David Howells | 642fb4d | 2006-01-06 00:11:41 -0800 | [diff] [blame] | 1 | /* file-nommu.c: no-MMU version of ramfs |
| 2 | * |
| 3 | * Copyright (C) 2005 Red Hat, Inc. All Rights Reserved. |
| 4 | * Written by David Howells (dhowells@redhat.com) |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/module.h> |
| 13 | #include <linux/fs.h> |
| 14 | #include <linux/pagemap.h> |
| 15 | #include <linux/highmem.h> |
| 16 | #include <linux/init.h> |
| 17 | #include <linux/string.h> |
| 18 | #include <linux/smp_lock.h> |
| 19 | #include <linux/backing-dev.h> |
| 20 | #include <linux/ramfs.h> |
| 21 | #include <linux/quotaops.h> |
| 22 | #include <linux/pagevec.h> |
| 23 | #include <linux/mman.h> |
| 24 | |
| 25 | #include <asm/uaccess.h> |
| 26 | #include "internal.h" |
| 27 | |
| 28 | static int ramfs_nommu_setattr(struct dentry *, struct iattr *); |
| 29 | |
| 30 | struct address_space_operations ramfs_aops = { |
| 31 | .readpage = simple_readpage, |
| 32 | .prepare_write = simple_prepare_write, |
| 33 | .commit_write = simple_commit_write |
| 34 | }; |
| 35 | |
| 36 | struct file_operations ramfs_file_operations = { |
| 37 | .mmap = ramfs_nommu_mmap, |
| 38 | .get_unmapped_area = ramfs_nommu_get_unmapped_area, |
| 39 | .read = generic_file_read, |
| 40 | .write = generic_file_write, |
| 41 | .fsync = simple_sync_file, |
| 42 | .sendfile = generic_file_sendfile, |
| 43 | .llseek = generic_file_llseek, |
| 44 | }; |
| 45 | |
| 46 | struct inode_operations ramfs_file_inode_operations = { |
| 47 | .setattr = ramfs_nommu_setattr, |
| 48 | .getattr = simple_getattr, |
| 49 | }; |
| 50 | |
| 51 | /*****************************************************************************/ |
| 52 | /* |
| 53 | * add a contiguous set of pages into a ramfs inode when it's truncated from |
| 54 | * size 0 on the assumption that it's going to be used for an mmap of shared |
| 55 | * memory |
| 56 | */ |
| 57 | static int ramfs_nommu_expand_for_mapping(struct inode *inode, size_t newsize) |
| 58 | { |
| 59 | struct pagevec lru_pvec; |
| 60 | unsigned long npages, xpages, loop, limit; |
| 61 | struct page *pages; |
| 62 | unsigned order; |
| 63 | void *data; |
| 64 | int ret; |
| 65 | |
| 66 | /* make various checks */ |
| 67 | order = get_order(newsize); |
| 68 | if (unlikely(order >= MAX_ORDER)) |
| 69 | goto too_big; |
| 70 | |
| 71 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; |
| 72 | if (limit != RLIM_INFINITY && newsize > limit) |
| 73 | goto fsize_exceeded; |
| 74 | |
| 75 | if (newsize > inode->i_sb->s_maxbytes) |
| 76 | goto too_big; |
| 77 | |
| 78 | i_size_write(inode, newsize); |
| 79 | |
| 80 | /* allocate enough contiguous pages to be able to satisfy the |
| 81 | * request */ |
| 82 | pages = alloc_pages(mapping_gfp_mask(inode->i_mapping), order); |
| 83 | if (!pages) |
| 84 | return -ENOMEM; |
| 85 | |
| 86 | /* split the high-order page into an array of single pages */ |
| 87 | xpages = 1UL << order; |
| 88 | npages = (newsize + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 89 | |
Nick Piggin | 8409751 | 2006-03-22 00:08:34 -0800 | [diff] [blame] | 90 | split_page(pages, order); |
David Howells | 642fb4d | 2006-01-06 00:11:41 -0800 | [diff] [blame] | 91 | |
| 92 | /* trim off any pages we don't actually require */ |
| 93 | for (loop = npages; loop < xpages; loop++) |
| 94 | __free_page(pages + loop); |
| 95 | |
| 96 | /* clear the memory we allocated */ |
| 97 | newsize = PAGE_SIZE * npages; |
| 98 | data = page_address(pages); |
| 99 | memset(data, 0, newsize); |
| 100 | |
| 101 | /* attach all the pages to the inode's address space */ |
| 102 | pagevec_init(&lru_pvec, 0); |
| 103 | for (loop = 0; loop < npages; loop++) { |
| 104 | struct page *page = pages + loop; |
| 105 | |
| 106 | ret = add_to_page_cache(page, inode->i_mapping, loop, GFP_KERNEL); |
| 107 | if (ret < 0) |
| 108 | goto add_error; |
| 109 | |
| 110 | if (!pagevec_add(&lru_pvec, page)) |
| 111 | __pagevec_lru_add(&lru_pvec); |
| 112 | |
| 113 | unlock_page(page); |
| 114 | } |
| 115 | |
| 116 | pagevec_lru_add(&lru_pvec); |
| 117 | return 0; |
| 118 | |
| 119 | fsize_exceeded: |
| 120 | send_sig(SIGXFSZ, current, 0); |
| 121 | too_big: |
| 122 | return -EFBIG; |
| 123 | |
| 124 | add_error: |
| 125 | page_cache_release(pages + loop); |
| 126 | for (loop++; loop < npages; loop++) |
| 127 | __free_page(pages + loop); |
| 128 | return ret; |
| 129 | } |
| 130 | |
| 131 | /*****************************************************************************/ |
| 132 | /* |
| 133 | * check that file shrinkage doesn't leave any VMAs dangling in midair |
| 134 | */ |
| 135 | static int ramfs_nommu_check_mappings(struct inode *inode, |
| 136 | size_t newsize, size_t size) |
| 137 | { |
| 138 | struct vm_area_struct *vma; |
| 139 | struct prio_tree_iter iter; |
| 140 | |
| 141 | /* search for VMAs that fall within the dead zone */ |
| 142 | vma_prio_tree_foreach(vma, &iter, &inode->i_mapping->i_mmap, |
| 143 | newsize >> PAGE_SHIFT, |
| 144 | (size + PAGE_SIZE - 1) >> PAGE_SHIFT |
| 145 | ) { |
| 146 | /* found one - only interested if it's shared out of the page |
| 147 | * cache */ |
| 148 | if (vma->vm_flags & VM_SHARED) |
| 149 | return -ETXTBSY; /* not quite true, but near enough */ |
| 150 | } |
| 151 | |
| 152 | return 0; |
| 153 | } |
| 154 | |
| 155 | /*****************************************************************************/ |
| 156 | /* |
| 157 | * |
| 158 | */ |
| 159 | static int ramfs_nommu_resize(struct inode *inode, loff_t newsize, loff_t size) |
| 160 | { |
| 161 | int ret; |
| 162 | |
| 163 | /* assume a truncate from zero size is going to be for the purposes of |
| 164 | * shared mmap */ |
| 165 | if (size == 0) { |
| 166 | if (unlikely(newsize >> 32)) |
| 167 | return -EFBIG; |
| 168 | |
| 169 | return ramfs_nommu_expand_for_mapping(inode, newsize); |
| 170 | } |
| 171 | |
| 172 | /* check that a decrease in size doesn't cut off any shared mappings */ |
| 173 | if (newsize < size) { |
| 174 | ret = ramfs_nommu_check_mappings(inode, newsize, size); |
| 175 | if (ret < 0) |
| 176 | return ret; |
| 177 | } |
| 178 | |
| 179 | ret = vmtruncate(inode, size); |
| 180 | |
| 181 | return ret; |
| 182 | } |
| 183 | |
| 184 | /*****************************************************************************/ |
| 185 | /* |
| 186 | * handle a change of attributes |
| 187 | * - we're specifically interested in a change of size |
| 188 | */ |
| 189 | static int ramfs_nommu_setattr(struct dentry *dentry, struct iattr *ia) |
| 190 | { |
| 191 | struct inode *inode = dentry->d_inode; |
| 192 | unsigned int old_ia_valid = ia->ia_valid; |
| 193 | int ret = 0; |
| 194 | |
| 195 | /* by providing our own setattr() method, we skip this quotaism */ |
| 196 | if ((old_ia_valid & ATTR_UID && ia->ia_uid != inode->i_uid) || |
| 197 | (old_ia_valid & ATTR_GID && ia->ia_gid != inode->i_gid)) |
| 198 | ret = DQUOT_TRANSFER(inode, ia) ? -EDQUOT : 0; |
| 199 | |
| 200 | /* pick out size-changing events */ |
| 201 | if (ia->ia_valid & ATTR_SIZE) { |
| 202 | loff_t size = i_size_read(inode); |
| 203 | if (ia->ia_size != size) { |
| 204 | ret = ramfs_nommu_resize(inode, ia->ia_size, size); |
| 205 | if (ret < 0 || ia->ia_valid == ATTR_SIZE) |
| 206 | goto out; |
| 207 | } else { |
| 208 | /* we skipped the truncate but must still update |
| 209 | * timestamps |
| 210 | */ |
| 211 | ia->ia_valid |= ATTR_MTIME|ATTR_CTIME; |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | ret = inode_setattr(inode, ia); |
| 216 | out: |
| 217 | ia->ia_valid = old_ia_valid; |
| 218 | return ret; |
| 219 | } |
| 220 | |
| 221 | /*****************************************************************************/ |
| 222 | /* |
| 223 | * try to determine where a shared mapping can be made |
| 224 | * - we require that: |
| 225 | * - the pages to be mapped must exist |
| 226 | * - the pages be physically contiguous in sequence |
| 227 | */ |
| 228 | unsigned long ramfs_nommu_get_unmapped_area(struct file *file, |
| 229 | unsigned long addr, unsigned long len, |
| 230 | unsigned long pgoff, unsigned long flags) |
| 231 | { |
| 232 | unsigned long maxpages, lpages, nr, loop, ret; |
| 233 | struct inode *inode = file->f_dentry->d_inode; |
| 234 | struct page **pages = NULL, **ptr, *page; |
| 235 | loff_t isize; |
| 236 | |
| 237 | if (!(flags & MAP_SHARED)) |
| 238 | return addr; |
| 239 | |
| 240 | /* the mapping mustn't extend beyond the EOF */ |
| 241 | lpages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 242 | isize = i_size_read(inode); |
| 243 | |
| 244 | ret = -EINVAL; |
| 245 | maxpages = (isize + PAGE_SIZE - 1) >> PAGE_SHIFT; |
| 246 | if (pgoff >= maxpages) |
| 247 | goto out; |
| 248 | |
| 249 | if (maxpages - pgoff < lpages) |
| 250 | goto out; |
| 251 | |
| 252 | /* gang-find the pages */ |
| 253 | ret = -ENOMEM; |
| 254 | pages = kzalloc(lpages * sizeof(struct page *), GFP_KERNEL); |
| 255 | if (!pages) |
| 256 | goto out; |
| 257 | |
| 258 | nr = find_get_pages(inode->i_mapping, pgoff, lpages, pages); |
| 259 | if (nr != lpages) |
| 260 | goto out; /* leave if some pages were missing */ |
| 261 | |
| 262 | /* check the pages for physical adjacency */ |
| 263 | ptr = pages; |
| 264 | page = *ptr++; |
| 265 | page++; |
| 266 | for (loop = lpages; loop > 1; loop--) |
| 267 | if (*ptr++ != page++) |
| 268 | goto out; |
| 269 | |
| 270 | /* okay - all conditions fulfilled */ |
| 271 | ret = (unsigned long) page_address(pages[0]); |
| 272 | |
| 273 | out: |
| 274 | if (pages) { |
| 275 | ptr = pages; |
| 276 | for (loop = lpages; loop > 0; loop--) |
| 277 | put_page(*ptr++); |
| 278 | kfree(pages); |
| 279 | } |
| 280 | |
| 281 | return ret; |
| 282 | } |
| 283 | |
| 284 | /*****************************************************************************/ |
| 285 | /* |
| 286 | * set up a mapping |
| 287 | */ |
| 288 | int ramfs_nommu_mmap(struct file *file, struct vm_area_struct *vma) |
| 289 | { |
| 290 | return 0; |
| 291 | } |