory. + */ struct shm_handle { - void * shm_object; /* vm_offset_t kva; */ + void * shm_object; /* named entry for this chunk*/ + memory_object_size_t shm_handle_size; /* size of this chunk */ + struct shm_handle *shm_handle_next; /* next chunk */ }; struct shmmap_state { @@ -160,13 +168,19 @@ static int shm_delete_mapping(struct proc *, struct shmmap_state *, int); #define DEFAULT_SHMMNI 32 #define DEFAULT_SHMSEG 8 #define DEFAULT_SHMALL 1024 -struct shminfo shminfo = { - DEFAULT_SHMMAX, - DEFAULT_SHMMIN, - DEFAULT_SHMMNI, + +struct shminfo shminfo = { + DEFAULT_SHMMAX, + DEFAULT_SHMMIN, + DEFAULT_SHMMNI, DEFAULT_SHMSEG, DEFAULT_SHMALL }; + +#define SHMID_IS_VALID(x) ((x) >= 0) +#define SHMID_UNALLOCATED (-1) +#define SHMID_SENTINEL (-2) + #endif /* __APPLE_API_PRIVATE */ void sysv_shm_lock_init(void); @@ -186,17 +200,17 @@ sysv_shmtime(void) * NOTE: Source and target may *NOT* overlap! (target is smaller) */ static void -shmid_ds_64to32(struct user_shmid_ds *in, struct shmid_ds *out) +shmid_ds_64to32(struct user_shmid_ds *in, struct user32_shmid_ds *out) { out->shm_perm = in->shm_perm; - out->shm_segsz = (size_t)in->shm_segsz; + out->shm_segsz = in->shm_segsz; out->shm_lpid = in->shm_lpid; out->shm_cpid = in->shm_cpid; out->shm_nattch = in->shm_nattch; out->shm_atime = in->shm_atime; out->shm_dtime = in->shm_dtime; out->shm_ctime = in->shm_ctime; - out->shm_internal = CAST_DOWN(void *,in->shm_internal); + out->shm_internal = CAST_DOWN_EXPLICIT(int,in->shm_internal); } /* @@ -205,16 +219,16 @@ shmid_ds_64to32(struct user_shmid_ds *in, struct shmid_ds *out) * the beginning. */ static void -shmid_ds_32to64(struct shmid_ds *in, struct user_shmid_ds *out) +shmid_ds_32to64(struct user32_shmid_ds *in, struct user_shmid_ds *out) { - out->shm_internal = CAST_USER_ADDR_T(in->shm_internal); + out->shm_internal = in->shm_internal; out->shm_ctime = in->shm_ctime; out->shm_dtime = in->shm_dtime; out->shm_atime = in->shm_atime; out->shm_nattch = in->shm_nattch; out->shm_cpid = in->shm_cpid; out->shm_lpid = in->shm_lpid; - out->shm_segsz = (user_size_t)in->shm_segsz; + out->shm_segsz = in->shm_segsz; out->shm_perm = in->shm_perm; } @@ -251,15 +265,18 @@ shm_find_segment_by_shmid(int shmid) static void shm_deallocate_segment(struct shmid_kernel *shmseg) { - struct shm_handle *shm_handle; + struct shm_handle *shm_handle, *shm_handle_next; mach_vm_size_t size; - shm_handle = CAST_DOWN(void *,shmseg->u.shm_internal); /* tunnel */ - size = mach_vm_round_page(shmseg->u.shm_segsz); - mach_memory_entry_port_release(shm_handle->shm_object); - shm_handle->shm_object = NULL; - FREE((caddr_t)shm_handle, M_SHM); + for (shm_handle = CAST_DOWN(void *,shmseg->u.shm_internal); /* tunnel */ + shm_handle != NULL; + shm_handle = shm_handle_next) { + shm_handle_next = shm_handle->shm_handle_next; + mach_memory_entry_port_release(shm_handle->shm_object); + FREE((caddr_t) shm_handle, M_SHM); + } shmseg->u.shm_internal = USER_ADDR_NULL; /* tunnel */ + size = mach_vm_round_page(shmseg->u.shm_segsz); shm_committed -= btoc(size); shm_nused--; shmseg->u.shm_perm.mode = SHMSEG_FREE; @@ -285,7 +302,7 @@ shm_delete_mapping(__unused struct proc *p, struct shmmap_state *shmmap_s, if (result != KERN_SUCCESS) return EINVAL; } - shmmap_s->shmid = -1; + shmmap_s->shmid = SHMID_UNALLOCATED; shmseg->u.shm_dtime = sysv_shmtime(); if ((--shmseg->u.shm_nattch <= 0) && (shmseg->u.shm_perm.mode & SHMSEG_REMOVED)) { @@ -296,7 +313,7 @@ shm_delete_mapping(__unused struct proc *p, struct shmmap_state *shmmap_s, } int -shmdt(struct proc *p, struct shmdt_args *uap, register_t *retval) +shmdt(struct proc *p, struct shmdt_args *uap, int32_t *retval) { #if CONFIG_MACF struct shmid_kernel *shmsegptr; @@ -309,23 +326,28 @@ shmdt(struct proc *p, struct shmdt_args *uap, register_t *retval) SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if ((shmdtret = shminit())) { + goto shmdt_out; } + shmmap_s = (struct shmmap_state *)p->vm_shm; if (shmmap_s == NULL) { shmdtret = EINVAL; goto shmdt_out; } - for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) - if (shmmap_s->shmid != -1 && - shmmap_s->va == (mach_vm_offset_t)uap->shmaddr) + for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) { + if (SHMID_IS_VALID(shmmap_s->shmid) && + shmmap_s->va == (mach_vm_offset_t)uap->shmaddr) { break; - if (i == shminfo.shmseg) { + } + } + + if (!SHMID_IS_VALID(shmmap_s->shmid)) { shmdtret = EINVAL; goto shmdt_out; } + #if CONFIG_MACF /* * XXX: It might be useful to move this into the shm_delete_mapping @@ -355,33 +377,55 @@ shmat(struct proc *p, struct shmat_args *uap, user_addr_t *retval) struct shm_handle *shm_handle; mach_vm_address_t attach_va; /* attach address in/out */ mach_vm_size_t map_size; /* size of map entry */ - vm_prot_t prot; - size_t size; + mach_vm_size_t mapped_size; + vm_prot_t prot; + size_t size; kern_return_t rv; - int shmat_ret = 0; + int shmat_ret; + int vm_flags; + + shmat_ret = 0; AUDIT_ARG(svipc_id, uap->shmid); AUDIT_ARG(svipc_addr, uap->shmaddr); SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if ((shmat_ret = shminit())) { + goto shmat_out; } shmmap_s = (struct shmmap_state *)p->vm_shm; - if (shmmap_s == NULL) { - size = shminfo.shmseg * sizeof(struct shmmap_state); - MALLOC(shmmap_s, struct shmmap_state *, size, M_SHM, M_WAITOK); + /* lazily allocate the shm map */ + + int nsegs = shminfo.shmseg; + if (nsegs <= 0) { + shmat_ret = EMFILE; + goto shmat_out; + } + + /* +1 for the sentinel */ + if (os_add_and_mul_overflow(nsegs, 1, sizeof(struct shmmap_state), &size)) { + shmat_ret = ENOMEM; + goto shmat_out; + } + + MALLOC(shmmap_s, struct shmmap_state *, size, M_SHM, M_WAITOK | M_NULL); if (shmmap_s == NULL) { shmat_ret = ENOMEM; goto shmat_out; } - for (i = 0; i < shminfo.shmseg; i++) - shmmap_s[i].shmid = -1; + + /* initialize the entries */ + for (i = 0; i < nsegs; i++) { + shmmap_s[i].shmid = SHMID_UNALLOCATED; + } + shmmap_s[i].shmid = SHMID_SENTINEL; + p->vm_shm = (caddr_t)shmmap_s; } + shmseg = shm_find_segment_by_shmid(uap->shmid); if (shmseg == NULL) { shmat_ret = EINVAL; @@ -403,12 +447,13 @@ shmat(struct proc *p, struct shmat_args *uap, user_addr_t *retval) goto shmat_out; } #endif - for (i = 0; i < shminfo.shmseg; i++) { - if (shmmap_s->shmid == -1) - break; + + /* find a free shmid */ + while (SHMID_IS_VALID(shmmap_s->shmid)) { shmmap_s++; } - if (i >= shminfo.shmseg) { + if (shmmap_s->shmid != SHMID_UNALLOCATED) { + /* no free shmids */ shmat_ret = EMFILE; goto shmat_out; } @@ -429,45 +474,84 @@ shmat(struct proc *p, struct shmat_args *uap, user_addr_t *retval) goto shmat_out; } - shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal); /* tunnel */ + if (flags & MAP_FIXED) { + vm_flags = VM_FLAGS_FIXED; + } else { + vm_flags = VM_FLAGS_ANYWHERE; + } + + mapped_size = 0; + + /* first reserve enough space... */ + rv = mach_vm_map_kernel(current_map(), + &attach_va, + map_size, + 0, + vm_flags, + VM_MAP_KERNEL_FLAGS_NONE, + VM_KERN_MEMORY_NONE, + IPC_PORT_NULL, + 0, + FALSE, + VM_PROT_NONE, + VM_PROT_NONE, + VM_INHERIT_NONE); + if (rv != KERN_SUCCESS) { + goto out; + } - rv = mach_vm_map(current_map(), /* process map */ - &attach_va, /* attach address */ - map_size, /* segment size */ - (mach_vm_offset_t)0, /* alignment mask */ - (flags & MAP_FIXED)? VM_FLAGS_FIXED: VM_FLAGS_ANYWHERE, + shmmap_s->va = attach_va; + + /* ... then map the shared memory over the reserved space */ + for (shm_handle = CAST_DOWN(void *, shmseg->u.shm_internal);/* tunnel */ + shm_handle != NULL; + shm_handle = shm_handle->shm_handle_next) { + + rv = vm_map_enter_mem_object( + current_map(), /* process map */ + &attach_va, /* attach address */ + shm_handle->shm_handle_size, /* segment size */ + (mach_vm_offset_t)0, /* alignment mask */ + VM_FLAGS_FIXED | VM_FLAGS_OVERWRITE, + VM_MAP_KERNEL_FLAGS_NONE, + VM_KERN_MEMORY_NONE, shm_handle->shm_object, (mach_vm_offset_t)0, FALSE, prot, prot, - VM_INHERIT_DEFAULT); - if (rv != KERN_SUCCESS) + VM_INHERIT_SHARE); + if (rv != KERN_SUCCESS) goto out; - rv = mach_vm_inherit(current_map(), attach_va, map_size, VM_INHERIT_SHARE); - if (rv != KERN_SUCCESS) { - (void)mach_vm_deallocate(current_map(), attach_va, map_size); - goto out; + mapped_size += shm_handle->shm_handle_size; + attach_va = attach_va + shm_handle->shm_handle_size; } - shmmap_s->va = attach_va; shmmap_s->shmid = uap->shmid; shmseg->u.shm_lpid = p->p_pid; shmseg->u.shm_atime = sysv_shmtime(); shmseg->u.shm_nattch++; - *retval = attach_va; /* XXX return -1 on error */ + *retval = shmmap_s->va; /* XXX return -1 on error */ shmat_ret = 0; goto shmat_out; out: + if (mapped_size > 0) { + (void) mach_vm_deallocate(current_map(), + shmmap_s->va, + mapped_size); + } switch (rv) { case KERN_INVALID_ADDRESS: case KERN_NO_SPACE: shmat_ret = ENOMEM; + break; case KERN_PROTECTION_FAILURE: shmat_ret = EACCES; + break; default: shmat_ret = EINVAL; + break; } shmat_out: SYSV_SHM_SUBSYS_UNLOCK(); @@ -489,13 +573,12 @@ oshmctl(__unused void *p, __unused void *uap, __unused void *retval) * ipcperm:EACCES */ int -shmctl(__unused struct proc *p, struct shmctl_args *uap, register_t *retval) +shmctl(__unused struct proc *p, struct shmctl_args *uap, int32_t *retval) { int error; kauth_cred_t cred = kauth_cred_get(); struct user_shmid_ds inbuf; struct shmid_kernel *shmseg; - size_t shmid_ds_sz = sizeof(struct user_shmid_ds); int shmctl_ret = 0; @@ -504,13 +587,10 @@ shmctl(__unused struct proc *p, struct shmctl_args *uap, register_t *retval) SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if ((shmctl_ret = shminit())) { + goto shmctl_out; } - if (!IS_64BIT_PROCESS(p)) - shmid_ds_sz = sizeof(struct shmid_ds); - shmseg = shm_find_segment_by_shmid(uap->shmid); if (shmseg == NULL) { shmctl_ret = EINVAL; @@ -538,11 +618,21 @@ shmctl(__unused struct proc *p, struct shmctl_args *uap, register_t *retval) } if (IS_64BIT_PROCESS(p)) { - error = copyout((caddr_t)&shmseg->u, uap->buf, sizeof(struct user_shmid_ds)); + struct user_shmid_ds shmid_ds; + memcpy(&shmid_ds, &shmseg->u, sizeof(struct user_shmid_ds)); + + /* Clear kernel reserved pointer before copying to user space */ + shmid_ds.shm_internal = USER_ADDR_NULL; + + error = copyout(&shmid_ds, uap->buf, sizeof(shmid_ds)); } else { - struct shmid_ds shmid_ds32; + struct user32_shmid_ds shmid_ds32 = {}; shmid_ds_64to32(&shmseg->u, &shmid_ds32); - error = copyout(&shmid_ds32, uap->buf, sizeof(struct shmid_ds)); + + /* Clear kernel reserved pointer before copying to user space */ + shmid_ds32.shm_internal = (user32_addr_t)0; + + error = copyout(&shmid_ds32, uap->buf, sizeof(shmid_ds32)); } if (error) { shmctl_ret = error; @@ -558,9 +648,10 @@ shmctl(__unused struct proc *p, struct shmctl_args *uap, register_t *retval) if (IS_64BIT_PROCESS(p)) { error = copyin(uap->buf, &inbuf, sizeof(struct user_shmid_ds)); } else { - error = copyin(uap->buf, &inbuf, sizeof(struct shmid_ds)); + struct user32_shmid_ds shmid_ds32; + error = copyin(uap->buf, &shmid_ds32, sizeof(shmid_ds32)); /* convert in place; ugly, but safe */ - shmid_ds_32to64((struct shmid_ds *)&inbuf, &inbuf); + shmid_ds_32to64(&shmid_ds32, &inbuf); } if (error) { shmctl_ret = error; @@ -657,17 +748,21 @@ shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, struct shmid_kernel *shmseg; struct shm_handle *shm_handle; kern_return_t kret; - mach_vm_offset_t user_addr; - mach_vm_size_t size; + mach_vm_size_t total_size, size, alloc_size; void * mem_object; + struct shm_handle *shm_handle_next, **shm_handle_next_p; - if (uap->size < (user_size_t)shminfo.shmmin || - uap->size > (user_size_t)shminfo.shmmax) + if (uap->size <= 0 || + uap->size < (user_size_t)shminfo.shmmin || + uap->size > (user_size_t)shminfo.shmmax) { return EINVAL; + } if (shm_nused >= shminfo.shmmni) /* any shmids left? */ return ENOSPC; - size = mach_vm_round_page(uap->size); - if ((user_ssize_t)(shm_committed + btoc(size)) > shminfo.shmall) + if (mach_vm_round_page_overflow(uap->size, &total_size)) { + return EINVAL; + } + if ((user_ssize_t)(shm_committed + btoc(total_size)) > shminfo.shmall) return ENOMEM; if (shm_last_free < 0) { for (i = 0; i < shminfo.shmmni; i++) @@ -681,41 +776,52 @@ shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, 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. + * XXX but we don't release the global lock !? */ - kret = mach_vm_allocate(current_map(), &user_addr, size, VM_FLAGS_ANYWHERE); - if (kret != KERN_SUCCESS) - goto out; - - kret = mach_make_memory_entry_64(current_map(), - (memory_object_size_t *)&size, - (memory_object_offset_t)user_addr, - VM_PROT_DEFAULT, - (ipc_port_t *)&mem_object, 0); - - if (kret != KERN_SUCCESS) - goto out; - - mach_vm_deallocate(current_map(), user_addr, size); - shmseg->u.shm_perm.mode = SHMSEG_ALLOCATED | SHMSEG_REMOVED; shmseg->u.shm_perm._key = uap->key; shmseg->u.shm_perm._seq = (shmseg->u.shm_perm._seq + 1) & 0x7fff; - MALLOC(shm_handle, struct shm_handle *, sizeof(struct shm_handle), M_SHM, M_WAITOK); - if (shm_handle == NULL) { - kret = KERN_NO_SPACE; - mach_memory_entry_port_release(mem_object); - mem_object = NULL; - goto out; + + shm_handle_next_p = NULL; + for (alloc_size = 0; + alloc_size < total_size; + alloc_size += size) { + size = MIN(total_size - alloc_size, ANON_MAX_SIZE); + kret = mach_make_memory_entry_64( + VM_MAP_NULL, + (memory_object_size_t *) &size, + (memory_object_offset_t) 0, + MAP_MEM_NAMED_CREATE | VM_PROT_DEFAULT, + (ipc_port_t *) &mem_object, 0); + if (kret != KERN_SUCCESS) + goto out; + + MALLOC(shm_handle, struct shm_handle *, sizeof(struct shm_handle), M_SHM, M_WAITOK); + if (shm_handle == NULL) { + kret = KERN_NO_SPACE; + mach_memory_entry_port_release(mem_object); + mem_object = NULL; + goto out; + } + shm_handle->shm_object = mem_object; + shm_handle->shm_handle_size = size; + shm_handle->shm_handle_next = NULL; + if (shm_handle_next_p == NULL) { + shmseg->u.shm_internal = CAST_USER_ADDR_T(shm_handle);/* tunnel */ + } else { + *shm_handle_next_p = shm_handle; + } + shm_handle_next_p = &shm_handle->shm_handle_next; } - shm_handle->shm_object = mem_object; + shmid = IXSEQ_TO_IPCID(segnum, shmseg->u.shm_perm); - shmseg->u.shm_internal = CAST_USER_ADDR_T(shm_handle); /* tunnel */ shmseg->u.shm_perm.cuid = shmseg->u.shm_perm.uid = kauth_cred_getuid(cred); - shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = cred->cr_gid; + shmseg->u.shm_perm.cgid = shmseg->u.shm_perm.gid = kauth_cred_getgid(cred); shmseg->u.shm_perm.mode = (shmseg->u.shm_perm.mode & SHMSEG_WANTED) | (mode & ACCESSPERMS) | SHMSEG_ALLOCATED; shmseg->u.shm_segsz = uap->size; @@ -741,6 +847,17 @@ shmget_allocate_segment(struct proc *p, struct shmget_args *uap, int mode, AUDIT_ARG(svipc_id, shmid); return 0; out: + if (kret != KERN_SUCCESS) { + for (shm_handle = CAST_DOWN(void *,shmseg->u.shm_internal); /* tunnel */ + shm_handle != NULL; + shm_handle = shm_handle_next) { + shm_handle_next = shm_handle->shm_handle_next; + mach_memory_entry_port_release(shm_handle->shm_object); + FREE((caddr_t) shm_handle, M_SHM); + } + shmseg->u.shm_internal = USER_ADDR_NULL; /* tunnel */ + } + switch (kret) { case KERN_INVALID_ADDRESS: case KERN_NO_SPACE: @@ -754,17 +871,17 @@ out: } int -shmget(struct proc *p, struct shmget_args *uap, register_t *retval) +shmget(struct proc *p, struct shmget_args *uap, int32_t *retval) { int segnum, mode, error; int shmget_ret = 0; - + /* Auditing is actually done in shmget_allocate_segment() */ SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if ((shmget_ret = shminit())) { + goto shmget_out; } mode = uap->shmflg & ACCESSPERMS; @@ -787,13 +904,31 @@ shmget(struct proc *p, struct shmget_args *uap, register_t *retval) shmget_out: SYSV_SHM_SUBSYS_UNLOCK(); return shmget_ret; - /*NOTREACHED*/ - } -/* XXX actually varargs. */ +/* + * shmsys + * + * Entry point for all SHM calls: shmat, oshmctl, shmdt, shmget, shmctl + * + * Parameters: p Process requesting the call + * uap User argument descriptor (see below) + * retval Return value of the selected shm call + * + * Indirect parameters: uap->which msg call to invoke (index in array of shm calls) + * uap->a2 User argument descriptor + * + * Returns: 0 Success + * !0 Not success + * + * Implicit returns: retval Return value of the selected shm call + * + * DEPRECATED: This interface should not be used to call the other SHM + * functions (shmat, oshmctl, shmdt, shmget, shmctl). The correct + * usage is to call the other SHM functions directly. + */ int -shmsys(struct proc *p, struct shmsys_args *uap, register_t *retval) +shmsys(struct proc *p, struct shmsys_args *uap, int32_t *retval) { /* The routine that we are dispatching already does this */ @@ -811,54 +946,76 @@ shmfork(struct proc *p1, struct proc *p2) { struct shmmap_state *shmmap_s; size_t size; - int i; - int shmfork_ret = 0; + int nsegs = 0; + int ret = 0; SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if (shminit()) { + ret = 1; + goto shmfork_out; + } + + struct shmmap_state *src = (struct shmmap_state *)p1->vm_shm; + assert(src); + + /* count number of shmid entries in src */ + for (struct shmmap_state *s = src; s->shmid != SHMID_SENTINEL; s++) { + nsegs++; + } + + if (os_add_and_mul_overflow(nsegs, 1, sizeof(struct shmmap_state), &size)) { + ret = 1; + goto shmfork_out; } - - size = shminfo.shmseg * sizeof(struct shmmap_state); MALLOC(shmmap_s, struct shmmap_state *, size, M_SHM, M_WAITOK); - if (shmmap_s != NULL) { - bcopy((caddr_t)p1->vm_shm, (caddr_t)shmmap_s, size); - p2->vm_shm = (caddr_t)shmmap_s; - for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) - if (shmmap_s->shmid != -1) - shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++; - shmfork_ret = 0; + if (shmmap_s == NULL) { + ret = 1; goto shmfork_out; } - shmfork_ret = 1; /* failed to copy to child - ENOMEM */ + bcopy(src, (caddr_t)shmmap_s, size); + p2->vm_shm = (caddr_t)shmmap_s; + for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) { + if (SHMID_IS_VALID(shmmap_s->shmid)) { + shmsegs[IPCID_TO_IX(shmmap_s->shmid)].u.shm_nattch++; + } + } + shmfork_out: SYSV_SHM_SUBSYS_UNLOCK(); - return shmfork_ret; + return ret; } -void -shmexit(struct proc *p) +static void +shmcleanup(struct proc *p, int deallocate) { struct shmmap_state *shmmap_s; - int i; - - shmmap_s = (struct shmmap_state *)p->vm_shm; SYSV_SHM_SUBSYS_LOCK(); - for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) - if (shmmap_s->shmid != -1) + + shmmap_s = (struct shmmap_state *)p->vm_shm; + for (; shmmap_s->shmid != SHMID_SENTINEL; shmmap_s++) { + if (SHMID_IS_VALID(shmmap_s->shmid)) { /* * XXX: Should the MAC framework enforce * check here as well. */ - shm_delete_mapping(p, shmmap_s, 1); + shm_delete_mapping(p, shmmap_s, deallocate); + } + } + FREE((caddr_t)p->vm_shm, M_SHM); p->vm_shm = NULL; SYSV_SHM_SUBSYS_UNLOCK(); } +void +shmexit(struct proc *p) +{ + shmcleanup(p, 1); +} + /* * shmexec() is like shmexit(), only it doesn't delete the mappings, * since the old address space has already been destroyed and the new @@ -868,24 +1025,14 @@ shmexit(struct proc *p) __private_extern__ void shmexec(struct proc *p) { - struct shmmap_state *shmmap_s; - int i; - - shmmap_s = (struct shmmap_state *)p->vm_shm; - SYSV_SHM_SUBSYS_LOCK(); - for (i = 0; i < shminfo.shmseg; i++, shmmap_s++) - if (shmmap_s->shmid != -1) - shm_delete_mapping(p, shmmap_s, 0); - FREE((caddr_t)p->vm_shm, M_SHM); - p->vm_shm = NULL; - SYSV_SHM_SUBSYS_UNLOCK(); + shmcleanup(p, 0); } -void -shminit(__unused void *dummy) +int +shminit(void) { + size_t sz; int i; - int s; if (!shm_inited) { /* @@ -895,12 +1042,13 @@ shminit(__unused void *dummy) * dictates this filed be a size_t, which is 64 bits when * running 64 bit binaries. */ - s = sizeof(struct shmid_kernel) * shminfo.shmmni; + if (os_mul_overflow(shminfo.shmmni, sizeof(struct shmid_kernel), &sz)) { + return ENOMEM; + } - MALLOC(shmsegs, struct shmid_kernel *, s, M_SHM, M_WAITOK); + MALLOC(shmsegs, struct shmid_kernel *, sz, M_SHM, M_WAITOK); if (shmsegs == NULL) { - /* XXX fail safely: leave shared memory uninited */ - return; + return ENOMEM; } for (i = 0; i < shminfo.shmmni; i++) { shmsegs[i].u.shm_perm.mode = SHMSEG_FREE; @@ -914,7 +1062,10 @@ shminit(__unused void *dummy) shm_committed = 0; shm_inited = 1; } + + return 0; } + /* Initialize the mutex governing access to the SysV shm subsystem */ __private_extern__ void sysv_shm_lock_init( void ) @@ -936,7 +1087,11 @@ sysctl_shminfo(__unused struct sysctl_oid *oidp, void *arg1, { int error = 0; int sysctl_shminfo_ret = 0; - uint64_t saved_shmmax; + int64_t saved_shmmax; + int64_t saved_shmmin; + int64_t saved_shmseg; + int64_t saved_shmmni; + int64_t saved_shmall; error = SYSCTL_OUT(req, arg1, sizeof(int64_t)); if (error || req->newptr == USER_ADDR_NULL) @@ -949,7 +1104,11 @@ sysctl_shminfo(__unused struct sysctl_oid *oidp, void *arg1, sysctl_shminfo_ret = EPERM; goto sysctl_shminfo_out; } - saved_shmmax = shminfo.shmmax; + saved_shmmax = shminfo.shmmax; + saved_shmmin = shminfo.shmmin; + saved_shmseg = shminfo.shmseg; + saved_shmmni = shminfo.shmmni; + saved_shmall = shminfo.shmall; if ((error = SYSCTL_IN(req, arg1, sizeof(int64_t))) != 0) { sysctl_shminfo_ret = error; @@ -958,12 +1117,43 @@ sysctl_shminfo(__unused struct sysctl_oid *oidp, void *arg1, if (arg1 == &shminfo.shmmax) { /* shmmax needs to be page-aligned */ - if (shminfo.shmmax & PAGE_MASK_64) { + if (shminfo.shmmax & PAGE_MASK_64 || shminfo.shmmax < 0) { shminfo.shmmax = saved_shmmax; sysctl_shminfo_ret = EINVAL; goto sysctl_shminfo_out; } } + else if (arg1 == &shminfo.shmmin) { + if (shminfo.shmmin < 0) { + shminfo.shmmin = saved_shmmin; + sysctl_shminfo_ret = EINVAL; + goto sysctl_shminfo_out; + } + } + else if (arg1 == &shminfo.shmseg) { + /* add a sanity check - 20847256 */ + if (shminfo.shmseg > INT32_MAX || shminfo.shmseg < 0) { + shminfo.shmseg = saved_shmseg; + sysctl_shminfo_ret = EINVAL; + goto sysctl_shminfo_out; + } + } + else if (arg1 == &shminfo.shmmni) { + /* add a sanity check - 20847256 */ + if (shminfo.shmmni > INT32_MAX || shminfo.shmmni < 0) { + shminfo.shmmni = saved_shmmni; + sysctl_shminfo_ret = EINVAL; + goto sysctl_shminfo_out; + } + } + else if (arg1 == &shminfo.shmall) { + /* add a sanity check - 20847256 */ + if (shminfo.shmall > INT32_MAX || shminfo.shmall < 0) { + shminfo.shmall = saved_shmall; + sysctl_shminfo_ret = EINVAL; + goto sysctl_shminfo_out; + } + } sysctl_shminfo_ret = 0; sysctl_shminfo_out: SYSV_SHM_SUBSYS_UNLOCK(); @@ -977,10 +1167,11 @@ IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1, int error; int cursor; union { - struct IPCS_command u32; + struct user32_IPCS_command u32; struct user_IPCS_command u64; } ipcs; - struct shmid_ds shmid_ds32; /* post conversion, 32 bit version */ + struct user32_shmid_ds shmid_ds32 = {}; /* post conversion, 32 bit version */ + struct user_shmid_ds shmid_ds; /* 64 bit version */ void *shmid_dsp; size_t ipcs_sz = sizeof(struct user_IPCS_command); size_t shmid_ds_sz = sizeof(struct user_shmid_ds); @@ -988,13 +1179,13 @@ IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1, SYSV_SHM_SUBSYS_LOCK(); - if (!shm_inited) { - shminit(NULL); + if ((error = shminit())) { + goto ipcs_shm_sysctl_out; } if (!IS_64BIT_PROCESS(p)) { - ipcs_sz = sizeof(struct IPCS_command); - shmid_ds_sz = sizeof(struct shmid_ds); + ipcs_sz = sizeof(struct user32_IPCS_command); + shmid_ds_sz = sizeof(struct user32_shmid_ds); } /* Copy in the command structure */ @@ -1052,7 +1243,18 @@ IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1, */ if (!IS_64BIT_PROCESS(p)) { shmid_ds_64to32(shmid_dsp, &shmid_ds32); + + /* Clear kernel reserved pointer before copying to user space */ + shmid_ds32.shm_internal = (user32_addr_t)0; + shmid_dsp = &shmid_ds32; + } else { + memcpy(&shmid_ds, shmid_dsp, sizeof(shmid_ds)); + + /* Clear kernel reserved pointer before copying to user space */ + shmid_ds.shm_internal = USER_ADDR_NULL; + + shmid_dsp = &shmid_ds; } error = copyout(shmid_dsp, ipcs.u64.ipcs_data, ipcs.u64.ipcs_datalen); if (!error) { @@ -1060,8 +1262,9 @@ IPCS_shm_sysctl(__unused struct sysctl_oid *oidp, __unused void *arg1, ipcs.u64.ipcs_cursor = cursor + 1; if (!IS_64BIT_PROCESS(p)) /* convert in place */ - ipcs.u32.ipcs_data = CAST_DOWN(void *,ipcs.u64.ipcs_data); - error = SYSCTL_OUT(req, &ipcs, ipcs_sz); + ipcs.u32.ipcs_data = CAST_DOWN_EXPLICIT(user32_addr_t,ipcs.u64.ipcs_data); + + error = SYSCTL_OUT(req, &ipcs, ipcs_sz); } break; @@ -1074,26 +1277,26 @@ ipcs_shm_sysctl_out: return(error); } -SYSCTL_NODE(_kern, KERN_SYSV, sysv, CTLFLAG_RW|CTLFLAG_LOCKED, 0, "SYSV"); +SYSCTL_NODE(_kern, KERN_SYSV, sysv, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSV"); -SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmax, CTLTYPE_QUAD | CTLFLAG_RW, +SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmax, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &shminfo.shmmax, 0, &sysctl_shminfo ,"Q","shmmax"); -SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmin, CTLTYPE_QUAD | CTLFLAG_RW, +SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmin, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &shminfo.shmmin, 0, &sysctl_shminfo ,"Q","shmmin"); -SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmni, CTLTYPE_QUAD | CTLFLAG_RW, +SYSCTL_PROC(_kern_sysv, OID_AUTO, shmmni, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &shminfo.shmmni, 0, &sysctl_shminfo ,"Q","shmmni"); -SYSCTL_PROC(_kern_sysv, OID_AUTO, shmseg, CTLTYPE_QUAD | CTLFLAG_RW, +SYSCTL_PROC(_kern_sysv, OID_AUTO, shmseg, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &shminfo.shmseg, 0, &sysctl_shminfo ,"Q","shmseg"); -SYSCTL_PROC(_kern_sysv, OID_AUTO, shmall, CTLTYPE_QUAD | CTLFLAG_RW, +SYSCTL_PROC(_kern_sysv, OID_AUTO, shmall, CTLTYPE_QUAD | CTLFLAG_RW | CTLFLAG_LOCKED, &shminfo.shmall, 0, &sysctl_shminfo ,"Q","shmall"); -SYSCTL_NODE(_kern_sysv, OID_AUTO, ipcs, CTLFLAG_RW|CTLFLAG_LOCKED, 0, "SYSVIPCS"); +SYSCTL_NODE(_kern_sysv, OID_AUTO, ipcs, CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, "SYSVIPCS"); -SYSCTL_PROC(_kern_sysv_ipcs, OID_AUTO, shm, CTLFLAG_RW|CTLFLAG_ANYBODY, +SYSCTL_PROC(_kern_sysv_ipcs, OID_AUTO, shm, CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_LOCKED, 0, 0, IPCS_shm_sysctl, "S,IPCS_shm_command", "ipcs shm command interface");