aboutsummaryrefslogtreecommitdiffstats
path: root/mm/shmem.c
diff options
context:
space:
mode:
Diffstat (limited to 'mm/shmem.c')
-rw-r--r--mm/shmem.c193
1 files changed, 58 insertions, 135 deletions
diff --git a/mm/shmem.c b/mm/shmem.c
index 4ce02e0673db..bd106361be4b 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -264,46 +264,55 @@ static int shmem_radix_tree_replace(struct address_space *mapping,
264} 264}
265 265
266/* 266/*
267 * Sometimes, before we decide whether to proceed or to fail, we must check
268 * that an entry was not already brought back from swap by a racing thread.
269 *
270 * Checking page is not enough: by the time a SwapCache page is locked, it
271 * might be reused, and again be SwapCache, using the same swap as before.
272 */
273static bool shmem_confirm_swap(struct address_space *mapping,
274 pgoff_t index, swp_entry_t swap)
275{
276 void *item;
277
278 rcu_read_lock();
279 item = radix_tree_lookup(&mapping->page_tree, index);
280 rcu_read_unlock();
281 return item == swp_to_radix_entry(swap);
282}
283
284/*
267 * Like add_to_page_cache_locked, but error if expected item has gone. 285 * Like add_to_page_cache_locked, but error if expected item has gone.
268 */ 286 */
269static int shmem_add_to_page_cache(struct page *page, 287static int shmem_add_to_page_cache(struct page *page,
270 struct address_space *mapping, 288 struct address_space *mapping,
271 pgoff_t index, gfp_t gfp, void *expected) 289 pgoff_t index, gfp_t gfp, void *expected)
272{ 290{
273 int error = 0; 291 int error;
274 292
275 VM_BUG_ON(!PageLocked(page)); 293 VM_BUG_ON(!PageLocked(page));
276 VM_BUG_ON(!PageSwapBacked(page)); 294 VM_BUG_ON(!PageSwapBacked(page));
277 295
296 page_cache_get(page);
297 page->mapping = mapping;
298 page->index = index;
299
300 spin_lock_irq(&mapping->tree_lock);
278 if (!expected) 301 if (!expected)
279 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); 302 error = radix_tree_insert(&mapping->page_tree, index, page);
303 else
304 error = shmem_radix_tree_replace(mapping, index, expected,
305 page);
280 if (!error) { 306 if (!error) {
281 page_cache_get(page); 307 mapping->nrpages++;
282 page->mapping = mapping; 308 __inc_zone_page_state(page, NR_FILE_PAGES);
283 page->index = index; 309 __inc_zone_page_state(page, NR_SHMEM);
284 310 spin_unlock_irq(&mapping->tree_lock);
285 spin_lock_irq(&mapping->tree_lock); 311 } else {
286 if (!expected) 312 page->mapping = NULL;
287 error = radix_tree_insert(&mapping->page_tree, 313 spin_unlock_irq(&mapping->tree_lock);
288 index, page); 314 page_cache_release(page);
289 else
290 error = shmem_radix_tree_replace(mapping, index,
291 expected, page);
292 if (!error) {
293 mapping->nrpages++;
294 __inc_zone_page_state(page, NR_FILE_PAGES);
295 __inc_zone_page_state(page, NR_SHMEM);
296 spin_unlock_irq(&mapping->tree_lock);
297 } else {
298 page->mapping = NULL;
299 spin_unlock_irq(&mapping->tree_lock);
300 page_cache_release(page);
301 }
302 if (!expected)
303 radix_tree_preload_end();
304 } 315 }
305 if (error)
306 mem_cgroup_uncharge_cache_page(page);
307 return error; 316 return error;
308} 317}
309 318
@@ -1124,9 +1133,9 @@ repeat:
1124 /* We have to do this with page locked to prevent races */ 1133 /* We have to do this with page locked to prevent races */
1125 lock_page(page); 1134 lock_page(page);
1126 if (!PageSwapCache(page) || page_private(page) != swap.val || 1135 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1127 page->mapping) { 1136 !shmem_confirm_swap(mapping, index, swap)) {
1128 error = -EEXIST; /* try again */ 1137 error = -EEXIST; /* try again */
1129 goto failed; 1138 goto unlock;
1130 } 1139 }
1131 if (!PageUptodate(page)) { 1140 if (!PageUptodate(page)) {
1132 error = -EIO; 1141 error = -EIO;
@@ -1142,9 +1151,12 @@ repeat:
1142 1151
1143 error = mem_cgroup_cache_charge(page, current->mm, 1152 error = mem_cgroup_cache_charge(page, current->mm,
1144 gfp & GFP_RECLAIM_MASK); 1153 gfp & GFP_RECLAIM_MASK);
1145 if (!error) 1154 if (!error) {
1146 error = shmem_add_to_page_cache(page, mapping, index, 1155 error = shmem_add_to_page_cache(page, mapping, index,
1147 gfp, swp_to_radix_entry(swap)); 1156 gfp, swp_to_radix_entry(swap));
1157 /* We already confirmed swap, and make no allocation */
1158 VM_BUG_ON(error);
1159 }
1148 if (error) 1160 if (error)
1149 goto failed; 1161 goto failed;
1150 1162
@@ -1181,11 +1193,18 @@ repeat:
1181 __set_page_locked(page); 1193 __set_page_locked(page);
1182 error = mem_cgroup_cache_charge(page, current->mm, 1194 error = mem_cgroup_cache_charge(page, current->mm,
1183 gfp & GFP_RECLAIM_MASK); 1195 gfp & GFP_RECLAIM_MASK);
1184 if (!error)
1185 error = shmem_add_to_page_cache(page, mapping, index,
1186 gfp, NULL);
1187 if (error) 1196 if (error)
1188 goto decused; 1197 goto decused;
1198 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1199 if (!error) {
1200 error = shmem_add_to_page_cache(page, mapping, index,
1201 gfp, NULL);
1202 radix_tree_preload_end();
1203 }
1204 if (error) {
1205 mem_cgroup_uncharge_cache_page(page);
1206 goto decused;
1207 }
1189 lru_cache_add_anon(page); 1208 lru_cache_add_anon(page);
1190 1209
1191 spin_lock(&info->lock); 1210 spin_lock(&info->lock);
@@ -1245,14 +1264,10 @@ decused:
1245unacct: 1264unacct:
1246 shmem_unacct_blocks(info->flags, 1); 1265 shmem_unacct_blocks(info->flags, 1);
1247failed: 1266failed:
1248 if (swap.val && error != -EINVAL) { 1267 if (swap.val && error != -EINVAL &&
1249 struct page *test = find_get_page(mapping, index); 1268 !shmem_confirm_swap(mapping, index, swap))
1250 if (test && !radix_tree_exceptional_entry(test)) 1269 error = -EEXIST;
1251 page_cache_release(test); 1270unlock:
1252 /* Have another try if the entry has changed */
1253 if (test != swp_to_radix_entry(swap))
1254 error = -EEXIST;
1255 }
1256 if (page) { 1271 if (page) {
1257 unlock_page(page); 1272 unlock_page(page);
1258 page_cache_release(page); 1273 page_cache_release(page);
@@ -1264,7 +1279,7 @@ failed:
1264 spin_unlock(&info->lock); 1279 spin_unlock(&info->lock);
1265 goto repeat; 1280 goto repeat;
1266 } 1281 }
1267 if (error == -EEXIST) 1282 if (error == -EEXIST) /* from above or from radix_tree_insert */
1268 goto repeat; 1283 goto repeat;
1269 return error; 1284 return error;
1270} 1285}
@@ -1692,98 +1707,6 @@ static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1692 return error; 1707 return error;
1693} 1708}
1694 1709
1695/*
1696 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1697 */
1698static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1699 pgoff_t index, pgoff_t end, int origin)
1700{
1701 struct page *page;
1702 struct pagevec pvec;
1703 pgoff_t indices[PAGEVEC_SIZE];
1704 bool done = false;
1705 int i;
1706
1707 pagevec_init(&pvec, 0);
1708 pvec.nr = 1; /* start small: we may be there already */
1709 while (!done) {
1710 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1711 pvec.nr, pvec.pages, indices);
1712 if (!pvec.nr) {
1713 if (origin == SEEK_DATA)
1714 index = end;
1715 break;
1716 }
1717 for (i = 0; i < pvec.nr; i++, index++) {
1718 if (index < indices[i]) {
1719 if (origin == SEEK_HOLE) {
1720 done = true;
1721 break;
1722 }
1723 index = indices[i];
1724 }
1725 page = pvec.pages[i];
1726 if (page && !radix_tree_exceptional_entry(page)) {
1727 if (!PageUptodate(page))
1728 page = NULL;
1729 }
1730 if (index >= end ||
1731 (page && origin == SEEK_DATA) ||
1732 (!page && origin == SEEK_HOLE)) {
1733 done = true;
1734 break;
1735 }
1736 }
1737 shmem_deswap_pagevec(&pvec);
1738 pagevec_release(&pvec);
1739 pvec.nr = PAGEVEC_SIZE;
1740 cond_resched();
1741 }
1742 return index;
1743}
1744
1745static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin)
1746{
1747 struct address_space *mapping;
1748 struct inode *inode;
1749 pgoff_t start, end;
1750 loff_t new_offset;
1751
1752 if (origin != SEEK_DATA && origin != SEEK_HOLE)
1753 return generic_file_llseek_size(file, offset, origin,
1754 MAX_LFS_FILESIZE);
1755 mapping = file->f_mapping;
1756 inode = mapping->host;
1757 mutex_lock(&inode->i_mutex);
1758 /* We're holding i_mutex so we can access i_size directly */
1759
1760 if (offset < 0)
1761 offset = -EINVAL;
1762 else if (offset >= inode->i_size)
1763 offset = -ENXIO;
1764 else {
1765 start = offset >> PAGE_CACHE_SHIFT;
1766 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1767 new_offset = shmem_seek_hole_data(mapping, start, end, origin);
1768 new_offset <<= PAGE_CACHE_SHIFT;
1769 if (new_offset > offset) {
1770 if (new_offset < inode->i_size)
1771 offset = new_offset;
1772 else if (origin == SEEK_DATA)
1773 offset = -ENXIO;
1774 else
1775 offset = inode->i_size;
1776 }
1777 }
1778
1779 if (offset >= 0 && offset != file->f_pos) {
1780 file->f_pos = offset;
1781 file->f_version = 0;
1782 }
1783 mutex_unlock(&inode->i_mutex);
1784 return offset;
1785}
1786
1787static long shmem_fallocate(struct file *file, int mode, loff_t offset, 1710static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1788 loff_t len) 1711 loff_t len)
1789{ 1712{
@@ -2787,7 +2710,7 @@ static const struct address_space_operations shmem_aops = {
2787static const struct file_operations shmem_file_operations = { 2710static const struct file_operations shmem_file_operations = {
2788 .mmap = shmem_mmap, 2711 .mmap = shmem_mmap,
2789#ifdef CONFIG_TMPFS 2712#ifdef CONFIG_TMPFS
2790 .llseek = shmem_file_llseek, 2713 .llseek = generic_file_llseek,
2791 .read = do_sync_read, 2714 .read = do_sync_read,
2792 .write = do_sync_write, 2715 .write = do_sync_write,
2793 .aio_read = shmem_file_aio_read, 2716 .aio_read = shmem_file_aio_read,