diff options
Diffstat (limited to 'sys')
46 files changed, 599 insertions, 395 deletions
diff --git a/sys/amd64/amd64/mem.c b/sys/amd64/amd64/mem.c index 413b7c74890e..851f2df0e6e1 100644 --- a/sys/amd64/amd64/mem.c +++ b/sys/amd64/amd64/mem.c @@ -105,8 +105,8 @@ memrw(struct cdev *dev, struct uio *uio, int flags) * PAGE_SIZE, the uiomove() call does not * access past the end of the direct map. */ - if (v >= DMAP_MIN_ADDRESS && - v < DMAP_MIN_ADDRESS + dmaplimit) { + if (v >= kva_layout.dmap_low && + v < kva_layout.dmap_high) { error = uiomove((void *)v, c, uio); break; } diff --git a/sys/amd64/amd64/minidump_machdep.c b/sys/amd64/amd64/minidump_machdep.c index 6d0917e16099..43bf81a991bf 100644 --- a/sys/amd64/amd64/minidump_machdep.c +++ b/sys/amd64/amd64/minidump_machdep.c @@ -186,7 +186,7 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) * tables, so care must be taken to read each entry only once. */ pmapsize = 0; - for (va = VM_MIN_KERNEL_ADDRESS; va < kva_end; ) { + for (va = kva_layout.km_low; va < kva_end; ) { /* * We always write a page, even if it is zero. Each * page written corresponds to 1GB of space @@ -279,9 +279,9 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) mdhdr.msgbufsize = mbp->msg_size; mdhdr.bitmapsize = round_page(BITSET_SIZE(vm_page_dump_pages)); mdhdr.pmapsize = pmapsize; - mdhdr.kernbase = VM_MIN_KERNEL_ADDRESS; - mdhdr.dmapbase = DMAP_MIN_ADDRESS; - mdhdr.dmapend = DMAP_MAX_ADDRESS; + mdhdr.kernbase = kva_layout.km_low; + mdhdr.dmapbase = kva_layout.dmap_low; + mdhdr.dmapend = kva_layout.dmap_high; mdhdr.dumpavailsize = round_page(sizeof(dump_avail)); dump_init_header(di, &kdh, KERNELDUMPMAGIC, KERNELDUMP_AMD64_VERSION, @@ -323,7 +323,7 @@ cpu_minidumpsys(struct dumperinfo *di, const struct minidumpstate *state) /* Dump kernel page directory pages */ bzero(fakepd, sizeof(fakepd)); - for (va = VM_MIN_KERNEL_ADDRESS; va < kva_end; va += NBPDP) { + for (va = kva_layout.km_low; va < kva_end; va += NBPDP) { ii = pmap_pml4e_index(va); pml4 = (uint64_t *)PHYS_TO_DMAP(KPML4phys) + ii; pdp = (uint64_t *)PHYS_TO_DMAP(*pml4 & PG_FRAME); diff --git a/sys/amd64/amd64/pmap.c b/sys/amd64/amd64/pmap.c index 9c985df13ddf..2c7777e608b9 100644 --- a/sys/amd64/amd64/pmap.c +++ b/sys/amd64/amd64/pmap.c @@ -415,7 +415,7 @@ SYSCTL_INT(_machdep, OID_AUTO, nkpt, CTLFLAG_RD, &nkpt, 0, static int ndmpdp; vm_paddr_t dmaplimit; -vm_offset_t kernel_vm_end = VM_MIN_KERNEL_ADDRESS; +vm_offset_t kernel_vm_end = VM_MIN_KERNEL_ADDRESS_LA48; pt_entry_t pg_nx; static SYSCTL_NODE(_vm, OID_AUTO, pmap, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, @@ -475,11 +475,36 @@ _Static_assert(DMPML4I + NDMPML4E <= KMSANSHADPML4I, "direct map overflow"); static pml4_entry_t *kernel_pml4; static u_int64_t DMPDphys; /* phys addr of direct mapped level 2 */ static u_int64_t DMPDPphys; /* phys addr of direct mapped level 3 */ +static u_int64_t DMPML4phys; /* ... level 4, for la57 */ static int ndmpdpphys; /* number of DMPDPphys pages */ vm_paddr_t kernphys; /* phys addr of start of bootstrap data */ vm_paddr_t KERNend; /* and the end */ +struct kva_layout_s kva_layout = { + .kva_min = KV4ADDR(PML4PML4I, 0, 0, 0), + .dmap_low = KV4ADDR(DMPML4I, 0, 0, 0), + .dmap_high = KV4ADDR(DMPML4I + NDMPML4E, 0, 0, 0), + .lm_low = KV4ADDR(LMSPML4I, 0, 0, 0), + .lm_high = KV4ADDR(LMEPML4I + 1, 0, 0, 0), + .km_low = KV4ADDR(KPML4BASE, 0, 0, 0), + .km_high = KV4ADDR(KPML4BASE + NKPML4E - 1, NPDPEPG - 1, + NPDEPG - 1, NPTEPG - 1), + .rec_pt = KV4ADDR(PML4PML4I, 0, 0, 0), +}; + +struct kva_layout_s kva_layout_la57 = { + .kva_min = KV5ADDR(NPML5EPG / 2, 0, 0, 0, 0), /* == rec_pt */ + .dmap_low = KV5ADDR(DMPML5I, 0, 0, 0, 0), + .dmap_high = KV5ADDR(DMPML5I + NDMPML5E, 0, 0, 0, 0), + .lm_low = KV4ADDR(LMSPML4I, 0, 0, 0), + .lm_high = KV4ADDR(LMEPML4I + 1, 0, 0, 0), + .km_low = KV4ADDR(KPML4BASE, 0, 0, 0), + .km_high = KV4ADDR(KPML4BASE + NKPML4E - 1, NPDPEPG - 1, + NPDEPG - 1, NPTEPG - 1), + .rec_pt = KV5ADDR(PML5PML5I, 0, 0, 0, 0), +}; + /* * pmap_mapdev support pre initialization (i.e. console) */ @@ -549,8 +574,8 @@ static int pmap_flags = PMAP_PDE_SUPERPAGE; /* flags for x86 pmaps */ static vmem_t *large_vmem; static u_int lm_ents; -#define PMAP_ADDRESS_IN_LARGEMAP(va) ((va) >= LARGEMAP_MIN_ADDRESS && \ - (va) < LARGEMAP_MIN_ADDRESS + NBPML4 * (u_long)lm_ents) +#define PMAP_ADDRESS_IN_LARGEMAP(va) ((va) >= kva_layout.lm_low && \ + (va) < kva_layout.lm_high) int pmap_pcid_enabled = 1; SYSCTL_INT(_vm_pmap, OID_AUTO, pcid_enabled, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, @@ -1336,7 +1361,7 @@ static pdp_entry_t *pmap_pti_pdpe(vm_offset_t va); static pd_entry_t *pmap_pti_pde(vm_offset_t va); static void pmap_pti_wire_pte(void *pte); static int pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, - bool remove_pt, struct spglist *free, struct rwlock **lockp); + bool demote_kpde, struct spglist *free, struct rwlock **lockp); static int pmap_remove_pte(pmap_t pmap, pt_entry_t *ptq, vm_offset_t sva, pd_entry_t ptepde, struct spglist *free, struct rwlock **lockp); static vm_page_t pmap_remove_pt_page(pmap_t pmap, vm_offset_t va); @@ -1722,7 +1747,7 @@ create_pagetables(vm_paddr_t *firstaddr) { pd_entry_t *pd_p; pdp_entry_t *pdp_p; - pml4_entry_t *p4_p; + pml4_entry_t *p4_p, *p4d_p; pml5_entry_t *p5_p; uint64_t DMPDkernphys; vm_paddr_t pax; @@ -1732,7 +1757,7 @@ create_pagetables(vm_paddr_t *firstaddr) vm_offset_t kasankernbase; int kasankpdpi, kasankpdi, nkasanpte; #endif - int i, j, ndm1g, nkpdpe, nkdmpde; + int i, j, ndm1g, nkpdpe, nkdmpde, ndmpml4phys; TSENTER(); /* Allocate page table pages for the direct map */ @@ -1740,15 +1765,30 @@ create_pagetables(vm_paddr_t *firstaddr) if (ndmpdp < 4) /* Minimum 4GB of dirmap */ ndmpdp = 4; ndmpdpphys = howmany(ndmpdp, NPDPEPG); - if (ndmpdpphys > NDMPML4E) { - /* - * Each NDMPML4E allows 512 GB, so limit to that, - * and then readjust ndmpdp and ndmpdpphys. - */ - printf("NDMPML4E limits system to %d GB\n", NDMPML4E * 512); - Maxmem = atop(NDMPML4E * NBPML4); - ndmpdpphys = NDMPML4E; - ndmpdp = NDMPML4E * NPDEPG; + if (la57) { + ndmpml4phys = howmany(ndmpdpphys, NPML4EPG); + if (ndmpml4phys > NDMPML5E) { + printf("NDMPML5E limits system to %ld GB\n", + (u_long)NDMPML5E * NBPML5 / 1024 / 1024 / 1024); + Maxmem = atop(NDMPML5E * NBPML5); + ndmpml4phys = NDMPML5E; + ndmpdpphys = ndmpml4phys * NPML4EPG; + ndmpdp = ndmpdpphys * NPDEPG; + } + DMPML4phys = allocpages(firstaddr, ndmpml4phys); + } else { + if (ndmpdpphys > NDMPML4E) { + /* + * Each NDMPML4E allows 512 GB, so limit to + * that, and then readjust ndmpdp and + * ndmpdpphys. + */ + printf("NDMPML4E limits system to %d GB\n", + NDMPML4E * 512); + Maxmem = atop(NDMPML4E * NBPML4); + ndmpdpphys = NDMPML4E; + ndmpdp = NDMPML4E * NPDEPG; + } } DMPDPphys = allocpages(firstaddr, ndmpdpphys); ndm1g = 0; @@ -1773,7 +1813,13 @@ create_pagetables(vm_paddr_t *firstaddr) dmaplimit = (vm_paddr_t)ndmpdp << PDPSHIFT; /* Allocate pages. */ + if (la57) { + KPML5phys = allocpages(firstaddr, 1); + p5_p = (pml5_entry_t *)KPML5phys; + } KPML4phys = allocpages(firstaddr, 1); + p4_p = (pml4_entry_t *)KPML4phys; + KPDPphys = allocpages(firstaddr, NKPML4E); #ifdef KASAN KASANPDPphys = allocpages(firstaddr, NKASANPML4E); @@ -1893,6 +1939,16 @@ create_pagetables(vm_paddr_t *firstaddr) } /* + * Connect the Direct Map slots up to the PML4. + * pml5 entries for DMAP are handled below in global pml5 loop. + */ + p4d_p = la57 ? (pml4_entry_t *)DMPML4phys : &p4_p[DMPML4I]; + for (i = 0; i < ndmpdpphys; i++) { + p4d_p[i] = (DMPDPphys + ptoa(i)) | X86_PG_RW | X86_PG_V | + pg_nx; + } + + /* * Instead of using a 1G page for the memory containing the kernel, * use 2M pages with read-only and no-execute permissions. (If using 1G * pages, this will partially overwrite the PDPEs above.) @@ -1911,11 +1967,6 @@ create_pagetables(vm_paddr_t *firstaddr) } } - /* And recursively map PML4 to itself in order to get PTmap */ - p4_p = (pml4_entry_t *)KPML4phys; - p4_p[PML4PML4I] = KPML4phys; - p4_p[PML4PML4I] |= X86_PG_RW | X86_PG_V | pg_nx; - #ifdef KASAN /* Connect the KASAN shadow map slots up to the PML4. */ for (i = 0; i < NKASANPML4E; i++) { @@ -1938,25 +1989,15 @@ create_pagetables(vm_paddr_t *firstaddr) } #endif - /* Connect the Direct Map slots up to the PML4. */ - for (i = 0; i < ndmpdpphys; i++) { - p4_p[DMPML4I + i] = DMPDPphys + ptoa(i); - p4_p[DMPML4I + i] |= X86_PG_RW | X86_PG_V | pg_nx; - } - /* Connect the KVA slots up to the PML4 */ for (i = 0; i < NKPML4E; i++) { p4_p[KPML4BASE + i] = KPDPphys + ptoa(i); p4_p[KPML4BASE + i] |= X86_PG_RW | X86_PG_V; } - kernel_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); - if (la57) { /* XXXKIB bootstrap KPML5phys page is lost */ - KPML5phys = allocpages(firstaddr, 1); - for (i = 0, p5_p = (pml5_entry_t *)KPML5phys; i < NPML5EPG; - i++) { + for (i = 0; i < NPML5EPG; i++) { if (i == PML5PML5I) { /* * Recursively map PML5 to itself in @@ -1964,6 +2005,10 @@ create_pagetables(vm_paddr_t *firstaddr) */ p5_p[i] = KPML5phys | X86_PG_RW | X86_PG_A | X86_PG_M | X86_PG_V | pg_nx; + } else if (i >= DMPML5I && i < DMPML5I + NDMPML5E) { + /* Connect DMAP pml4 pages to PML5. */ + p5_p[i] = (DMPML4phys + ptoa(i - DMPML5I)) | + X86_PG_RW | X86_PG_V | pg_nx; } else if (i == pmap_pml5e_index(UPT_MAX_ADDRESS)) { p5_p[i] = KPML4phys | X86_PG_RW | X86_PG_A | X86_PG_M | X86_PG_V; @@ -1971,6 +2016,10 @@ create_pagetables(vm_paddr_t *firstaddr) p5_p[i] = 0; } } + } else { + /* Recursively map PML4 to itself in order to get PTmap */ + p4_p[PML4PML4I] = KPML4phys; + p4_p[PML4PML4I] |= X86_PG_RW | X86_PG_V | pg_nx; } TSEXIT(); } @@ -2024,7 +2073,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) */ virtual_avail = (vm_offset_t)KERNSTART + round_2mpage(KERNend - (vm_paddr_t)kernphys); - virtual_end = VM_MAX_KERNEL_ADDRESS; + virtual_end = kva_layout.km_high; /* * Enable PG_G global pages, then switch to the kernel page @@ -2046,9 +2095,13 @@ pmap_bootstrap(vm_paddr_t *firstaddr) * Initialize the kernel pmap (which is statically allocated). * Count bootstrap data as being resident in case any of this data is * later unmapped (using pmap_remove()) and freed. + * + * DMAP_TO_PHYS()/PHYS_TO_DMAP() are functional only after + * kva_layout is fixed. */ PMAP_LOCK_INIT(kernel_pmap); if (la57) { + kva_layout = kva_layout_la57; vtoptem = ((1ul << (NPTEPGSHIFT + NPDEPGSHIFT + NPDPEPGSHIFT + NPML4EPGSHIFT + NPML5EPGSHIFT)) - 1) << 3; PTmap = (vm_offset_t)P5Tmap; @@ -2059,6 +2112,7 @@ pmap_bootstrap(vm_paddr_t *firstaddr) kernel_pmap->pm_cr3 = KPML5phys; pmap_pt_page_count_adj(kernel_pmap, 1); /* top-level page */ } else { + kernel_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(KPML4phys); kernel_pmap->pm_pmltop = kernel_pml4; kernel_pmap->pm_cr3 = KPML4phys; } @@ -2420,6 +2474,7 @@ pmap_init(void) { struct pmap_preinit_mapping *ppim; vm_page_t m, mpte; + pml4_entry_t *pml4e; int error, i, ret, skz63; /* L1TF, reserve page @0 unconditionally */ @@ -2559,18 +2614,19 @@ pmap_init(void) printf("pmap: large map %u PML4 slots (%lu GB)\n", lm_ents, (u_long)lm_ents * (NBPML4 / 1024 / 1024 / 1024)); if (lm_ents != 0) { - large_vmem = vmem_create("large", LARGEMAP_MIN_ADDRESS, - (vmem_size_t)lm_ents * NBPML4, PAGE_SIZE, 0, M_WAITOK); + large_vmem = vmem_create("large", kva_layout.lm_low, + (vmem_size_t)kva_layout.lm_high - kva_layout.lm_low, + PAGE_SIZE, 0, M_WAITOK); if (large_vmem == NULL) { printf("pmap: cannot create large map\n"); lm_ents = 0; } for (i = 0; i < lm_ents; i++) { m = pmap_large_map_getptp_unlocked(); - /* XXXKIB la57 */ - kernel_pml4[LMSPML4I + i] = X86_PG_V | - X86_PG_RW | X86_PG_A | X86_PG_M | pg_nx | - VM_PAGE_TO_PHYS(m); + pml4e = pmap_pml4e(kernel_pmap, kva_layout.lm_low + + (u_long)i * NBPML4); + *pml4e = X86_PG_V | X86_PG_RW | X86_PG_A | X86_PG_M | + pg_nx | VM_PAGE_TO_PHYS(m); } } } @@ -3899,7 +3955,7 @@ pmap_kextract(vm_offset_t va) pd_entry_t pde; vm_paddr_t pa; - if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) { + if (va >= kva_layout.dmap_low && va < kva_layout.dmap_high) { pa = DMAP_TO_PHYS(va); } else if (PMAP_ADDRESS_IN_LARGEMAP(va)) { pa = pmap_large_map_kextract(va); @@ -4040,7 +4096,7 @@ pmap_qremove(vm_offset_t sva, int count) * enough to one of those pmap_enter() calls for it to * be caught up in a promotion. */ - KASSERT(va >= VM_MIN_KERNEL_ADDRESS, ("usermode va %lx", va)); + KASSERT(va >= kva_layout.km_low, ("usermode va %lx", va)); KASSERT((*vtopde(va) & X86_PG_PS) == 0, ("pmap_qremove on promoted va %#lx", va)); @@ -4328,21 +4384,13 @@ void pmap_pinit_pml5(vm_page_t pml5pg) { pml5_entry_t *pm_pml5; + int i; pm_pml5 = (pml5_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pml5pg)); - - /* - * Add pml5 entry at top of KVA pointing to existing pml4 table, - * entering all existing kernel mappings into level 5 table. - */ - pm_pml5[pmap_pml5e_index(UPT_MAX_ADDRESS)] = KPML4phys | X86_PG_V | - X86_PG_RW | X86_PG_A | X86_PG_M; - - /* - * Install self-referential address mapping entry. - */ - pm_pml5[PML5PML5I] = VM_PAGE_TO_PHYS(pml5pg) | - X86_PG_RW | X86_PG_V | X86_PG_M | X86_PG_A; + for (i = 0; i < NPML5EPG / 2; i++) + pm_pml5[i] = 0; + for (; i < NPML5EPG; i++) + pm_pml5[i] = kernel_pmap->pm_pmltop[i]; } static void @@ -4899,8 +4947,8 @@ pmap_release(pmap_t pmap) m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)pmap->pm_pmltop)); if (pmap_is_la57(pmap)) { - pmap->pm_pmltop[pmap_pml5e_index(UPT_MAX_ADDRESS)] = 0; - pmap->pm_pmltop[PML5PML5I] = 0; + for (i = NPML5EPG / 2; i < NPML5EPG; i++) + pmap->pm_pmltop[i] = 0; } else { for (i = 0; i < NKPML4E; i++) /* KVA */ pmap->pm_pmltop[KPML4BASE + i] = 0; @@ -4942,7 +4990,7 @@ pmap_release(pmap_t pmap) static int kvm_size(SYSCTL_HANDLER_ARGS) { - unsigned long ksize = VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS; + unsigned long ksize = kva_layout.km_high - kva_layout.km_low; return sysctl_handle_long(oidp, &ksize, 0, req); } @@ -4953,7 +5001,7 @@ SYSCTL_PROC(_vm, OID_AUTO, kvm_size, CTLTYPE_LONG | CTLFLAG_RD | CTLFLAG_MPSAFE, static int kvm_free(SYSCTL_HANDLER_ARGS) { - unsigned long kfree = VM_MAX_KERNEL_ADDRESS - kernel_vm_end; + unsigned long kfree = kva_layout.km_high - kernel_vm_end; return sysctl_handle_long(oidp, &kfree, 0, req); } @@ -5031,7 +5079,7 @@ pmap_page_array_startup(long pages) vm_page_array_size = pages; - start = VM_MIN_KERNEL_ADDRESS; + start = kva_layout.km_low; end = start + pages * sizeof(struct vm_page); for (va = start; va < end; va += NBPDR) { pfn = first_page + (va - start) / sizeof(struct vm_page); @@ -6067,8 +6115,8 @@ pmap_demote_pde_mpte(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * so the direct map region is the only part of the * kernel address space that must be handled here. */ - KASSERT(!in_kernel || (va >= DMAP_MIN_ADDRESS && - va < DMAP_MAX_ADDRESS), + KASSERT(!in_kernel || (va >= kva_layout.dmap_low && + va < kva_layout.dmap_high), ("pmap_demote_pde: No saved mpte for va %#lx", va)); /* @@ -6165,8 +6213,7 @@ pmap_demote_pde_mpte(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * pmap_remove_kernel_pde: Remove a kernel superpage mapping. */ static void -pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, - bool remove_pt) +pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va) { pd_entry_t newpde; vm_paddr_t mptepa; @@ -6174,12 +6221,8 @@ pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, KASSERT(pmap == kernel_pmap, ("pmap %p is not kernel_pmap", pmap)); PMAP_LOCK_ASSERT(pmap, MA_OWNED); - if (remove_pt) - mpte = pmap_remove_pt_page(pmap, va); - else - mpte = vm_radix_lookup(&pmap->pm_root, pmap_pde_pindex(va)); - if (mpte == NULL) - panic("pmap_remove_kernel_pde: Missing pt page."); + mpte = pmap_remove_pt_page(pmap, va); + KASSERT(mpte != NULL, ("pmap_remove_kernel_pde: missing pt page")); mptepa = VM_PAGE_TO_PHYS(mpte); newpde = mptepa | X86_PG_M | X86_PG_A | X86_PG_RW | X86_PG_V; @@ -6209,7 +6252,7 @@ pmap_remove_kernel_pde(pmap_t pmap, pd_entry_t *pde, vm_offset_t va, * pmap_remove_pde: do the things to unmap a superpage in a process */ static int -pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, +pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool demote_kpde, struct spglist *free, struct rwlock **lockp) { struct md_page *pvh; @@ -6249,9 +6292,7 @@ pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, pmap_delayed_invl_page(m); } } - if (pmap == kernel_pmap) { - pmap_remove_kernel_pde(pmap, pdq, sva, remove_pt); - } else { + if (pmap != kernel_pmap) { mpte = pmap_remove_pt_page(pmap, sva); if (mpte != NULL) { KASSERT(vm_page_any_valid(mpte), @@ -6262,6 +6303,14 @@ pmap_remove_pde(pmap_t pmap, pd_entry_t *pdq, vm_offset_t sva, bool remove_pt, mpte->ref_count = 0; pmap_add_delayed_free_list(mpte, free, false); } + } else if (demote_kpde) { + pmap_remove_kernel_pde(pmap, pdq, sva); + } else { + mpte = vm_radix_lookup(&pmap->pm_root, pmap_pde_pindex(sva)); + if (vm_page_any_valid(mpte)) { + mpte->valid = 0; + pmap_zero_page(mpte); + } } return (pmap_unuse_pt(pmap, sva, *pmap_pdpe(pmap, sva), free)); } @@ -7183,7 +7232,7 @@ pmap_enter(pmap_t pmap, vm_offset_t va, vm_page_t m, vm_prot_t prot, PG_RW = pmap_rw_bit(pmap); va = trunc_page(va); - KASSERT(va <= VM_MAX_KERNEL_ADDRESS, ("pmap_enter: toobig")); + KASSERT(va <= kva_layout.km_high, ("pmap_enter: toobig")); KASSERT(va < UPT_MIN_ADDRESS || va >= UPT_MAX_ADDRESS, ("pmap_enter: invalid to pmap_enter page table pages (va: 0x%lx)", va)); @@ -7573,8 +7622,8 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde, u_int flags, * the mapping is not from kernel_pmap, then * a reserved PT page could be freed. */ - (void)pmap_remove_pde(pmap, pde, va, - pmap != kernel_pmap, &free, lockp); + (void)pmap_remove_pde(pmap, pde, va, false, &free, + lockp); if ((oldpde & PG_G) == 0) pmap_invalidate_pde_page(pmap, va, oldpde); } else { @@ -7584,10 +7633,9 @@ pmap_enter_pde(pmap_t pmap, vm_offset_t va, pd_entry_t newpde, u_int flags, * before any changes to mappings are * made. Abort on failure. */ - mt = PHYS_TO_VM_PAGE(*pde & PG_FRAME); - if (pmap_insert_pt_page(pmap, mt, false, false)) { - if (pdpg != NULL) - pdpg->ref_count--; + mt = PHYS_TO_VM_PAGE(oldpde & PG_FRAME); + if (pmap_insert_pt_page(pmap, mt, false, + false)) { CTR1(KTR_PMAP, "pmap_enter_pde: cannot ins kern ptp va %#lx", va); @@ -9550,7 +9598,7 @@ pmap_unmapdev(void *p, vm_size_t size) va = (vm_offset_t)p; /* If we gave a direct map region in pmap_mapdev, do nothing */ - if (va >= DMAP_MIN_ADDRESS && va < DMAP_MAX_ADDRESS) + if (va >= kva_layout.dmap_low && va < kva_layout.dmap_high) return; offset = va & PAGE_MASK; size = round_page(offset + size); @@ -9649,6 +9697,8 @@ pmap_demote_pdpe(pmap_t pmap, pdp_entry_t *pdpe, vm_offset_t va, vm_page_t m) void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) { + if (m->md.pat_mode == ma) + return; m->md.pat_mode = ma; @@ -9668,6 +9718,9 @@ pmap_page_set_memattr_noflush(vm_page_t m, vm_memattr_t ma) { int error; + if (m->md.pat_mode == ma) + return; + m->md.pat_mode = ma; if ((m->flags & PG_FICTITIOUS) != 0) @@ -9724,7 +9777,7 @@ pmap_change_prot(vm_offset_t va, vm_size_t size, vm_prot_t prot) int error; /* Only supported within the kernel map. */ - if (va < VM_MIN_KERNEL_ADDRESS) + if (va < kva_layout.km_low) return (EINVAL); PMAP_LOCK(kernel_pmap); @@ -9755,7 +9808,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, * Only supported on kernel virtual addresses, including the direct * map but excluding the recursive map. */ - if (base < DMAP_MIN_ADDRESS) + if (base < kva_layout.dmap_low) return (EINVAL); /* @@ -9778,7 +9831,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pte_bits |= X86_PG_RW; } if ((prot & VM_PROT_EXECUTE) == 0 || - va < VM_MIN_KERNEL_ADDRESS) { + va < kva_layout.km_low) { pde_bits |= pg_nx; pte_bits |= pg_nx; } @@ -9874,7 +9927,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pdpe, pde_bits, pde_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pdpe & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -9904,7 +9957,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pde, pde_bits, pde_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pde & PG_PS_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -9932,7 +9985,7 @@ pmap_change_props_locked(vm_offset_t va, vm_size_t size, vm_prot_t prot, pmap_pte_props(pte, pte_bits, pte_mask); changed = true; } - if (tmpva >= VM_MIN_KERNEL_ADDRESS && + if (tmpva >= kva_layout.km_low && (*pte & PG_FRAME) < dmaplimit) { if (pa_start == pa_end) { /* Start physical address run. */ @@ -10904,8 +10957,8 @@ pmap_large_unmap(void *svaa, vm_size_t len) struct spglist spgf; sva = (vm_offset_t)svaa; - if (len == 0 || sva + len < sva || (sva >= DMAP_MIN_ADDRESS && - sva + len <= DMAP_MIN_ADDRESS + dmaplimit)) + if (len == 0 || sva + len < sva || (sva >= kva_layout.dmap_low && + sva + len < kva_layout.dmap_high)) return; SLIST_INIT(&spgf); @@ -11151,11 +11204,10 @@ pmap_large_map_wb(void *svap, vm_size_t len) sva = (vm_offset_t)svap; eva = sva + len; pmap_large_map_wb_fence(); - if (sva >= DMAP_MIN_ADDRESS && eva <= DMAP_MIN_ADDRESS + dmaplimit) { + if (sva >= kva_layout.dmap_low && eva < kva_layout.dmap_high) { pmap_large_map_flush_range(sva, len); } else { - KASSERT(sva >= LARGEMAP_MIN_ADDRESS && - eva <= LARGEMAP_MIN_ADDRESS + lm_ents * NBPML4, + KASSERT(sva >= kva_layout.lm_low && eva < kva_layout.lm_high, ("pmap_large_map_wb: not largemap %#lx %#lx", sva, len)); pmap_large_map_wb_large(sva, eva); } @@ -11196,8 +11248,8 @@ pmap_pti_init(void) VM_OBJECT_WLOCK(pti_obj); pml4_pg = pmap_pti_alloc_page(); pti_pml4 = (pml4_entry_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(pml4_pg)); - for (va = VM_MIN_KERNEL_ADDRESS; va <= VM_MAX_KERNEL_ADDRESS && - va >= VM_MIN_KERNEL_ADDRESS && va > NBPML4; va += NBPML4) { + for (va = kva_layout.km_low; va <= kva_layout.km_high && + va >= kva_layout.km_low && va > NBPML4; va += NBPML4) { pdpe = pmap_pti_pdpe(va); pmap_pti_wire_pte(pdpe); } diff --git a/sys/amd64/amd64/trap.c b/sys/amd64/amd64/trap.c index 09ac0a67dbef..eefddad2f142 100644 --- a/sys/amd64/amd64/trap.c +++ b/sys/amd64/amd64/trap.c @@ -769,7 +769,7 @@ trap_pfault(struct trapframe *frame, bool usermode, int *signo, int *ucode) return (-1); } } - if (eva >= VM_MIN_KERNEL_ADDRESS) { + if (eva >= kva_layout.km_low) { /* * Don't allow user-mode faults in kernel address space. */ diff --git a/sys/amd64/include/param.h b/sys/amd64/include/param.h index 8db314fa034d..1bbb302259d6 100644 --- a/sys/amd64/include/param.h +++ b/sys/amd64/include/param.h @@ -146,8 +146,9 @@ #define amd64_btop(x) ((unsigned long)(x) >> PAGE_SHIFT) #define amd64_ptob(x) ((unsigned long)(x) << PAGE_SHIFT) -#define INKERNEL(va) (((va) >= DMAP_MIN_ADDRESS && (va) < DMAP_MAX_ADDRESS) \ - || ((va) >= VM_MIN_KERNEL_ADDRESS && (va) < VM_MAX_KERNEL_ADDRESS)) +#define INKERNEL(va) \ + (((va) >= kva_layout.dmap_low && (va) < kva_layout.dmap_high) || \ + ((va) >= kva_layout.km_low && (va) < kva_layout.km_high)) #ifdef SMP #define SC_TABLESIZE 1024 /* Must be power of 2. */ diff --git a/sys/amd64/include/pmap.h b/sys/amd64/include/pmap.h index 7d3e91bcd9b9..08e96027a5ed 100644 --- a/sys/amd64/include/pmap.h +++ b/sys/amd64/include/pmap.h @@ -169,11 +169,12 @@ * the recursive page table map. */ #define NDMPML4E 8 +#define NDMPML5E 32 /* - * These values control the layout of virtual memory. The starting address - * of the direct map, which is controlled by DMPML4I, must be a multiple of - * its size. (See the PHYS_TO_DMAP() and DMAP_TO_PHYS() macros.) + * These values control the layout of virtual memory. The starting + * address of the direct map is controlled by DMPML4I on LA48 and + * DMPML5I on LA57. * * Note: KPML4I is the index of the (single) level 4 page that maps * the KVA that holds KERNBASE, while KPML4BASE is the index of the @@ -191,6 +192,7 @@ #define KPML4BASE (NPML4EPG-NKPML4E) /* KVM at highest addresses */ #define DMPML4I rounddown(KPML4BASE-NDMPML4E, NDMPML4E) /* Below KVM */ +#define DMPML5I (NPML5EPG / 2 + 1) #define KPML4I (NPML4EPG-1) #define KPDPI (NPDPEPG-2) /* kernbase at -2GB */ @@ -548,6 +550,18 @@ pmap_pml5e_index(vm_offset_t va) return ((va >> PML5SHIFT) & ((1ul << NPML5EPGSHIFT) - 1)); } +struct kva_layout_s { + vm_offset_t kva_min; + vm_offset_t dmap_low; /* DMAP_MIN_ADDRESS */ + vm_offset_t dmap_high; /* DMAP_MAX_ADDRESS */ + vm_offset_t lm_low; /* LARGEMAP_MIN_ADDRESS */ + vm_offset_t lm_high; /* LARGEMAP_MAX_ADDRESS */ + vm_offset_t km_low; /* VM_MIN_KERNEL_ADDRESS */ + vm_offset_t km_high; /* VM_MAX_KERNEL_ADDRESS */ + vm_offset_t rec_pt; +}; +extern struct kva_layout_s kva_layout; + #endif /* !LOCORE */ #endif /* !_MACHINE_PMAP_H_ */ diff --git a/sys/amd64/include/vmparam.h b/sys/amd64/include/vmparam.h index 0cd9bb4fa7a4..59053665dc40 100644 --- a/sys/amd64/include/vmparam.h +++ b/sys/amd64/include/vmparam.h @@ -163,6 +163,7 @@ * Virtual addresses of things. Derived from the page directory and * page table indexes from pmap.h for precision. * + * LA48: * 0x0000000000000000 - 0x00007fffffffffff user map * 0x0000800000000000 - 0xffff7fffffffffff does not exist (hole) * 0xffff800000000000 - 0xffff804020100fff recursive page table (512GB slot) @@ -175,18 +176,29 @@ * 0xfffffc0000000000 - 0xfffffdffffffffff 2TB KMSAN shadow map, optional * 0xfffffe0000000000 - 0xffffffffffffffff 2TB kernel map * + * LA57: + * 0x0000000000000000 - 0x00ffffffffffffff user map + * 0x0100000000000000 - 0xf0ffffffffffffff does not exist (hole) + * 0xff00000000000000 - 0xff00ffffffffffff recursive page table (2048TB slot) + * 0xff01000000000000 - 0xff20ffffffffffff direct map (32 x 2048TB slots) + * 0xff21000000000000 - 0xffff807fffffffff unused + * 0xffff808000000000 - 0xffff847fffffffff large map (can be tuned up) + * 0xffff848000000000 - 0xfffff77fffffffff unused (large map extends there) + * 0xfffff60000000000 - 0xfffff7ffffffffff 2TB KMSAN origin map, optional + * 0xfffff78000000000 - 0xfffff7bfffffffff 512GB KASAN shadow map, optional + * 0xfffff80000000000 - 0xfffffbffffffffff 4TB unused + * 0xfffffc0000000000 - 0xfffffdffffffffff 2TB KMSAN shadow map, optional + * 0xfffffe0000000000 - 0xffffffffffffffff 2TB kernel map + * * Within the kernel map: * * 0xfffffe0000000000 vm_page_array * 0xffffffff80000000 KERNBASE */ -#define VM_MIN_KERNEL_ADDRESS KV4ADDR(KPML4BASE, 0, 0, 0) -#define VM_MAX_KERNEL_ADDRESS KV4ADDR(KPML4BASE + NKPML4E - 1, \ - NPDPEPG-1, NPDEPG-1, NPTEPG-1) - -#define DMAP_MIN_ADDRESS KV4ADDR(DMPML4I, 0, 0, 0) -#define DMAP_MAX_ADDRESS KV4ADDR(DMPML4I + NDMPML4E, 0, 0, 0) +#define VM_MIN_KERNEL_ADDRESS_LA48 KV4ADDR(KPML4BASE, 0, 0, 0) +#define VM_MIN_KERNEL_ADDRESS kva_layout.km_low +#define VM_MAX_KERNEL_ADDRESS kva_layout.km_high #define KASAN_MIN_ADDRESS KV4ADDR(KASANPML4I, 0, 0, 0) #define KASAN_MAX_ADDRESS KV4ADDR(KASANPML4I + NKASANPML4E, 0, 0, 0) @@ -199,9 +211,6 @@ #define KMSAN_ORIG_MAX_ADDRESS KV4ADDR(KMSANORIGPML4I + NKMSANORIGPML4E, \ 0, 0, 0) -#define LARGEMAP_MIN_ADDRESS KV4ADDR(LMSPML4I, 0, 0, 0) -#define LARGEMAP_MAX_ADDRESS KV4ADDR(LMEPML4I + 1, 0, 0, 0) - /* * Formally kernel mapping starts at KERNBASE, but kernel linker * script leaves first PDE reserved. For legacy BIOS boot, kernel is @@ -239,21 +248,21 @@ * vt fb startup needs to be reworked. */ #define PHYS_IN_DMAP(pa) (dmaplimit == 0 || (pa) < dmaplimit) -#define VIRT_IN_DMAP(va) ((va) >= DMAP_MIN_ADDRESS && \ - (va) < (DMAP_MIN_ADDRESS + dmaplimit)) +#define VIRT_IN_DMAP(va) \ + ((va) >= kva_layout.dmap_low && (va) < kva_layout.dmap_high) #define PMAP_HAS_DMAP 1 -#define PHYS_TO_DMAP(x) ({ \ +#define PHYS_TO_DMAP(x) __extension__ ({ \ KASSERT(PHYS_IN_DMAP(x), \ ("physical address %#jx not covered by the DMAP", \ (uintmax_t)x)); \ - (x) | DMAP_MIN_ADDRESS; }) + (x) + kva_layout.dmap_low; }) -#define DMAP_TO_PHYS(x) ({ \ +#define DMAP_TO_PHYS(x) __extension__ ({ \ KASSERT(VIRT_IN_DMAP(x), \ ("virtual address %#jx not covered by the DMAP", \ (uintmax_t)x)); \ - (x) & ~DMAP_MIN_ADDRESS; }) + (x) - kva_layout.dmap_low; }) /* * amd64 maps the page array into KVA so that it can be more easily @@ -274,7 +283,7 @@ */ #ifndef VM_KMEM_SIZE_MAX #define VM_KMEM_SIZE_MAX ((VM_MAX_KERNEL_ADDRESS - \ - VM_MIN_KERNEL_ADDRESS + 1) * 3 / 5) + kva_layout.km_low + 1) * 3 / 5) #endif /* initial pagein size of beginning of executable file */ diff --git a/sys/arm/arm/pmap-v6.c b/sys/arm/arm/pmap-v6.c index 92eb0589f80b..78883296c5b7 100644 --- a/sys/arm/arm/pmap-v6.c +++ b/sys/arm/arm/pmap-v6.c @@ -5767,7 +5767,7 @@ pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) CTR5(KTR_PMAP, "%s: page %p - 0x%08X oma: %d, ma: %d", __func__, m, VM_PAGE_TO_PHYS(m), oma, ma); - if ((m->flags & PG_FICTITIOUS) != 0) + if (ma == oma || (m->flags & PG_FICTITIOUS) != 0) return; #if 0 /* @@ -5784,22 +5784,20 @@ pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) * If page is not mapped by sf buffer, map the page * transient and do invalidation. */ - if (ma != oma) { - pa = VM_PAGE_TO_PHYS(m); - sched_pin(); - pc = get_pcpu(); - cmap2_pte2p = pc->pc_cmap2_pte2p; - mtx_lock(&pc->pc_cmap_lock); - if (pte2_load(cmap2_pte2p) != 0) - panic("%s: CMAP2 busy", __func__); - pte2_store(cmap2_pte2p, PTE2_KERN_NG(pa, PTE2_AP_KRW, - vm_memattr_to_pte2(ma))); - dcache_wbinv_poc((vm_offset_t)pc->pc_cmap2_addr, pa, PAGE_SIZE); - pte2_clear(cmap2_pte2p); - tlb_flush((vm_offset_t)pc->pc_cmap2_addr); - sched_unpin(); - mtx_unlock(&pc->pc_cmap_lock); - } + pa = VM_PAGE_TO_PHYS(m); + sched_pin(); + pc = get_pcpu(); + cmap2_pte2p = pc->pc_cmap2_pte2p; + mtx_lock(&pc->pc_cmap_lock); + if (pte2_load(cmap2_pte2p) != 0) + panic("%s: CMAP2 busy", __func__); + pte2_store(cmap2_pte2p, PTE2_KERN_NG(pa, PTE2_AP_KRW, + vm_memattr_to_pte2(ma))); + dcache_wbinv_poc((vm_offset_t)pc->pc_cmap2_addr, pa, PAGE_SIZE); + pte2_clear(cmap2_pte2p); + tlb_flush((vm_offset_t)pc->pc_cmap2_addr); + sched_unpin(); + mtx_unlock(&pc->pc_cmap_lock); } /* diff --git a/sys/arm64/arm64/pmap.c b/sys/arm64/arm64/pmap.c index d2e56a270f54..a09da794e77d 100644 --- a/sys/arm64/arm64/pmap.c +++ b/sys/arm64/arm64/pmap.c @@ -497,7 +497,8 @@ static bool pmap_pv_insert_l3c(pmap_t pmap, vm_offset_t va, vm_page_t m, struct rwlock **lockp); static void pmap_remove_kernel_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va); static int pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, - pd_entry_t l1e, struct spglist *free, struct rwlock **lockp); + pd_entry_t l1e, bool demote_kl2e, struct spglist *free, + struct rwlock **lockp); static int pmap_remove_l3(pmap_t pmap, pt_entry_t *l3, vm_offset_t sva, pd_entry_t l2e, struct spglist *free, struct rwlock **lockp); static bool pmap_remove_l3c(pmap_t pmap, pt_entry_t *l3p, vm_offset_t va, @@ -3847,8 +3848,7 @@ pmap_remove_kernel_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va) PMAP_LOCK_ASSERT(pmap, MA_OWNED); ml3 = pmap_remove_pt_page(pmap, va); - if (ml3 == NULL) - panic("pmap_remove_kernel_l2: Missing pt page"); + KASSERT(ml3 != NULL, ("pmap_remove_kernel_l2: missing pt page")); ml3pa = VM_PAGE_TO_PHYS(ml3); newl2 = PHYS_TO_PTE(ml3pa) | L2_TABLE; @@ -3873,8 +3873,8 @@ pmap_remove_kernel_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t va) * pmap_remove_l2: Do the things to unmap a level 2 superpage. */ static int -pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, - pd_entry_t l1e, struct spglist *free, struct rwlock **lockp) +pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, pd_entry_t l1e, + bool demote_kl2e, struct spglist *free, struct rwlock **lockp) { struct md_page *pvh; pt_entry_t old_l2; @@ -3910,9 +3910,7 @@ pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, vm_page_aflag_clear(mt, PGA_WRITEABLE); } } - if (pmap == kernel_pmap) { - pmap_remove_kernel_l2(pmap, l2, sva); - } else { + if (pmap != kernel_pmap) { ml3 = pmap_remove_pt_page(pmap, sva); if (ml3 != NULL) { KASSERT(vm_page_any_valid(ml3), @@ -3923,6 +3921,14 @@ pmap_remove_l2(pmap_t pmap, pt_entry_t *l2, vm_offset_t sva, ml3->ref_count = 0; pmap_add_delayed_free_list(ml3, free, false); } + } else if (demote_kl2e) { + pmap_remove_kernel_l2(pmap, l2, sva); + } else { + ml3 = vm_radix_lookup(&pmap->pm_root, pmap_l2_pindex(sva)); + if (vm_page_any_valid(ml3)) { + ml3->valid = 0; + pmap_zero_page(ml3); + } } return (pmap_unuse_pt(pmap, sva, l1e, free)); } @@ -4232,7 +4238,7 @@ pmap_remove1(pmap_t pmap, vm_offset_t sva, vm_offset_t eva, bool map_delete) if ((l3_paddr & ATTR_DESCR_MASK) == L2_BLOCK) { if (sva + L2_SIZE == va_next && eva >= va_next) { pmap_remove_l2(pmap, l2, sva, pmap_load(l1), - &free, &lock); + true, &free, &lock); continue; } else if (pmap_demote_l2_locked(pmap, l2, sva, &lock) == NULL) @@ -5747,33 +5753,51 @@ pmap_enter_l2(pmap_t pmap, vm_offset_t va, pd_entry_t new_l2, u_int flags, } } SLIST_INIT(&free); - if ((old_l2 & ATTR_DESCR_MASK) == L2_BLOCK) + if ((old_l2 & ATTR_DESCR_MASK) == L2_BLOCK) { (void)pmap_remove_l2(pmap, l2, va, - pmap_load(pmap_l1(pmap, va)), &free, lockp); - else + pmap_load(pmap_l1(pmap, va)), false, &free, lockp); + } else { + if (ADDR_IS_KERNEL(va)) { + /* + * Try to save the ptp in the trie + * before any changes to mappings are + * made. Abort on failure. + */ + mt = PTE_TO_VM_PAGE(old_l2); + if (pmap_insert_pt_page(pmap, mt, false, + false)) { + CTR1(KTR_PMAP, + "pmap_enter_l2: cannot ins kern ptp va %#lx", + va); + return (KERN_RESOURCE_SHORTAGE); + } + /* + * Both pmap_remove_l2() and + * pmap_remove_l3_range() will zero fill + * the L3 kernel page table page. + */ + } pmap_remove_l3_range(pmap, old_l2, va, va + L2_SIZE, &free, lockp); + if (ADDR_IS_KERNEL(va)) { + /* + * The TLB could have an intermediate + * entry for the L3 kernel page table + * page, so request an invalidation at + * all levels after clearing the + * L2_TABLE entry. + */ + pmap_clear(l2); + pmap_s1_invalidate_page(pmap, va, false); + } + } + KASSERT(pmap_load(l2) == 0, + ("pmap_enter_l2: non-zero L2 entry %p", l2)); if (!ADDR_IS_KERNEL(va)) { vm_page_free_pages_toq(&free, true); - KASSERT(pmap_load(l2) == 0, - ("pmap_enter_l2: non-zero L2 entry %p", l2)); } else { KASSERT(SLIST_EMPTY(&free), ("pmap_enter_l2: freed kernel page table page")); - - /* - * Both pmap_remove_l2() and pmap_remove_l3_range() - * will leave the kernel page table page zero filled. - * Nonetheless, the TLB could have an intermediate - * entry for the kernel page table page, so request - * an invalidation at all levels after clearing - * the L2_TABLE entry. - */ - mt = PTE_TO_VM_PAGE(pmap_load(l2)); - if (pmap_insert_pt_page(pmap, mt, false, false)) - panic("pmap_enter_l2: trie insert failed"); - pmap_clear(l2); - pmap_s1_invalidate_page(pmap, va, false); } } @@ -8045,6 +8069,8 @@ pmap_unmapbios(void *p, vm_size_t size) void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) { + if (m->md.pv_memattr == ma) + return; m->md.pv_memattr = ma; @@ -8424,8 +8450,8 @@ pmap_demote_l2_abort(pmap_t pmap, vm_offset_t va, pt_entry_t *l2, struct spglist free; SLIST_INIT(&free); - (void)pmap_remove_l2(pmap, l2, va, pmap_load(pmap_l1(pmap, va)), &free, - lockp); + (void)pmap_remove_l2(pmap, l2, va, pmap_load(pmap_l1(pmap, va)), true, + &free, lockp); vm_page_free_pages_toq(&free, true); } diff --git a/sys/compat/linuxkpi/common/include/linux/slab.h b/sys/compat/linuxkpi/common/include/linux/slab.h index f3a840d9bf4b..efa5c8cb67b3 100644 --- a/sys/compat/linuxkpi/common/include/linux/slab.h +++ b/sys/compat/linuxkpi/common/include/linux/slab.h @@ -45,7 +45,7 @@ MALLOC_DECLARE(M_KMALLOC); -#define kvzalloc(size, flags) kmalloc(size, (flags) | __GFP_ZERO) +#define kvzalloc(size, flags) kvmalloc(size, (flags) | __GFP_ZERO) #define kvcalloc(n, size, flags) kvmalloc_array(n, size, (flags) | __GFP_ZERO) #define kzalloc(size, flags) kmalloc(size, (flags) | __GFP_ZERO) #define kzalloc_node(size, flags, node) kmalloc_node(size, (flags) | __GFP_ZERO, node) diff --git a/sys/compat/linuxkpi/common/src/linux_page.c b/sys/compat/linuxkpi/common/src/linux_page.c index ebb92eacbf9a..628af17df853 100644 --- a/sys/compat/linuxkpi/common/src/linux_page.c +++ b/sys/compat/linuxkpi/common/src/linux_page.c @@ -106,6 +106,7 @@ linux_alloc_pages(gfp_t flags, unsigned int order) if ((flags & M_ZERO) != 0) req |= VM_ALLOC_ZERO; + if (order == 0 && (flags & GFP_DMA32) == 0) { page = vm_page_alloc_noobj(req); if (page == NULL) @@ -113,6 +114,10 @@ linux_alloc_pages(gfp_t flags, unsigned int order) } else { vm_paddr_t pmax = (flags & GFP_DMA32) ? BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR; + + if ((flags & __GFP_NORETRY) != 0) + req |= VM_ALLOC_NORECLAIM; + retry: page = vm_page_alloc_noobj_contig(req, npages, 0, pmax, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); diff --git a/sys/dev/usb/controller/xhci_pci.c b/sys/dev/usb/controller/xhci_pci.c index b50e33ea36ce..d5cfd228a429 100644 --- a/sys/dev/usb/controller/xhci_pci.c +++ b/sys/dev/usb/controller/xhci_pci.c @@ -99,6 +99,11 @@ xhci_pci_match(device_t self) return ("AMD Starship USB 3.0 controller"); case 0x149c1022: return ("AMD Matisse USB 3.0 controller"); + case 0x15b61022: + case 0x15b71022: + return ("AMD Raphael/Granite Ridge USB 3.1 controller"); + case 0x15b81022: + return ("AMD Raphael/Granite Ridge USB 2.0 controller"); case 0x15e01022: case 0x15e11022: return ("AMD Raven USB 3.1 controller"); @@ -109,6 +114,8 @@ xhci_pci_match(device_t self) return ("AMD 300 Series USB 3.1 controller"); case 0x43d51022: return ("AMD 400 Series USB 3.1 controller"); + case 0x43f71022: + return ("AMD 600 Series USB 3.2 controller"); case 0x78121022: case 0x78141022: case 0x79141022: diff --git a/sys/fs/fdescfs/fdesc_vnops.c b/sys/fs/fdescfs/fdesc_vnops.c index 676ea5de12b8..58a22b8bdc50 100644 --- a/sys/fs/fdescfs/fdesc_vnops.c +++ b/sys/fs/fdescfs/fdesc_vnops.c @@ -547,6 +547,8 @@ fdesc_readdir(struct vop_readdir_args *ap) fmp = VFSTOFDESC(ap->a_vp->v_mount); if (ap->a_ncookies != NULL) *ap->a_ncookies = 0; + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 0; off = (int)uio->uio_offset; if (off != uio->uio_offset || off < 0 || (u_int)off % UIO_MX != 0 || @@ -559,7 +561,12 @@ fdesc_readdir(struct vop_readdir_args *ap) fcnt = i - 2; /* The first two nodes are `.' and `..' */ FILEDESC_SLOCK(fdp); - while (i < fdp->fd_nfiles + 2 && uio->uio_resid >= UIO_MX) { + while (uio->uio_resid >= UIO_MX) { + if (i >= fdp->fd_nfiles + 2) { + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 1; + break; + } bzero((caddr_t)dp, UIO_MX); switch (i) { case 0: /* `.' */ diff --git a/sys/fs/msdosfs/msdosfs_vnops.c b/sys/fs/msdosfs/msdosfs_vnops.c index 5db61c8951f6..33e0d94954d7 100644 --- a/sys/fs/msdosfs/msdosfs_vnops.c +++ b/sys/fs/msdosfs/msdosfs_vnops.c @@ -1521,6 +1521,9 @@ msdosfs_readdir(struct vop_readdir_args *ap) ap->a_vp, uio, ap->a_cred, ap->a_eofflag); #endif + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 0; + /* * msdosfs_readdir() won't operate properly on regular files since * it does i/o only with the filesystem vnode, and hence can @@ -1614,8 +1617,11 @@ msdosfs_readdir(struct vop_readdir_args *ap) on = (offset - bias) & pmp->pm_crbomask; n = min(pmp->pm_bpcluster - on, uio->uio_resid); diff = dep->de_FileSize - (offset - bias); - if (diff <= 0) - break; + if (diff <= 0) { + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 1; + goto out; + } n = min(n, diff); error = pcbmap(dep, lbn, &bn, &cn, &blsize); if (error) @@ -1646,6 +1652,8 @@ msdosfs_readdir(struct vop_readdir_args *ap) */ if (dentp->deName[0] == SLOT_EMPTY) { brelse(bp); + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 1; goto out; } /* @@ -1743,15 +1751,6 @@ out: uio->uio_offset = off; - /* - * Set the eofflag (NFS uses it) - */ - if (ap->a_eofflag) { - if (dep->de_FileSize - (offset - bias) <= 0) - *ap->a_eofflag = 1; - else - *ap->a_eofflag = 0; - } return (error); } diff --git a/sys/fs/p9fs/p9fs_vnops.c b/sys/fs/p9fs/p9fs_vnops.c index 56bf766ef801..227e2b93883e 100644 --- a/sys/fs/p9fs/p9fs_vnops.c +++ b/sys/fs/p9fs/p9fs_vnops.c @@ -1784,6 +1784,9 @@ p9fs_readdir(struct vop_readdir_args *ap) return (EBADF); } + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 0; + io_buffer = uma_zalloc(p9fs_io_buffer_zone, M_WAITOK); /* We haven't reached the end yet. read more. */ @@ -1801,8 +1804,11 @@ p9fs_readdir(struct vop_readdir_args *ap) count = p9_client_readdir(vofid, (char *)io_buffer, diroffset, count); - if (count == 0) + if (count == 0) { + if (ap->a_eofflag != NULL) + *ap->a_eofflag = 1; break; + } if (count < 0) { error = EIO; diff --git a/sys/i386/conf/GENERIC b/sys/i386/conf/GENERIC index e7d460af21d4..f577cd07ac7c 100644 --- a/sys/i386/conf/GENERIC +++ b/sys/i386/conf/GENERIC @@ -17,6 +17,8 @@ # in NOTES. # +#NO_UNIVERSE + cpu I486_CPU cpu I586_CPU cpu I686_CPU diff --git a/sys/i386/conf/GENERIC-NODEBUG b/sys/i386/conf/GENERIC-NODEBUG index ea07613a796f..a93304481b5f 100644 --- a/sys/i386/conf/GENERIC-NODEBUG +++ b/sys/i386/conf/GENERIC-NODEBUG @@ -25,6 +25,8 @@ # in NOTES. # +#NO_UNIVERSE + include GENERIC include "std.nodebug" diff --git a/sys/i386/conf/LINT b/sys/i386/conf/LINT index 41207eb63cb9..2e947202f723 100644 --- a/sys/i386/conf/LINT +++ b/sys/i386/conf/LINT @@ -1,3 +1,4 @@ +#NO_UNIVERSE include "../../conf/NOTES" include "../../x86/conf/NOTES" diff --git a/sys/i386/conf/MINIMAL b/sys/i386/conf/MINIMAL index 2a06eb84bff8..8019617ca4d4 100644 --- a/sys/i386/conf/MINIMAL +++ b/sys/i386/conf/MINIMAL @@ -31,6 +31,8 @@ # in NOTES. # +#NO_UNIVERSE + cpu I486_CPU cpu I586_CPU cpu I686_CPU diff --git a/sys/i386/conf/PAE b/sys/i386/conf/PAE index a39d32d77106..72af9e9a9eec 100644 --- a/sys/i386/conf/PAE +++ b/sys/i386/conf/PAE @@ -2,6 +2,8 @@ # PAE -- Generic kernel configuration file for FreeBSD/i386 PAE # +#NO_UNIVERSE + include GENERIC ident PAE-GENERIC diff --git a/sys/i386/i386/pmap.c b/sys/i386/i386/pmap.c index 465b4d0f365b..b44f5e08bbcf 100644 --- a/sys/i386/i386/pmap.c +++ b/sys/i386/i386/pmap.c @@ -876,14 +876,16 @@ __CONCAT(PMTYPE, init_pat)(void) #ifdef PMAP_PAE_COMP static void * -pmap_pdpt_allocf(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *flags, - int wait) +pmap_pdpt_allocf(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *sflagsp, + int flags) { /* Inform UMA that this allocator uses kernel_map/object. */ - *flags = UMA_SLAB_KERNEL; + *sflagsp = UMA_SLAB_KERNEL; + /* contig allocations cannot be NEVERFREED */ + flags &= ~M_NEVERFREED; return ((void *)kmem_alloc_contig_domainset(DOMAINSET_FIXED(domain), - bytes, wait, 0x0ULL, 0xffffffffULL, 1, 0, VM_MEMATTR_DEFAULT)); + bytes, flags, 0x0ULL, 0xffffffffULL, 1, 0, VM_MEMATTR_DEFAULT)); } #endif @@ -5617,6 +5619,8 @@ __CONCAT(PMTYPE, unmapdev)(void *p, vm_size_t size) static void __CONCAT(PMTYPE, page_set_memattr)(vm_page_t m, vm_memattr_t ma) { + if (m->md.pat_mode == ma) + return; m->md.pat_mode = ma; if ((m->flags & PG_FICTITIOUS) != 0) diff --git a/sys/kern/kern_descrip.c b/sys/kern/kern_descrip.c index 93bdd41d1515..a27ab33b34da 100644 --- a/sys/kern/kern_descrip.c +++ b/sys/kern/kern_descrip.c @@ -557,8 +557,10 @@ open_to_fde_flags(int open_flags, bool sticky_orb) { .f = O_CLOFORK, .t = UF_FOCLOSE }, { .f = O_RESOLVE_BENEATH, .t = UF_RESOLVE_BENEATH }, }; +#if defined(__clang__) && __clang_major__ >= 19 _Static_assert(open_to_fde_flags_s[nitems(open_to_fde_flags_s) - 1].f == O_RESOLVE_BENEATH, "O_RESOLVE_BENEATH must be last, for sticky_orb"); +#endif return (flags_trans(open_to_fde_flags_s, nitems(open_to_fde_flags_s) - (sticky_orb ? 0 : 1), open_flags)); diff --git a/sys/kern/subr_asan.c b/sys/kern/subr_asan.c index 0edb631d1475..464efda1e91a 100644 --- a/sys/kern/subr_asan.c +++ b/sys/kern/subr_asan.c @@ -263,8 +263,7 @@ kasan_mark(const void *addr, size_t size, size_t redzsize, uint8_t code) if (__predict_false(!kasan_enabled)) return; - if ((vm_offset_t)addr >= DMAP_MIN_ADDRESS && - (vm_offset_t)addr < DMAP_MAX_ADDRESS) + if (kasan_md_unsupported((vm_offset_t)addr)) return; KASSERT((vm_offset_t)addr >= VM_MIN_KERNEL_ADDRESS && diff --git a/sys/kern/subr_trap.c b/sys/kern/subr_trap.c index 18388ae5f232..bac7d0080c71 100644 --- a/sys/kern/subr_trap.c +++ b/sys/kern/subr_trap.c @@ -338,8 +338,9 @@ ast_handler(struct thread *td, struct trapframe *framep, bool dtor) td->td_ast = 0; } - CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, td->td_proc->p_pid, - td->td_proc->p_comm); + CTR3(KTR_SYSC, "ast: thread %p (pid %d, %s)", td, + td->td_proc == NULL ? -1 : td->td_proc->p_pid, + td->td_proc == NULL ? "" : td->td_proc->p_comm); KASSERT(framep == NULL || TRAPF_USERMODE(framep), ("ast in kernel mode")); diff --git a/sys/net/if_lagg.c b/sys/net/if_lagg.c index 9867a718e148..5b52bfa80e3b 100644 --- a/sys/net/if_lagg.c +++ b/sys/net/if_lagg.c @@ -718,6 +718,7 @@ lagg_capabilities(struct lagg_softc *sc) sc->sc_ifp->if_capenable = ena; sc->sc_ifp->if_capenable2 = ena2; sc->sc_ifp->if_hwassist = hwa; + (void)if_hw_tsomax_update(sc->sc_ifp, &hw_tsomax); getmicrotime(&sc->sc_ifp->if_lastchange); if (sc->sc_ifflags & IFF_DEBUG) diff --git a/sys/net/pfvar.h b/sys/net/pfvar.h index 36fab1a03ee6..452a8eb4024b 100644 --- a/sys/net/pfvar.h +++ b/sys/net/pfvar.h @@ -1370,7 +1370,6 @@ struct pf_kruleset { struct pf_krulequeue queues[2]; struct { struct pf_krulequeue *ptr; - struct pf_krule **ptr_array; u_int32_t rcount; u_int32_t ticket; int open; @@ -2500,7 +2499,7 @@ int pfr_match_addr(struct pfr_ktable *, struct pf_addr *, sa_family_t); void pfr_update_stats(struct pfr_ktable *, struct pf_addr *, sa_family_t, u_int64_t, int, int, int); int pfr_pool_get(struct pfr_ktable *, int *, struct pf_addr *, sa_family_t, - pf_addr_filter_func_t); + pf_addr_filter_func_t, bool); void pfr_dynaddr_update(struct pfr_ktable *, struct pfi_dynaddr *); struct pfr_ktable * pfr_attach_table(struct pf_kruleset *, char *); @@ -2534,6 +2533,8 @@ int pfr_ina_rollback(struct pfr_table *, u_int32_t, int *, int); int pfr_ina_commit(struct pfr_table *, u_int32_t, int *, int *, int); int pfr_ina_define(struct pfr_table *, struct pfr_addr *, int, int *, int *, u_int32_t, int); +struct pfr_ktable + *pfr_ktable_select_active(struct pfr_ktable *); MALLOC_DECLARE(PFI_MTYPE); VNET_DECLARE(struct pfi_kkif *, pfi_all); @@ -2712,7 +2713,6 @@ u_short pf_map_addr(u_int8_t, struct pf_krule *, u_short pf_map_addr_sn(u_int8_t, struct pf_krule *, struct pf_addr *, struct pf_addr *, struct pfi_kkif **nkif, struct pf_addr *, - struct pf_ksrc_node **, struct pf_srchash **, struct pf_kpool *, pf_sn_types_t); int pf_get_transaddr_af(struct pf_krule *, struct pf_pdesc *); diff --git a/sys/netinet6/raw_ip6.c b/sys/netinet6/raw_ip6.c index 0379ef7c789a..c90a1213bd66 100644 --- a/sys/netinet6/raw_ip6.c +++ b/sys/netinet6/raw_ip6.c @@ -765,8 +765,7 @@ rip6_bind(struct socket *so, struct sockaddr *nam, struct thread *td) } if (ifa != NULL && ((struct in6_ifaddr *)ifa)->ia6_flags & - (IN6_IFF_ANYCAST|IN6_IFF_NOTREADY| - IN6_IFF_DETACHED|IN6_IFF_DEPRECATED)) { + (IN6_IFF_NOTREADY|IN6_IFF_DETACHED|IN6_IFF_DEPRECATED)) { NET_EPOCH_EXIT(et); return (EADDRNOTAVAIL); } diff --git a/sys/netlink/netlink_message_parser.h b/sys/netlink/netlink_message_parser.h index 8492ecb3021b..720317ed74f3 100644 --- a/sys/netlink/netlink_message_parser.h +++ b/sys/netlink/netlink_message_parser.h @@ -209,7 +209,8 @@ int nlattr_get_nested(struct nlattr *nla, struct nl_pstate *npt, int nlattr_get_nested_ptr(struct nlattr *nla, struct nl_pstate *npt, const void *arg, void *target); -bool nlmsg_report_err_msg(struct nl_pstate *npt, const char *fmt, ...); +bool nlmsg_report_err_msg(struct nl_pstate *npt, const char *fmt, ...) + __printflike(2, 3); #define NLMSG_REPORT_ERR_MSG(_npt, _fmt, ...) { \ nlmsg_report_err_msg(_npt, _fmt, ## __VA_ARGS__); \ diff --git a/sys/netpfil/ipfilter/netinet/ip_fil_freebsd.c b/sys/netpfil/ipfilter/netinet/ip_fil_freebsd.c index 04850549db98..6eb6cf2a7a47 100644 --- a/sys/netpfil/ipfilter/netinet/ip_fil_freebsd.c +++ b/sys/netpfil/ipfilter/netinet/ip_fil_freebsd.c @@ -463,13 +463,14 @@ ipf_send_ip(fr_info_t *fin, mb_t *m) int ipf_send_icmp_err(int type, fr_info_t *fin, int dst) { - int err, hlen, xtra, iclen, ohlen, avail, code; + int err, hlen, xtra, iclen, ohlen, avail; struct in_addr dst4; struct icmp *icmp; struct mbuf *m; i6addr_t dst6; void *ifp; #ifdef USE_INET6 + int code; ip6_t *ip6; #endif ip_t *ip, *ip2; @@ -477,8 +478,8 @@ ipf_send_icmp_err(int type, fr_info_t *fin, int dst) if ((type < 0) || (type >= ICMP_MAXTYPE)) return (-1); - code = fin->fin_icode; #ifdef USE_INET6 + code = fin->fin_icode; /* See NetBSD ip_fil_netbsd.c r1.4: */ if ((code < 0) || (code >= sizeof(icmptoicmp6unreach)/sizeof(int))) return (-1); diff --git a/sys/netpfil/pf/if_pflog.c b/sys/netpfil/pf/if_pflog.c index 0a84f9d680ac..cb96d2fcc44c 100644 --- a/sys/netpfil/pf/if_pflog.c +++ b/sys/netpfil/pf/if_pflog.c @@ -284,9 +284,9 @@ pflog_packet(uint8_t action, u_int8_t reason, * state lock, since this leads to unsafe LOR. * These conditions are very very rare, however. */ - if (trigger->log & PF_LOG_SOCKET_LOOKUP && !pd->lookup.done && lookupsafe) + if (trigger->log & PF_LOG_USER && !pd->lookup.done && lookupsafe) pd->lookup.done = pf_socket_lookup(pd); - if (pd->lookup.done > 0) + if (trigger->log & PF_LOG_USER && pd->lookup.done > 0) hdr.uid = pd->lookup.uid; else hdr.uid = -1; diff --git a/sys/netpfil/pf/if_pfsync.c b/sys/netpfil/pf/if_pfsync.c index 2391edaf1a5a..4e03584b8f85 100644 --- a/sys/netpfil/pf/if_pfsync.c +++ b/sys/netpfil/pf/if_pfsync.c @@ -532,6 +532,7 @@ pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version) struct pf_kpooladdr *rpool_first; int error; uint8_t rt = 0; + int n = 0; PF_RULES_RASSERT(); @@ -557,10 +558,12 @@ pfsync_state_import(union pfsync_state_union *sp, int flags, int msg_version) */ if (sp->pfs_1301.rule != htonl(-1) && sp->pfs_1301.anchor == htonl(-1) && (flags & (PFSYNC_SI_IOCTL | PFSYNC_SI_CKSUM)) && ntohl(sp->pfs_1301.rule) < - pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) - r = pf_main_ruleset.rules[ - PF_RULESET_FILTER].active.ptr_array[ntohl(sp->pfs_1301.rule)]; - else + pf_main_ruleset.rules[PF_RULESET_FILTER].active.rcount) { + TAILQ_FOREACH(r, pf_main_ruleset.rules[ + PF_RULESET_FILTER].active.ptr, entries) + if (ntohl(sp->pfs_1301.rule) == n++) + break; + } else r = &V_pf_default_rule; /* diff --git a/sys/netpfil/pf/pf.c b/sys/netpfil/pf/pf.c index d5f01e5c4956..63d513fb1956 100644 --- a/sys/netpfil/pf/pf.c +++ b/sys/netpfil/pf/pf.c @@ -5901,18 +5901,17 @@ pf_test_rule(struct pf_krule **rm, struct pf_kstate **sm, M_SETFIB(pd->m, pd->act.rtableid); if (r->rt) { - struct pf_ksrc_node *sn = NULL; - struct pf_srchash *snh = NULL; /* * Set act.rt here instead of in pf_rule_to_actions() because * it is applied only from the last pass rule. */ pd->act.rt = r->rt; - /* Don't use REASON_SET, pf_map_addr increases the reason counters */ - ctx.reason = pf_map_addr_sn(pd->af, r, pd->src, &pd->act.rt_addr, - &pd->act.rt_kif, NULL, &sn, &snh, &(r->route), PF_SN_ROUTE); - if (ctx.reason != 0) + if ((transerror = pf_map_addr_sn(pd->af, r, pd->src, + &pd->act.rt_addr, &pd->act.rt_kif, NULL, &(r->route), + PF_SN_ROUTE)) != PFRES_MATCH) { + REASON_SET(&ctx.reason, transerror); goto cleanup; + } } if (pd->virtual_proto != PF_VPROTO_FRAGMENT && @@ -6056,9 +6055,16 @@ pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx, /* src node for translation rule */ if (ctx->nr != NULL) { KASSERT(ctx->nat_pool != NULL, ("%s: nat_pool is NULL", __func__)); + /* + * The NAT addresses are chosen during ruleset parsing. + * The new afto code stores post-nat addresses in nsaddr. + * The old nat code (also used for new nat-to rules) creates + * state keys and stores addresses in them. + */ if ((ctx->nat_pool->opts & PF_POOL_STICKYADDR) && (sn_reason = pf_insert_src_node(sns, snhs, ctx->nr, - &ctx->sk->addr[pd->sidx], pd->af, &ctx->nk->addr[1], NULL, + ctx->sk ? &(ctx->sk->addr[pd->sidx]) : pd->src, pd->af, + ctx->nk ? &(ctx->nk->addr[1]) : &(pd->nsaddr), NULL, PF_SN_NAT)) != 0 ) { REASON_SET(&ctx->reason, sn_reason); goto csfailed; @@ -6213,7 +6219,7 @@ pf_create_state(struct pf_krule *r, struct pf_test_ctx *ctx, if (ctx->tag > 0) s->tag = ctx->tag; if (pd->proto == IPPROTO_TCP && (tcp_get_flags(th) & (TH_SYN|TH_ACK)) == - TH_SYN && r->keep_state == PF_STATE_SYNPROXY) { + TH_SYN && r->keep_state == PF_STATE_SYNPROXY && pd->dir == PF_IN) { pf_set_protostate(s, PF_PEER_SRC, PF_TCPS_PROXY_SRC); pf_undo_nat(ctx->nr, pd, bip_sum); s->src.seqhi = arc4random(); @@ -9062,6 +9068,9 @@ pf_route(struct pf_krule *r, struct ifnet *oifp, goto bad; } + if (r->rt == PF_DUPTO) + skip_test = true; + if (pd->dir == PF_IN && !skip_test) { if (pf_test(AF_INET, PF_OUT, PFIL_FWD, ifp, &m0, inp, &pd->act) != PF_PASS) { @@ -9364,6 +9373,9 @@ pf_route6(struct pf_krule *r, struct ifnet *oifp, goto bad; } + if (r->rt == PF_DUPTO) + skip_test = true; + if (pd->dir == PF_IN && !skip_test) { if (pf_test(AF_INET6, PF_OUT, PFIL_FWD | PF_PFIL_NOREFRAGMENT, ifp, &m0, inp, &pd->act) != PF_PASS) { diff --git a/sys/netpfil/pf/pf.h b/sys/netpfil/pf/pf.h index 2009d2907985..cfff58064922 100644 --- a/sys/netpfil/pf/pf.h +++ b/sys/netpfil/pf/pf.h @@ -140,7 +140,7 @@ enum { PF_ADDR_ADDRMASK, PF_ADDR_NOROUTE, PF_ADDR_DYNIFTL, #define PF_LOG 0x01 #define PF_LOG_ALL 0x02 -#define PF_LOG_SOCKET_LOOKUP 0x04 +#define PF_LOG_USER 0x04 #define PF_LOG_FORCE 0x08 #define PF_LOG_MATCHES 0x10 @@ -490,6 +490,7 @@ struct pf_osfp_ioctl { #define PF_ANCHOR_NAME_SIZE 64 #define PF_ANCHOR_MAXPATH (MAXPATHLEN - PF_ANCHOR_NAME_SIZE - 1) +#define PF_OPTIMIZER_TABLE_PFX "__automatic_" struct pf_rule { struct pf_rule_addr src; diff --git a/sys/netpfil/pf/pf_ioctl.c b/sys/netpfil/pf/pf_ioctl.c index c96741023db9..3caa0d2e3b11 100644 --- a/sys/netpfil/pf/pf_ioctl.c +++ b/sys/netpfil/pf/pf_ioctl.c @@ -1274,7 +1274,9 @@ pf_hash_rule_addr(MD5_CTX *ctx, struct pf_rule_addr *pfr) PF_MD5_UPD(pfr, addr.iflags); break; case PF_ADDR_TABLE: - PF_MD5_UPD(pfr, addr.v.tblname); + if (strncmp(pfr->addr.v.tblname, PF_OPTIMIZER_TABLE_PFX, + strlen(PF_OPTIMIZER_TABLE_PFX))) + PF_MD5_UPD(pfr, addr.v.tblname); break; case PF_ADDR_ADDRMASK: /* XXX ignore af? */ @@ -1357,7 +1359,7 @@ static int pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) { struct pf_kruleset *rs; - struct pf_krule *rule, **old_array, *old_rule; + struct pf_krule *rule, *old_rule; struct pf_krulequeue *old_rules; struct pf_krule_global *old_tree; int error; @@ -1382,13 +1384,10 @@ pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) /* Swap rules, keep the old. */ old_rules = rs->rules[rs_num].active.ptr; old_rcount = rs->rules[rs_num].active.rcount; - old_array = rs->rules[rs_num].active.ptr_array; old_tree = rs->rules[rs_num].active.tree; rs->rules[rs_num].active.ptr = rs->rules[rs_num].inactive.ptr; - rs->rules[rs_num].active.ptr_array = - rs->rules[rs_num].inactive.ptr_array; rs->rules[rs_num].active.tree = rs->rules[rs_num].inactive.tree; rs->rules[rs_num].active.rcount = @@ -1418,7 +1417,6 @@ pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) } rs->rules[rs_num].inactive.ptr = old_rules; - rs->rules[rs_num].inactive.ptr_array = old_array; rs->rules[rs_num].inactive.tree = NULL; /* important for pf_ioctl_addrule */ rs->rules[rs_num].inactive.rcount = old_rcount; @@ -1431,9 +1429,6 @@ pf_commit_rules(u_int32_t ticket, int rs_num, char *anchor) while ((rule = TAILQ_FIRST(old_rules)) != NULL) pf_unlink_rule_locked(old_rules, rule); PF_UNLNKDRULES_UNLOCK(); - if (rs->rules[rs_num].inactive.ptr_array) - free(rs->rules[rs_num].inactive.ptr_array, M_TEMP); - rs->rules[rs_num].inactive.ptr_array = NULL; rs->rules[rs_num].inactive.rcount = 0; rs->rules[rs_num].inactive.open = 0; pf_remove_if_empty_kruleset(rs); @@ -1456,24 +1451,11 @@ pf_setup_pfsync_matching(struct pf_kruleset *rs) if (rs_cnt == PF_RULESET_SCRUB) continue; - if (rs->rules[rs_cnt].inactive.ptr_array) - free(rs->rules[rs_cnt].inactive.ptr_array, M_TEMP); - rs->rules[rs_cnt].inactive.ptr_array = NULL; - if (rs->rules[rs_cnt].inactive.rcount) { - rs->rules[rs_cnt].inactive.ptr_array = - mallocarray(rs->rules[rs_cnt].inactive.rcount, - sizeof(struct pf_rule **), - M_TEMP, M_NOWAIT); - - if (!rs->rules[rs_cnt].inactive.ptr_array) - return (ENOMEM); - } - - TAILQ_FOREACH(rule, rs->rules[rs_cnt].inactive.ptr, - entries) { - pf_hash_rule_rolling(&ctx, rule); - (rs->rules[rs_cnt].inactive.ptr_array)[rule->nr] = rule; + TAILQ_FOREACH(rule, rs->rules[rs_cnt].inactive.ptr, + entries) { + pf_hash_rule_rolling(&ctx, rule); + } } } @@ -2059,6 +2041,19 @@ pf_ioctl_getrules(struct pfioc_rule *pr) return (0); } +static int +pf_validate_range(uint8_t op, uint16_t port[2]) +{ + uint16_t a = ntohs(port[0]); + uint16_t b = ntohs(port[1]); + + if ((op == PF_OP_RRG && a > b) || /* 34:12, i.e. none */ + (op == PF_OP_IRG && a >= b) || /* 34><12, i.e. none */ + (op == PF_OP_XRG && a > b)) /* 34<>22, i.e. all */ + return 1; + return 0; +} + int pf_ioctl_addrule(struct pf_krule *rule, uint32_t ticket, uint32_t pool_ticket, const char *anchor, const char *anchor_call, @@ -2078,6 +2073,11 @@ pf_ioctl_addrule(struct pf_krule *rule, uint32_t ticket, #define ERROUT(x) ERROUT_FUNCTION(errout, x) + if (pf_validate_range(rule->src.port_op, rule->src.port)) + ERROUT(EINVAL); + if (pf_validate_range(rule->dst.port_op, rule->dst.port)) + ERROUT(EINVAL); + if (rule->ifname[0]) kif = pf_kkif_create(M_WAITOK); if (rule->rcv_ifname[0]) diff --git a/sys/netpfil/pf/pf_lb.c b/sys/netpfil/pf/pf_lb.c index 308d76c46e5b..26f7ab41eef4 100644 --- a/sys/netpfil/pf/pf_lb.c +++ b/sys/netpfil/pf/pf_lb.c @@ -80,7 +80,6 @@ static enum pf_test_status pf_step_into_translation_anchor(int, struct pf_test_c struct pf_krule *); static int pf_get_sport(struct pf_pdesc *, struct pf_krule *, struct pf_addr *, uint16_t *, uint16_t, uint16_t, - struct pf_ksrc_node **, struct pf_srchash **, struct pf_kpool *, struct pf_udp_mapping **, pf_sn_types_t); static bool pf_islinklocal(const sa_family_t, const struct pf_addr *); @@ -291,10 +290,8 @@ pf_match_translation(int rs_num, struct pf_test_ctx *ctx) } static int -pf_get_sport(struct pf_pdesc *pd, struct pf_krule *r, - struct pf_addr *naddr, uint16_t *nport, uint16_t low, - uint16_t high, struct pf_ksrc_node **sn, - struct pf_srchash **sh, struct pf_kpool *rpool, +pf_get_sport(struct pf_pdesc *pd, struct pf_krule *r, struct pf_addr *naddr, + uint16_t *nport, uint16_t low, uint16_t high, struct pf_kpool *rpool, struct pf_udp_mapping **udp_mapping, pf_sn_types_t sn_type) { struct pf_state_key_cmp key; @@ -322,19 +319,24 @@ pf_get_sport(struct pf_pdesc *pd, struct pf_krule *r, pf_addrcpy(&udp_source.addr, &pd->nsaddr, pd->af); udp_source.port = pd->nsport; if (udp_mapping) { + struct pf_ksrc_node *sn = NULL; + struct pf_srchash *sh = NULL; *udp_mapping = pf_udp_mapping_find(&udp_source); if (*udp_mapping) { pf_addrcpy(naddr, &(*udp_mapping)->endpoints[1].addr, pd->af); *nport = (*udp_mapping)->endpoints[1].port; - /* Try to find a src_node as per pf_map_addr(). */ - if (*sn == NULL && rpool->opts & PF_POOL_STICKYADDR && + /* + * Try to find a src_node as per pf_map_addr(). + * XXX: Why? This code seems to do nothing. + */ + if (rpool->opts & PF_POOL_STICKYADDR && (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) - *sn = pf_find_src_node(&pd->nsaddr, r, - pd->af, sh, sn_type, false); - if (*sn != NULL) - PF_SRC_NODE_UNLOCK(*sn); + sn = pf_find_src_node(&pd->nsaddr, r, + pd->af, &sh, sn_type, false); + if (sn != NULL) + PF_SRC_NODE_UNLOCK(sn); return (0); } else { *udp_mapping = pf_udp_mapping_create(pd->af, &pd->nsaddr, @@ -346,7 +348,7 @@ pf_get_sport(struct pf_pdesc *pd, struct pf_krule *r, } if (pf_map_addr_sn(pd->naf, r, &pd->nsaddr, naddr, NULL, &init_addr, - sn, sh, rpool, sn_type)) + rpool, sn_type)) goto failed; if (pd->proto == IPPROTO_ICMP) { @@ -470,9 +472,8 @@ pf_get_sport(struct pf_pdesc *pd, struct pf_krule *r, * pick a different source address since we're out * of free port choices for the current one. */ - (*sn) = NULL; if (pf_map_addr_sn(pd->naf, r, &pd->nsaddr, naddr, NULL, - &init_addr, sn, sh, rpool, sn_type)) + &init_addr, rpool, sn_type)) return (1); break; case PF_POOL_NONE: @@ -503,7 +504,6 @@ pf_islinklocal(const sa_family_t af, const struct pf_addr *addr) static int pf_get_mape_sport(struct pf_pdesc *pd, struct pf_krule *r, struct pf_addr *naddr, uint16_t *nport, - struct pf_ksrc_node **sn, struct pf_srchash **sh, struct pf_udp_mapping **udp_mapping, struct pf_kpool *rpool) { uint16_t psmask, low, highmask; @@ -523,16 +523,14 @@ pf_get_mape_sport(struct pf_pdesc *pd, struct pf_krule *r, for (i = cut; i <= ahigh; i++) { low = (i << ashift) | psmask; - if (!pf_get_sport(pd, r, - naddr, nport, low, low | highmask, sn, sh, rpool, - udp_mapping, PF_SN_NAT)) + if (!pf_get_sport(pd, r, naddr, nport, low, low | highmask, + rpool, udp_mapping, PF_SN_NAT)) return (0); } for (i = cut - 1; i > 0; i--) { low = (i << ashift) | psmask; - if (!pf_get_sport(pd, r, - naddr, nport, low, low | highmask, sn, sh, rpool, - udp_mapping, PF_SN_NAT)) + if (!pf_get_sport(pd, r, naddr, nport, low, low | highmask, + rpool, udp_mapping, PF_SN_NAT)) return (0); } return (1); @@ -545,6 +543,7 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, { u_short reason = PFRES_MATCH; struct pf_addr *raddr = NULL, *rmask = NULL; + struct pfr_ktable *kt; uint64_t hashidx; int cnt; @@ -600,29 +599,25 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, pf_poolmask(naddr, raddr, rmask, saddr, af); break; case PF_POOL_RANDOM: - if (rpool->cur->addr.type == PF_ADDR_TABLE) { - cnt = rpool->cur->addr.p.tbl->pfrkt_cnt; - if (cnt == 0) - rpool->tblidx = 0; + if (rpool->cur->addr.type == PF_ADDR_TABLE || + rpool->cur->addr.type == PF_ADDR_DYNIFTL) { + if (rpool->cur->addr.type == PF_ADDR_TABLE) + kt = rpool->cur->addr.p.tbl; else - rpool->tblidx = (int)arc4random_uniform(cnt); - memset(&rpool->counter, 0, sizeof(rpool->counter)); - if (pfr_pool_get(rpool->cur->addr.p.tbl, - &rpool->tblidx, &rpool->counter, af, NULL)) { + kt = rpool->cur->addr.p.dyn->pfid_kt; + kt = pfr_ktable_select_active(kt); + if (kt == NULL) { reason = PFRES_MAPFAILED; goto done_pool_mtx; /* unsupported */ } - pf_addrcpy(naddr, &rpool->counter, af); - } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) { - cnt = rpool->cur->addr.p.dyn->pfid_kt->pfrkt_cnt; + cnt = kt->pfrkt_cnt; if (cnt == 0) rpool->tblidx = 0; else rpool->tblidx = (int)arc4random_uniform(cnt); memset(&rpool->counter, 0, sizeof(rpool->counter)); - if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt, - &rpool->tblidx, &rpool->counter, af, - pf_islinklocal)) { + if (pfr_pool_get(kt, &rpool->tblidx, &rpool->counter, + af, pf_islinklocal, false)) { reason = PFRES_MAPFAILED; goto done_pool_mtx; /* unsupported */ } @@ -671,29 +666,25 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, hashidx = pf_hash(saddr, (struct pf_addr *)&hash, &rpool->key, af); - if (rpool->cur->addr.type == PF_ADDR_TABLE) { - cnt = rpool->cur->addr.p.tbl->pfrkt_cnt; - if (cnt == 0) - rpool->tblidx = 0; + if (rpool->cur->addr.type == PF_ADDR_TABLE || + rpool->cur->addr.type == PF_ADDR_DYNIFTL) { + if (rpool->cur->addr.type == PF_ADDR_TABLE) + kt = rpool->cur->addr.p.tbl; else - rpool->tblidx = (int)(hashidx % cnt); - memset(&rpool->counter, 0, sizeof(rpool->counter)); - if (pfr_pool_get(rpool->cur->addr.p.tbl, - &rpool->tblidx, &rpool->counter, af, NULL)) { + kt = rpool->cur->addr.p.dyn->pfid_kt; + kt = pfr_ktable_select_active(kt); + if (kt == NULL) { reason = PFRES_MAPFAILED; goto done_pool_mtx; /* unsupported */ } - pf_addrcpy(naddr, &rpool->counter, af); - } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) { - cnt = rpool->cur->addr.p.dyn->pfid_kt->pfrkt_cnt; + cnt = kt->pfrkt_cnt; if (cnt == 0) rpool->tblidx = 0; else rpool->tblidx = (int)(hashidx % cnt); memset(&rpool->counter, 0, sizeof(rpool->counter)); - if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt, - &rpool->tblidx, &rpool->counter, af, - pf_islinklocal)) { + if (pfr_pool_get(kt, &rpool->tblidx, &rpool->counter, + af, pf_islinklocal, false)) { reason = PFRES_MAPFAILED; goto done_pool_mtx; /* unsupported */ } @@ -710,11 +701,12 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, if (rpool->cur->addr.type == PF_ADDR_TABLE) { if (!pfr_pool_get(rpool->cur->addr.p.tbl, - &rpool->tblidx, &rpool->counter, af, NULL)) + &rpool->tblidx, &rpool->counter, af, NULL, true)) goto get_addr; } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) { if (!pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt, - &rpool->tblidx, &rpool->counter, af, pf_islinklocal)) + &rpool->tblidx, &rpool->counter, af, pf_islinklocal, + true)) goto get_addr; } else if (pf_match_addr(0, raddr, rmask, &rpool->counter, af)) goto get_addr; @@ -724,9 +716,10 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, rpool->cur = TAILQ_FIRST(&rpool->list); else rpool->cur = TAILQ_NEXT(rpool->cur, entries); + rpool->tblidx = -1; if (rpool->cur->addr.type == PF_ADDR_TABLE) { if (pfr_pool_get(rpool->cur->addr.p.tbl, - &rpool->tblidx, &rpool->counter, af, NULL)) { + &rpool->tblidx, &rpool->counter, af, NULL, true)) { /* table contains no address of type 'af' */ if (rpool->cur != acur) goto try_next; @@ -734,9 +727,9 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, goto done_pool_mtx; } } else if (rpool->cur->addr.type == PF_ADDR_DYNIFTL) { - rpool->tblidx = -1; if (pfr_pool_get(rpool->cur->addr.p.dyn->pfid_kt, - &rpool->tblidx, &rpool->counter, af, pf_islinklocal)) { + &rpool->tblidx, &rpool->counter, af, pf_islinklocal, + true)) { /* table contains no address of type 'af' */ if (rpool->cur != acur) goto try_next; @@ -764,48 +757,41 @@ pf_map_addr(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, done_pool_mtx: mtx_unlock(&rpool->mtx); - if (reason) { - counter_u64_add(V_pf_status.counters[reason], 1); - } - return (reason); } u_short pf_map_addr_sn(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, struct pf_addr *naddr, struct pfi_kkif **nkif, struct pf_addr *init_addr, - struct pf_ksrc_node **sn, struct pf_srchash **sh, struct pf_kpool *rpool, - pf_sn_types_t sn_type) + struct pf_kpool *rpool, pf_sn_types_t sn_type) { + struct pf_ksrc_node *sn = NULL; + struct pf_srchash *sh = NULL; u_short reason = 0; - KASSERT(*sn == NULL, ("*sn not NULL")); - /* * If this is a sticky-address rule, try to find an existing src_node. - * Request the sh to be unlocked if sn was not found, as we never - * insert a new sn when parsing the ruleset. */ if (rpool->opts & PF_POOL_STICKYADDR && (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) - *sn = pf_find_src_node(saddr, r, af, sh, sn_type, false); + sn = pf_find_src_node(saddr, r, af, &sh, sn_type, false); - if (*sn != NULL) { - PF_SRC_NODE_LOCK_ASSERT(*sn); + if (sn != NULL) { + PF_SRC_NODE_LOCK_ASSERT(sn); /* If the supplied address is the same as the current one we've * been asked before, so tell the caller that there's no other * address to be had. */ - if (PF_AEQ(naddr, &(*sn)->raddr, af)) { + if (PF_AEQ(naddr, &(sn->raddr), af)) { reason = PFRES_MAPFAILED; goto done; } - pf_addrcpy(naddr, &(*sn)->raddr, af); + pf_addrcpy(naddr, &(sn->raddr), af); if (nkif) - *nkif = (*sn)->rkif; + *nkif = sn->rkif; if (V_pf_status.debug >= PF_DEBUG_NOISY) { - printf("pf_map_addr: src tracking maps "); + printf("%s: src tracking maps ", __func__); pf_print_host(saddr, 0, af); printf(" to "); pf_print_host(naddr, 0, af); @@ -820,14 +806,16 @@ pf_map_addr_sn(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, * Source node has not been found. Find a new address and store it * in variables given by the caller. */ - if (pf_map_addr(af, r, saddr, naddr, nkif, init_addr, rpool) != 0) { - /* pf_map_addr() sets reason counters on its own */ + if ((reason = pf_map_addr(af, r, saddr, naddr, nkif, init_addr, + rpool)) != 0) { + if (V_pf_status.debug >= PF_DEBUG_MISC) + printf("%s: pf_map_addr has failed\n", __func__); goto done; } if (V_pf_status.debug >= PF_DEBUG_NOISY && (rpool->opts & PF_POOL_TYPEMASK) != PF_POOL_NONE) { - printf("pf_map_addr: selected address "); + printf("%s: selected address ", __func__); pf_print_host(naddr, 0, af); if (nkif) printf("@%s", (*nkif)->pfik_name); @@ -835,12 +823,8 @@ pf_map_addr_sn(sa_family_t af, struct pf_krule *r, struct pf_addr *saddr, } done: - if ((*sn) != NULL) - PF_SRC_NODE_UNLOCK(*sn); - - if (reason) { - counter_u64_add(V_pf_status.counters[reason], 1); - } + if (sn != NULL) + PF_SRC_NODE_UNLOCK(sn); return (reason); } @@ -890,8 +874,6 @@ pf_get_transaddr(struct pf_test_ctx *ctx, struct pf_krule *r, { struct pf_pdesc *pd = ctx->pd; struct pf_addr *naddr; - struct pf_ksrc_node *sn = NULL; - struct pf_srchash *sh = NULL; uint16_t *nportp; uint16_t low, high; u_short reason; @@ -919,8 +901,8 @@ pf_get_transaddr(struct pf_test_ctx *ctx, struct pf_krule *r, high = rpool->proxy_port[1]; } if (rpool->mape.offset > 0) { - if (pf_get_mape_sport(pd, r, naddr, nportp, &sn, - &sh, &ctx->udp_mapping, rpool)) { + if (pf_get_mape_sport(pd, r, naddr, nportp, + &ctx->udp_mapping, rpool)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: MAP-E port allocation (%u/%u/%u)" " failed\n", @@ -930,8 +912,8 @@ pf_get_transaddr(struct pf_test_ctx *ctx, struct pf_krule *r, reason = PFRES_MAPFAILED; goto notrans; } - } else if (pf_get_sport(pd, r, naddr, nportp, low, high, &sn, - &sh, rpool, &ctx->udp_mapping, PF_SN_NAT)) { + } else if (pf_get_sport(pd, r, naddr, nportp, low, high, + rpool, &ctx->udp_mapping, PF_SN_NAT)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: NAT proxy port allocation (%u-%u) failed\n", rpool->proxy_port[0], rpool->proxy_port[1])); @@ -1017,7 +999,7 @@ pf_get_transaddr(struct pf_test_ctx *ctx, struct pf_krule *r, uint16_t cut, low, high, nport; reason = pf_map_addr_sn(pd->af, r, &pd->nsaddr, naddr, NULL, - NULL, &sn, &sh, rpool, PF_SN_NAT); + NULL, rpool, PF_SN_NAT); if (reason != 0) goto notrans; if ((rpool->opts & PF_POOL_TYPEMASK) == PF_POOL_BITMASK) @@ -1134,8 +1116,6 @@ pf_get_transaddr_af(struct pf_krule *r, struct pf_pdesc *pd) struct pf_addr ndaddr, nsaddr, naddr; u_int16_t nport = 0; int prefixlen = 96; - struct pf_srchash *sh = NULL; - struct pf_ksrc_node *sns = NULL; bzero(&nsaddr, sizeof(nsaddr)); bzero(&ndaddr, sizeof(ndaddr)); @@ -1154,9 +1134,8 @@ pf_get_transaddr_af(struct pf_krule *r, struct pf_pdesc *pd) panic("pf_get_transaddr_af: no nat pool for source address"); /* get source address and port */ - if (pf_get_sport(pd, r, &nsaddr, &nport, - r->nat.proxy_port[0], r->nat.proxy_port[1], &sns, &sh, &r->nat, - NULL, PF_SN_NAT)) { + if (pf_get_sport(pd, r, &nsaddr, &nport, r->nat.proxy_port[0], + r->nat.proxy_port[1], &r->nat, NULL, PF_SN_NAT)) { DPFPRINTF(PF_DEBUG_MISC, ("pf: af-to NAT proxy port allocation (%u-%u) failed", r->nat.proxy_port[0], r->nat.proxy_port[1])); @@ -1182,7 +1161,7 @@ pf_get_transaddr_af(struct pf_krule *r, struct pf_pdesc *pd) /* get the destination address and port */ if (! TAILQ_EMPTY(&r->rdr.list)) { if (pf_map_addr_sn(pd->naf, r, &nsaddr, &naddr, NULL, NULL, - &sns, NULL, &r->rdr, PF_SN_NAT)) + &r->rdr, PF_SN_NAT)) return (-1); if (r->rdr.proxy_port[0]) pd->ndport = htons(r->rdr.proxy_port[0]); diff --git a/sys/netpfil/pf/pf_table.c b/sys/netpfil/pf/pf_table.c index 43e4366845a2..9c0151b7da2b 100644 --- a/sys/netpfil/pf/pf_table.c +++ b/sys/netpfil/pf/pf_table.c @@ -819,10 +819,10 @@ pfr_create_kentry(struct pfr_addr *ad, bool counters) static void pfr_destroy_kentries(struct pfr_kentryworkq *workq) { - struct pfr_kentry *p, *q; + struct pfr_kentry *p; - for (p = SLIST_FIRST(workq); p != NULL; p = q) { - q = SLIST_NEXT(p, pfrke_workq); + while ((p = SLIST_FIRST(workq)) != NULL) { + SLIST_REMOVE_HEAD(workq, pfrke_workq); pfr_destroy_kentry(p); } } @@ -1680,8 +1680,7 @@ pfr_ina_commit(struct pfr_table *trs, u_int32_t ticket, int *nadd, } if (!(flags & PFR_FLAG_DUMMY)) { - for (p = SLIST_FIRST(&workq); p != NULL; p = q) { - q = SLIST_NEXT(p, pfrkt_workq); + SLIST_FOREACH_SAFE(p, &workq, pfrkt_workq, q) { pfr_commit_ktable(p, tzero); } rs->topen = 0; @@ -1710,7 +1709,7 @@ pfr_commit_ktable(struct pfr_ktable *kt, time_t tzero) } else if (kt->pfrkt_flags & PFR_TFLAG_ACTIVE) { /* kt might contain addresses */ struct pfr_kentryworkq addrq, addq, changeq, delq, garbageq; - struct pfr_kentry *p, *q, *next; + struct pfr_kentry *p, *q; struct pfr_addr ad; pfr_enqueue_addrs(shadow, &addrq, NULL, 0); @@ -1720,7 +1719,8 @@ pfr_commit_ktable(struct pfr_ktable *kt, time_t tzero) SLIST_INIT(&delq); SLIST_INIT(&garbageq); pfr_clean_node_mask(shadow, &addrq); - SLIST_FOREACH_SAFE(p, &addrq, pfrke_workq, next) { + while ((p = SLIST_FIRST(&addrq)) != NULL) { + SLIST_REMOVE_HEAD(&addrq, pfrke_workq); pfr_copyout_addr(&ad, p); q = pfr_lookup_addr(kt, &ad, 1); if (q != NULL) { @@ -1864,8 +1864,7 @@ pfr_setflags_ktables(struct pfr_ktableworkq *workq) { struct pfr_ktable *p, *q; - for (p = SLIST_FIRST(workq); p; p = q) { - q = SLIST_NEXT(p, pfrkt_workq); + SLIST_FOREACH_SAFE(p, workq, pfrkt_workq, q) { pfr_setflags_ktable(p, p->pfrkt_nflags); } } @@ -2015,10 +2014,10 @@ pfr_create_ktable(struct pfr_table *tbl, time_t tzero, int attachruleset) static void pfr_destroy_ktables(struct pfr_ktableworkq *workq, int flushaddr) { - struct pfr_ktable *p, *q; + struct pfr_ktable *p; - for (p = SLIST_FIRST(workq); p; p = q) { - q = SLIST_NEXT(p, pfrkt_workq); + while ((p = SLIST_FIRST(workq)) != NULL) { + SLIST_REMOVE_HEAD(workq, pfrkt_workq); pfr_destroy_ktable(p, flushaddr); } } @@ -2074,17 +2073,16 @@ pfr_lookup_table(struct pfr_table *tbl) (struct pfr_ktable *)tbl)); } -int -pfr_match_addr(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af) +static struct pfr_kentry * +pfr_kentry_byaddr(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af, + int exact) { struct pfr_kentry *ke = NULL; - int match; PF_RULES_RASSERT(); - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) - kt = kt->pfrkt_root; - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) + kt = pfr_ktable_select_active(kt); + if (kt == NULL) return (0); switch (af) { @@ -2121,11 +2119,26 @@ pfr_match_addr(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af) default: unhandled_af(af); } + if (exact && ke && KENTRY_NETWORK(ke)) + ke = NULL; + + return (ke); +} + +int +pfr_match_addr(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af) +{ + struct pfr_kentry *ke = NULL; + int match; + + ke = pfr_kentry_byaddr(kt, a, af, 0); + match = (ke && !ke->pfrke_not); if (match) pfr_kstate_counter_add(&kt->pfrkt_match, 1); else pfr_kstate_counter_add(&kt->pfrkt_nomatch, 1); + return (match); } @@ -2135,9 +2148,8 @@ pfr_update_stats(struct pfr_ktable *kt, struct pf_addr *a, sa_family_t af, { struct pfr_kentry *ke = NULL; - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) - kt = kt->pfrkt_root; - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) + kt = pfr_ktable_select_active(kt); + if (kt == NULL) return; switch (af) { @@ -2281,7 +2293,7 @@ pfr_detach_table(struct pfr_ktable *kt) int pfr_pool_get(struct pfr_ktable *kt, int *pidx, struct pf_addr *counter, - sa_family_t af, pf_addr_filter_func_t filter) + sa_family_t af, pf_addr_filter_func_t filter, bool loop_once) { struct pf_addr *addr, cur, mask, umask_addr; union sockaddr_union uaddr, umask; @@ -2306,9 +2318,8 @@ pfr_pool_get(struct pfr_ktable *kt, int *pidx, struct pf_addr *counter, unhandled_af(af); } - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) - kt = kt->pfrkt_root; - if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) + kt = pfr_ktable_select_active(kt); + if (kt == NULL) return (-1); idx = *pidx; @@ -2327,7 +2338,7 @@ _next_block: ke = pfr_kentry_byidx(kt, idx, af); if (ke == NULL) { /* we don't have this idx, try looping */ - if (loop || (ke = pfr_kentry_byidx(kt, 0, af)) == NULL) { + if ((loop || loop_once) || (ke = pfr_kentry_byidx(kt, 0, af)) == NULL) { pfr_kstate_counter_add(&kt->pfrkt_nomatch, 1); return (1); } @@ -2455,3 +2466,14 @@ pfr_dynaddr_update(struct pfr_ktable *kt, struct pfi_dynaddr *dyn) unhandled_af(dyn->pfid_af); } } + +struct pfr_ktable * +pfr_ktable_select_active(struct pfr_ktable *kt) +{ + if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE) && kt->pfrkt_root != NULL) + kt = kt->pfrkt_root; + if (!(kt->pfrkt_flags & PFR_TFLAG_ACTIVE)) + return (NULL); + + return (kt); +} diff --git a/sys/powerpc/aim/mmu_oea.c b/sys/powerpc/aim/mmu_oea.c index 7746b668265d..ae17b3289593 100644 --- a/sys/powerpc/aim/mmu_oea.c +++ b/sys/powerpc/aim/mmu_oea.c @@ -1469,6 +1469,9 @@ moea_page_set_memattr(vm_page_t m, vm_memattr_t ma) pmap_t pmap; u_int lo; + if (m->md.mdpg_cache_attrs == ma) + return; + if ((m->oflags & VPO_UNMANAGED) != 0) { m->md.mdpg_cache_attrs = ma; return; diff --git a/sys/powerpc/aim/mmu_oea64.c b/sys/powerpc/aim/mmu_oea64.c index 79cea408bb5f..796b1719b8ba 100644 --- a/sys/powerpc/aim/mmu_oea64.c +++ b/sys/powerpc/aim/mmu_oea64.c @@ -2134,6 +2134,9 @@ moea64_page_set_memattr(vm_page_t m, vm_memattr_t ma) CTR3(KTR_PMAP, "%s: pa=%#jx, ma=%#x", __func__, (uintmax_t)VM_PAGE_TO_PHYS(m), ma); + if (m->md.mdpg_cache_attrs == ma) + return; + if ((m->oflags & VPO_UNMANAGED) != 0) { m->md.mdpg_cache_attrs = ma; return; diff --git a/sys/powerpc/aim/mmu_radix.c b/sys/powerpc/aim/mmu_radix.c index 45f7bef8bcc9..a12142fc2d7b 100644 --- a/sys/powerpc/aim/mmu_radix.c +++ b/sys/powerpc/aim/mmu_radix.c @@ -5937,6 +5937,10 @@ mmu_radix_page_set_memattr(vm_page_t m, vm_memattr_t ma) { CTR3(KTR_PMAP, "%s(%p, %#x)", __func__, m, ma); + + if (m->md.mdpg_cache_attrs == ma) + return; + m->md.mdpg_cache_attrs = ma; /* diff --git a/sys/powerpc/include/pcb.h b/sys/powerpc/include/pcb.h index 050ada6b0f64..0230cf78aba7 100644 --- a/sys/powerpc/include/pcb.h +++ b/sys/powerpc/include/pcb.h @@ -66,16 +66,8 @@ struct pcb { #define PCB_VECREGS 0x200 /* Process had Altivec registers initialized */ struct fpu { union { -#if _BYTE_ORDER == _BIG_ENDIAN - double fpr; - uint32_t vsr[4]; -#else uint32_t vsr[4]; - struct { - double padding; - double fpr; - }; -#endif + double fpr; } fpr[32]; double fpscr; /* FPSCR stored as double for easier access */ } pcb_fpu; /* Floating point processor */ diff --git a/sys/powerpc/include/ucontext.h b/sys/powerpc/include/ucontext.h index d35c6c773fe0..dc87edd578bc 100644 --- a/sys/powerpc/include/ucontext.h +++ b/sys/powerpc/include/ucontext.h @@ -41,6 +41,7 @@ typedef struct __mcontext { int mc_flags; #define _MC_FP_VALID 0x01 #define _MC_AV_VALID 0x02 +#define _MC_VS_VALID 0x04 int mc_onstack; /* saved onstack flag */ int mc_len; /* sizeof(__mcontext) */ __uint64_t mc_avec[32*2]; /* vector register file */ @@ -56,6 +57,7 @@ typedef struct __mcontext32 { int mc_flags; #define _MC_FP_VALID 0x01 #define _MC_AV_VALID 0x02 +#define _MC_VS_VALID 0x04 int mc_onstack; /* saved onstack flag */ int mc_len; /* sizeof(__mcontext) */ uint64_t mc_avec[32*2]; /* vector register file */ diff --git a/sys/powerpc/powerpc/exec_machdep.c b/sys/powerpc/powerpc/exec_machdep.c index 1893d79f29a8..8a33d0f589a7 100644 --- a/sys/powerpc/powerpc/exec_machdep.c +++ b/sys/powerpc/powerpc/exec_machdep.c @@ -214,10 +214,10 @@ sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask) sfpsize = sizeof(sf); #ifdef __powerpc64__ /* - * 64-bit PPC defines a 288 byte scratch region - * below the stack. + * 64-bit PPC defines a 512 byte red zone below + * the existing stack (ELF ABI v2 ยง2.2.2.4) */ - rndfsize = 288 + roundup(sizeof(sf), 48); + rndfsize = 512 + roundup(sizeof(sf), 48); #else rndfsize = roundup(sizeof(sf), 16); #endif @@ -349,13 +349,6 @@ sys_sigreturn(struct thread *td, struct sigreturn_args *uap) if (error != 0) return (error); - /* - * Save FPU state if needed. User may have changed it on - * signal handler - */ - if (uc.uc_mcontext.mc_srr1 & PSL_FP) - save_fpu(td); - kern_sigprocmask(td, SIG_SETMASK, &uc.uc_sigmask, NULL, 0); CTR3(KTR_SIG, "sigreturn: return td=%p pc=%#x sp=%#x", @@ -432,6 +425,7 @@ grab_mcontext(struct thread *td, mcontext_t *mcp, int flags) } if (pcb->pcb_flags & PCB_VSX) { + mcp->mc_flags |= _MC_VS_VALID; for (i = 0; i < 32; i++) memcpy(&mcp->mc_vsxfpreg[i], &pcb->pcb_fpu.fpr[i].vsr[2], sizeof(double)); @@ -481,6 +475,7 @@ set_mcontext(struct thread *td, mcontext_t *mcp) struct pcb *pcb; struct trapframe *tf; register_t tls; + register_t msr; int i; pcb = td->td_pcb; @@ -531,6 +526,22 @@ set_mcontext(struct thread *td, mcontext_t *mcp) tf->srr1 &= ~(PSL_FP | PSL_VSX | PSL_VEC); pcb->pcb_flags &= ~(PCB_FPU | PCB_VSX | PCB_VEC); + /* + * Ensure the FPU is also disabled in hardware. + * + * Without this, it's possible for the register reload to fail if we + * don't switch to a FPU disabled context before resuming the original + * thread. Specifically, if the FPU/VSX unavailable exception is never + * hit, then whatever data is still in the FP/VSX registers when + * sigresume is callled will used by the resumed thread, instead of the + * previously saved data from the mcontext. + */ + critical_enter(); + msr = mfmsr() & ~(PSL_FP | PSL_VSX | PSL_VEC); + isync(); + mtmsr(msr); + critical_exit(); + if (mcp->mc_flags & _MC_FP_VALID) { /* enable_fpu() will happen lazily on a fault */ pcb->pcb_flags |= PCB_FPREGS; @@ -539,8 +550,12 @@ set_mcontext(struct thread *td, mcontext_t *mcp) for (i = 0; i < 32; i++) { memcpy(&pcb->pcb_fpu.fpr[i].fpr, &mcp->mc_fpreg[i], sizeof(double)); - memcpy(&pcb->pcb_fpu.fpr[i].vsr[2], - &mcp->mc_vsxfpreg[i], sizeof(double)); + } + if (mcp->mc_flags & _MC_VS_VALID) { + for (i = 0; i < 32; i++) { + memcpy(&pcb->pcb_fpu.fpr[i].vsr[2], + &mcp->mc_vsxfpreg[i], sizeof(double)); + } } } diff --git a/sys/powerpc/powerpc/fpu.c b/sys/powerpc/powerpc/fpu.c index 0eaff2ea4932..cc8f22f7dda3 100644 --- a/sys/powerpc/powerpc/fpu.c +++ b/sys/powerpc/powerpc/fpu.c @@ -64,8 +64,19 @@ save_fpu_int(struct thread *td) * Save the floating-point registers and FPSCR to the PCB */ if (pcb->pcb_flags & PCB_VSX) { - #define SFP(n) __asm ("stxvw4x " #n ", 0,%0" \ +#if _BYTE_ORDER == _BIG_ENDIAN + #define SFP(n) __asm("stxvw4x " #n ", 0,%0" \ :: "b"(&pcb->pcb_fpu.fpr[n])); +#else + /* + * stxvw2x will swap words within the FP double word on LE systems, + * leading to corruption if VSX is used to store state and FP is + * subsequently used to restore state. + * Use stxvd2x instead. + */ + #define SFP(n) __asm("stxvd2x " #n ", 0,%0" \ + :: "b"(&pcb->pcb_fpu.fpr[n])); +#endif SFP(0); SFP(1); SFP(2); SFP(3); SFP(4); SFP(5); SFP(6); SFP(7); SFP(8); SFP(9); SFP(10); SFP(11); @@ -76,7 +87,7 @@ save_fpu_int(struct thread *td) SFP(28); SFP(29); SFP(30); SFP(31); #undef SFP } else { - #define SFP(n) __asm ("stfd " #n ", 0(%0)" \ + #define SFP(n) __asm("stfd " #n ", 0(%0)" \ :: "b"(&pcb->pcb_fpu.fpr[n].fpr)); SFP(0); SFP(1); SFP(2); SFP(3); SFP(4); SFP(5); SFP(6); SFP(7); @@ -149,8 +160,19 @@ enable_fpu(struct thread *td) :: "b"(&pcb->pcb_fpu.fpscr)); if (pcb->pcb_flags & PCB_VSX) { - #define LFP(n) __asm ("lxvw4x " #n ", 0,%0" \ +#if _BYTE_ORDER == _BIG_ENDIAN + #define LFP(n) __asm("lxvw4x " #n ", 0,%0" \ + :: "b"(&pcb->pcb_fpu.fpr[n])); +#else + /* + * lxvw4x will swap words within the FP double word on LE systems, + * leading to corruption if FP is used to store state and VSX is + * subsequently used to restore state. + * Use lxvd2x instead. + */ + #define LFP(n) __asm("lxvd2x " #n ", 0,%0" \ :: "b"(&pcb->pcb_fpu.fpr[n])); +#endif LFP(0); LFP(1); LFP(2); LFP(3); LFP(4); LFP(5); LFP(6); LFP(7); LFP(8); LFP(9); LFP(10); LFP(11); @@ -161,7 +183,7 @@ enable_fpu(struct thread *td) LFP(28); LFP(29); LFP(30); LFP(31); #undef LFP } else { - #define LFP(n) __asm ("lfd " #n ", 0(%0)" \ + #define LFP(n) __asm("lfd " #n ", 0(%0)" \ :: "b"(&pcb->pcb_fpu.fpr[n].fpr)); LFP(0); LFP(1); LFP(2); LFP(3); LFP(4); LFP(5); LFP(6); LFP(7); diff --git a/sys/riscv/riscv/pmap.c b/sys/riscv/riscv/pmap.c index 5d15bd671285..26efaecc64d1 100644 --- a/sys/riscv/riscv/pmap.c +++ b/sys/riscv/riscv/pmap.c @@ -4838,6 +4838,8 @@ pmap_unmapbios(void *p, vm_size_t size) void pmap_page_set_memattr(vm_page_t m, vm_memattr_t ma) { + if (m->md.pv_memattr == ma) + return; m->md.pv_memattr = ma; diff --git a/sys/vm/vm_domainset.c b/sys/vm/vm_domainset.c index 7b8bf4c77663..b44bdb96b0d4 100644 --- a/sys/vm/vm_domainset.c +++ b/sys/vm/vm_domainset.c @@ -131,8 +131,7 @@ static void vm_domainset_iter_next(struct vm_domainset_iter *di, int *domain) { - KASSERT(di->di_n > 0, - ("vm_domainset_iter_first: Invalid n %d", di->di_n)); + KASSERT(di->di_n > 0, ("%s: Invalid n %d", __func__, di->di_n)); switch (di->di_policy) { case DOMAINSET_POLICY_FIRSTTOUCH: /* @@ -149,11 +148,10 @@ vm_domainset_iter_next(struct vm_domainset_iter *di, int *domain) vm_domainset_iter_prefer(di, domain); break; default: - panic("vm_domainset_iter_first: Unknown policy %d", - di->di_policy); + panic("%s: Unknown policy %d", __func__, di->di_policy); } KASSERT(*domain < vm_ndomains, - ("vm_domainset_iter_next: Invalid domain %d", *domain)); + ("%s: Invalid domain %d", __func__, *domain)); } static void @@ -189,13 +187,11 @@ vm_domainset_iter_first(struct vm_domainset_iter *di, int *domain) di->di_n = di->di_domain->ds_cnt; break; default: - panic("vm_domainset_iter_first: Unknown policy %d", - di->di_policy); + panic("%s: Unknown policy %d", __func__, di->di_policy); } - KASSERT(di->di_n > 0, - ("vm_domainset_iter_first: Invalid n %d", di->di_n)); + KASSERT(di->di_n > 0, ("%s: Invalid n %d", __func__, di->di_n)); KASSERT(*domain < vm_ndomains, - ("vm_domainset_iter_first: Invalid domain %d", *domain)); + ("%s: Invalid domain %d", __func__, *domain)); } void diff --git a/sys/vm/vm_kern.c b/sys/vm/vm_kern.c index 875c22d27628..e7d7b6726d2c 100644 --- a/sys/vm/vm_kern.c +++ b/sys/vm/vm_kern.c @@ -110,11 +110,18 @@ u_int exec_map_entry_size; u_int exec_map_entries; SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD, - SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address"); +#if defined(__amd64__) + &kva_layout.km_low, 0, +#else + SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, +#endif + "Min kernel address"); SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD, #if defined(__arm__) &vm_max_kernel_address, 0, +#elif defined(__amd64__) + &kva_layout.km_high, 0, #else SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS, #endif |