Merge patch series "riscv: Fix set_memory_XX() and set_direct_map_XX()"

Alexandre Ghiti <alexghiti@rivosinc.com> says:

Those 2 patches fix the set_memory_XX() and set_direct_map_XX() APIs, which
in turn fix STRICT_KERNEL_RWX and memfd_secret(). Those were broken since the
permission changes were not applied to the linear mapping because the linear
mapping is mapped using hugepages and walk_page_range_novma() does not split
such mappings.

To fix that, patch 1 disables PGD mappings in the linear mapping as it is
hard to propagate changes at this level in *all* the page tables, this has the
downside of disabling PMD mapping for sv32 and PUD (1GB) mapping for sv39 in
the linear mapping (for specific kernels, we could add a Kconfig to enable
ARCH_HAS_SET_DIRECT_MAP and STRICT_KERNEL_RWX if needed, I'm pretty sure we'll
discuss that).

patch 2 implements the split of the huge linear mappings so that
walk_page_range_novma() can properly apply the permissions. The whole split is
protected with mmap_sem in write mode, but I'm wondering if that's enough,
any opinion on that is appreciated.

* b4-shazam-merge:
  riscv: Fix set_memory_XX() and set_direct_map_XX() by splitting huge linear mappings
  riscv: Don't use PGD entries for the linear mapping

Link: https://lore.kernel.org/r/20231108075930.7157-1-alexghiti@rivosinc.comSigned-off-by: Palmer Dabbelt <palmer@rivosinc.com>

arch/riscv/mm/init.c

@@ -666,16 +666,16 @@ void __init create_pgd_mapping(pgd_t *pgdp,
 static uintptr_t __init best_map_size(phys_addr_t pa, uintptr_t va,
 				      phys_addr_t size)
 {
-	if (!(pa & (PGDIR_SIZE - 1)) && !(va & (PGDIR_SIZE - 1)) && size >= PGDIR_SIZE)
-		return PGDIR_SIZE;
-
-	if (!(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
+	if (pgtable_l5_enabled &&
+	    !(pa & (P4D_SIZE - 1)) && !(va & (P4D_SIZE - 1)) && size >= P4D_SIZE)
 		return P4D_SIZE;
 
-	if (!(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
+	if (pgtable_l4_enabled &&
+	    !(pa & (PUD_SIZE - 1)) && !(va & (PUD_SIZE - 1)) && size >= PUD_SIZE)
 		return PUD_SIZE;
 
-	if (!(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
+	if (IS_ENABLED(CONFIG_64BIT) &&
+	    !(pa & (PMD_SIZE - 1)) && !(va & (PMD_SIZE - 1)) && size >= PMD_SIZE)
 		return PMD_SIZE;
 
 	return PAGE_SIZE;

arch/riscv/mm/pageattr.c

@@ -5,6 +5,7 @@
 
 #include <linux/pagewalk.h>
 #include <linux/pgtable.h>
+#include <linux/vmalloc.h>
 #include <asm/tlbflush.h>
 #include <asm/bitops.h>
 #include <asm/set_memory.h>
@@ -25,19 +26,6 @@ static unsigned long set_pageattr_masks(unsigned long val, struct mm_walk *walk)
 	return new_val;
 }
 
-static int pageattr_pgd_entry(pgd_t *pgd, unsigned long addr,
-			      unsigned long next, struct mm_walk *walk)
-{
-	pgd_t val = READ_ONCE(*pgd);
-
-	if (pgd_leaf(val)) {
-		val = __pgd(set_pageattr_masks(pgd_val(val), walk));
-		set_pgd(pgd, val);
-	}
-
-	return 0;
-}
-
 static int pageattr_p4d_entry(p4d_t *p4d, unsigned long addr,
 			      unsigned long next, struct mm_walk *walk)
 {
@@ -96,7 +84,6 @@ static int pageattr_pte_hole(unsigned long addr, unsigned long next,
 }
 
 static const struct mm_walk_ops pageattr_ops = {
-	.pgd_entry = pageattr_pgd_entry,
 	.p4d_entry = pageattr_p4d_entry,
 	.pud_entry = pageattr_pud_entry,
 	.pmd_entry = pageattr_pmd_entry,
@@ -105,12 +92,181 @@ static const struct mm_walk_ops pageattr_ops = {
 	.walk_lock = PGWALK_RDLOCK,
 };
 
+#ifdef CONFIG_64BIT
+static int __split_linear_mapping_pmd(pud_t *pudp,
+				      unsigned long vaddr, unsigned long end)
+{
+	pmd_t *pmdp;
+	unsigned long next;
+
+	pmdp = pmd_offset(pudp, vaddr);
+
+	do {
+		next = pmd_addr_end(vaddr, end);
+
+		if (next - vaddr >= PMD_SIZE &&
+		    vaddr <= (vaddr & PMD_MASK) && end >= next)
+			continue;
+
+		if (pmd_leaf(*pmdp)) {
+			struct page *pte_page;
+			unsigned long pfn = _pmd_pfn(*pmdp);
+			pgprot_t prot = __pgprot(pmd_val(*pmdp) & ~_PAGE_PFN_MASK);
+			pte_t *ptep_new;
+			int i;
+
+			pte_page = alloc_page(GFP_KERNEL);
+			if (!pte_page)
+				return -ENOMEM;
+
+			ptep_new = (pte_t *)page_address(pte_page);
+			for (i = 0; i < PTRS_PER_PTE; ++i, ++ptep_new)
+				set_pte(ptep_new, pfn_pte(pfn + i, prot));
+
+			smp_wmb();
+
+			set_pmd(pmdp, pfn_pmd(page_to_pfn(pte_page), PAGE_TABLE));
+		}
+	} while (pmdp++, vaddr = next, vaddr != end);
+
+	return 0;
+}
+
+static int __split_linear_mapping_pud(p4d_t *p4dp,
+				      unsigned long vaddr, unsigned long end)
+{
+	pud_t *pudp;
+	unsigned long next;
+	int ret;
+
+	pudp = pud_offset(p4dp, vaddr);
+
+	do {
+		next = pud_addr_end(vaddr, end);
+
+		if (next - vaddr >= PUD_SIZE &&
+		    vaddr <= (vaddr & PUD_MASK) && end >= next)
+			continue;
+
+		if (pud_leaf(*pudp)) {
+			struct page *pmd_page;
+			unsigned long pfn = _pud_pfn(*pudp);
+			pgprot_t prot = __pgprot(pud_val(*pudp) & ~_PAGE_PFN_MASK);
+			pmd_t *pmdp_new;
+			int i;
+
+			pmd_page = alloc_page(GFP_KERNEL);
+			if (!pmd_page)
+				return -ENOMEM;
+
+			pmdp_new = (pmd_t *)page_address(pmd_page);
+			for (i = 0; i < PTRS_PER_PMD; ++i, ++pmdp_new)
+				set_pmd(pmdp_new,
+					pfn_pmd(pfn + ((i * PMD_SIZE) >> PAGE_SHIFT), prot));
+
+			smp_wmb();
+
+			set_pud(pudp, pfn_pud(page_to_pfn(pmd_page), PAGE_TABLE));
+		}
+
+		ret = __split_linear_mapping_pmd(pudp, vaddr, next);
+		if (ret)
+			return ret;
+	} while (pudp++, vaddr = next, vaddr != end);
+
+	return 0;
+}
+
+static int __split_linear_mapping_p4d(pgd_t *pgdp,
+				      unsigned long vaddr, unsigned long end)
+{
+	p4d_t *p4dp;
+	unsigned long next;
+	int ret;
+
+	p4dp = p4d_offset(pgdp, vaddr);
+
+	do {
+		next = p4d_addr_end(vaddr, end);
+
+		/*
+		 * If [vaddr; end] contains [vaddr & P4D_MASK; next], we don't
+		 * need to split, we'll change the protections on the whole P4D.
+		 */
+		if (next - vaddr >= P4D_SIZE &&
+		    vaddr <= (vaddr & P4D_MASK) && end >= next)
+			continue;
+
+		if (p4d_leaf(*p4dp)) {
+			struct page *pud_page;
+			unsigned long pfn = _p4d_pfn(*p4dp);
+			pgprot_t prot = __pgprot(p4d_val(*p4dp) & ~_PAGE_PFN_MASK);
+			pud_t *pudp_new;
+			int i;
+
+			pud_page = alloc_page(GFP_KERNEL);
+			if (!pud_page)
+				return -ENOMEM;
+
+			/*
+			 * Fill the pud level with leaf puds that have the same
+			 * protections as the leaf p4d.
+			 */
+			pudp_new = (pud_t *)page_address(pud_page);
+			for (i = 0; i < PTRS_PER_PUD; ++i, ++pudp_new)
+				set_pud(pudp_new,
+					pfn_pud(pfn + ((i * PUD_SIZE) >> PAGE_SHIFT), prot));
+
+			/*
+			 * Make sure the pud filling is not reordered with the
+			 * p4d store which could result in seeing a partially
+			 * filled pud level.
+			 */
+			smp_wmb();
+
+			set_p4d(p4dp, pfn_p4d(page_to_pfn(pud_page), PAGE_TABLE));
+		}
+
+		ret = __split_linear_mapping_pud(p4dp, vaddr, next);
+		if (ret)
+			return ret;
+	} while (p4dp++, vaddr = next, vaddr != end);
+
+	return 0;
+}
+
+static int __split_linear_mapping_pgd(pgd_t *pgdp,
+				      unsigned long vaddr,
+				      unsigned long end)
+{
+	unsigned long next;
+	int ret;
+
+	do {
+		next = pgd_addr_end(vaddr, end);
+		/* We never use PGD mappings for the linear mapping */
+		ret = __split_linear_mapping_p4d(pgdp, vaddr, next);
+		if (ret)
+			return ret;
+	} while (pgdp++, vaddr = next, vaddr != end);
+
+	return 0;
+}
+
+static int split_linear_mapping(unsigned long start, unsigned long end)
+{
+	return __split_linear_mapping_pgd(pgd_offset_k(start), start, end);
+}
+#endif	/* CONFIG_64BIT */
+
 static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask,
 			pgprot_t clear_mask)
 {
 	int ret;
 	unsigned long start = addr;
 	unsigned long end = start + PAGE_SIZE * numpages;
+	unsigned long __maybe_unused lm_start;
+	unsigned long __maybe_unused lm_end;
 	struct pageattr_masks masks = {
 		.set_mask = set_mask,
 		.clear_mask = clear_mask
@@ -120,11 +276,67 @@ static int __set_memory(unsigned long addr, int numpages, pgprot_t set_mask,
 		return 0;
 
 	mmap_write_lock(&init_mm);
+
+#ifdef CONFIG_64BIT
+	/*
+	 * We are about to change the permissions of a kernel mapping, we must
+	 * apply the same changes to its linear mapping alias, which may imply
+	 * splitting a huge mapping.
+	 */
+
+	if (is_vmalloc_or_module_addr((void *)start)) {
+		struct vm_struct *area = NULL;
+		int i, page_start;
+
+		area = find_vm_area((void *)start);
+		page_start = (start - (unsigned long)area->addr) >> PAGE_SHIFT;
+
+		for (i = page_start; i < page_start + numpages; ++i) {
+			lm_start = (unsigned long)page_address(area->pages[i]);
+			lm_end = lm_start + PAGE_SIZE;
+
+			ret = split_linear_mapping(lm_start, lm_end);
+			if (ret)
+				goto unlock;
+
+			ret = walk_page_range_novma(&init_mm, lm_start, lm_end,
+						    &pageattr_ops, NULL, &masks);
+			if (ret)
+				goto unlock;
+		}
+	} else if (is_kernel_mapping(start) || is_linear_mapping(start)) {
+		lm_start = (unsigned long)lm_alias(start);
+		lm_end = (unsigned long)lm_alias(end);
+
+		ret = split_linear_mapping(lm_start, lm_end);
+		if (ret)
+			goto unlock;
+
+		ret = walk_page_range_novma(&init_mm, lm_start, lm_end,
+					    &pageattr_ops, NULL, &masks);
+		if (ret)
+			goto unlock;
+	}
+
+	ret =  walk_page_range_novma(&init_mm, start, end, &pageattr_ops, NULL,
+				     &masks);
+
+unlock:
+	mmap_write_unlock(&init_mm);
+
+	/*
+	 * We can't use flush_tlb_kernel_range() here as we may have split a
+	 * hugepage that is larger than that, so let's flush everything.
+	 */
+	flush_tlb_all();
+#else
 	ret =  walk_page_range_novma(&init_mm, start, end, &pageattr_ops, NULL,
 				     &masks);
+
 	mmap_write_unlock(&init_mm);
 
 	flush_tlb_kernel_range(start, end);
+#endif
 
 	return ret;
 }
@@ -159,36 +371,14 @@ int set_memory_nx(unsigned long addr, int numpages)
 
 int set_direct_map_invalid_noflush(struct page *page)
 {
-	int ret;
-	unsigned long start = (unsigned long)page_address(page);
-	unsigned long end = start + PAGE_SIZE;
-	struct pageattr_masks masks = {
-		.set_mask = __pgprot(0),
-		.clear_mask = __pgprot(_PAGE_PRESENT)
-	};
-
-	mmap_read_lock(&init_mm);
-	ret = walk_page_range(&init_mm, start, end, &pageattr_ops, &masks);
-	mmap_read_unlock(&init_mm);
-
-	return ret;
+	return __set_memory((unsigned long)page_address(page), 1,
+			    __pgprot(0), __pgprot(_PAGE_PRESENT));
 }
 
 int set_direct_map_default_noflush(struct page *page)
 {
-	int ret;
-	unsigned long start = (unsigned long)page_address(page);
-	unsigned long end = start + PAGE_SIZE;
-	struct pageattr_masks masks = {
-		.set_mask = PAGE_KERNEL,
-		.clear_mask = __pgprot(0)
-	};
-
-	mmap_read_lock(&init_mm);
-	ret = walk_page_range(&init_mm, start, end, &pageattr_ops, &masks);
-	mmap_read_unlock(&init_mm);
-
-	return ret;
+	return __set_memory((unsigned long)page_address(page), 1,
+			    PAGE_KERNEL, __pgprot(0));
 }
 
 #ifdef CONFIG_DEBUG_PAGEALLOC