| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "mos/io/io.hpp" |
| 4 | #include "mos/mm/mm.hpp" |
| 5 | #include "mos/mm/paging/table_ops.hpp" |
| 6 | #include "mos/platform/platform.hpp" |
| 7 | #include "mos/syslog/printk.hpp" |
| 8 | |
| 9 | #include <mos/mm/cow.hpp> |
| 10 | #include <mos/mm/mm_types.h> |
| 11 | #include <mos/mm/mmap.hpp> |
| 12 | #include <mos/mm/paging/paging.hpp> |
| 13 | #include <mos/mos_global.h> |
| 14 | #include <mos/tasks/process.hpp> |
| 15 | #include <mos/tasks/task_types.hpp> |
| 16 | |
| 17 | /** |
| 18 | * @brief Check if the mmap flags are valid |
| 19 | * |
| 20 | * @param hint_addr The hint address |
| 21 | * @param mmap_flags The mmap flags |
| 22 | */ |
| 23 | static bool mmap_verify_arguments(ptr_t *hint_addr, MMapFlags mmap_flags) |
| 24 | { |
| 25 | if ((*hint_addr % MOS_PAGE_SIZE) != 0) |
| 26 | { |
| 27 | pr_warn("hint address must be page-aligned" ); |
| 28 | return false; |
| 29 | } |
| 30 | const bool shared = mmap_flags & MMAP_SHARED; // when forked, shared between parent and child |
| 31 | const bool map_private = mmap_flags & MMAP_PRIVATE; // when forked, make it Copy-On-Write |
| 32 | |
| 33 | if (shared == map_private) |
| 34 | { |
| 35 | pr_warn("mmap_file: shared and private are mutually exclusive, and one of them must be specified" ); |
| 36 | return false; |
| 37 | } |
| 38 | |
| 39 | if (mmap_flags & MMAP_EXACT) |
| 40 | { |
| 41 | // always use the hint address if MMAP_EXACT is specified |
| 42 | return true; |
| 43 | } |
| 44 | else |
| 45 | { |
| 46 | // if no hint address is specified, use the default |
| 47 | if (*hint_addr == 0) |
| 48 | *hint_addr = MOS_ADDR_USER_MMAP; |
| 49 | } |
| 50 | |
| 51 | return true; |
| 52 | } |
| 53 | |
| 54 | ptr_t mmap_anonymous(MMContext *ctx, ptr_t hint_addr, MMapFlags flags, VMFlags VMFlags, size_t n_pages) |
| 55 | { |
| 56 | if (!mmap_verify_arguments(hint_addr: &hint_addr, mmap_flags: flags)) |
| 57 | return 0; |
| 58 | |
| 59 | auto vmap = cow_allocate_zeroed_pages(handle: ctx, npages: n_pages, vaddr: hint_addr, flags: VMFlags, exact: flags & MMAP_EXACT); |
| 60 | |
| 61 | if (vmap.isErr()) |
| 62 | return vmap.getErr(); |
| 63 | |
| 64 | pr_dinfo2(vmm, "allocated %zd pages at " PTR_FMT, vmap->npages, vmap->vaddr); |
| 65 | |
| 66 | const vmap_type_t type = (flags & MMAP_SHARED) ? VMAP_TYPE_SHARED : VMAP_TYPE_PRIVATE; |
| 67 | vmap_finalise_init(vmap: vmap.get(), content: VMAP_MMAP, type); |
| 68 | return vmap->vaddr; |
| 69 | } |
| 70 | |
| 71 | ptr_t mmap_file(MMContext *ctx, ptr_t hint_addr, MMapFlags flags, VMFlags VMFlags, size_t n_pages, IO *io, off_t offset) |
| 72 | { |
| 73 | if (!mmap_verify_arguments(hint_addr: &hint_addr, mmap_flags: flags)) |
| 74 | return 0; |
| 75 | |
| 76 | if (offset % MOS_PAGE_SIZE != 0) |
| 77 | { |
| 78 | pr_warn("mmap_file: offset must be page-aligned" ); |
| 79 | return 0; |
| 80 | } |
| 81 | |
| 82 | const vmap_type_t type = (flags & MMAP_SHARED) ? VMAP_TYPE_SHARED : VMAP_TYPE_PRIVATE; |
| 83 | |
| 84 | mm_lock_context_pair(ctx1: ctx); |
| 85 | auto vmap = mm_get_free_vaddr_locked(mmctx: ctx, n_pages, base_vaddr: hint_addr, exact: flags & MMAP_EXACT); |
| 86 | mm_unlock_context_pair(ctx1: ctx); |
| 87 | |
| 88 | if (vmap.isErr()) |
| 89 | { |
| 90 | pr_warn("mmap_file: no free virtual address space" ); |
| 91 | return 0; |
| 92 | } |
| 93 | |
| 94 | vmap->vmflags = VMFlags; |
| 95 | vmap->type = type; |
| 96 | |
| 97 | if (!io->map(vmap: vmap.get(), offset)) |
| 98 | { |
| 99 | vmap_destroy(vmap: vmap.get()); |
| 100 | pr_warn("mmap_file: could not map the file: io_mmap() failed" ); |
| 101 | return 0; |
| 102 | } |
| 103 | |
| 104 | vmap_finalise_init(vmap: vmap.get(), content: VMAP_FILE, type); |
| 105 | return vmap->vaddr; |
| 106 | } |
| 107 | |
| 108 | bool munmap(ptr_t addr, size_t size) |
| 109 | { |
| 110 | spinlock_acquire(¤t_process->mm->mm_lock); |
| 111 | vmap_t *const whole_map = vmap_obtain(current_process->mm, vaddr: addr); |
| 112 | if (unlikely(!whole_map)) |
| 113 | { |
| 114 | spinlock_release(¤t_process->mm->mm_lock); |
| 115 | pr_warn("munmap: could not find the vmap" ); |
| 116 | return false; |
| 117 | } |
| 118 | |
| 119 | // will unmap all pages containing the range, even if they are not fully contained |
| 120 | const ptr_t range_start = ALIGN_DOWN_TO_PAGE(addr); |
| 121 | const ptr_t range_end = ALIGN_UP_TO_PAGE(addr + size); |
| 122 | |
| 123 | const size_t start_pgoff = (range_start - whole_map->vaddr) / MOS_PAGE_SIZE; |
| 124 | const size_t end_pgoff = (range_end - whole_map->vaddr) / MOS_PAGE_SIZE; |
| 125 | |
| 126 | vmap_t *const range_map = vmap_split_for_range(vmap: whole_map, rstart_pgoff: start_pgoff, rend_pgoff: end_pgoff); |
| 127 | if (unlikely(!range_map)) |
| 128 | { |
| 129 | pr_warn("munmap: could not split the vmap" ); |
| 130 | spinlock_release(¤t_process->mm->mm_lock); |
| 131 | spinlock_release(&whole_map->lock); |
| 132 | return false; |
| 133 | } |
| 134 | |
| 135 | vmap_destroy(vmap: range_map); |
| 136 | spinlock_release(¤t_process->mm->mm_lock); |
| 137 | spinlock_release(&whole_map->lock); |
| 138 | return true; |
| 139 | } |
| 140 | |
| 141 | bool vm_protect(MMContext *mmctx, ptr_t addr, size_t size, VMFlags perm) |
| 142 | { |
| 143 | MOS_ASSERT(addr % MOS_PAGE_SIZE == 0); |
| 144 | size = ALIGN_UP_TO_PAGE(size); |
| 145 | |
| 146 | spinlock_acquire(&mmctx->mm_lock); |
| 147 | vmap_t *const first_part = vmap_obtain(mmctx, vaddr: addr); |
| 148 | const size_t addr_pgoff = (addr - first_part->vaddr) / MOS_PAGE_SIZE; |
| 149 | |
| 150 | // |
| 151 | // first | second | third |
| 152 | // ^ ^ |
| 153 | // | | |
| 154 | // addr addr + size |
| 155 | // |
| 156 | |
| 157 | vmap_t *const to_protect = __extension__({ |
| 158 | vmap_t *vmap = first_part; |
| 159 | // if 'addr_pgoff' is 0, then the first part is the one we want to protect, |
| 160 | // otherwise we need to split it to get the vmap that starts at 'addr' |
| 161 | if (addr_pgoff) |
| 162 | { |
| 163 | vmap = vmap_split(vmap: first_part, split: addr_pgoff); |
| 164 | spinlock_release(&first_part->lock); // release the lock on the first part, we don't need it |
| 165 | } |
| 166 | |
| 167 | vmap; |
| 168 | }); |
| 169 | |
| 170 | const size_t size_pgoff = size / MOS_PAGE_SIZE; |
| 171 | if (size_pgoff < to_protect->npages) |
| 172 | { |
| 173 | // if there is a third part |
| 174 | vmap_t *const part3 = vmap_split(vmap: to_protect, split: size_pgoff); |
| 175 | spinlock_release(&part3->lock); // release the lock on the third part, we don't need it |
| 176 | } |
| 177 | |
| 178 | if (to_protect->io) |
| 179 | { |
| 180 | if (!to_protect->io->VerifyMMapPermissions(flags: perm, is_private: to_protect->type == VMAP_TYPE_PRIVATE)) |
| 181 | { |
| 182 | spinlock_release(&to_protect->lock); // permission denied |
| 183 | spinlock_release(&mmctx->mm_lock); |
| 184 | return false; |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | const bool read_lost = to_protect->vmflags.test(b: VM_READ) && !perm.test(b: VM_READ); // if we lose read permission |
| 189 | const bool write_lost = to_protect->vmflags.test(b: VM_WRITE) && !perm.test(b: VM_WRITE); // if we lose write permission |
| 190 | const bool exec_lost = to_protect->vmflags.test(b: VM_EXEC) && !perm.test(b: VM_EXEC); // if we lose exec permission |
| 191 | |
| 192 | VMFlags mask = VM_NONE; |
| 193 | if (read_lost) |
| 194 | { |
| 195 | mask |= VM_READ; |
| 196 | pr_warn("read permission lost, this is not supported yet" ); |
| 197 | } |
| 198 | |
| 199 | if (write_lost) |
| 200 | mask |= VM_WRITE; |
| 201 | if (exec_lost) |
| 202 | mask |= VM_EXEC; |
| 203 | |
| 204 | // remove permissions immediately |
| 205 | mm_do_mask_flags(max: mmctx->pgd, vaddr: to_protect->vaddr, n_pages: to_protect->npages, to_remove: mask); |
| 206 | |
| 207 | // do not add permissions immediately, we will let the page fault handler do it |
| 208 | // e.g. write permission granted only when the page is written to (and proper e.g. CoW) |
| 209 | // can be done. |
| 210 | |
| 211 | // let page fault handler do the real flags update |
| 212 | to_protect->vmflags = perm | VM_USER; |
| 213 | |
| 214 | spinlock_release(&to_protect->lock); |
| 215 | spinlock_release(&mmctx->mm_lock); |
| 216 | return true; |
| 217 | } |
| 218 | |