diff --git a/sys/amd64/amd64/sys_machdep.c b/sys/amd64/amd64/sys_machdep.c index a3716a354448..b715663ea925 100644 --- a/sys/amd64/amd64/sys_machdep.c +++ b/sys/amd64/amd64/sys_machdep.c @@ -1,811 +1,813 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2003 Peter Wemm. * Copyright (c) 1990 The Regents of the University of California. * 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. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``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 REGENTS OR CONTRIBUTORS 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. * * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" +#include "opt_ktrace.h" #include #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include /* for kernel_map */ #include #include #include #include #include #include #include #include #include #include static void user_ldt_deref(struct proc_ldt *pldt); static void user_ldt_derefl(struct proc_ldt *pldt); #define MAX_LD 8192 int max_ldt_segment = 512; SYSCTL_INT(_machdep, OID_AUTO, max_ldt_segment, CTLFLAG_RDTUN, &max_ldt_segment, 0, "Maximum number of allowed LDT segments in the single address space"); static void max_ldt_segment_init(void *arg __unused) { if (max_ldt_segment <= 0) max_ldt_segment = 1; if (max_ldt_segment > MAX_LD) max_ldt_segment = MAX_LD; } SYSINIT(maxldt, SI_SUB_VM_CONF, SI_ORDER_ANY, max_ldt_segment_init, NULL); #ifndef _SYS_SYSPROTO_H_ struct sysarch_args { int op; char *parms; }; #endif int sysarch_ldt(struct thread *td, struct sysarch_args *uap, int uap_space) { struct i386_ldt_args *largs, la; struct user_segment_descriptor *lp; int error = 0; /* * XXXKIB check that the BSM generation code knows to encode * the op argument. */ AUDIT_ARG_CMD(uap->op); if (uap_space == UIO_USERSPACE) { error = copyin(uap->parms, &la, sizeof(struct i386_ldt_args)); if (error != 0) return (error); largs = &la; } else largs = (struct i386_ldt_args *)uap->parms; switch (uap->op) { case I386_GET_LDT: error = amd64_get_ldt(td, largs); break; case I386_SET_LDT: if (largs->descs != NULL && largs->num > max_ldt_segment) return (EINVAL); set_pcb_flags(td->td_pcb, PCB_FULL_IRET); if (largs->descs != NULL) { lp = malloc(largs->num * sizeof(struct user_segment_descriptor), M_TEMP, M_WAITOK); error = copyin(largs->descs, lp, largs->num * sizeof(struct user_segment_descriptor)); if (error == 0) error = amd64_set_ldt(td, largs, lp); free(lp, M_TEMP); } else { error = amd64_set_ldt(td, largs, NULL); } break; } return (error); } void update_gdt_gsbase(struct thread *td, uint32_t base) { struct user_segment_descriptor *sd; if (td != curthread) return; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); critical_enter(); sd = PCPU_GET(gs32p); sd->sd_lobase = base & 0xffffff; sd->sd_hibase = (base >> 24) & 0xff; critical_exit(); } void update_gdt_fsbase(struct thread *td, uint32_t base) { struct user_segment_descriptor *sd; if (td != curthread) return; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); critical_enter(); sd = PCPU_GET(fs32p); sd->sd_lobase = base & 0xffffff; sd->sd_hibase = (base >> 24) & 0xff; critical_exit(); } int sysarch(struct thread *td, struct sysarch_args *uap) { struct pcb *pcb; struct vm_map *map; uint32_t i386base; uint64_t a64base; struct i386_ioperm_args iargs; struct i386_get_xfpustate i386xfpu; struct i386_set_pkru i386pkru; struct amd64_get_xfpustate a64xfpu; struct amd64_set_pkru a64pkru; int error; #ifdef CAPABILITY_MODE /* * When adding new operations, add a new case statement here to * explicitly indicate whether or not the operation is safe to * perform in capability mode. */ if (IN_CAPABILITY_MODE(td)) { switch (uap->op) { case I386_GET_LDT: case I386_SET_LDT: case I386_GET_IOPERM: case I386_GET_FSBASE: case I386_SET_FSBASE: case I386_GET_GSBASE: case I386_SET_GSBASE: case I386_GET_XFPUSTATE: case I386_SET_PKRU: case I386_CLEAR_PKRU: case AMD64_GET_FSBASE: case AMD64_SET_FSBASE: case AMD64_GET_GSBASE: case AMD64_SET_GSBASE: case AMD64_GET_XFPUSTATE: case AMD64_SET_PKRU: case AMD64_CLEAR_PKRU: break; case I386_SET_IOPERM: default: #ifdef KTRACE if (KTRPOINT(td, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); #endif return (ECAPMODE); } } #endif if (uap->op == I386_GET_LDT || uap->op == I386_SET_LDT) return (sysarch_ldt(td, uap, UIO_USERSPACE)); error = 0; pcb = td->td_pcb; /* * XXXKIB check that the BSM generation code knows to encode * the op argument. */ AUDIT_ARG_CMD(uap->op); switch (uap->op) { case I386_GET_IOPERM: case I386_SET_IOPERM: if ((error = copyin(uap->parms, &iargs, sizeof(struct i386_ioperm_args))) != 0) return (error); break; case I386_GET_XFPUSTATE: if ((error = copyin(uap->parms, &i386xfpu, sizeof(struct i386_get_xfpustate))) != 0) return (error); a64xfpu.addr = (void *)(uintptr_t)i386xfpu.addr; a64xfpu.len = i386xfpu.len; break; case I386_SET_PKRU: case I386_CLEAR_PKRU: if ((error = copyin(uap->parms, &i386pkru, sizeof(struct i386_set_pkru))) != 0) return (error); a64pkru.addr = (void *)(uintptr_t)i386pkru.addr; a64pkru.len = i386pkru.len; a64pkru.keyidx = i386pkru.keyidx; a64pkru.flags = i386pkru.flags; break; case AMD64_GET_XFPUSTATE: if ((error = copyin(uap->parms, &a64xfpu, sizeof(struct amd64_get_xfpustate))) != 0) return (error); break; case AMD64_SET_PKRU: case AMD64_CLEAR_PKRU: if ((error = copyin(uap->parms, &a64pkru, sizeof(struct amd64_set_pkru))) != 0) return (error); break; default: break; } switch (uap->op) { case I386_GET_IOPERM: error = amd64_get_ioperm(td, &iargs); if (error == 0) error = copyout(&iargs, uap->parms, sizeof(struct i386_ioperm_args)); break; case I386_SET_IOPERM: error = amd64_set_ioperm(td, &iargs); break; case I386_GET_FSBASE: update_pcb_bases(pcb); i386base = pcb->pcb_fsbase; error = copyout(&i386base, uap->parms, sizeof(i386base)); break; case I386_SET_FSBASE: error = copyin(uap->parms, &i386base, sizeof(i386base)); if (!error) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_fsbase = i386base; td->td_frame->tf_fs = _ufssel; update_gdt_fsbase(td, i386base); } break; case I386_GET_GSBASE: update_pcb_bases(pcb); i386base = pcb->pcb_gsbase; error = copyout(&i386base, uap->parms, sizeof(i386base)); break; case I386_SET_GSBASE: error = copyin(uap->parms, &i386base, sizeof(i386base)); if (!error) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_gsbase = i386base; td->td_frame->tf_gs = _ugssel; update_gdt_gsbase(td, i386base); } break; case AMD64_GET_FSBASE: update_pcb_bases(pcb); error = copyout(&pcb->pcb_fsbase, uap->parms, sizeof(pcb->pcb_fsbase)); break; case AMD64_SET_FSBASE: error = copyin(uap->parms, &a64base, sizeof(a64base)); if (!error) { if (a64base < VM_MAXUSER_ADDRESS) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_fsbase = a64base; td->td_frame->tf_fs = _ufssel; } else error = EINVAL; } break; case AMD64_GET_GSBASE: update_pcb_bases(pcb); error = copyout(&pcb->pcb_gsbase, uap->parms, sizeof(pcb->pcb_gsbase)); break; case AMD64_SET_GSBASE: error = copyin(uap->parms, &a64base, sizeof(a64base)); if (!error) { if (a64base < VM_MAXUSER_ADDRESS) { set_pcb_flags(pcb, PCB_FULL_IRET); pcb->pcb_gsbase = a64base; td->td_frame->tf_gs = _ugssel; } else error = EINVAL; } break; case I386_GET_XFPUSTATE: case AMD64_GET_XFPUSTATE: if (a64xfpu.len > cpu_max_ext_state_size - sizeof(struct savefpu)) return (EINVAL); fpugetregs(td); error = copyout((char *)(get_pcb_user_save_td(td) + 1), a64xfpu.addr, a64xfpu.len); break; case I386_SET_PKRU: case AMD64_SET_PKRU: /* * Read-lock the map to synchronize with parallel * pmap_vmspace_copy() on fork. */ map = &td->td_proc->p_vmspace->vm_map; vm_map_lock_read(map); error = pmap_pkru_set(PCPU_GET(curpmap), (vm_offset_t)a64pkru.addr, (vm_offset_t)a64pkru.addr + a64pkru.len, a64pkru.keyidx, a64pkru.flags); vm_map_unlock_read(map); break; case I386_CLEAR_PKRU: case AMD64_CLEAR_PKRU: if (a64pkru.flags != 0 || a64pkru.keyidx != 0) { error = EINVAL; break; } map = &td->td_proc->p_vmspace->vm_map; vm_map_lock_read(map); error = pmap_pkru_clear(PCPU_GET(curpmap), (vm_offset_t)a64pkru.addr, (vm_offset_t)a64pkru.addr + a64pkru.len); vm_map_unlock_read(map); break; default: error = EINVAL; break; } return (error); } int amd64_set_ioperm(struct thread *td, struct i386_ioperm_args *uap) { char *iomap; struct amd64tss *tssp; struct system_segment_descriptor *tss_sd; struct pcb *pcb; u_int i; int error; if ((error = priv_check(td, PRIV_IO)) != 0) return (error); if ((error = securelevel_gt(td->td_ucred, 0)) != 0) return (error); if (uap->start > uap->start + uap->length || uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); /* * XXX * While this is restricted to root, we should probably figure out * whether any other driver is using this i/o address, as so not to * cause confusion. This probably requires a global 'usage registry'. */ pcb = td->td_pcb; if (pcb->pcb_tssp == NULL) { tssp = kmem_malloc(ctob(IOPAGES + 1), M_WAITOK); pmap_pti_add_kva((vm_offset_t)tssp, (vm_offset_t)tssp + ctob(IOPAGES + 1), false); iomap = (char *)&tssp[1]; memset(iomap, 0xff, IOPERM_BITMAP_SIZE); critical_enter(); /* Takes care of tss_rsp0. */ memcpy(tssp, PCPU_PTR(common_tss), sizeof(struct amd64tss)); tssp->tss_iobase = sizeof(*tssp); pcb->pcb_tssp = tssp; tss_sd = PCPU_GET(tss); tss_sd->sd_lobase = (u_long)tssp & 0xffffff; tss_sd->sd_hibase = ((u_long)tssp >> 24) & 0xfffffffffful; tss_sd->sd_type = SDT_SYSTSS; ltr(GSEL(GPROC0_SEL, SEL_KPL)); PCPU_SET(tssp, tssp); critical_exit(); } else iomap = (char *)&pcb->pcb_tssp[1]; for (i = uap->start; i < uap->start + uap->length; i++) { if (uap->enable) iomap[i >> 3] &= ~(1 << (i & 7)); else iomap[i >> 3] |= (1 << (i & 7)); } return (error); } int amd64_get_ioperm(struct thread *td, struct i386_ioperm_args *uap) { int i, state; char *iomap; if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); if (td->td_pcb->pcb_tssp == NULL) { uap->length = 0; goto done; } iomap = (char *)&td->td_pcb->pcb_tssp[1]; i = uap->start; state = (iomap[i >> 3] >> (i & 7)) & 1; uap->enable = !state; uap->length = 1; for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) break; uap->length++; } done: return (0); } /* * Update the GDT entry pointing to the LDT to point to the LDT of the * current process. */ static void set_user_ldt(struct mdproc *mdp) { *PCPU_GET(ldt) = mdp->md_ldt_sd; lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); } static void set_user_ldt_rv(void *arg) { struct proc *orig, *target; struct proc_ldt *ldt; orig = arg; target = curthread->td_proc; ldt = (void *)atomic_load_acq_ptr((uintptr_t *)&orig->p_md.md_ldt); if (target->p_md.md_ldt != ldt) return; set_user_ldt(&target->p_md); } struct proc_ldt * user_ldt_alloc(struct proc *p, int force) { struct proc_ldt *pldt, *new_ldt; struct mdproc *mdp; struct soft_segment_descriptor sldt; vm_offset_t sva; vm_size_t sz; mtx_assert(&dt_lock, MA_OWNED); mdp = &p->p_md; if (!force && mdp->md_ldt != NULL) return (mdp->md_ldt); mtx_unlock(&dt_lock); new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK); sz = max_ldt_segment * sizeof(struct user_segment_descriptor); new_ldt->ldt_base = kmem_malloc(sz, M_WAITOK | M_ZERO); sva = (uintptr_t)new_ldt->ldt_base; pmap_pti_add_kva(sva, sva + sz, false); new_ldt->ldt_refcnt = 1; sldt.ssd_base = sva; sldt.ssd_limit = sz - 1; sldt.ssd_type = SDT_SYSLDT; sldt.ssd_dpl = SEL_KPL; sldt.ssd_p = 1; sldt.ssd_long = 0; sldt.ssd_def32 = 0; sldt.ssd_gran = 0; mtx_lock(&dt_lock); pldt = mdp->md_ldt; if (pldt != NULL && !force) { pmap_pti_remove_kva(sva, sva + sz); kmem_free(new_ldt->ldt_base, sz); free(new_ldt, M_SUBPROC); return (pldt); } if (pldt != NULL) { bcopy(pldt->ldt_base, new_ldt->ldt_base, max_ldt_segment * sizeof(struct user_segment_descriptor)); user_ldt_derefl(pldt); } critical_enter(); ssdtosyssd(&sldt, &p->p_md.md_ldt_sd); atomic_thread_fence_rel(); mdp->md_ldt = new_ldt; critical_exit(); smp_rendezvous(NULL, set_user_ldt_rv, NULL, p); return (mdp->md_ldt); } void user_ldt_free(struct thread *td) { struct proc *p = td->td_proc; struct mdproc *mdp = &p->p_md; struct proc_ldt *pldt; mtx_lock(&dt_lock); if ((pldt = mdp->md_ldt) == NULL) { mtx_unlock(&dt_lock); return; } critical_enter(); mdp->md_ldt = NULL; atomic_thread_fence_rel(); bzero(&mdp->md_ldt_sd, sizeof(mdp->md_ldt_sd)); if (td == curthread) lldt(GSEL(GNULL_SEL, SEL_KPL)); critical_exit(); user_ldt_deref(pldt); } static void user_ldt_derefl(struct proc_ldt *pldt) { vm_offset_t sva; vm_size_t sz; if (--pldt->ldt_refcnt == 0) { sva = (vm_offset_t)pldt->ldt_base; sz = max_ldt_segment * sizeof(struct user_segment_descriptor); pmap_pti_remove_kva(sva, sva + sz); kmem_free(pldt->ldt_base, sz); free(pldt, M_SUBPROC); } } static void user_ldt_deref(struct proc_ldt *pldt) { mtx_assert(&dt_lock, MA_OWNED); user_ldt_derefl(pldt); mtx_unlock(&dt_lock); } /* * Note for the authors of compat layers (linux, etc): copyout() in * the function below is not a problem since it presents data in * arch-specific format (i.e. i386-specific in this case), not in * the OS-specific one. */ int amd64_get_ldt(struct thread *td, struct i386_ldt_args *uap) { struct proc_ldt *pldt; struct user_segment_descriptor *lp; uint64_t *data; u_int i, num; int error; #ifdef DEBUG printf("amd64_get_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif pldt = td->td_proc->p_md.md_ldt; if (pldt == NULL || uap->start >= max_ldt_segment || uap->num == 0) { td->td_retval[0] = 0; return (0); } num = min(uap->num, max_ldt_segment - uap->start); lp = &((struct user_segment_descriptor *)(pldt->ldt_base))[uap->start]; data = malloc(num * sizeof(struct user_segment_descriptor), M_TEMP, M_WAITOK); mtx_lock(&dt_lock); for (i = 0; i < num; i++) data[i] = ((volatile uint64_t *)lp)[i]; mtx_unlock(&dt_lock); error = copyout(data, uap->descs, num * sizeof(struct user_segment_descriptor)); free(data, M_TEMP); if (error == 0) td->td_retval[0] = num; return (error); } int amd64_set_ldt(struct thread *td, struct i386_ldt_args *uap, struct user_segment_descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; struct user_segment_descriptor *dp; struct proc *p; u_int largest_ld, i; int error; #ifdef DEBUG printf("amd64_set_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif mdp = &td->td_proc->p_md; error = 0; set_pcb_flags(td->td_pcb, PCB_FULL_IRET); p = td->td_proc; if (descs == NULL) { /* Free descriptors */ if (uap->start == 0 && uap->num == 0) uap->num = max_ldt_segment; if (uap->num == 0) return (EINVAL); if ((pldt = mdp->md_ldt) == NULL || uap->start >= max_ldt_segment) return (0); largest_ld = uap->start + uap->num; if (largest_ld > max_ldt_segment) largest_ld = max_ldt_segment; if (largest_ld < uap->start) return (EINVAL); mtx_lock(&dt_lock); for (i = uap->start; i < largest_ld; i++) ((volatile uint64_t *)(pldt->ldt_base))[i] = 0; mtx_unlock(&dt_lock); return (0); } if (!(uap->start == LDT_AUTO_ALLOC && uap->num == 1)) { /* verify range of descriptors to modify */ largest_ld = uap->start + uap->num; if (uap->start >= max_ldt_segment || largest_ld > max_ldt_segment || largest_ld < uap->start) return (EINVAL); } /* Check descriptors for access violations */ for (i = 0; i < uap->num; i++) { dp = &descs[i]; switch (dp->sd_type) { case SDT_SYSNULL: /* system null */ dp->sd_p = 0; break; case SDT_SYS286TSS: case SDT_SYSLDT: case SDT_SYS286BSY: case SDT_SYS286CGT: case SDT_SYSTASKGT: case SDT_SYS286IGT: case SDT_SYS286TGT: case SDT_SYSNULL2: case SDT_SYSTSS: case SDT_SYSNULL3: case SDT_SYSBSY: case SDT_SYSCGT: case SDT_SYSNULL4: case SDT_SYSIGT: case SDT_SYSTGT: return (EACCES); /* memory segment types */ case SDT_MEMEC: /* memory execute only conforming */ case SDT_MEMEAC: /* memory execute only accessed conforming */ case SDT_MEMERC: /* memory execute read conforming */ case SDT_MEMERAC: /* memory execute read accessed conforming */ /* Must be "present" if executable and conforming. */ if (dp->sd_p == 0) return (EACCES); break; case SDT_MEMRO: /* memory read only */ case SDT_MEMROA: /* memory read only accessed */ case SDT_MEMRW: /* memory read write */ case SDT_MEMRWA: /* memory read write accessed */ case SDT_MEMROD: /* memory read only expand dwn limit */ case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ case SDT_MEMRWD: /* memory read write expand dwn limit */ case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ case SDT_MEME: /* memory execute only */ case SDT_MEMEA: /* memory execute only accessed */ case SDT_MEMER: /* memory execute read */ case SDT_MEMERA: /* memory execute read accessed */ break; default: return(EINVAL); } /* Only user (ring-3) descriptors may be present. */ if ((dp->sd_p != 0) && (dp->sd_dpl != SEL_UPL)) return (EACCES); } if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { /* Allocate a free slot */ mtx_lock(&dt_lock); pldt = user_ldt_alloc(p, 0); if (pldt == NULL) { mtx_unlock(&dt_lock); return (ENOMEM); } /* * start scanning a bit up to leave room for NVidia and * Wine, which still user the "Blat" method of allocation. */ i = 16; dp = &((struct user_segment_descriptor *)(pldt->ldt_base))[i]; for (; i < max_ldt_segment; ++i, ++dp) { if (dp->sd_type == SDT_SYSNULL) break; } if (i >= max_ldt_segment) { mtx_unlock(&dt_lock); return (ENOSPC); } uap->start = i; error = amd64_set_ldt_data(td, i, 1, descs); mtx_unlock(&dt_lock); } else { largest_ld = uap->start + uap->num; if (largest_ld > max_ldt_segment) return (EINVAL); mtx_lock(&dt_lock); if (user_ldt_alloc(p, 0) != NULL) { error = amd64_set_ldt_data(td, uap->start, uap->num, descs); } mtx_unlock(&dt_lock); } if (error == 0) td->td_retval[0] = uap->start; return (error); } int amd64_set_ldt_data(struct thread *td, int start, int num, struct user_segment_descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; volatile uint64_t *dst, *src; int i; mtx_assert(&dt_lock, MA_OWNED); mdp = &td->td_proc->p_md; pldt = mdp->md_ldt; dst = (volatile uint64_t *)(pldt->ldt_base); src = (volatile uint64_t *)descs; for (i = 0; i < num; i++) dst[start + i] = src[i]; return (0); } diff --git a/sys/arm/arm/sys_machdep.c b/sys/arm/arm/sys_machdep.c index fc424d0fad39..718c5a06318e 100644 --- a/sys/arm/arm/sys_machdep.c +++ b/sys/arm/arm/sys_machdep.c @@ -1,220 +1,222 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1990 The Regents of the University of California. * 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. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``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 REGENTS OR CONTRIBUTORS 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. * * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" +#include "opt_ktrace.h" #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #ifndef _SYS_SYSPROTO_H_ struct sysarch_args { int op; char *parms; }; #endif /* Prototypes */ static int arm32_sync_icache (struct thread *, void *); static int arm32_drain_writebuf(struct thread *, void *); static int sync_icache(uintptr_t addr, size_t len) { size_t size; vm_offset_t rv; /* Align starting address to cacheline size */ len += addr & cpuinfo.dcache_line_mask; addr &= ~cpuinfo.dcache_line_mask; /* Break whole range to pages. */ do { size = PAGE_SIZE - (addr & PAGE_MASK); size = min(size, len); rv = dcache_wb_pou_checked(addr, size); if (rv == 1) /* see dcache_wb_pou_checked() */ rv = icache_inv_pou_checked(addr, size); if (rv != 1) { if (!useracc((void *)addr, size, VM_PROT_READ)) { /* Invalid access */ return (rv); } /* Valid but unmapped page - skip it. */ } len -= size; addr += size; } while (len > 0); /* Invalidate branch predictor buffer. */ bpb_inv_all(); return (1); } static int arm32_sync_icache(struct thread *td, void *args) { struct arm_sync_icache_args ua; int error; ksiginfo_t ksi; vm_offset_t rv; if ((error = copyin(args, &ua, sizeof(ua))) != 0) return (error); if (ua.len == 0) { td->td_retval[0] = 0; return (0); } /* * Validate arguments. Address and length are unsigned, * so we can use wrapped overflow check. */ if (((ua.addr + ua.len) < ua.addr) || ((ua.addr + ua.len) > VM_MAXUSER_ADDRESS)) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGSEGV; ksi.ksi_code = SEGV_ACCERR; ksi.ksi_addr = (void *)max(ua.addr, VM_MAXUSER_ADDRESS); trapsignal(td, &ksi); return (EINVAL); } rv = sync_icache(ua.addr, ua.len); if (rv != 1) { ksiginfo_init_trap(&ksi); ksi.ksi_signo = SIGSEGV; ksi.ksi_code = SEGV_MAPERR; ksi.ksi_addr = (void *)rv; trapsignal(td, &ksi); return (EINVAL); } td->td_retval[0] = 0; return (0); } static int arm32_drain_writebuf(struct thread *td, void *args) { /* No args. */ dsb(); cpu_l2cache_drain_writebuf(); td->td_retval[0] = 0; return (0); } static int arm32_set_tp(struct thread *td, void *args) { set_tls(args); return (0); } static int arm32_get_tp(struct thread *td, void *args) { td->td_retval[0] = (register_t)get_tls(); return (0); } int sysarch(struct thread *td, struct sysarch_args *uap) { int error; #ifdef CAPABILITY_MODE /* * When adding new operations, add a new case statement here to * explicitly indicate whether or not the operation is safe to * perform in capability mode. */ if (IN_CAPABILITY_MODE(td)) { switch (uap->op) { case ARM_SYNC_ICACHE: case ARM_DRAIN_WRITEBUF: case ARM_SET_TP: case ARM_GET_TP: case ARM_GET_VFPSTATE: break; default: #ifdef KTRACE if (KTRPOINT(td, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); #endif return (ECAPMODE); } } #endif switch (uap->op) { case ARM_SYNC_ICACHE: error = arm32_sync_icache(td, uap->parms); break; case ARM_DRAIN_WRITEBUF: error = arm32_drain_writebuf(td, uap->parms); break; case ARM_SET_TP: error = arm32_set_tp(td, uap->parms); break; case ARM_GET_TP: error = arm32_get_tp(td, uap->parms); break; case ARM_GET_VFPSTATE: error = arm_get_vfpstate(td, uap->parms); break; default: error = EINVAL; break; } return (error); } diff --git a/sys/i386/i386/sys_machdep.c b/sys/i386/i386/sys_machdep.c index eeff4b280696..04456b55da52 100644 --- a/sys/i386/i386/sys_machdep.c +++ b/sys/i386/i386/sys_machdep.c @@ -1,807 +1,809 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1990 The Regents of the University of California. * 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. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``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 REGENTS OR CONTRIBUTORS 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. * * from: @(#)sys_machdep.c 5.5 (Berkeley) 1/19/91 */ #include __FBSDID("$FreeBSD$"); #include "opt_capsicum.h" #include "opt_kstack_pages.h" +#include "opt_ktrace.h" #include #include #include +#include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* for kernel_map */ #define MAX_LD 8192 #define LD_PER_PAGE 512 #define NEW_MAX_LD(num) rounddown2(num + LD_PER_PAGE, LD_PER_PAGE) #define SIZE_FROM_LARGEST_LD(num) (NEW_MAX_LD(num) << 3) #define NULL_LDT_BASE ((caddr_t)NULL) #ifdef SMP static void set_user_ldt_rv(void *arg); #endif static int i386_set_ldt_data(struct thread *, int start, int num, union descriptor *descs); static int i386_ldt_grow(struct thread *td, int len); void fill_based_sd(struct segment_descriptor *sdp, uint32_t base) { sdp->sd_lobase = base & 0xffffff; sdp->sd_hibase = (base >> 24) & 0xff; sdp->sd_lolimit = 0xffff; /* 4GB limit, wraps around */ sdp->sd_hilimit = 0xf; sdp->sd_type = SDT_MEMRWA; sdp->sd_dpl = SEL_UPL; sdp->sd_p = 1; sdp->sd_xx = 0; sdp->sd_def32 = 1; sdp->sd_gran = 1; } /* * Construct special descriptors for "base" selectors. Store them in * the PCB for later use by cpu_switch(). Store them in the GDT for * more immediate use. The GDT entries are part of the current * context. Callers must load related segment registers to complete * setting up the current context. */ void set_fsbase(struct thread *td, uint32_t base) { struct segment_descriptor sd; fill_based_sd(&sd, base); critical_enter(); td->td_pcb->pcb_fsd = sd; if (td == curthread) PCPU_GET(fsgs_gdt)[0] = sd; critical_exit(); } void set_gsbase(struct thread *td, uint32_t base) { struct segment_descriptor sd; fill_based_sd(&sd, base); critical_enter(); td->td_pcb->pcb_gsd = sd; if (td == curthread) PCPU_GET(fsgs_gdt)[1] = sd; critical_exit(); } #ifndef _SYS_SYSPROTO_H_ struct sysarch_args { int op; char *parms; }; #endif int sysarch(struct thread *td, struct sysarch_args *uap) { int error; union descriptor *lp; union { struct i386_ldt_args largs; struct i386_ioperm_args iargs; struct i386_get_xfpustate xfpu; } kargs; uint32_t base; struct segment_descriptor *sdp; AUDIT_ARG_CMD(uap->op); #ifdef CAPABILITY_MODE /* * When adding new operations, add a new case statement here to * explicitly indicate whether or not the operation is safe to * perform in capability mode. */ if (IN_CAPABILITY_MODE(td)) { switch (uap->op) { case I386_GET_LDT: case I386_SET_LDT: case I386_GET_IOPERM: case I386_GET_FSBASE: case I386_SET_FSBASE: case I386_GET_GSBASE: case I386_SET_GSBASE: case I386_GET_XFPUSTATE: break; case I386_SET_IOPERM: default: #ifdef KTRACE if (KTRPOINT(td, KTR_CAPFAIL)) ktrcapfail(CAPFAIL_SYSCALL, NULL, NULL); #endif return (ECAPMODE); } } #endif switch (uap->op) { case I386_GET_IOPERM: case I386_SET_IOPERM: if ((error = copyin(uap->parms, &kargs.iargs, sizeof(struct i386_ioperm_args))) != 0) return (error); break; case I386_GET_LDT: case I386_SET_LDT: if ((error = copyin(uap->parms, &kargs.largs, sizeof(struct i386_ldt_args))) != 0) return (error); break; case I386_GET_XFPUSTATE: if ((error = copyin(uap->parms, &kargs.xfpu, sizeof(struct i386_get_xfpustate))) != 0) return (error); break; default: break; } switch (uap->op) { case I386_GET_LDT: error = i386_get_ldt(td, &kargs.largs); break; case I386_SET_LDT: if (kargs.largs.descs != NULL) { if (kargs.largs.num > MAX_LD) return (EINVAL); lp = malloc(kargs.largs.num * sizeof(union descriptor), M_TEMP, M_WAITOK); error = copyin(kargs.largs.descs, lp, kargs.largs.num * sizeof(union descriptor)); if (error == 0) error = i386_set_ldt(td, &kargs.largs, lp); free(lp, M_TEMP); } else { error = i386_set_ldt(td, &kargs.largs, NULL); } break; case I386_GET_IOPERM: error = i386_get_ioperm(td, &kargs.iargs); if (error == 0) error = copyout(&kargs.iargs, uap->parms, sizeof(struct i386_ioperm_args)); break; case I386_SET_IOPERM: error = i386_set_ioperm(td, &kargs.iargs); break; case I386_VM86: error = vm86_sysarch(td, uap->parms); break; case I386_GET_FSBASE: sdp = &td->td_pcb->pcb_fsd; base = sdp->sd_hibase << 24 | sdp->sd_lobase; error = copyout(&base, uap->parms, sizeof(base)); break; case I386_SET_FSBASE: error = copyin(uap->parms, &base, sizeof(base)); if (error == 0) { /* * Construct the special descriptor for fsbase * and arrange for doreti to load its selector * soon enough. */ set_fsbase(td, base); td->td_frame->tf_fs = GSEL(GUFS_SEL, SEL_UPL); } break; case I386_GET_GSBASE: sdp = &td->td_pcb->pcb_gsd; base = sdp->sd_hibase << 24 | sdp->sd_lobase; error = copyout(&base, uap->parms, sizeof(base)); break; case I386_SET_GSBASE: error = copyin(uap->parms, &base, sizeof(base)); if (error == 0) { /* * Construct the special descriptor for gsbase. * The selector is loaded immediately, since we * normally only reload %gs on context switches. */ set_gsbase(td, base); load_gs(GSEL(GUGS_SEL, SEL_UPL)); } break; case I386_GET_XFPUSTATE: if (kargs.xfpu.len > cpu_max_ext_state_size - sizeof(union savefpu)) return (EINVAL); npxgetregs(td); error = copyout((char *)(get_pcb_user_save_td(td) + 1), kargs.xfpu.addr, kargs.xfpu.len); break; default: error = EINVAL; break; } return (error); } int i386_extend_pcb(struct thread *td) { int i, offset; u_long *addr; struct pcb_ext *ext; struct soft_segment_descriptor ssd = { 0, /* segment base address (overwritten) */ ctob(IOPAGES + 1) - 1, /* length */ SDT_SYS386TSS, /* segment type */ 0, /* priority level */ 1, /* descriptor present */ 0, 0, 0, /* default 32 size */ 0 /* granularity */ }; ext = pmap_trm_alloc(ctob(IOPAGES + 1), M_WAITOK | M_ZERO); /* -16 is so we can convert a trapframe into vm86trapframe inplace */ ext->ext_tss.tss_ss0 = GSEL(GDATA_SEL, SEL_KPL); /* * The last byte of the i/o map must be followed by an 0xff byte. * We arbitrarily allocate 16 bytes here, to keep the starting * address on a doubleword boundary. */ offset = PAGE_SIZE - 16; ext->ext_tss.tss_ioopt = (offset - ((unsigned)&ext->ext_tss - (unsigned)ext)) << 16; ext->ext_iomap = (caddr_t)ext + offset; ext->ext_vm86.vm86_intmap = (caddr_t)ext + offset - 32; addr = (u_long *)ext->ext_vm86.vm86_intmap; for (i = 0; i < (ctob(IOPAGES) + 32 + 16) / sizeof(u_long); i++) *addr++ = ~0; ssd.ssd_base = (unsigned)&ext->ext_tss; ssd.ssd_limit -= ((unsigned)&ext->ext_tss - (unsigned)ext); ssdtosd(&ssd, &ext->ext_tssd); KASSERT(td == curthread, ("giving TSS to !curthread")); KASSERT(td->td_pcb->pcb_ext == 0, ("already have a TSS!")); /* Switch to the new TSS. */ critical_enter(); ext->ext_tss.tss_esp0 = PCPU_GET(trampstk); td->td_pcb->pcb_ext = ext; PCPU_SET(private_tss, 1); *PCPU_GET(tss_gdt) = ext->ext_tssd; ltr(GSEL(GPROC0_SEL, SEL_KPL)); critical_exit(); return 0; } int i386_set_ioperm(struct thread *td, struct i386_ioperm_args *uap) { char *iomap; u_int i; int error; if ((error = priv_check(td, PRIV_IO)) != 0) return (error); if ((error = securelevel_gt(td->td_ucred, 0)) != 0) return (error); /* * XXX * While this is restricted to root, we should probably figure out * whether any other driver is using this i/o address, as so not to * cause confusion. This probably requires a global 'usage registry'. */ if (td->td_pcb->pcb_ext == 0) if ((error = i386_extend_pcb(td)) != 0) return (error); iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; if (uap->start > uap->start + uap->length || uap->start + uap->length > IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); for (i = uap->start; i < uap->start + uap->length; i++) { if (uap->enable) iomap[i >> 3] &= ~(1 << (i & 7)); else iomap[i >> 3] |= (1 << (i & 7)); } return (error); } int i386_get_ioperm(struct thread *td, struct i386_ioperm_args *uap) { int i, state; char *iomap; if (uap->start >= IOPAGES * PAGE_SIZE * NBBY) return (EINVAL); if (td->td_pcb->pcb_ext == 0) { uap->length = 0; goto done; } iomap = (char *)td->td_pcb->pcb_ext->ext_iomap; i = uap->start; state = (iomap[i >> 3] >> (i & 7)) & 1; uap->enable = !state; uap->length = 1; for (i = uap->start + 1; i < IOPAGES * PAGE_SIZE * NBBY; i++) { if (state != ((iomap[i >> 3] >> (i & 7)) & 1)) break; uap->length++; } done: return (0); } /* * Update the GDT entry pointing to the LDT to point to the LDT of the * current process. Manage dt_lock holding/unholding autonomously. */ static void set_user_ldt_locked(struct mdproc *mdp) { struct proc_ldt *pldt; int gdt_idx; mtx_assert(&dt_lock, MA_OWNED); pldt = mdp->md_ldt; gdt_idx = GUSERLDT_SEL; gdt_idx += PCPU_GET(cpuid) * NGDT; /* always 0 on UP */ gdt[gdt_idx].sd = pldt->ldt_sd; lldt(GSEL(GUSERLDT_SEL, SEL_KPL)); PCPU_SET(currentldt, GSEL(GUSERLDT_SEL, SEL_KPL)); } void set_user_ldt(struct mdproc *mdp) { mtx_lock_spin(&dt_lock); set_user_ldt_locked(mdp); mtx_unlock_spin(&dt_lock); } #ifdef SMP static void set_user_ldt_rv(void *arg) { struct proc *p; p = curproc; if (arg == p->p_vmspace) set_user_ldt(&p->p_md); } #endif /* * dt_lock must be held. Returns with dt_lock held. */ struct proc_ldt * user_ldt_alloc(struct mdproc *mdp, int len) { struct proc_ldt *pldt, *new_ldt; mtx_assert(&dt_lock, MA_OWNED); mtx_unlock_spin(&dt_lock); new_ldt = malloc(sizeof(struct proc_ldt), M_SUBPROC, M_WAITOK); new_ldt->ldt_len = len = NEW_MAX_LD(len); new_ldt->ldt_base = pmap_trm_alloc(len * sizeof(union descriptor), M_WAITOK | M_ZERO); new_ldt->ldt_refcnt = 1; new_ldt->ldt_active = 0; mtx_lock_spin(&dt_lock); gdt_segs[GUSERLDT_SEL].ssd_base = (unsigned)new_ldt->ldt_base; gdt_segs[GUSERLDT_SEL].ssd_limit = len * sizeof(union descriptor) - 1; ssdtosd(&gdt_segs[GUSERLDT_SEL], &new_ldt->ldt_sd); if ((pldt = mdp->md_ldt) != NULL) { if (len > pldt->ldt_len) len = pldt->ldt_len; bcopy(pldt->ldt_base, new_ldt->ldt_base, len * sizeof(union descriptor)); } else bcopy(ldt, new_ldt->ldt_base, sizeof(union descriptor) * NLDT); return (new_ldt); } /* * Must be called with dt_lock held. Returns with dt_lock unheld. */ void user_ldt_free(struct thread *td) { struct mdproc *mdp; struct proc_ldt *pldt; mtx_assert(&dt_lock, MA_OWNED); mdp = &td->td_proc->p_md; if ((pldt = mdp->md_ldt) == NULL) { mtx_unlock_spin(&dt_lock); return; } if (td == curthread) { lldt(_default_ldt); PCPU_SET(currentldt, _default_ldt); } mdp->md_ldt = NULL; user_ldt_deref(pldt); } void user_ldt_deref(struct proc_ldt *pldt) { mtx_assert(&dt_lock, MA_OWNED); if (--pldt->ldt_refcnt == 0) { mtx_unlock_spin(&dt_lock); pmap_trm_free(pldt->ldt_base, pldt->ldt_len * sizeof(union descriptor)); free(pldt, M_SUBPROC); } else mtx_unlock_spin(&dt_lock); } /* * Note for the authors of compat layers (linux, etc): copyout() in * the function below is not a problem since it presents data in * arch-specific format (i.e. i386-specific in this case), not in * the OS-specific one. */ int i386_get_ldt(struct thread *td, struct i386_ldt_args *uap) { struct proc_ldt *pldt; char *data; u_int nldt, num; int error; #ifdef DEBUG printf("i386_get_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif num = min(uap->num, MAX_LD); data = malloc(num * sizeof(union descriptor), M_TEMP, M_WAITOK); mtx_lock_spin(&dt_lock); pldt = td->td_proc->p_md.md_ldt; nldt = pldt != NULL ? pldt->ldt_len : NLDT; if (uap->start >= nldt) { num = 0; } else { num = min(num, nldt - uap->start); bcopy(pldt != NULL ? &((union descriptor *)(pldt->ldt_base))[uap->start] : &ldt[uap->start], data, num * sizeof(union descriptor)); } mtx_unlock_spin(&dt_lock); error = copyout(data, uap->descs, num * sizeof(union descriptor)); if (error == 0) td->td_retval[0] = num; free(data, M_TEMP); return (error); } int i386_set_ldt(struct thread *td, struct i386_ldt_args *uap, union descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; union descriptor *dp; u_int largest_ld, i; int error; #ifdef DEBUG printf("i386_set_ldt: start=%u num=%u descs=%p\n", uap->start, uap->num, (void *)uap->descs); #endif error = 0; mdp = &td->td_proc->p_md; if (descs == NULL) { /* Free descriptors */ if (uap->start == 0 && uap->num == 0) { /* * Treat this as a special case, so userland needn't * know magic number NLDT. */ uap->start = NLDT; uap->num = MAX_LD - NLDT; } mtx_lock_spin(&dt_lock); if ((pldt = mdp->md_ldt) == NULL || uap->start >= pldt->ldt_len) { mtx_unlock_spin(&dt_lock); return (0); } largest_ld = uap->start + uap->num; if (largest_ld > pldt->ldt_len) largest_ld = pldt->ldt_len; for (i = uap->start; i < largest_ld; i++) atomic_store_rel_64(&((uint64_t *)(pldt->ldt_base))[i], 0); mtx_unlock_spin(&dt_lock); return (0); } if (uap->start != LDT_AUTO_ALLOC || uap->num != 1) { /* verify range of descriptors to modify */ largest_ld = uap->start + uap->num; if (uap->start >= MAX_LD || largest_ld > MAX_LD) return (EINVAL); } /* Check descriptors for access violations */ for (i = 0; i < uap->num; i++) { dp = &descs[i]; switch (dp->sd.sd_type) { case SDT_SYSNULL: /* system null */ dp->sd.sd_p = 0; break; case SDT_SYS286TSS: /* system 286 TSS available */ case SDT_SYSLDT: /* system local descriptor table */ case SDT_SYS286BSY: /* system 286 TSS busy */ case SDT_SYSTASKGT: /* system task gate */ case SDT_SYS286IGT: /* system 286 interrupt gate */ case SDT_SYS286TGT: /* system 286 trap gate */ case SDT_SYSNULL2: /* undefined by Intel */ case SDT_SYS386TSS: /* system 386 TSS available */ case SDT_SYSNULL3: /* undefined by Intel */ case SDT_SYS386BSY: /* system 386 TSS busy */ case SDT_SYSNULL4: /* undefined by Intel */ case SDT_SYS386IGT: /* system 386 interrupt gate */ case SDT_SYS386TGT: /* system 386 trap gate */ case SDT_SYS286CGT: /* system 286 call gate */ case SDT_SYS386CGT: /* system 386 call gate */ return (EACCES); /* memory segment types */ case SDT_MEMEC: /* memory execute only conforming */ case SDT_MEMEAC: /* memory execute only accessed conforming */ case SDT_MEMERC: /* memory execute read conforming */ case SDT_MEMERAC: /* memory execute read accessed conforming */ /* Must be "present" if executable and conforming. */ if (dp->sd.sd_p == 0) return (EACCES); break; case SDT_MEMRO: /* memory read only */ case SDT_MEMROA: /* memory read only accessed */ case SDT_MEMRW: /* memory read write */ case SDT_MEMRWA: /* memory read write accessed */ case SDT_MEMROD: /* memory read only expand dwn limit */ case SDT_MEMRODA: /* memory read only expand dwn lim accessed */ case SDT_MEMRWD: /* memory read write expand dwn limit */ case SDT_MEMRWDA: /* memory read write expand dwn lim acessed */ case SDT_MEME: /* memory execute only */ case SDT_MEMEA: /* memory execute only accessed */ case SDT_MEMER: /* memory execute read */ case SDT_MEMERA: /* memory execute read accessed */ break; default: return (EINVAL); } /* Only user (ring-3) descriptors may be present. */ if (dp->sd.sd_p != 0 && dp->sd.sd_dpl != SEL_UPL) return (EACCES); } if (uap->start == LDT_AUTO_ALLOC && uap->num == 1) { /* Allocate a free slot */ mtx_lock_spin(&dt_lock); if ((pldt = mdp->md_ldt) == NULL) { if ((error = i386_ldt_grow(td, NLDT + 1))) { mtx_unlock_spin(&dt_lock); return (error); } pldt = mdp->md_ldt; } again: /* * start scanning a bit up to leave room for NVidia and * Wine, which still user the "Blat" method of allocation. */ dp = &((union descriptor *)(pldt->ldt_base))[NLDT]; for (i = NLDT; i < pldt->ldt_len; ++i) { if (dp->sd.sd_type == SDT_SYSNULL) break; dp++; } if (i >= pldt->ldt_len) { if ((error = i386_ldt_grow(td, pldt->ldt_len+1))) { mtx_unlock_spin(&dt_lock); return (error); } goto again; } uap->start = i; error = i386_set_ldt_data(td, i, 1, descs); mtx_unlock_spin(&dt_lock); } else { largest_ld = uap->start + uap->num; mtx_lock_spin(&dt_lock); if (!(error = i386_ldt_grow(td, largest_ld))) { error = i386_set_ldt_data(td, uap->start, uap->num, descs); } mtx_unlock_spin(&dt_lock); } if (error == 0) td->td_retval[0] = uap->start; return (error); } static int i386_set_ldt_data(struct thread *td, int start, int num, union descriptor *descs) { struct mdproc *mdp; struct proc_ldt *pldt; uint64_t *dst, *src; int i; mtx_assert(&dt_lock, MA_OWNED); mdp = &td->td_proc->p_md; pldt = mdp->md_ldt; dst = (uint64_t *)(pldt->ldt_base); src = (uint64_t *)descs; /* * Atomic(9) is used only to get 64bit atomic store with * cmpxchg8b when available. There is no op without release * semantic. */ for (i = 0; i < num; i++) atomic_store_rel_64(&dst[start + i], src[i]); return (0); } static int i386_ldt_grow(struct thread *td, int len) { struct mdproc *mdp; struct proc_ldt *new_ldt, *pldt; caddr_t old_ldt_base; int old_ldt_len; mtx_assert(&dt_lock, MA_OWNED); if (len > MAX_LD) return (ENOMEM); if (len < NLDT + 1) len = NLDT + 1; mdp = &td->td_proc->p_md; old_ldt_base = NULL_LDT_BASE; old_ldt_len = 0; /* Allocate a user ldt. */ if ((pldt = mdp->md_ldt) == NULL || len > pldt->ldt_len) { new_ldt = user_ldt_alloc(mdp, len); if (new_ldt == NULL) return (ENOMEM); pldt = mdp->md_ldt; if (pldt != NULL) { if (new_ldt->ldt_len <= pldt->ldt_len) { /* * We just lost the race for allocation, so * free the new object and return. */ mtx_unlock_spin(&dt_lock); pmap_trm_free(new_ldt->ldt_base, new_ldt->ldt_len * sizeof(union descriptor)); free(new_ldt, M_SUBPROC); mtx_lock_spin(&dt_lock); return (0); } /* * We have to substitute the current LDT entry for * curproc with the new one since its size grew. */ old_ldt_base = pldt->ldt_base; old_ldt_len = pldt->ldt_len; pldt->ldt_sd = new_ldt->ldt_sd; pldt->ldt_base = new_ldt->ldt_base; pldt->ldt_len = new_ldt->ldt_len; } else mdp->md_ldt = pldt = new_ldt; #ifdef SMP /* * Signal other cpus to reload ldt. We need to unlock dt_lock * here because other CPU will contest on it since their * curthreads won't hold the lock and will block when trying * to acquire it. */ mtx_unlock_spin(&dt_lock); smp_rendezvous(NULL, set_user_ldt_rv, NULL, td->td_proc->p_vmspace); #else set_user_ldt_locked(&td->td_proc->p_md); mtx_unlock_spin(&dt_lock); #endif if (old_ldt_base != NULL_LDT_BASE) { pmap_trm_free(old_ldt_base, old_ldt_len * sizeof(union descriptor)); free(new_ldt, M_SUBPROC); } mtx_lock_spin(&dt_lock); } return (0); }