Chris Mason | 6cbd557 | 2007-06-12 09:07:21 -0400 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public |
| 6 | * License v2 as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, |
| 9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 11 | * General Public License for more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public |
| 14 | * License along with this program; if not, write to the |
| 15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 16 | * Boston, MA 021110-1307, USA. |
| 17 | */ |
| 18 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 19 | #include <linux/fs.h> |
| 20 | #include <linux/pagemap.h> |
| 21 | #include <linux/highmem.h> |
| 22 | #include <linux/time.h> |
| 23 | #include <linux/init.h> |
| 24 | #include <linux/string.h> |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 25 | #include <linux/backing-dev.h> |
| 26 | #include <linux/mpage.h> |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 27 | #include <linux/falloc.h> |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 28 | #include <linux/swap.h> |
| 29 | #include <linux/writeback.h> |
| 30 | #include <linux/statfs.h> |
| 31 | #include <linux/compat.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 32 | #include <linux/slab.h> |
Filipe Brandenburger | 55e301f | 2013-01-29 06:04:50 +0000 | [diff] [blame] | 33 | #include <linux/btrfs.h> |
Christoph Hellwig | e2e40f2 | 2015-02-22 08:58:50 -0800 | [diff] [blame] | 34 | #include <linux/uio.h> |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 35 | #include "ctree.h" |
| 36 | #include "disk-io.h" |
| 37 | #include "transaction.h" |
| 38 | #include "btrfs_inode.h" |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 39 | #include "print-tree.h" |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 40 | #include "tree-log.h" |
| 41 | #include "locking.h" |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 42 | #include "volumes.h" |
Josef Bacik | fcebe45 | 2014-05-13 17:30:47 -0700 | [diff] [blame] | 43 | #include "qgroup.h" |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 44 | |
Miao Xie | 9247f31 | 2012-11-26 09:24:43 +0000 | [diff] [blame] | 45 | static struct kmem_cache *btrfs_inode_defrag_cachep; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 46 | /* |
| 47 | * when auto defrag is enabled we |
| 48 | * queue up these defrag structs to remember which |
| 49 | * inodes need defragging passes |
| 50 | */ |
| 51 | struct inode_defrag { |
| 52 | struct rb_node rb_node; |
| 53 | /* objectid */ |
| 54 | u64 ino; |
| 55 | /* |
| 56 | * transid where the defrag was added, we search for |
| 57 | * extents newer than this |
| 58 | */ |
| 59 | u64 transid; |
| 60 | |
| 61 | /* root objectid */ |
| 62 | u64 root; |
| 63 | |
| 64 | /* last offset we were able to defrag */ |
| 65 | u64 last_offset; |
| 66 | |
| 67 | /* if we've wrapped around back to zero once already */ |
| 68 | int cycled; |
| 69 | }; |
| 70 | |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 71 | static int __compare_inode_defrag(struct inode_defrag *defrag1, |
| 72 | struct inode_defrag *defrag2) |
| 73 | { |
| 74 | if (defrag1->root > defrag2->root) |
| 75 | return 1; |
| 76 | else if (defrag1->root < defrag2->root) |
| 77 | return -1; |
| 78 | else if (defrag1->ino > defrag2->ino) |
| 79 | return 1; |
| 80 | else if (defrag1->ino < defrag2->ino) |
| 81 | return -1; |
| 82 | else |
| 83 | return 0; |
| 84 | } |
| 85 | |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 86 | /* pop a record for an inode into the defrag tree. The lock |
| 87 | * must be held already |
| 88 | * |
| 89 | * If you're inserting a record for an older transid than an |
| 90 | * existing record, the transid already in the tree is lowered |
| 91 | * |
| 92 | * If an existing record is found the defrag item you |
| 93 | * pass in is freed |
| 94 | */ |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 95 | static int __btrfs_add_inode_defrag(struct inode *inode, |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 96 | struct inode_defrag *defrag) |
| 97 | { |
| 98 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 99 | struct inode_defrag *entry; |
| 100 | struct rb_node **p; |
| 101 | struct rb_node *parent = NULL; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 102 | int ret; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 103 | |
| 104 | p = &root->fs_info->defrag_inodes.rb_node; |
| 105 | while (*p) { |
| 106 | parent = *p; |
| 107 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
| 108 | |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 109 | ret = __compare_inode_defrag(defrag, entry); |
| 110 | if (ret < 0) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 111 | p = &parent->rb_left; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 112 | else if (ret > 0) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 113 | p = &parent->rb_right; |
| 114 | else { |
| 115 | /* if we're reinserting an entry for |
| 116 | * an old defrag run, make sure to |
| 117 | * lower the transid of our existing record |
| 118 | */ |
| 119 | if (defrag->transid < entry->transid) |
| 120 | entry->transid = defrag->transid; |
| 121 | if (defrag->last_offset > entry->last_offset) |
| 122 | entry->last_offset = defrag->last_offset; |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 123 | return -EEXIST; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 124 | } |
| 125 | } |
Josef Bacik | 72ac3c0 | 2012-05-23 14:13:11 -0400 | [diff] [blame] | 126 | set_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 127 | rb_link_node(&defrag->rb_node, parent, p); |
| 128 | rb_insert_color(&defrag->rb_node, &root->fs_info->defrag_inodes); |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 129 | return 0; |
| 130 | } |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 131 | |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 132 | static inline int __need_auto_defrag(struct btrfs_root *root) |
| 133 | { |
| 134 | if (!btrfs_test_opt(root, AUTO_DEFRAG)) |
| 135 | return 0; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 136 | |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 137 | if (btrfs_fs_closing(root->fs_info)) |
| 138 | return 0; |
| 139 | |
| 140 | return 1; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 141 | } |
| 142 | |
| 143 | /* |
| 144 | * insert a defrag record for this inode if auto defrag is |
| 145 | * enabled |
| 146 | */ |
| 147 | int btrfs_add_inode_defrag(struct btrfs_trans_handle *trans, |
| 148 | struct inode *inode) |
| 149 | { |
| 150 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 151 | struct inode_defrag *defrag; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 152 | u64 transid; |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 153 | int ret; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 154 | |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 155 | if (!__need_auto_defrag(root)) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 156 | return 0; |
| 157 | |
Josef Bacik | 72ac3c0 | 2012-05-23 14:13:11 -0400 | [diff] [blame] | 158 | if (test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 159 | return 0; |
| 160 | |
| 161 | if (trans) |
| 162 | transid = trans->transid; |
| 163 | else |
| 164 | transid = BTRFS_I(inode)->root->last_trans; |
| 165 | |
Miao Xie | 9247f31 | 2012-11-26 09:24:43 +0000 | [diff] [blame] | 166 | defrag = kmem_cache_zalloc(btrfs_inode_defrag_cachep, GFP_NOFS); |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 167 | if (!defrag) |
| 168 | return -ENOMEM; |
| 169 | |
David Sterba | a4689d2 | 2011-05-31 17:08:14 +0000 | [diff] [blame] | 170 | defrag->ino = btrfs_ino(inode); |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 171 | defrag->transid = transid; |
| 172 | defrag->root = root->root_key.objectid; |
| 173 | |
| 174 | spin_lock(&root->fs_info->defrag_inodes_lock); |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 175 | if (!test_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags)) { |
| 176 | /* |
| 177 | * If we set IN_DEFRAG flag and evict the inode from memory, |
| 178 | * and then re-read this inode, this new inode doesn't have |
| 179 | * IN_DEFRAG flag. At the case, we may find the existed defrag. |
| 180 | */ |
| 181 | ret = __btrfs_add_inode_defrag(inode, defrag); |
| 182 | if (ret) |
| 183 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
| 184 | } else { |
Miao Xie | 9247f31 | 2012-11-26 09:24:43 +0000 | [diff] [blame] | 185 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 186 | } |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 187 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
Wanlong Gao | a0f98dd | 2011-07-18 12:19:35 +0000 | [diff] [blame] | 188 | return 0; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 189 | } |
| 190 | |
| 191 | /* |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 192 | * Requeue the defrag object. If there is a defrag object that points to |
| 193 | * the same inode in the tree, we will merge them together (by |
| 194 | * __btrfs_add_inode_defrag()) and free the one that we want to requeue. |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 195 | */ |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 196 | static void btrfs_requeue_inode_defrag(struct inode *inode, |
| 197 | struct inode_defrag *defrag) |
Miao Xie | 8ddc473 | 2012-11-26 09:25:38 +0000 | [diff] [blame] | 198 | { |
| 199 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 200 | int ret; |
| 201 | |
| 202 | if (!__need_auto_defrag(root)) |
| 203 | goto out; |
| 204 | |
| 205 | /* |
| 206 | * Here we don't check the IN_DEFRAG flag, because we need merge |
| 207 | * them together. |
| 208 | */ |
| 209 | spin_lock(&root->fs_info->defrag_inodes_lock); |
| 210 | ret = __btrfs_add_inode_defrag(inode, defrag); |
| 211 | spin_unlock(&root->fs_info->defrag_inodes_lock); |
| 212 | if (ret) |
| 213 | goto out; |
| 214 | return; |
| 215 | out: |
| 216 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
| 217 | } |
| 218 | |
| 219 | /* |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 220 | * pick the defragable inode that we want, if it doesn't exist, we will get |
| 221 | * the next one. |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 222 | */ |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 223 | static struct inode_defrag * |
| 224 | btrfs_pick_defrag_inode(struct btrfs_fs_info *fs_info, u64 root, u64 ino) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 225 | { |
| 226 | struct inode_defrag *entry = NULL; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 227 | struct inode_defrag tmp; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 228 | struct rb_node *p; |
| 229 | struct rb_node *parent = NULL; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 230 | int ret; |
| 231 | |
| 232 | tmp.ino = ino; |
| 233 | tmp.root = root; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 234 | |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 235 | spin_lock(&fs_info->defrag_inodes_lock); |
| 236 | p = fs_info->defrag_inodes.rb_node; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 237 | while (p) { |
| 238 | parent = p; |
| 239 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
| 240 | |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 241 | ret = __compare_inode_defrag(&tmp, entry); |
| 242 | if (ret < 0) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 243 | p = parent->rb_left; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 244 | else if (ret > 0) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 245 | p = parent->rb_right; |
| 246 | else |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 247 | goto out; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 248 | } |
| 249 | |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 250 | if (parent && __compare_inode_defrag(&tmp, entry) > 0) { |
| 251 | parent = rb_next(parent); |
| 252 | if (parent) |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 253 | entry = rb_entry(parent, struct inode_defrag, rb_node); |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 254 | else |
| 255 | entry = NULL; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 256 | } |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 257 | out: |
| 258 | if (entry) |
| 259 | rb_erase(parent, &fs_info->defrag_inodes); |
| 260 | spin_unlock(&fs_info->defrag_inodes_lock); |
| 261 | return entry; |
| 262 | } |
| 263 | |
| 264 | void btrfs_cleanup_defrag_inodes(struct btrfs_fs_info *fs_info) |
| 265 | { |
| 266 | struct inode_defrag *defrag; |
| 267 | struct rb_node *node; |
| 268 | |
| 269 | spin_lock(&fs_info->defrag_inodes_lock); |
| 270 | node = rb_first(&fs_info->defrag_inodes); |
| 271 | while (node) { |
| 272 | rb_erase(node, &fs_info->defrag_inodes); |
| 273 | defrag = rb_entry(node, struct inode_defrag, rb_node); |
| 274 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
| 275 | |
David Sterba | 351810c | 2015-01-08 15:20:54 +0100 | [diff] [blame] | 276 | cond_resched_lock(&fs_info->defrag_inodes_lock); |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 277 | |
| 278 | node = rb_first(&fs_info->defrag_inodes); |
| 279 | } |
| 280 | spin_unlock(&fs_info->defrag_inodes_lock); |
| 281 | } |
| 282 | |
| 283 | #define BTRFS_DEFRAG_BATCH 1024 |
| 284 | |
| 285 | static int __btrfs_run_defrag_inode(struct btrfs_fs_info *fs_info, |
| 286 | struct inode_defrag *defrag) |
| 287 | { |
| 288 | struct btrfs_root *inode_root; |
| 289 | struct inode *inode; |
| 290 | struct btrfs_key key; |
| 291 | struct btrfs_ioctl_defrag_range_args range; |
| 292 | int num_defrag; |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 293 | int index; |
| 294 | int ret; |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 295 | |
| 296 | /* get the inode */ |
| 297 | key.objectid = defrag->root; |
David Sterba | 962a298 | 2014-06-04 18:41:45 +0200 | [diff] [blame] | 298 | key.type = BTRFS_ROOT_ITEM_KEY; |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 299 | key.offset = (u64)-1; |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 300 | |
| 301 | index = srcu_read_lock(&fs_info->subvol_srcu); |
| 302 | |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 303 | inode_root = btrfs_read_fs_root_no_name(fs_info, &key); |
| 304 | if (IS_ERR(inode_root)) { |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 305 | ret = PTR_ERR(inode_root); |
| 306 | goto cleanup; |
| 307 | } |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 308 | |
| 309 | key.objectid = defrag->ino; |
David Sterba | 962a298 | 2014-06-04 18:41:45 +0200 | [diff] [blame] | 310 | key.type = BTRFS_INODE_ITEM_KEY; |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 311 | key.offset = 0; |
| 312 | inode = btrfs_iget(fs_info->sb, &key, inode_root, NULL); |
| 313 | if (IS_ERR(inode)) { |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 314 | ret = PTR_ERR(inode); |
| 315 | goto cleanup; |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 316 | } |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 317 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 318 | |
| 319 | /* do a chunk of defrag */ |
| 320 | clear_bit(BTRFS_INODE_IN_DEFRAG, &BTRFS_I(inode)->runtime_flags); |
| 321 | memset(&range, 0, sizeof(range)); |
| 322 | range.len = (u64)-1; |
| 323 | range.start = defrag->last_offset; |
Miao Xie | b66f00d | 2012-11-26 09:27:29 +0000 | [diff] [blame] | 324 | |
| 325 | sb_start_write(fs_info->sb); |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 326 | num_defrag = btrfs_defrag_file(inode, NULL, &range, defrag->transid, |
| 327 | BTRFS_DEFRAG_BATCH); |
Miao Xie | b66f00d | 2012-11-26 09:27:29 +0000 | [diff] [blame] | 328 | sb_end_write(fs_info->sb); |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 329 | /* |
| 330 | * if we filled the whole defrag batch, there |
| 331 | * must be more work to do. Queue this defrag |
| 332 | * again |
| 333 | */ |
| 334 | if (num_defrag == BTRFS_DEFRAG_BATCH) { |
| 335 | defrag->last_offset = range.start; |
| 336 | btrfs_requeue_inode_defrag(inode, defrag); |
| 337 | } else if (defrag->last_offset && !defrag->cycled) { |
| 338 | /* |
| 339 | * we didn't fill our defrag batch, but |
| 340 | * we didn't start at zero. Make sure we loop |
| 341 | * around to the start of the file. |
| 342 | */ |
| 343 | defrag->last_offset = 0; |
| 344 | defrag->cycled = 1; |
| 345 | btrfs_requeue_inode_defrag(inode, defrag); |
| 346 | } else { |
| 347 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
| 348 | } |
| 349 | |
| 350 | iput(inode); |
| 351 | return 0; |
Liu Bo | 6f1c360 | 2013-01-29 03:22:10 +0000 | [diff] [blame] | 352 | cleanup: |
| 353 | srcu_read_unlock(&fs_info->subvol_srcu, index); |
| 354 | kmem_cache_free(btrfs_inode_defrag_cachep, defrag); |
| 355 | return ret; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 356 | } |
| 357 | |
| 358 | /* |
| 359 | * run through the list of inodes in the FS that need |
| 360 | * defragging |
| 361 | */ |
| 362 | int btrfs_run_defrag_inodes(struct btrfs_fs_info *fs_info) |
| 363 | { |
| 364 | struct inode_defrag *defrag; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 365 | u64 first_ino = 0; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 366 | u64 root_objectid = 0; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 367 | |
| 368 | atomic_inc(&fs_info->defrag_running); |
Dulshani Gunawardhana | 6787125 | 2013-10-31 10:33:04 +0530 | [diff] [blame] | 369 | while (1) { |
Miao Xie | dc81cdc | 2013-02-20 23:32:52 -0700 | [diff] [blame] | 370 | /* Pause the auto defragger. */ |
| 371 | if (test_bit(BTRFS_FS_STATE_REMOUNTING, |
| 372 | &fs_info->fs_state)) |
| 373 | break; |
| 374 | |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 375 | if (!__need_auto_defrag(fs_info->tree_root)) |
| 376 | break; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 377 | |
| 378 | /* find an inode to defrag */ |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 379 | defrag = btrfs_pick_defrag_inode(fs_info, root_objectid, |
| 380 | first_ino); |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 381 | if (!defrag) { |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 382 | if (root_objectid || first_ino) { |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 383 | root_objectid = 0; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 384 | first_ino = 0; |
| 385 | continue; |
| 386 | } else { |
| 387 | break; |
| 388 | } |
| 389 | } |
| 390 | |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 391 | first_ino = defrag->ino + 1; |
Miao Xie | 762f226 | 2012-05-24 18:58:27 +0800 | [diff] [blame] | 392 | root_objectid = defrag->root; |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 393 | |
Miao Xie | 26176e7 | 2012-11-26 09:26:20 +0000 | [diff] [blame] | 394 | __btrfs_run_defrag_inode(fs_info, defrag); |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 395 | } |
Chris Mason | 4cb5300 | 2011-05-24 15:35:30 -0400 | [diff] [blame] | 396 | atomic_dec(&fs_info->defrag_running); |
| 397 | |
| 398 | /* |
| 399 | * during unmount, we use the transaction_wait queue to |
| 400 | * wait for the defragger to stop |
| 401 | */ |
| 402 | wake_up(&fs_info->transaction_wait); |
| 403 | return 0; |
| 404 | } |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 405 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 406 | /* simple helper to fault in pages and copy. This should go away |
| 407 | * and be replaced with calls into generic code. |
| 408 | */ |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 409 | static noinline int btrfs_copy_from_user(loff_t pos, int num_pages, |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 410 | size_t write_bytes, |
Chris Mason | a1b32a5 | 2008-09-05 16:09:51 -0400 | [diff] [blame] | 411 | struct page **prepared_pages, |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 412 | struct iov_iter *i) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 413 | { |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 414 | size_t copied = 0; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 415 | size_t total_copied = 0; |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 416 | int pg = 0; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 417 | int offset = pos & (PAGE_CACHE_SIZE - 1); |
| 418 | |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 419 | while (write_bytes > 0) { |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 420 | size_t count = min_t(size_t, |
| 421 | PAGE_CACHE_SIZE - offset, write_bytes); |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 422 | struct page *page = prepared_pages[pg]; |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 423 | /* |
| 424 | * Copy data from userspace to the current page |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 425 | */ |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 426 | copied = iov_iter_copy_from_user_atomic(page, i, offset, count); |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 427 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 428 | /* Flush processor's dcache for this page */ |
| 429 | flush_dcache_page(page); |
Chris Mason | 31339ac | 2011-03-07 11:10:24 -0500 | [diff] [blame] | 430 | |
| 431 | /* |
| 432 | * if we get a partial write, we can end up with |
| 433 | * partially up to date pages. These add |
| 434 | * a lot of complexity, so make sure they don't |
| 435 | * happen by forcing this copy to be retried. |
| 436 | * |
| 437 | * The rest of the btrfs_file_write code will fall |
| 438 | * back to page at a time copies after we return 0. |
| 439 | */ |
| 440 | if (!PageUptodate(page) && copied < count) |
| 441 | copied = 0; |
| 442 | |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 443 | iov_iter_advance(i, copied); |
| 444 | write_bytes -= copied; |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 445 | total_copied += copied; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 446 | |
Al Viro | b30ac0f | 2014-04-03 14:29:04 -0400 | [diff] [blame] | 447 | /* Return to btrfs_file_write_iter to fault page */ |
Josef Bacik | 9f570b8 | 2011-01-25 12:42:37 -0500 | [diff] [blame] | 448 | if (unlikely(copied == 0)) |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 449 | break; |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 450 | |
David Sterba | ee39b43 | 2014-09-30 01:33:33 +0200 | [diff] [blame] | 451 | if (copied < PAGE_CACHE_SIZE - offset) { |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 452 | offset += copied; |
| 453 | } else { |
| 454 | pg++; |
| 455 | offset = 0; |
| 456 | } |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 457 | } |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 458 | return total_copied; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 459 | } |
| 460 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 461 | /* |
| 462 | * unlocks pages after btrfs_file_write is done with them |
| 463 | */ |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 464 | static void btrfs_drop_pages(struct page **pages, size_t num_pages) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 465 | { |
| 466 | size_t i; |
| 467 | for (i = 0; i < num_pages; i++) { |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 468 | /* page checked is some magic around finding pages that |
| 469 | * have been modified without going through btrfs_set_page_dirty |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 470 | * clear it here. There should be no need to mark the pages |
| 471 | * accessed as prepare_pages should have marked them accessed |
| 472 | * in prepare_pages via find_or_create_page() |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 473 | */ |
Chris Mason | 4a09675 | 2008-07-21 10:29:44 -0400 | [diff] [blame] | 474 | ClearPageChecked(pages[i]); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 475 | unlock_page(pages[i]); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 476 | page_cache_release(pages[i]); |
| 477 | } |
| 478 | } |
| 479 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 480 | /* |
| 481 | * after copy_from_user, pages need to be dirtied and we need to make |
| 482 | * sure holes are created between the current EOF and the start of |
| 483 | * any next extents (if required). |
| 484 | * |
| 485 | * this also makes the decision about creating an inline extent vs |
| 486 | * doing real data extents, marking pages dirty and delalloc as required. |
| 487 | */ |
Josef Bacik | be1a12a | 2011-04-06 13:05:22 -0400 | [diff] [blame] | 488 | int btrfs_dirty_pages(struct btrfs_root *root, struct inode *inode, |
Eric Sandeen | 48a3b63 | 2013-04-25 20:41:01 +0000 | [diff] [blame] | 489 | struct page **pages, size_t num_pages, |
| 490 | loff_t pos, size_t write_bytes, |
| 491 | struct extent_state **cached) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 492 | { |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 493 | int err = 0; |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 494 | int i; |
Chris Mason | db94535 | 2007-10-15 16:15:53 -0400 | [diff] [blame] | 495 | u64 num_bytes; |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 496 | u64 start_pos; |
| 497 | u64 end_of_last_block; |
| 498 | u64 end_pos = pos + write_bytes; |
| 499 | loff_t isize = i_size_read(inode); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 500 | |
Chris Mason | 5f39d39 | 2007-10-15 16:14:19 -0400 | [diff] [blame] | 501 | start_pos = pos & ~((u64)root->sectorsize - 1); |
Qu Wenruo | fda2832 | 2013-02-26 08:10:22 +0000 | [diff] [blame] | 502 | num_bytes = ALIGN(write_bytes + pos - start_pos, root->sectorsize); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 503 | |
Chris Mason | db94535 | 2007-10-15 16:15:53 -0400 | [diff] [blame] | 504 | end_of_last_block = start_pos + num_bytes - 1; |
Josef Bacik | 2ac55d4 | 2010-02-03 19:33:23 +0000 | [diff] [blame] | 505 | err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block, |
Josef Bacik | be1a12a | 2011-04-06 13:05:22 -0400 | [diff] [blame] | 506 | cached); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 507 | if (err) |
| 508 | return err; |
Josef Bacik | 9ed74f2 | 2009-09-11 16:12:44 -0400 | [diff] [blame] | 509 | |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 510 | for (i = 0; i < num_pages; i++) { |
| 511 | struct page *p = pages[i]; |
| 512 | SetPageUptodate(p); |
| 513 | ClearPageChecked(p); |
| 514 | set_page_dirty(p); |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 515 | } |
Josef Bacik | 9f570b8 | 2011-01-25 12:42:37 -0500 | [diff] [blame] | 516 | |
| 517 | /* |
| 518 | * we've only changed i_size in ram, and we haven't updated |
| 519 | * the disk i_size. There is no need to log the inode |
| 520 | * at this time. |
| 521 | */ |
| 522 | if (end_pos > isize) |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 523 | i_size_write(inode, end_pos); |
Yan, Zheng | a22285a | 2010-05-16 10:48:46 -0400 | [diff] [blame] | 524 | return 0; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 525 | } |
| 526 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 527 | /* |
| 528 | * this drops all the extents in the cache that intersect the range |
| 529 | * [start, end]. Existing extents are split as required. |
| 530 | */ |
Josef Bacik | 7014cdb | 2012-08-30 20:06:49 -0400 | [diff] [blame] | 531 | void btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end, |
| 532 | int skip_pinned) |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 533 | { |
| 534 | struct extent_map *em; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 535 | struct extent_map *split = NULL; |
| 536 | struct extent_map *split2 = NULL; |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 537 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
Yan | 39b5637 | 2008-02-15 10:40:50 -0500 | [diff] [blame] | 538 | u64 len = end - start + 1; |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 539 | u64 gen; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 540 | int ret; |
| 541 | int testend = 1; |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 542 | unsigned long flags; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 543 | int compressed = 0; |
Josef Bacik | 09a2a8f9 | 2013-04-05 16:51:15 -0400 | [diff] [blame] | 544 | bool modified; |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 545 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 546 | WARN_ON(end < start); |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 547 | if (end == (u64)-1) { |
Yan | 39b5637 | 2008-02-15 10:40:50 -0500 | [diff] [blame] | 548 | len = (u64)-1; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 549 | testend = 0; |
| 550 | } |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 551 | while (1) { |
Josef Bacik | 7014cdb | 2012-08-30 20:06:49 -0400 | [diff] [blame] | 552 | int no_splits = 0; |
| 553 | |
Josef Bacik | 09a2a8f9 | 2013-04-05 16:51:15 -0400 | [diff] [blame] | 554 | modified = false; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 555 | if (!split) |
David Sterba | 172ddd6 | 2011-04-21 00:48:27 +0200 | [diff] [blame] | 556 | split = alloc_extent_map(); |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 557 | if (!split2) |
David Sterba | 172ddd6 | 2011-04-21 00:48:27 +0200 | [diff] [blame] | 558 | split2 = alloc_extent_map(); |
Josef Bacik | 7014cdb | 2012-08-30 20:06:49 -0400 | [diff] [blame] | 559 | if (!split || !split2) |
| 560 | no_splits = 1; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 561 | |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 562 | write_lock(&em_tree->lock); |
Yan | 39b5637 | 2008-02-15 10:40:50 -0500 | [diff] [blame] | 563 | em = lookup_extent_mapping(em_tree, start, len); |
Chris Mason | d1310b2 | 2008-01-24 16:13:08 -0500 | [diff] [blame] | 564 | if (!em) { |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 565 | write_unlock(&em_tree->lock); |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 566 | break; |
Chris Mason | d1310b2 | 2008-01-24 16:13:08 -0500 | [diff] [blame] | 567 | } |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 568 | flags = em->flags; |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 569 | gen = em->generation; |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 570 | if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) { |
Yan, Zheng | 55ef689 | 2009-11-12 09:36:44 +0000 | [diff] [blame] | 571 | if (testend && em->start + em->len >= start + len) { |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 572 | free_extent_map(em); |
Chris Mason | a1ed835 | 2009-09-11 12:27:37 -0400 | [diff] [blame] | 573 | write_unlock(&em_tree->lock); |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 574 | break; |
| 575 | } |
Yan, Zheng | 55ef689 | 2009-11-12 09:36:44 +0000 | [diff] [blame] | 576 | start = em->start + em->len; |
| 577 | if (testend) |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 578 | len = start + len - (em->start + em->len); |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 579 | free_extent_map(em); |
Chris Mason | a1ed835 | 2009-09-11 12:27:37 -0400 | [diff] [blame] | 580 | write_unlock(&em_tree->lock); |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 581 | continue; |
| 582 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 583 | compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags); |
Chris Mason | 3ce7e67 | 2008-07-31 15:42:54 -0400 | [diff] [blame] | 584 | clear_bit(EXTENT_FLAG_PINNED, &em->flags); |
Liu Bo | 3b27759 | 2013-03-15 08:46:39 -0600 | [diff] [blame] | 585 | clear_bit(EXTENT_FLAG_LOGGING, &flags); |
Josef Bacik | 09a2a8f9 | 2013-04-05 16:51:15 -0400 | [diff] [blame] | 586 | modified = !list_empty(&em->list); |
Josef Bacik | 7014cdb | 2012-08-30 20:06:49 -0400 | [diff] [blame] | 587 | if (no_splits) |
| 588 | goto next; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 589 | |
Josef Bacik | ee20a98 | 2013-07-11 10:34:59 -0400 | [diff] [blame] | 590 | if (em->start < start) { |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 591 | split->start = em->start; |
| 592 | split->len = start - em->start; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 593 | |
Josef Bacik | ee20a98 | 2013-07-11 10:34:59 -0400 | [diff] [blame] | 594 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
| 595 | split->orig_start = em->orig_start; |
| 596 | split->block_start = em->block_start; |
| 597 | |
| 598 | if (compressed) |
| 599 | split->block_len = em->block_len; |
| 600 | else |
| 601 | split->block_len = split->len; |
| 602 | split->orig_block_len = max(split->block_len, |
| 603 | em->orig_block_len); |
| 604 | split->ram_bytes = em->ram_bytes; |
| 605 | } else { |
| 606 | split->orig_start = split->start; |
| 607 | split->block_len = 0; |
| 608 | split->block_start = em->block_start; |
| 609 | split->orig_block_len = 0; |
| 610 | split->ram_bytes = split->len; |
| 611 | } |
| 612 | |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 613 | split->generation = gen; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 614 | split->bdev = em->bdev; |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 615 | split->flags = flags; |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 616 | split->compress_type = em->compress_type; |
Filipe Manana | 176840b | 2014-02-25 14:15:13 +0000 | [diff] [blame] | 617 | replace_extent_mapping(em_tree, em, split, modified); |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 618 | free_extent_map(split); |
| 619 | split = split2; |
| 620 | split2 = NULL; |
| 621 | } |
Josef Bacik | ee20a98 | 2013-07-11 10:34:59 -0400 | [diff] [blame] | 622 | if (testend && em->start + em->len > start + len) { |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 623 | u64 diff = start + len - em->start; |
| 624 | |
| 625 | split->start = start + len; |
| 626 | split->len = em->start + em->len - (start + len); |
| 627 | split->bdev = em->bdev; |
Zheng Yan | 5b21f2e | 2008-09-26 10:05:38 -0400 | [diff] [blame] | 628 | split->flags = flags; |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 629 | split->compress_type = em->compress_type; |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 630 | split->generation = gen; |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 631 | |
Josef Bacik | ee20a98 | 2013-07-11 10:34:59 -0400 | [diff] [blame] | 632 | if (em->block_start < EXTENT_MAP_LAST_BYTE) { |
| 633 | split->orig_block_len = max(em->block_len, |
| 634 | em->orig_block_len); |
| 635 | |
| 636 | split->ram_bytes = em->ram_bytes; |
| 637 | if (compressed) { |
| 638 | split->block_len = em->block_len; |
| 639 | split->block_start = em->block_start; |
| 640 | split->orig_start = em->orig_start; |
| 641 | } else { |
| 642 | split->block_len = split->len; |
| 643 | split->block_start = em->block_start |
| 644 | + diff; |
| 645 | split->orig_start = em->orig_start; |
| 646 | } |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 647 | } else { |
Josef Bacik | ee20a98 | 2013-07-11 10:34:59 -0400 | [diff] [blame] | 648 | split->ram_bytes = split->len; |
| 649 | split->orig_start = split->start; |
| 650 | split->block_len = 0; |
| 651 | split->block_start = em->block_start; |
| 652 | split->orig_block_len = 0; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 653 | } |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 654 | |
Filipe Manana | 176840b | 2014-02-25 14:15:13 +0000 | [diff] [blame] | 655 | if (extent_map_in_tree(em)) { |
| 656 | replace_extent_mapping(em_tree, em, split, |
| 657 | modified); |
| 658 | } else { |
| 659 | ret = add_extent_mapping(em_tree, split, |
| 660 | modified); |
| 661 | ASSERT(ret == 0); /* Logic error */ |
| 662 | } |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 663 | free_extent_map(split); |
| 664 | split = NULL; |
| 665 | } |
Josef Bacik | 7014cdb | 2012-08-30 20:06:49 -0400 | [diff] [blame] | 666 | next: |
Filipe Manana | 176840b | 2014-02-25 14:15:13 +0000 | [diff] [blame] | 667 | if (extent_map_in_tree(em)) |
| 668 | remove_extent_mapping(em_tree, em); |
Chris Mason | 890871b | 2009-09-02 16:24:52 -0400 | [diff] [blame] | 669 | write_unlock(&em_tree->lock); |
Chris Mason | d1310b2 | 2008-01-24 16:13:08 -0500 | [diff] [blame] | 670 | |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 671 | /* once for us */ |
| 672 | free_extent_map(em); |
| 673 | /* once for the tree*/ |
| 674 | free_extent_map(em); |
| 675 | } |
Chris Mason | 3b95151 | 2008-04-17 11:29:12 -0400 | [diff] [blame] | 676 | if (split) |
| 677 | free_extent_map(split); |
| 678 | if (split2) |
| 679 | free_extent_map(split2); |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 680 | } |
| 681 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 682 | /* |
| 683 | * this is very complex, but the basic idea is to drop all extents |
| 684 | * in the range start - end. hint_block is filled in with a block number |
| 685 | * that would be a good hint to the block allocator for this file. |
| 686 | * |
| 687 | * If an extent intersects the range but is not entirely inside the range |
| 688 | * it is either truncated or split. Anything entirely inside the range |
| 689 | * is deleted from the tree. |
| 690 | */ |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 691 | int __btrfs_drop_extents(struct btrfs_trans_handle *trans, |
| 692 | struct btrfs_root *root, struct inode *inode, |
| 693 | struct btrfs_path *path, u64 start, u64 end, |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 694 | u64 *drop_end, int drop_cache, |
| 695 | int replace_extent, |
| 696 | u32 extent_item_size, |
| 697 | int *key_inserted) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 698 | { |
Chris Mason | 00f5c79 | 2007-11-30 10:09:33 -0500 | [diff] [blame] | 699 | struct extent_buffer *leaf; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 700 | struct btrfs_file_extent_item *fi; |
Chris Mason | 00f5c79 | 2007-11-30 10:09:33 -0500 | [diff] [blame] | 701 | struct btrfs_key key; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 702 | struct btrfs_key new_key; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 703 | u64 ino = btrfs_ino(inode); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 704 | u64 search_start = start; |
| 705 | u64 disk_bytenr = 0; |
| 706 | u64 num_bytes = 0; |
| 707 | u64 extent_offset = 0; |
| 708 | u64 extent_end = 0; |
| 709 | int del_nr = 0; |
| 710 | int del_slot = 0; |
| 711 | int extent_type; |
Chris Mason | ccd467d | 2007-06-28 15:57:36 -0400 | [diff] [blame] | 712 | int recow; |
Chris Mason | 00f5c79 | 2007-11-30 10:09:33 -0500 | [diff] [blame] | 713 | int ret; |
Chris Mason | dc7fdde | 2012-04-27 14:31:29 -0400 | [diff] [blame] | 714 | int modify_tree = -1; |
Miao Xie | 27cdeb7 | 2014-04-02 19:51:05 +0800 | [diff] [blame] | 715 | int update_refs; |
Josef Bacik | c3308f8 | 2012-09-14 14:51:22 -0400 | [diff] [blame] | 716 | int found = 0; |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 717 | int leafs_visited = 0; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 718 | |
Chris Mason | a1ed835 | 2009-09-11 12:27:37 -0400 | [diff] [blame] | 719 | if (drop_cache) |
| 720 | btrfs_drop_extent_cache(inode, start, end - 1, 0); |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 721 | |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 722 | if (start >= BTRFS_I(inode)->disk_i_size && !replace_extent) |
Chris Mason | dc7fdde | 2012-04-27 14:31:29 -0400 | [diff] [blame] | 723 | modify_tree = 0; |
| 724 | |
Miao Xie | 27cdeb7 | 2014-04-02 19:51:05 +0800 | [diff] [blame] | 725 | update_refs = (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
| 726 | root == root->fs_info->tree_root); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 727 | while (1) { |
Chris Mason | ccd467d | 2007-06-28 15:57:36 -0400 | [diff] [blame] | 728 | recow = 0; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 729 | ret = btrfs_lookup_file_extent(trans, root, path, ino, |
Chris Mason | dc7fdde | 2012-04-27 14:31:29 -0400 | [diff] [blame] | 730 | search_start, modify_tree); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 731 | if (ret < 0) |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 732 | break; |
| 733 | if (ret > 0 && path->slots[0] > 0 && search_start == start) { |
| 734 | leaf = path->nodes[0]; |
| 735 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1); |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 736 | if (key.objectid == ino && |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 737 | key.type == BTRFS_EXTENT_DATA_KEY) |
| 738 | path->slots[0]--; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 739 | } |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 740 | ret = 0; |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 741 | leafs_visited++; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 742 | next_slot: |
| 743 | leaf = path->nodes[0]; |
| 744 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
| 745 | BUG_ON(del_nr > 0); |
| 746 | ret = btrfs_next_leaf(root, path); |
| 747 | if (ret < 0) |
| 748 | break; |
| 749 | if (ret > 0) { |
| 750 | ret = 0; |
| 751 | break; |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 752 | } |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 753 | leafs_visited++; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 754 | leaf = path->nodes[0]; |
| 755 | recow = 1; |
| 756 | } |
| 757 | |
| 758 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
Filipe Manana | aeafbf8 | 2015-11-06 13:33:33 +0000 | [diff] [blame] | 759 | |
| 760 | if (key.objectid > ino) |
| 761 | break; |
| 762 | if (WARN_ON_ONCE(key.objectid < ino) || |
| 763 | key.type < BTRFS_EXTENT_DATA_KEY) { |
| 764 | ASSERT(del_nr == 0); |
| 765 | path->slots[0]++; |
| 766 | goto next_slot; |
| 767 | } |
| 768 | if (key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end) |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 769 | break; |
| 770 | |
| 771 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 772 | struct btrfs_file_extent_item); |
| 773 | extent_type = btrfs_file_extent_type(leaf, fi); |
| 774 | |
| 775 | if (extent_type == BTRFS_FILE_EXTENT_REG || |
| 776 | extent_type == BTRFS_FILE_EXTENT_PREALLOC) { |
| 777 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
| 778 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); |
| 779 | extent_offset = btrfs_file_extent_offset(leaf, fi); |
| 780 | extent_end = key.offset + |
| 781 | btrfs_file_extent_num_bytes(leaf, fi); |
| 782 | } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
| 783 | extent_end = key.offset + |
Chris Mason | 514ac8a | 2014-01-03 21:07:00 -0800 | [diff] [blame] | 784 | btrfs_file_extent_inline_len(leaf, |
| 785 | path->slots[0], fi); |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 786 | } else { |
Filipe Manana | aeafbf8 | 2015-11-06 13:33:33 +0000 | [diff] [blame] | 787 | /* can't happen */ |
| 788 | BUG(); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 789 | } |
| 790 | |
Filipe Manana | fc19c5e | 2014-04-29 13:18:40 +0100 | [diff] [blame] | 791 | /* |
| 792 | * Don't skip extent items representing 0 byte lengths. They |
| 793 | * used to be created (bug) if while punching holes we hit |
| 794 | * -ENOSPC condition. So if we find one here, just ensure we |
| 795 | * delete it, otherwise we would insert a new file extent item |
| 796 | * with the same key (offset) as that 0 bytes length file |
| 797 | * extent item in the call to setup_items_for_insert() later |
| 798 | * in this function. |
| 799 | */ |
| 800 | if (extent_end == key.offset && extent_end >= search_start) |
| 801 | goto delete_extent_item; |
| 802 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 803 | if (extent_end <= search_start) { |
| 804 | path->slots[0]++; |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 805 | goto next_slot; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 806 | } |
| 807 | |
Josef Bacik | c3308f8 | 2012-09-14 14:51:22 -0400 | [diff] [blame] | 808 | found = 1; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 809 | search_start = max(key.offset, start); |
Chris Mason | dc7fdde | 2012-04-27 14:31:29 -0400 | [diff] [blame] | 810 | if (recow || !modify_tree) { |
| 811 | modify_tree = -1; |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 812 | btrfs_release_path(path); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 813 | continue; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 814 | } |
Chris Mason | 771ed68 | 2008-11-06 22:02:51 -0500 | [diff] [blame] | 815 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 816 | /* |
| 817 | * | - range to drop - | |
| 818 | * | -------- extent -------- | |
| 819 | */ |
| 820 | if (start > key.offset && end < extent_end) { |
| 821 | BUG_ON(del_nr > 0); |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 822 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
David Sterba | 3f9e3df | 2014-04-15 18:50:17 +0200 | [diff] [blame] | 823 | ret = -EOPNOTSUPP; |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 824 | break; |
| 825 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 826 | |
| 827 | memcpy(&new_key, &key, sizeof(new_key)); |
| 828 | new_key.offset = start; |
| 829 | ret = btrfs_duplicate_item(trans, root, path, |
| 830 | &new_key); |
| 831 | if (ret == -EAGAIN) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 832 | btrfs_release_path(path); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 833 | continue; |
| 834 | } |
| 835 | if (ret < 0) |
| 836 | break; |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 837 | |
Chris Mason | 5f39d39 | 2007-10-15 16:14:19 -0400 | [diff] [blame] | 838 | leaf = path->nodes[0]; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 839 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, |
| 840 | struct btrfs_file_extent_item); |
| 841 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 842 | start - key.offset); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 843 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 844 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 845 | struct btrfs_file_extent_item); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 846 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 847 | extent_offset += start - key.offset; |
| 848 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
| 849 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 850 | extent_end - start); |
| 851 | btrfs_mark_buffer_dirty(leaf); |
Chris Mason | db94535 | 2007-10-15 16:15:53 -0400 | [diff] [blame] | 852 | |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 853 | if (update_refs && disk_bytenr > 0) { |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 854 | ret = btrfs_inc_extent_ref(trans, root, |
| 855 | disk_bytenr, num_bytes, 0, |
| 856 | root->root_key.objectid, |
| 857 | new_key.objectid, |
Filipe Manana | b06c4bf | 2015-10-23 07:52:54 +0100 | [diff] [blame] | 858 | start - extent_offset); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 859 | BUG_ON(ret); /* -ENOMEM */ |
Zheng Yan | 31840ae | 2008-09-23 13:14:14 -0400 | [diff] [blame] | 860 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 861 | key.offset = start; |
| 862 | } |
| 863 | /* |
| 864 | * | ---- range to drop ----- | |
| 865 | * | -------- extent -------- | |
| 866 | */ |
| 867 | if (start <= key.offset && end < extent_end) { |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 868 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
David Sterba | 3f9e3df | 2014-04-15 18:50:17 +0200 | [diff] [blame] | 869 | ret = -EOPNOTSUPP; |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 870 | break; |
| 871 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 872 | |
| 873 | memcpy(&new_key, &key, sizeof(new_key)); |
| 874 | new_key.offset = end; |
Daniel Dressler | b7a0365 | 2014-11-12 13:43:09 +0900 | [diff] [blame] | 875 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 876 | |
| 877 | extent_offset += end - key.offset; |
| 878 | btrfs_set_file_extent_offset(leaf, fi, extent_offset); |
| 879 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 880 | extent_end - end); |
| 881 | btrfs_mark_buffer_dirty(leaf); |
Josef Bacik | 2671485 | 2012-08-29 12:24:27 -0400 | [diff] [blame] | 882 | if (update_refs && disk_bytenr > 0) |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 883 | inode_sub_bytes(inode, end - key.offset); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 884 | break; |
Zheng Yan | 31840ae | 2008-09-23 13:14:14 -0400 | [diff] [blame] | 885 | } |
| 886 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 887 | search_start = extent_end; |
| 888 | /* |
| 889 | * | ---- range to drop ----- | |
| 890 | * | -------- extent -------- | |
| 891 | */ |
| 892 | if (start > key.offset && end >= extent_end) { |
| 893 | BUG_ON(del_nr > 0); |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 894 | if (extent_type == BTRFS_FILE_EXTENT_INLINE) { |
David Sterba | 3f9e3df | 2014-04-15 18:50:17 +0200 | [diff] [blame] | 895 | ret = -EOPNOTSUPP; |
Liu Bo | 00fdf13 | 2014-03-10 18:56:07 +0800 | [diff] [blame] | 896 | break; |
| 897 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 898 | |
| 899 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 900 | start - key.offset); |
| 901 | btrfs_mark_buffer_dirty(leaf); |
Josef Bacik | 2671485 | 2012-08-29 12:24:27 -0400 | [diff] [blame] | 902 | if (update_refs && disk_bytenr > 0) |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 903 | inode_sub_bytes(inode, extent_end - start); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 904 | if (end == extent_end) |
| 905 | break; |
| 906 | |
| 907 | path->slots[0]++; |
| 908 | goto next_slot; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 909 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 910 | |
| 911 | /* |
| 912 | * | ---- range to drop ----- | |
| 913 | * | ------ extent ------ | |
| 914 | */ |
| 915 | if (start <= key.offset && end >= extent_end) { |
Filipe Manana | fc19c5e | 2014-04-29 13:18:40 +0100 | [diff] [blame] | 916 | delete_extent_item: |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 917 | if (del_nr == 0) { |
| 918 | del_slot = path->slots[0]; |
| 919 | del_nr = 1; |
| 920 | } else { |
| 921 | BUG_ON(del_slot + del_nr != path->slots[0]); |
| 922 | del_nr++; |
| 923 | } |
| 924 | |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 925 | if (update_refs && |
| 926 | extent_type == BTRFS_FILE_EXTENT_INLINE) { |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 927 | inode_sub_bytes(inode, |
| 928 | extent_end - key.offset); |
| 929 | extent_end = ALIGN(extent_end, |
| 930 | root->sectorsize); |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 931 | } else if (update_refs && disk_bytenr > 0) { |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 932 | ret = btrfs_free_extent(trans, root, |
| 933 | disk_bytenr, num_bytes, 0, |
| 934 | root->root_key.objectid, |
| 935 | key.objectid, key.offset - |
Filipe Manana | b06c4bf | 2015-10-23 07:52:54 +0100 | [diff] [blame] | 936 | extent_offset); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 937 | BUG_ON(ret); /* -ENOMEM */ |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 938 | inode_sub_bytes(inode, |
| 939 | extent_end - key.offset); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 940 | } |
| 941 | |
| 942 | if (end == extent_end) |
| 943 | break; |
| 944 | |
| 945 | if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) { |
| 946 | path->slots[0]++; |
| 947 | goto next_slot; |
| 948 | } |
| 949 | |
| 950 | ret = btrfs_del_items(trans, root, path, del_slot, |
| 951 | del_nr); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 952 | if (ret) { |
| 953 | btrfs_abort_transaction(trans, root, ret); |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 954 | break; |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 955 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 956 | |
| 957 | del_nr = 0; |
| 958 | del_slot = 0; |
| 959 | |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 960 | btrfs_release_path(path); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 961 | continue; |
| 962 | } |
| 963 | |
| 964 | BUG_ON(1); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 965 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 966 | |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 967 | if (!ret && del_nr > 0) { |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 968 | /* |
| 969 | * Set path->slots[0] to first slot, so that after the delete |
| 970 | * if items are move off from our leaf to its immediate left or |
| 971 | * right neighbor leafs, we end up with a correct and adjusted |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 972 | * path->slots[0] for our insertion (if replace_extent != 0). |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 973 | */ |
| 974 | path->slots[0] = del_slot; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 975 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 976 | if (ret) |
| 977 | btrfs_abort_transaction(trans, root, ret); |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 978 | } |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 979 | |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 980 | leaf = path->nodes[0]; |
| 981 | /* |
| 982 | * If btrfs_del_items() was called, it might have deleted a leaf, in |
| 983 | * which case it unlocked our path, so check path->locks[0] matches a |
| 984 | * write lock. |
| 985 | */ |
| 986 | if (!ret && replace_extent && leafs_visited == 1 && |
| 987 | (path->locks[0] == BTRFS_WRITE_LOCK_BLOCKING || |
| 988 | path->locks[0] == BTRFS_WRITE_LOCK) && |
| 989 | btrfs_leaf_free_space(root, leaf) >= |
| 990 | sizeof(struct btrfs_item) + extent_item_size) { |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 991 | |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 992 | key.objectid = ino; |
| 993 | key.type = BTRFS_EXTENT_DATA_KEY; |
| 994 | key.offset = start; |
| 995 | if (!del_nr && path->slots[0] < btrfs_header_nritems(leaf)) { |
| 996 | struct btrfs_key slot_key; |
| 997 | |
| 998 | btrfs_item_key_to_cpu(leaf, &slot_key, path->slots[0]); |
| 999 | if (btrfs_comp_cpu_keys(&key, &slot_key) > 0) |
| 1000 | path->slots[0]++; |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 1001 | } |
Filipe David Borba Manana | d5f3752 | 2014-02-09 23:45:12 +0000 | [diff] [blame] | 1002 | setup_items_for_insert(root, path, &key, |
| 1003 | &extent_item_size, |
| 1004 | extent_item_size, |
| 1005 | sizeof(struct btrfs_item) + |
| 1006 | extent_item_size, 1); |
| 1007 | *key_inserted = 1; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1008 | } |
| 1009 | |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 1010 | if (!replace_extent || !(*key_inserted)) |
| 1011 | btrfs_release_path(path); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 1012 | if (drop_end) |
Josef Bacik | c3308f8 | 2012-09-14 14:51:22 -0400 | [diff] [blame] | 1013 | *drop_end = found ? min(end, extent_end) : end; |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 1014 | return ret; |
| 1015 | } |
| 1016 | |
| 1017 | int btrfs_drop_extents(struct btrfs_trans_handle *trans, |
| 1018 | struct btrfs_root *root, struct inode *inode, u64 start, |
Josef Bacik | 2671485 | 2012-08-29 12:24:27 -0400 | [diff] [blame] | 1019 | u64 end, int drop_cache) |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 1020 | { |
| 1021 | struct btrfs_path *path; |
| 1022 | int ret; |
| 1023 | |
| 1024 | path = btrfs_alloc_path(); |
| 1025 | if (!path) |
| 1026 | return -ENOMEM; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 1027 | ret = __btrfs_drop_extents(trans, root, inode, path, start, end, NULL, |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 1028 | drop_cache, 0, 0, NULL); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1029 | btrfs_free_path(path); |
| 1030 | return ret; |
| 1031 | } |
| 1032 | |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1033 | static int extent_mergeable(struct extent_buffer *leaf, int slot, |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1034 | u64 objectid, u64 bytenr, u64 orig_offset, |
| 1035 | u64 *start, u64 *end) |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1036 | { |
| 1037 | struct btrfs_file_extent_item *fi; |
| 1038 | struct btrfs_key key; |
| 1039 | u64 extent_end; |
| 1040 | |
| 1041 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) |
| 1042 | return 0; |
| 1043 | |
| 1044 | btrfs_item_key_to_cpu(leaf, &key, slot); |
| 1045 | if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY) |
| 1046 | return 0; |
| 1047 | |
| 1048 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
| 1049 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG || |
| 1050 | btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr || |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1051 | btrfs_file_extent_offset(leaf, fi) != key.offset - orig_offset || |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1052 | btrfs_file_extent_compression(leaf, fi) || |
| 1053 | btrfs_file_extent_encryption(leaf, fi) || |
| 1054 | btrfs_file_extent_other_encoding(leaf, fi)) |
| 1055 | return 0; |
| 1056 | |
| 1057 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
| 1058 | if ((*start && *start != key.offset) || (*end && *end != extent_end)) |
| 1059 | return 0; |
| 1060 | |
| 1061 | *start = key.offset; |
| 1062 | *end = extent_end; |
| 1063 | return 1; |
| 1064 | } |
| 1065 | |
| 1066 | /* |
| 1067 | * Mark extent in the range start - end as written. |
| 1068 | * |
| 1069 | * This changes extent type from 'pre-allocated' to 'regular'. If only |
| 1070 | * part of extent is marked as written, the extent will be split into |
| 1071 | * two or three. |
| 1072 | */ |
| 1073 | int btrfs_mark_extent_written(struct btrfs_trans_handle *trans, |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1074 | struct inode *inode, u64 start, u64 end) |
| 1075 | { |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1076 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1077 | struct extent_buffer *leaf; |
| 1078 | struct btrfs_path *path; |
| 1079 | struct btrfs_file_extent_item *fi; |
| 1080 | struct btrfs_key key; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1081 | struct btrfs_key new_key; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1082 | u64 bytenr; |
| 1083 | u64 num_bytes; |
| 1084 | u64 extent_end; |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 1085 | u64 orig_offset; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1086 | u64 other_start; |
| 1087 | u64 other_end; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1088 | u64 split; |
| 1089 | int del_nr = 0; |
| 1090 | int del_slot = 0; |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1091 | int recow; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1092 | int ret; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1093 | u64 ino = btrfs_ino(inode); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1094 | |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1095 | path = btrfs_alloc_path(); |
Mark Fasheh | d8926bb | 2011-07-13 10:38:47 -0700 | [diff] [blame] | 1096 | if (!path) |
| 1097 | return -ENOMEM; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1098 | again: |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1099 | recow = 0; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1100 | split = start; |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1101 | key.objectid = ino; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1102 | key.type = BTRFS_EXTENT_DATA_KEY; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1103 | key.offset = split; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1104 | |
| 1105 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
Josef Bacik | 4141573 | 2011-03-16 13:59:32 -0400 | [diff] [blame] | 1106 | if (ret < 0) |
| 1107 | goto out; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1108 | if (ret > 0 && path->slots[0] > 0) |
| 1109 | path->slots[0]--; |
| 1110 | |
| 1111 | leaf = path->nodes[0]; |
| 1112 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1113 | BUG_ON(key.objectid != ino || key.type != BTRFS_EXTENT_DATA_KEY); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1114 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1115 | struct btrfs_file_extent_item); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1116 | BUG_ON(btrfs_file_extent_type(leaf, fi) != |
| 1117 | BTRFS_FILE_EXTENT_PREALLOC); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1118 | extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi); |
| 1119 | BUG_ON(key.offset > start || extent_end < end); |
| 1120 | |
| 1121 | bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
| 1122 | num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); |
Yan Zheng | 5d4f98a | 2009-06-10 10:45:14 -0400 | [diff] [blame] | 1123 | orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1124 | memcpy(&new_key, &key, sizeof(new_key)); |
| 1125 | |
| 1126 | if (start == key.offset && end < extent_end) { |
| 1127 | other_start = 0; |
| 1128 | other_end = start; |
| 1129 | if (extent_mergeable(leaf, path->slots[0] - 1, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1130 | ino, bytenr, orig_offset, |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1131 | &other_start, &other_end)) { |
| 1132 | new_key.offset = end; |
Daniel Dressler | b7a0365 | 2014-11-12 13:43:09 +0900 | [diff] [blame] | 1133 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1134 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1135 | struct btrfs_file_extent_item); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1136 | btrfs_set_file_extent_generation(leaf, fi, |
| 1137 | trans->transid); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1138 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1139 | extent_end - end); |
| 1140 | btrfs_set_file_extent_offset(leaf, fi, |
| 1141 | end - orig_offset); |
| 1142 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, |
| 1143 | struct btrfs_file_extent_item); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1144 | btrfs_set_file_extent_generation(leaf, fi, |
| 1145 | trans->transid); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1146 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1147 | end - other_start); |
| 1148 | btrfs_mark_buffer_dirty(leaf); |
| 1149 | goto out; |
| 1150 | } |
| 1151 | } |
| 1152 | |
| 1153 | if (start > key.offset && end == extent_end) { |
| 1154 | other_start = end; |
| 1155 | other_end = 0; |
| 1156 | if (extent_mergeable(leaf, path->slots[0] + 1, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1157 | ino, bytenr, orig_offset, |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1158 | &other_start, &other_end)) { |
| 1159 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1160 | struct btrfs_file_extent_item); |
| 1161 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1162 | start - key.offset); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1163 | btrfs_set_file_extent_generation(leaf, fi, |
| 1164 | trans->transid); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1165 | path->slots[0]++; |
| 1166 | new_key.offset = start; |
Daniel Dressler | b7a0365 | 2014-11-12 13:43:09 +0900 | [diff] [blame] | 1167 | btrfs_set_item_key_safe(root->fs_info, path, &new_key); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1168 | |
| 1169 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1170 | struct btrfs_file_extent_item); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1171 | btrfs_set_file_extent_generation(leaf, fi, |
| 1172 | trans->transid); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1173 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1174 | other_end - start); |
| 1175 | btrfs_set_file_extent_offset(leaf, fi, |
| 1176 | start - orig_offset); |
| 1177 | btrfs_mark_buffer_dirty(leaf); |
| 1178 | goto out; |
| 1179 | } |
| 1180 | } |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1181 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1182 | while (start > key.offset || end < extent_end) { |
| 1183 | if (key.offset == start) |
| 1184 | split = end; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1185 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1186 | new_key.offset = split; |
| 1187 | ret = btrfs_duplicate_item(trans, root, path, &new_key); |
| 1188 | if (ret == -EAGAIN) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1189 | btrfs_release_path(path); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1190 | goto again; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1191 | } |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 1192 | if (ret < 0) { |
| 1193 | btrfs_abort_transaction(trans, root, ret); |
| 1194 | goto out; |
| 1195 | } |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1196 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1197 | leaf = path->nodes[0]; |
| 1198 | fi = btrfs_item_ptr(leaf, path->slots[0] - 1, |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1199 | struct btrfs_file_extent_item); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1200 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1201 | btrfs_set_file_extent_num_bytes(leaf, fi, |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1202 | split - key.offset); |
| 1203 | |
| 1204 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1205 | struct btrfs_file_extent_item); |
| 1206 | |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1207 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1208 | btrfs_set_file_extent_offset(leaf, fi, split - orig_offset); |
| 1209 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1210 | extent_end - split); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1211 | btrfs_mark_buffer_dirty(leaf); |
| 1212 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1213 | ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, |
| 1214 | root->root_key.objectid, |
Filipe Manana | b06c4bf | 2015-10-23 07:52:54 +0100 | [diff] [blame] | 1215 | ino, orig_offset); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 1216 | BUG_ON(ret); /* -ENOMEM */ |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1217 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1218 | if (split == start) { |
| 1219 | key.offset = start; |
| 1220 | } else { |
| 1221 | BUG_ON(start != key.offset); |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1222 | path->slots[0]--; |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1223 | extent_end = end; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1224 | } |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1225 | recow = 1; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1226 | } |
| 1227 | |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1228 | other_start = end; |
| 1229 | other_end = 0; |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1230 | if (extent_mergeable(leaf, path->slots[0] + 1, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1231 | ino, bytenr, orig_offset, |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1232 | &other_start, &other_end)) { |
| 1233 | if (recow) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1234 | btrfs_release_path(path); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1235 | goto again; |
| 1236 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1237 | extent_end = other_end; |
| 1238 | del_slot = path->slots[0] + 1; |
| 1239 | del_nr++; |
| 1240 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
| 1241 | 0, root->root_key.objectid, |
Filipe Manana | b06c4bf | 2015-10-23 07:52:54 +0100 | [diff] [blame] | 1242 | ino, orig_offset); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 1243 | BUG_ON(ret); /* -ENOMEM */ |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1244 | } |
| 1245 | other_start = 0; |
| 1246 | other_end = start; |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1247 | if (extent_mergeable(leaf, path->slots[0] - 1, |
Li Zefan | 33345d01 | 2011-04-20 10:31:50 +0800 | [diff] [blame] | 1248 | ino, bytenr, orig_offset, |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1249 | &other_start, &other_end)) { |
| 1250 | if (recow) { |
David Sterba | b3b4aa7 | 2011-04-21 01:20:15 +0200 | [diff] [blame] | 1251 | btrfs_release_path(path); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1252 | goto again; |
| 1253 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1254 | key.offset = other_start; |
| 1255 | del_slot = path->slots[0]; |
| 1256 | del_nr++; |
| 1257 | ret = btrfs_free_extent(trans, root, bytenr, num_bytes, |
| 1258 | 0, root->root_key.objectid, |
Filipe Manana | b06c4bf | 2015-10-23 07:52:54 +0100 | [diff] [blame] | 1259 | ino, orig_offset); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 1260 | BUG_ON(ret); /* -ENOMEM */ |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1261 | } |
| 1262 | if (del_nr == 0) { |
Shaohua Li | 3f6fae9 | 2010-02-11 07:43:00 +0000 | [diff] [blame] | 1263 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 1264 | struct btrfs_file_extent_item); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1265 | btrfs_set_file_extent_type(leaf, fi, |
| 1266 | BTRFS_FILE_EXTENT_REG); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1267 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1268 | btrfs_mark_buffer_dirty(leaf); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1269 | } else { |
Shaohua Li | 3f6fae9 | 2010-02-11 07:43:00 +0000 | [diff] [blame] | 1270 | fi = btrfs_item_ptr(leaf, del_slot - 1, |
| 1271 | struct btrfs_file_extent_item); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1272 | btrfs_set_file_extent_type(leaf, fi, |
| 1273 | BTRFS_FILE_EXTENT_REG); |
Josef Bacik | 224ecce | 2012-08-16 16:32:06 -0400 | [diff] [blame] | 1274 | btrfs_set_file_extent_generation(leaf, fi, trans->transid); |
Yan, Zheng | 6c7d54a | 2010-01-15 08:43:09 +0000 | [diff] [blame] | 1275 | btrfs_set_file_extent_num_bytes(leaf, fi, |
| 1276 | extent_end - key.offset); |
| 1277 | btrfs_mark_buffer_dirty(leaf); |
| 1278 | |
| 1279 | ret = btrfs_del_items(trans, root, path, del_slot, del_nr); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 1280 | if (ret < 0) { |
| 1281 | btrfs_abort_transaction(trans, root, ret); |
| 1282 | goto out; |
| 1283 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1284 | } |
Yan, Zheng | 920bbbf | 2009-11-12 09:34:08 +0000 | [diff] [blame] | 1285 | out: |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 1286 | btrfs_free_path(path); |
| 1287 | return 0; |
| 1288 | } |
| 1289 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1290 | /* |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1291 | * on error we return an unlocked page and the error value |
| 1292 | * on success we return a locked page and 0 |
| 1293 | */ |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1294 | static int prepare_uptodate_page(struct inode *inode, |
| 1295 | struct page *page, u64 pos, |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1296 | bool force_uptodate) |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1297 | { |
| 1298 | int ret = 0; |
| 1299 | |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1300 | if (((pos & (PAGE_CACHE_SIZE - 1)) || force_uptodate) && |
| 1301 | !PageUptodate(page)) { |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1302 | ret = btrfs_readpage(NULL, page); |
| 1303 | if (ret) |
| 1304 | return ret; |
| 1305 | lock_page(page); |
| 1306 | if (!PageUptodate(page)) { |
| 1307 | unlock_page(page); |
| 1308 | return -EIO; |
| 1309 | } |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1310 | if (page->mapping != inode->i_mapping) { |
| 1311 | unlock_page(page); |
| 1312 | return -EAGAIN; |
| 1313 | } |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1314 | } |
| 1315 | return 0; |
| 1316 | } |
| 1317 | |
| 1318 | /* |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1319 | * this just gets pages into the page cache and locks them down. |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1320 | */ |
Miao Xie | b37392e | 2013-12-10 19:25:03 +0800 | [diff] [blame] | 1321 | static noinline int prepare_pages(struct inode *inode, struct page **pages, |
| 1322 | size_t num_pages, loff_t pos, |
| 1323 | size_t write_bytes, bool force_uptodate) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1324 | { |
| 1325 | int i; |
| 1326 | unsigned long index = pos >> PAGE_CACHE_SHIFT; |
Josef Bacik | 3b16a4e | 2011-09-21 15:05:58 -0400 | [diff] [blame] | 1327 | gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); |
Filipe David Borba Manana | fc28b62 | 2013-12-13 19:39:34 +0000 | [diff] [blame] | 1328 | int err = 0; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1329 | int faili; |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 1330 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1331 | for (i = 0; i < num_pages; i++) { |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1332 | again: |
Josef Bacik | a94733d | 2011-07-11 10:47:06 -0400 | [diff] [blame] | 1333 | pages[i] = find_or_create_page(inode->i_mapping, index + i, |
Johannes Weiner | e3a41a5 | 2012-01-10 15:07:55 -0800 | [diff] [blame] | 1334 | mask | __GFP_WRITE); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1335 | if (!pages[i]) { |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1336 | faili = i - 1; |
| 1337 | err = -ENOMEM; |
| 1338 | goto fail; |
| 1339 | } |
| 1340 | |
| 1341 | if (i == 0) |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1342 | err = prepare_uptodate_page(inode, pages[i], pos, |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1343 | force_uptodate); |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1344 | if (!err && i == num_pages - 1) |
| 1345 | err = prepare_uptodate_page(inode, pages[i], |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1346 | pos + write_bytes, false); |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1347 | if (err) { |
| 1348 | page_cache_release(pages[i]); |
Chris Mason | bb1591b4 | 2015-12-14 15:40:44 -0800 | [diff] [blame^] | 1349 | if (err == -EAGAIN) { |
| 1350 | err = 0; |
| 1351 | goto again; |
| 1352 | } |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1353 | faili = i - 1; |
| 1354 | goto fail; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1355 | } |
Chris Mason | ccd467d | 2007-06-28 15:57:36 -0400 | [diff] [blame] | 1356 | wait_on_page_writeback(pages[i]); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1357 | } |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 1358 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1359 | return 0; |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1360 | fail: |
| 1361 | while (faili >= 0) { |
| 1362 | unlock_page(pages[faili]); |
| 1363 | page_cache_release(pages[faili]); |
| 1364 | faili--; |
| 1365 | } |
| 1366 | return err; |
| 1367 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1368 | } |
| 1369 | |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1370 | /* |
| 1371 | * This function locks the extent and properly waits for data=ordered extents |
| 1372 | * to finish before allowing the pages to be modified if need. |
| 1373 | * |
| 1374 | * The return value: |
| 1375 | * 1 - the extent is locked |
| 1376 | * 0 - the extent is not locked, and everything is OK |
| 1377 | * -EAGAIN - need re-prepare the pages |
| 1378 | * the other < 0 number - Something wrong happens |
| 1379 | */ |
| 1380 | static noinline int |
| 1381 | lock_and_cleanup_extent_if_need(struct inode *inode, struct page **pages, |
| 1382 | size_t num_pages, loff_t pos, |
| 1383 | u64 *lockstart, u64 *lockend, |
| 1384 | struct extent_state **cached_state) |
| 1385 | { |
| 1386 | u64 start_pos; |
| 1387 | u64 last_pos; |
| 1388 | int i; |
| 1389 | int ret = 0; |
| 1390 | |
| 1391 | start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1); |
| 1392 | last_pos = start_pos + ((u64)num_pages << PAGE_CACHE_SHIFT) - 1; |
| 1393 | |
| 1394 | if (start_pos < inode->i_size) { |
| 1395 | struct btrfs_ordered_extent *ordered; |
| 1396 | lock_extent_bits(&BTRFS_I(inode)->io_tree, |
| 1397 | start_pos, last_pos, 0, cached_state); |
Miao Xie | b88935b | 2014-03-06 13:54:58 +0800 | [diff] [blame] | 1398 | ordered = btrfs_lookup_ordered_range(inode, start_pos, |
| 1399 | last_pos - start_pos + 1); |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1400 | if (ordered && |
| 1401 | ordered->file_offset + ordered->len > start_pos && |
| 1402 | ordered->file_offset <= last_pos) { |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1403 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
| 1404 | start_pos, last_pos, |
| 1405 | cached_state, GFP_NOFS); |
| 1406 | for (i = 0; i < num_pages; i++) { |
| 1407 | unlock_page(pages[i]); |
| 1408 | page_cache_release(pages[i]); |
| 1409 | } |
Miao Xie | b88935b | 2014-03-06 13:54:58 +0800 | [diff] [blame] | 1410 | btrfs_start_ordered_extent(inode, ordered, 1); |
| 1411 | btrfs_put_ordered_extent(ordered); |
| 1412 | return -EAGAIN; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1413 | } |
| 1414 | if (ordered) |
| 1415 | btrfs_put_ordered_extent(ordered); |
| 1416 | |
| 1417 | clear_extent_bit(&BTRFS_I(inode)->io_tree, start_pos, |
| 1418 | last_pos, EXTENT_DIRTY | EXTENT_DELALLOC | |
| 1419 | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, |
| 1420 | 0, 0, cached_state, GFP_NOFS); |
| 1421 | *lockstart = start_pos; |
| 1422 | *lockend = last_pos; |
| 1423 | ret = 1; |
| 1424 | } |
| 1425 | |
| 1426 | for (i = 0; i < num_pages; i++) { |
| 1427 | if (clear_page_dirty_for_io(pages[i])) |
| 1428 | account_page_redirty(pages[i]); |
| 1429 | set_page_extent_mapped(pages[i]); |
| 1430 | WARN_ON(!PageLocked(pages[i])); |
| 1431 | } |
| 1432 | |
| 1433 | return ret; |
| 1434 | } |
| 1435 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1436 | static noinline int check_can_nocow(struct inode *inode, loff_t pos, |
| 1437 | size_t *write_bytes) |
| 1438 | { |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1439 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 1440 | struct btrfs_ordered_extent *ordered; |
| 1441 | u64 lockstart, lockend; |
| 1442 | u64 num_bytes; |
| 1443 | int ret; |
| 1444 | |
Filipe Manana | 9ea24bb | 2014-10-29 11:57:59 +0000 | [diff] [blame] | 1445 | ret = btrfs_start_write_no_snapshoting(root); |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1446 | if (!ret) |
| 1447 | return -ENOSPC; |
| 1448 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1449 | lockstart = round_down(pos, root->sectorsize); |
Miao Xie | c933956 | 2014-02-27 13:58:04 +0800 | [diff] [blame] | 1450 | lockend = round_up(pos + *write_bytes, root->sectorsize) - 1; |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1451 | |
| 1452 | while (1) { |
| 1453 | lock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); |
| 1454 | ordered = btrfs_lookup_ordered_range(inode, lockstart, |
| 1455 | lockend - lockstart + 1); |
| 1456 | if (!ordered) { |
| 1457 | break; |
| 1458 | } |
| 1459 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); |
| 1460 | btrfs_start_ordered_extent(inode, ordered, 1); |
| 1461 | btrfs_put_ordered_extent(ordered); |
| 1462 | } |
| 1463 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1464 | num_bytes = lockend - lockstart + 1; |
Josef Bacik | 0036158 | 2013-08-14 14:02:47 -0400 | [diff] [blame] | 1465 | ret = can_nocow_extent(inode, lockstart, &num_bytes, NULL, NULL, NULL); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1466 | if (ret <= 0) { |
| 1467 | ret = 0; |
Filipe Manana | 9ea24bb | 2014-10-29 11:57:59 +0000 | [diff] [blame] | 1468 | btrfs_end_write_no_snapshoting(root); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1469 | } else { |
Miao Xie | c933956 | 2014-02-27 13:58:04 +0800 | [diff] [blame] | 1470 | *write_bytes = min_t(size_t, *write_bytes , |
| 1471 | num_bytes - pos + lockstart); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1472 | } |
| 1473 | |
| 1474 | unlock_extent(&BTRFS_I(inode)->io_tree, lockstart, lockend); |
| 1475 | |
| 1476 | return ret; |
| 1477 | } |
| 1478 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1479 | static noinline ssize_t __btrfs_buffered_write(struct file *file, |
| 1480 | struct iov_iter *i, |
| 1481 | loff_t pos) |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 1482 | { |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1483 | struct inode *inode = file_inode(file); |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 1484 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 1485 | struct page **pages = NULL; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1486 | struct extent_state *cached_state = NULL; |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1487 | u64 release_bytes = 0; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1488 | u64 lockstart; |
| 1489 | u64 lockend; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1490 | size_t num_written = 0; |
| 1491 | int nrptrs; |
Tsutomu Itoh | c914923 | 2011-03-30 00:57:23 +0000 | [diff] [blame] | 1492 | int ret = 0; |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1493 | bool only_release_metadata = false; |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1494 | bool force_page_uptodate = false; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1495 | bool need_unlock; |
Chris Mason | cb843a6 | 2008-10-03 12:30:02 -0400 | [diff] [blame] | 1496 | |
David Sterba | ed6078f | 2014-06-05 01:59:57 +0200 | [diff] [blame] | 1497 | nrptrs = min(DIV_ROUND_UP(iov_iter_count(i), PAGE_CACHE_SIZE), |
| 1498 | PAGE_CACHE_SIZE / (sizeof(struct page *))); |
Wu Fengguang | 142349f | 2011-12-16 12:32:57 -0500 | [diff] [blame] | 1499 | nrptrs = min(nrptrs, current->nr_dirtied_pause - current->nr_dirtied); |
| 1500 | nrptrs = max(nrptrs, 8); |
David Sterba | 31e818f | 2015-02-20 18:00:26 +0100 | [diff] [blame] | 1501 | pages = kmalloc_array(nrptrs, sizeof(struct page *), GFP_KERNEL); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1502 | if (!pages) |
| 1503 | return -ENOMEM; |
Chris Mason | ab93dbe | 2009-10-01 12:29:10 -0400 | [diff] [blame] | 1504 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1505 | while (iov_iter_count(i) > 0) { |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1506 | size_t offset = pos & (PAGE_CACHE_SIZE - 1); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1507 | size_t write_bytes = min(iov_iter_count(i), |
Josef Bacik | 11c65dc | 2010-05-23 11:07:21 -0400 | [diff] [blame] | 1508 | nrptrs * (size_t)PAGE_CACHE_SIZE - |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 1509 | offset); |
David Sterba | ed6078f | 2014-06-05 01:59:57 +0200 | [diff] [blame] | 1510 | size_t num_pages = DIV_ROUND_UP(write_bytes + offset, |
| 1511 | PAGE_CACHE_SIZE); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1512 | size_t reserve_bytes; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1513 | size_t dirty_pages; |
| 1514 | size_t copied; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1515 | |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 1516 | WARN_ON(num_pages > nrptrs); |
Chris Mason | 1832a6d | 2007-12-21 16:27:21 -0500 | [diff] [blame] | 1517 | |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1518 | /* |
| 1519 | * Fault pages before locking them in prepare_pages |
| 1520 | * to avoid recursive lock |
| 1521 | */ |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1522 | if (unlikely(iov_iter_fault_in_readable(i, write_bytes))) { |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1523 | ret = -EFAULT; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1524 | break; |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1525 | } |
| 1526 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1527 | reserve_bytes = num_pages << PAGE_CACHE_SHIFT; |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1528 | |
| 1529 | if (BTRFS_I(inode)->flags & (BTRFS_INODE_NODATACOW | |
| 1530 | BTRFS_INODE_PREALLOC)) { |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1531 | ret = check_can_nocow(inode, pos, &write_bytes); |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1532 | if (ret < 0) |
| 1533 | break; |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1534 | if (ret > 0) { |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1535 | /* |
| 1536 | * For nodata cow case, no need to reserve |
| 1537 | * data space. |
| 1538 | */ |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1539 | only_release_metadata = true; |
| 1540 | /* |
| 1541 | * our prealloc extent may be smaller than |
| 1542 | * write_bytes, so scale down. |
| 1543 | */ |
David Sterba | ed6078f | 2014-06-05 01:59:57 +0200 | [diff] [blame] | 1544 | num_pages = DIV_ROUND_UP(write_bytes + offset, |
| 1545 | PAGE_CACHE_SIZE); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1546 | reserve_bytes = num_pages << PAGE_CACHE_SHIFT; |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1547 | goto reserve_metadata; |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1548 | } |
| 1549 | } |
Qu Wenruo | 7cf5b97 | 2015-09-08 17:25:55 +0800 | [diff] [blame] | 1550 | ret = btrfs_check_data_free_space(inode, pos, write_bytes); |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1551 | if (ret < 0) |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1552 | break; |
Chris Mason | 1832a6d | 2007-12-21 16:27:21 -0500 | [diff] [blame] | 1553 | |
Qu Wenruo | d9d8b2a | 2015-09-08 17:22:43 +0800 | [diff] [blame] | 1554 | reserve_metadata: |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1555 | ret = btrfs_delalloc_reserve_metadata(inode, reserve_bytes); |
| 1556 | if (ret) { |
| 1557 | if (!only_release_metadata) |
Qu Wenruo | 7cf5b97 | 2015-09-08 17:25:55 +0800 | [diff] [blame] | 1558 | btrfs_free_reserved_data_space(inode, pos, |
| 1559 | write_bytes); |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1560 | else |
Filipe Manana | 9ea24bb | 2014-10-29 11:57:59 +0000 | [diff] [blame] | 1561 | btrfs_end_write_no_snapshoting(root); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1562 | break; |
| 1563 | } |
| 1564 | |
| 1565 | release_bytes = reserve_bytes; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1566 | need_unlock = false; |
| 1567 | again: |
Josef Bacik | 4a64001 | 2011-01-25 15:10:08 -0500 | [diff] [blame] | 1568 | /* |
| 1569 | * This is going to setup the pages array with the number of |
| 1570 | * pages we want, so we don't really need to worry about the |
| 1571 | * contents of pages from loop to loop |
| 1572 | */ |
Miao Xie | b37392e | 2013-12-10 19:25:03 +0800 | [diff] [blame] | 1573 | ret = prepare_pages(inode, pages, num_pages, |
| 1574 | pos, write_bytes, |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1575 | force_page_uptodate); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1576 | if (ret) |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1577 | break; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1578 | |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1579 | ret = lock_and_cleanup_extent_if_need(inode, pages, num_pages, |
| 1580 | pos, &lockstart, &lockend, |
| 1581 | &cached_state); |
| 1582 | if (ret < 0) { |
| 1583 | if (ret == -EAGAIN) |
| 1584 | goto again; |
| 1585 | break; |
| 1586 | } else if (ret > 0) { |
| 1587 | need_unlock = true; |
| 1588 | ret = 0; |
| 1589 | } |
| 1590 | |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1591 | copied = btrfs_copy_from_user(pos, num_pages, |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1592 | write_bytes, pages, i); |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1593 | |
| 1594 | /* |
| 1595 | * if we have trouble faulting in the pages, fall |
| 1596 | * back to one page at a time |
| 1597 | */ |
| 1598 | if (copied < write_bytes) |
| 1599 | nrptrs = 1; |
| 1600 | |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1601 | if (copied == 0) { |
| 1602 | force_page_uptodate = true; |
Chris Mason | b1bf862 | 2011-02-28 09:52:08 -0500 | [diff] [blame] | 1603 | dirty_pages = 0; |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1604 | } else { |
| 1605 | force_page_uptodate = false; |
David Sterba | ed6078f | 2014-06-05 01:59:57 +0200 | [diff] [blame] | 1606 | dirty_pages = DIV_ROUND_UP(copied + offset, |
| 1607 | PAGE_CACHE_SIZE); |
Josef Bacik | b6316429 | 2011-09-30 15:23:54 -0400 | [diff] [blame] | 1608 | } |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1609 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1610 | /* |
| 1611 | * If we had a short copy we need to release the excess delaloc |
| 1612 | * bytes we reserved. We need to increment outstanding_extents |
| 1613 | * because btrfs_delalloc_release_space will decrement it, but |
| 1614 | * we still have an outstanding extent for the chunk we actually |
| 1615 | * managed to copy. |
| 1616 | */ |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1617 | if (num_pages > dirty_pages) { |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1618 | release_bytes = (num_pages - dirty_pages) << |
| 1619 | PAGE_CACHE_SHIFT; |
Josef Bacik | 9e0baf6 | 2011-07-15 15:16:44 +0000 | [diff] [blame] | 1620 | if (copied > 0) { |
| 1621 | spin_lock(&BTRFS_I(inode)->lock); |
| 1622 | BTRFS_I(inode)->outstanding_extents++; |
| 1623 | spin_unlock(&BTRFS_I(inode)->lock); |
| 1624 | } |
Qu Wenruo | 485290a | 2015-10-29 17:28:46 +0800 | [diff] [blame] | 1625 | if (only_release_metadata) { |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1626 | btrfs_delalloc_release_metadata(inode, |
| 1627 | release_bytes); |
Qu Wenruo | 485290a | 2015-10-29 17:28:46 +0800 | [diff] [blame] | 1628 | } else { |
| 1629 | u64 __pos; |
| 1630 | |
| 1631 | __pos = round_down(pos, root->sectorsize) + |
| 1632 | (dirty_pages << PAGE_CACHE_SHIFT); |
| 1633 | btrfs_delalloc_release_space(inode, __pos, |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1634 | release_bytes); |
Qu Wenruo | 485290a | 2015-10-29 17:28:46 +0800 | [diff] [blame] | 1635 | } |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1636 | } |
| 1637 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1638 | release_bytes = dirty_pages << PAGE_CACHE_SHIFT; |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1639 | |
| 1640 | if (copied > 0) |
Josef Bacik | be1a12a | 2011-04-06 13:05:22 -0400 | [diff] [blame] | 1641 | ret = btrfs_dirty_pages(root, inode, pages, |
| 1642 | dirty_pages, pos, copied, |
| 1643 | NULL); |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1644 | if (need_unlock) |
| 1645 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
| 1646 | lockstart, lockend, &cached_state, |
| 1647 | GFP_NOFS); |
Miao Xie | f1de968 | 2014-01-09 10:06:10 +0800 | [diff] [blame] | 1648 | if (ret) { |
| 1649 | btrfs_drop_pages(pages, num_pages); |
Miao Xie | 376cc68 | 2013-12-10 19:25:04 +0800 | [diff] [blame] | 1650 | break; |
Miao Xie | f1de968 | 2014-01-09 10:06:10 +0800 | [diff] [blame] | 1651 | } |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1652 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1653 | release_bytes = 0; |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1654 | if (only_release_metadata) |
Filipe Manana | 9ea24bb | 2014-10-29 11:57:59 +0000 | [diff] [blame] | 1655 | btrfs_end_write_no_snapshoting(root); |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1656 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1657 | if (only_release_metadata && copied > 0) { |
David Sterba | f64c7b1 | 2015-02-24 19:07:26 +0100 | [diff] [blame] | 1658 | lockstart = round_down(pos, root->sectorsize); |
| 1659 | lockend = lockstart + |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1660 | (dirty_pages << PAGE_CACHE_SHIFT) - 1; |
| 1661 | |
| 1662 | set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, |
| 1663 | lockend, EXTENT_NORESERVE, NULL, |
| 1664 | NULL, GFP_NOFS); |
| 1665 | only_release_metadata = false; |
| 1666 | } |
| 1667 | |
Miao Xie | f1de968 | 2014-01-09 10:06:10 +0800 | [diff] [blame] | 1668 | btrfs_drop_pages(pages, num_pages); |
| 1669 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1670 | cond_resched(); |
| 1671 | |
Namjae Jeon | d0e1d66 | 2012-12-11 16:00:21 -0800 | [diff] [blame] | 1672 | balance_dirty_pages_ratelimited(inode->i_mapping); |
David Sterba | 707e8a0 | 2014-06-04 19:22:26 +0200 | [diff] [blame] | 1673 | if (dirty_pages < (root->nodesize >> PAGE_CACHE_SHIFT) + 1) |
Liu Bo | b53d3f5 | 2012-11-14 14:34:34 +0000 | [diff] [blame] | 1674 | btrfs_btree_balance_dirty(root); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1675 | |
Xin Zhong | 914ee29 | 2010-12-09 09:30:14 +0000 | [diff] [blame] | 1676 | pos += copied; |
| 1677 | num_written += copied; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1678 | } |
Chris Mason | 5b92ee7 | 2008-01-03 13:46:11 -0500 | [diff] [blame] | 1679 | |
Chris Mason | 8c2383c | 2007-06-18 09:57:58 -0400 | [diff] [blame] | 1680 | kfree(pages); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1681 | |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1682 | if (release_bytes) { |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1683 | if (only_release_metadata) { |
Filipe Manana | 9ea24bb | 2014-10-29 11:57:59 +0000 | [diff] [blame] | 1684 | btrfs_end_write_no_snapshoting(root); |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1685 | btrfs_delalloc_release_metadata(inode, release_bytes); |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1686 | } else { |
Qu Wenruo | 7cf5b97 | 2015-09-08 17:25:55 +0800 | [diff] [blame] | 1687 | btrfs_delalloc_release_space(inode, pos, release_bytes); |
Miao Xie | 8257b2d | 2014-03-06 13:38:19 +0800 | [diff] [blame] | 1688 | } |
Josef Bacik | 7ee9e44 | 2013-06-21 16:37:03 -0400 | [diff] [blame] | 1689 | } |
| 1690 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1691 | return num_written ? num_written : ret; |
| 1692 | } |
| 1693 | |
| 1694 | static ssize_t __btrfs_direct_write(struct kiocb *iocb, |
Al Viro | 0ae5e4d | 2014-03-03 22:09:39 -0500 | [diff] [blame] | 1695 | struct iov_iter *from, |
Al Viro | 0c94933 | 2014-03-22 06:51:37 -0400 | [diff] [blame] | 1696 | loff_t pos) |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1697 | { |
| 1698 | struct file *file = iocb->ki_filp; |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 1699 | struct inode *inode = file_inode(file); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1700 | ssize_t written; |
| 1701 | ssize_t written_buffered; |
| 1702 | loff_t endbyte; |
| 1703 | int err; |
| 1704 | |
Al Viro | 0c94933 | 2014-03-22 06:51:37 -0400 | [diff] [blame] | 1705 | written = generic_file_direct_write(iocb, from, pos); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1706 | |
Al Viro | 0c94933 | 2014-03-22 06:51:37 -0400 | [diff] [blame] | 1707 | if (written < 0 || !iov_iter_count(from)) |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1708 | return written; |
| 1709 | |
| 1710 | pos += written; |
Al Viro | 0ae5e4d | 2014-03-03 22:09:39 -0500 | [diff] [blame] | 1711 | written_buffered = __btrfs_buffered_write(file, from, pos); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1712 | if (written_buffered < 0) { |
| 1713 | err = written_buffered; |
| 1714 | goto out; |
| 1715 | } |
Filipe Manana | 075bdbd | 2014-10-09 21:18:55 +0100 | [diff] [blame] | 1716 | /* |
| 1717 | * Ensure all data is persisted. We want the next direct IO read to be |
| 1718 | * able to read what was just written. |
| 1719 | */ |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1720 | endbyte = pos + written_buffered - 1; |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 1721 | err = btrfs_fdatawrite_range(inode, pos, endbyte); |
Filipe Manana | 075bdbd | 2014-10-09 21:18:55 +0100 | [diff] [blame] | 1722 | if (err) |
| 1723 | goto out; |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 1724 | err = filemap_fdatawait_range(inode->i_mapping, pos, endbyte); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1725 | if (err) |
| 1726 | goto out; |
| 1727 | written += written_buffered; |
Al Viro | 867c4f9 | 2014-02-11 19:31:06 -0500 | [diff] [blame] | 1728 | iocb->ki_pos = pos + written_buffered; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1729 | invalidate_mapping_pages(file->f_mapping, pos >> PAGE_CACHE_SHIFT, |
| 1730 | endbyte >> PAGE_CACHE_SHIFT); |
| 1731 | out: |
| 1732 | return written ? written : err; |
| 1733 | } |
| 1734 | |
Josef Bacik | 6c760c0 | 2012-11-09 10:53:21 -0500 | [diff] [blame] | 1735 | static void update_time_for_write(struct inode *inode) |
| 1736 | { |
| 1737 | struct timespec now; |
| 1738 | |
| 1739 | if (IS_NOCMTIME(inode)) |
| 1740 | return; |
| 1741 | |
| 1742 | now = current_fs_time(inode->i_sb); |
| 1743 | if (!timespec_equal(&inode->i_mtime, &now)) |
| 1744 | inode->i_mtime = now; |
| 1745 | |
| 1746 | if (!timespec_equal(&inode->i_ctime, &now)) |
| 1747 | inode->i_ctime = now; |
| 1748 | |
| 1749 | if (IS_I_VERSION(inode)) |
| 1750 | inode_inc_iversion(inode); |
| 1751 | } |
| 1752 | |
Al Viro | b30ac0f | 2014-04-03 14:29:04 -0400 | [diff] [blame] | 1753 | static ssize_t btrfs_file_write_iter(struct kiocb *iocb, |
| 1754 | struct iov_iter *from) |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1755 | { |
| 1756 | struct file *file = iocb->ki_filp; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1757 | struct inode *inode = file_inode(file); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1758 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Miao Xie | 0c1a98c | 2011-09-11 10:52:24 -0400 | [diff] [blame] | 1759 | u64 start_pos; |
Qu Wenruo | 3ac0d7b | 2014-03-27 02:51:58 +0000 | [diff] [blame] | 1760 | u64 end_pos; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1761 | ssize_t num_written = 0; |
Josef Bacik | b812ce2 | 2012-11-16 13:56:32 -0500 | [diff] [blame] | 1762 | bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1763 | ssize_t err; |
| 1764 | loff_t pos; |
| 1765 | size_t count; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1766 | |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1767 | mutex_lock(&inode->i_mutex); |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1768 | err = generic_write_checks(iocb, from); |
| 1769 | if (err <= 0) { |
| 1770 | mutex_unlock(&inode->i_mutex); |
| 1771 | return err; |
| 1772 | } |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1773 | |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 1774 | current->backing_dev_info = inode_to_bdi(inode); |
Jan Kara | 5fa8e0a | 2015-05-21 16:05:53 +0200 | [diff] [blame] | 1775 | err = file_remove_privs(file); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1776 | if (err) { |
| 1777 | mutex_unlock(&inode->i_mutex); |
| 1778 | goto out; |
| 1779 | } |
| 1780 | |
| 1781 | /* |
| 1782 | * If BTRFS flips readonly due to some impossible error |
| 1783 | * (fs_info->fs_state now has BTRFS_SUPER_FLAG_ERROR), |
| 1784 | * although we have opened a file as writable, we have |
| 1785 | * to stop this write operation to ensure FS consistency. |
| 1786 | */ |
Miao Xie | 87533c4 | 2013-01-29 10:14:48 +0000 | [diff] [blame] | 1787 | if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state)) { |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1788 | mutex_unlock(&inode->i_mutex); |
| 1789 | err = -EROFS; |
| 1790 | goto out; |
| 1791 | } |
| 1792 | |
Josef Bacik | 6c760c0 | 2012-11-09 10:53:21 -0500 | [diff] [blame] | 1793 | /* |
| 1794 | * We reserve space for updating the inode when we reserve space for the |
| 1795 | * extent we are going to write, so we will enospc out there. We don't |
| 1796 | * need to start yet another transaction to update the inode as we will |
| 1797 | * update the inode when we finish writing whatever data we write. |
| 1798 | */ |
| 1799 | update_time_for_write(inode); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1800 | |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1801 | pos = iocb->ki_pos; |
| 1802 | count = iov_iter_count(from); |
Miao Xie | 0c1a98c | 2011-09-11 10:52:24 -0400 | [diff] [blame] | 1803 | start_pos = round_down(pos, root->sectorsize); |
| 1804 | if (start_pos > i_size_read(inode)) { |
Qu Wenruo | 3ac0d7b | 2014-03-27 02:51:58 +0000 | [diff] [blame] | 1805 | /* Expand hole size to cover write data, preventing empty gap */ |
Qu Wenruo | c5f7d0b | 2014-04-15 10:41:00 +0800 | [diff] [blame] | 1806 | end_pos = round_up(pos + count, root->sectorsize); |
Qu Wenruo | 3ac0d7b | 2014-03-27 02:51:58 +0000 | [diff] [blame] | 1807 | err = btrfs_cont_expand(inode, i_size_read(inode), end_pos); |
Miao Xie | 0c1a98c | 2011-09-11 10:52:24 -0400 | [diff] [blame] | 1808 | if (err) { |
| 1809 | mutex_unlock(&inode->i_mutex); |
| 1810 | goto out; |
| 1811 | } |
| 1812 | } |
| 1813 | |
Josef Bacik | b812ce2 | 2012-11-16 13:56:32 -0500 | [diff] [blame] | 1814 | if (sync) |
| 1815 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
| 1816 | |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1817 | if (iocb->ki_flags & IOCB_DIRECT) { |
Al Viro | b30ac0f | 2014-04-03 14:29:04 -0400 | [diff] [blame] | 1818 | num_written = __btrfs_direct_write(iocb, from, pos); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1819 | } else { |
Al Viro | b30ac0f | 2014-04-03 14:29:04 -0400 | [diff] [blame] | 1820 | num_written = __btrfs_buffered_write(file, from, pos); |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1821 | if (num_written > 0) |
Al Viro | 867c4f9 | 2014-02-11 19:31:06 -0500 | [diff] [blame] | 1822 | iocb->ki_pos = pos + num_written; |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1823 | } |
| 1824 | |
| 1825 | mutex_unlock(&inode->i_mutex); |
Chris Mason | 2ff3e9b | 2007-10-29 14:36:41 -0400 | [diff] [blame] | 1826 | |
Chris Mason | 5a3f23d | 2009-03-31 13:27:11 -0400 | [diff] [blame] | 1827 | /* |
Josef Bacik | 6c760c0 | 2012-11-09 10:53:21 -0500 | [diff] [blame] | 1828 | * We also have to set last_sub_trans to the current log transid, |
| 1829 | * otherwise subsequent syncs to a file that's been synced in this |
| 1830 | * transaction will appear to have already occured. |
Chris Mason | 5a3f23d | 2009-03-31 13:27:11 -0400 | [diff] [blame] | 1831 | */ |
Filipe Manana | 2f2ff0e | 2015-03-20 17:19:46 +0000 | [diff] [blame] | 1832 | spin_lock(&BTRFS_I(inode)->lock); |
Josef Bacik | 6c760c0 | 2012-11-09 10:53:21 -0500 | [diff] [blame] | 1833 | BTRFS_I(inode)->last_sub_trans = root->log_transid; |
Filipe Manana | 2f2ff0e | 2015-03-20 17:19:46 +0000 | [diff] [blame] | 1834 | spin_unlock(&BTRFS_I(inode)->lock); |
Christoph Hellwig | 02afc27 | 2013-09-04 15:04:40 +0200 | [diff] [blame] | 1835 | if (num_written > 0) { |
Josef Bacik | d0215f3 | 2011-01-25 14:57:24 -0500 | [diff] [blame] | 1836 | err = generic_write_sync(file, pos, num_written); |
Dan Carpenter | 45d4f85 | 2014-04-03 14:47:17 -0700 | [diff] [blame] | 1837 | if (err < 0) |
Chris Mason | 2ff3e9b | 2007-10-29 14:36:41 -0400 | [diff] [blame] | 1838 | num_written = err; |
| 1839 | } |
Miao Xie | 0a3404d | 2013-01-28 12:34:55 +0000 | [diff] [blame] | 1840 | |
Josef Bacik | b812ce2 | 2012-11-16 13:56:32 -0500 | [diff] [blame] | 1841 | if (sync) |
| 1842 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
Miao Xie | 0a3404d | 2013-01-28 12:34:55 +0000 | [diff] [blame] | 1843 | out: |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1844 | current->backing_dev_info = NULL; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1845 | return num_written ? num_written : err; |
| 1846 | } |
| 1847 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 1848 | int btrfs_release_file(struct inode *inode, struct file *filp) |
Mingming | e1b81e6 | 2008-05-27 10:55:43 -0400 | [diff] [blame] | 1849 | { |
Sage Weil | 6bf13c0 | 2008-06-10 10:07:39 -0400 | [diff] [blame] | 1850 | if (filp->private_data) |
| 1851 | btrfs_ioctl_trans_end(filp); |
Chris Mason | f6dc45c | 2014-08-20 07:15:33 -0700 | [diff] [blame] | 1852 | /* |
| 1853 | * ordered_data_close is set by settattr when we are about to truncate |
| 1854 | * a file from a non-zero size to a zero size. This tries to |
| 1855 | * flush down new bytes that may have been written if the |
| 1856 | * application were using truncate to replace a file in place. |
| 1857 | */ |
| 1858 | if (test_and_clear_bit(BTRFS_INODE_ORDERED_DATA_CLOSE, |
| 1859 | &BTRFS_I(inode)->runtime_flags)) |
| 1860 | filemap_flush(inode->i_mapping); |
Mingming | e1b81e6 | 2008-05-27 10:55:43 -0400 | [diff] [blame] | 1861 | return 0; |
| 1862 | } |
| 1863 | |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1864 | static int start_ordered_ops(struct inode *inode, loff_t start, loff_t end) |
| 1865 | { |
| 1866 | int ret; |
| 1867 | |
| 1868 | atomic_inc(&BTRFS_I(inode)->sync_writers); |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 1869 | ret = btrfs_fdatawrite_range(inode, start, end); |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1870 | atomic_dec(&BTRFS_I(inode)->sync_writers); |
| 1871 | |
| 1872 | return ret; |
| 1873 | } |
| 1874 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 1875 | /* |
| 1876 | * fsync call for both files and directories. This logs the inode into |
| 1877 | * the tree log instead of forcing full commits whenever possible. |
| 1878 | * |
| 1879 | * It needs to call filemap_fdatawait so that all ordered extent updates are |
| 1880 | * in the metadata btree are up to date for copying to the log. |
| 1881 | * |
| 1882 | * It drops the inode mutex before doing the tree log commit. This is an |
| 1883 | * important optimization for directories because holding the mutex prevents |
| 1884 | * new operations on the dir while we write to disk. |
| 1885 | */ |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 1886 | int btrfs_sync_file(struct file *file, loff_t start, loff_t end, int datasync) |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1887 | { |
Christoph Hellwig | 7ea8085 | 2010-05-26 17:53:25 +0200 | [diff] [blame] | 1888 | struct dentry *dentry = file->f_path.dentry; |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 1889 | struct inode *inode = d_inode(dentry); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1890 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1891 | struct btrfs_trans_handle *trans; |
Miao Xie | 8b050d3 | 2014-02-20 18:08:58 +0800 | [diff] [blame] | 1892 | struct btrfs_log_ctx ctx; |
| 1893 | int ret = 0; |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 1894 | bool full_sync = 0; |
David Sterba | 9dcbeed | 2015-11-09 11:44:45 +0100 | [diff] [blame] | 1895 | u64 len; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1896 | |
David Sterba | 9dcbeed | 2015-11-09 11:44:45 +0100 | [diff] [blame] | 1897 | /* |
| 1898 | * The range length can be represented by u64, we have to do the typecasts |
| 1899 | * to avoid signed overflow if it's [0, LLONG_MAX] eg. from fsync() |
| 1900 | */ |
| 1901 | len = (u64)end - (u64)start + 1; |
liubo | 1abe9b8 | 2011-03-24 11:18:59 +0000 | [diff] [blame] | 1902 | trace_btrfs_sync_file(file, datasync); |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 1903 | |
Miao Xie | 90abccf | 2012-09-13 04:53:47 -0600 | [diff] [blame] | 1904 | /* |
| 1905 | * We write the dirty pages in the range and wait until they complete |
| 1906 | * out of the ->i_mutex. If so, we can flush the dirty pages by |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 1907 | * multi-task, and make the performance up. See |
| 1908 | * btrfs_wait_ordered_range for an explanation of the ASYNC check. |
Miao Xie | 90abccf | 2012-09-13 04:53:47 -0600 | [diff] [blame] | 1909 | */ |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1910 | ret = start_ordered_ops(inode, start, end); |
Miao Xie | 90abccf | 2012-09-13 04:53:47 -0600 | [diff] [blame] | 1911 | if (ret) |
| 1912 | return ret; |
| 1913 | |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 1914 | mutex_lock(&inode->i_mutex); |
Miao Xie | 2ecb792 | 2012-09-06 04:04:27 -0600 | [diff] [blame] | 1915 | atomic_inc(&root->log_batch); |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 1916 | full_sync = test_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
| 1917 | &BTRFS_I(inode)->runtime_flags); |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1918 | /* |
| 1919 | * We might have have had more pages made dirty after calling |
| 1920 | * start_ordered_ops and before acquiring the inode's i_mutex. |
| 1921 | */ |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 1922 | if (full_sync) { |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1923 | /* |
| 1924 | * For a full sync, we need to make sure any ordered operations |
| 1925 | * start and finish before we start logging the inode, so that |
| 1926 | * all extents are persisted and the respective file extent |
| 1927 | * items are in the fs/subvol btree. |
| 1928 | */ |
Filipe Manana | b659ef0 | 2015-03-31 14:16:52 +0100 | [diff] [blame] | 1929 | ret = btrfs_wait_ordered_range(inode, start, len); |
Filipe Manana | 669249e | 2014-09-02 11:09:58 +0100 | [diff] [blame] | 1930 | } else { |
| 1931 | /* |
| 1932 | * Start any new ordered operations before starting to log the |
| 1933 | * inode. We will wait for them to finish in btrfs_sync_log(). |
| 1934 | * |
| 1935 | * Right before acquiring the inode's mutex, we might have new |
| 1936 | * writes dirtying pages, which won't immediately start the |
| 1937 | * respective ordered operations - that is done through the |
| 1938 | * fill_delalloc callbacks invoked from the writepage and |
| 1939 | * writepages address space operations. So make sure we start |
| 1940 | * all ordered operations before starting to log our inode. Not |
| 1941 | * doing this means that while logging the inode, writeback |
| 1942 | * could start and invoke writepage/writepages, which would call |
| 1943 | * the fill_delalloc callbacks (cow_file_range, |
| 1944 | * submit_compressed_extents). These callbacks add first an |
| 1945 | * extent map to the modified list of extents and then create |
| 1946 | * the respective ordered operation, which means in |
| 1947 | * tree-log.c:btrfs_log_inode() we might capture all existing |
| 1948 | * ordered operations (with btrfs_get_logged_extents()) before |
| 1949 | * the fill_delalloc callback adds its ordered operation, and by |
| 1950 | * the time we visit the modified list of extent maps (with |
| 1951 | * btrfs_log_changed_extents()), we see and process the extent |
| 1952 | * map they created. We then use the extent map to construct a |
| 1953 | * file extent item for logging without waiting for the |
| 1954 | * respective ordered operation to finish - this file extent |
| 1955 | * item points to a disk location that might not have yet been |
| 1956 | * written to, containing random data - so after a crash a log |
| 1957 | * replay will make our inode have file extent items that point |
| 1958 | * to disk locations containing invalid data, as we returned |
| 1959 | * success to userspace without waiting for the respective |
| 1960 | * ordered operation to finish, because it wasn't captured by |
| 1961 | * btrfs_get_logged_extents(). |
| 1962 | */ |
| 1963 | ret = start_ordered_ops(inode, start, end); |
| 1964 | } |
| 1965 | if (ret) { |
| 1966 | mutex_unlock(&inode->i_mutex); |
| 1967 | goto out; |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 1968 | } |
Miao Xie | 2ecb792 | 2012-09-06 04:04:27 -0600 | [diff] [blame] | 1969 | atomic_inc(&root->log_batch); |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 1970 | |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 1971 | /* |
Filipe Manana | 3a8b36f | 2015-03-01 20:36:00 +0000 | [diff] [blame] | 1972 | * If the last transaction that changed this file was before the current |
| 1973 | * transaction and we have the full sync flag set in our inode, we can |
| 1974 | * bail out now without any syncing. |
| 1975 | * |
| 1976 | * Note that we can't bail out if the full sync flag isn't set. This is |
| 1977 | * because when the full sync flag is set we start all ordered extents |
| 1978 | * and wait for them to fully complete - when they complete they update |
| 1979 | * the inode's last_trans field through: |
| 1980 | * |
| 1981 | * btrfs_finish_ordered_io() -> |
| 1982 | * btrfs_update_inode_fallback() -> |
| 1983 | * btrfs_update_inode() -> |
| 1984 | * btrfs_set_inode_last_trans() |
| 1985 | * |
| 1986 | * So we are sure that last_trans is up to date and can do this check to |
| 1987 | * bail out safely. For the fast path, when the full sync flag is not |
| 1988 | * set in our inode, we can not do it because we start only our ordered |
| 1989 | * extents and don't wait for them to complete (that is when |
| 1990 | * btrfs_finish_ordered_io runs), so here at this point their last_trans |
| 1991 | * value might be less than or equals to fs_info->last_trans_committed, |
| 1992 | * and setting a speculative last_trans for an inode when a buffered |
| 1993 | * write is made (such as fs_info->generation + 1 for example) would not |
| 1994 | * be reliable since after setting the value and before fsync is called |
| 1995 | * any number of transactions can start and commit (transaction kthread |
| 1996 | * commits the current transaction periodically), and a transaction |
| 1997 | * commit does not start nor waits for ordered extents to complete. |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 1998 | */ |
Josef Bacik | a4abeea | 2011-04-11 17:25:13 -0400 | [diff] [blame] | 1999 | smp_mb(); |
Josef Bacik | 22ee698 | 2012-05-29 16:57:49 -0400 | [diff] [blame] | 2000 | if (btrfs_inode_in_log(inode, root->fs_info->generation) || |
Filipe Manana | b659ef0 | 2015-03-31 14:16:52 +0100 | [diff] [blame] | 2001 | (BTRFS_I(inode)->last_trans <= |
| 2002 | root->fs_info->last_trans_committed && |
| 2003 | (full_sync || |
| 2004 | !btrfs_have_ordered_extents_in_range(inode, start, len)))) { |
Josef Bacik | 5dc562c | 2012-08-17 13:14:17 -0400 | [diff] [blame] | 2005 | /* |
| 2006 | * We'v had everything committed since the last time we were |
| 2007 | * modified so clear this flag in case it was set for whatever |
| 2008 | * reason, it's no longer relevant. |
| 2009 | */ |
| 2010 | clear_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
| 2011 | &BTRFS_I(inode)->runtime_flags); |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 2012 | mutex_unlock(&inode->i_mutex); |
Josef Bacik | 15ee9bc | 2007-08-10 16:22:09 -0400 | [diff] [blame] | 2013 | goto out; |
| 2014 | } |
Josef Bacik | 15ee9bc | 2007-08-10 16:22:09 -0400 | [diff] [blame] | 2015 | |
| 2016 | /* |
Chris Mason | a52d9a8 | 2007-08-27 16:49:44 -0400 | [diff] [blame] | 2017 | * ok we haven't committed the transaction yet, lets do a commit |
| 2018 | */ |
Dan Carpenter | 6f902af | 2010-05-29 09:49:07 +0000 | [diff] [blame] | 2019 | if (file->private_data) |
Sage Weil | 6bf13c0 | 2008-06-10 10:07:39 -0400 | [diff] [blame] | 2020 | btrfs_ioctl_trans_end(file); |
| 2021 | |
Josef Bacik | 5039edd | 2014-01-15 13:34:13 -0500 | [diff] [blame] | 2022 | /* |
| 2023 | * We use start here because we will need to wait on the IO to complete |
| 2024 | * in btrfs_sync_log, which could require joining a transaction (for |
| 2025 | * example checking cross references in the nocow path). If we use join |
| 2026 | * here we could get into a situation where we're waiting on IO to |
| 2027 | * happen that is blocked on a transaction trying to commit. With start |
| 2028 | * we inc the extwriter counter, so we wait for all extwriters to exit |
| 2029 | * before we start blocking join'ers. This comment is to keep somebody |
| 2030 | * from thinking they are super smart and changing this to |
| 2031 | * btrfs_join_transaction *cough*Josef*cough*. |
| 2032 | */ |
Yan, Zheng | a22285a | 2010-05-16 10:48:46 -0400 | [diff] [blame] | 2033 | trans = btrfs_start_transaction(root, 0); |
| 2034 | if (IS_ERR(trans)) { |
| 2035 | ret = PTR_ERR(trans); |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 2036 | mutex_unlock(&inode->i_mutex); |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2037 | goto out; |
| 2038 | } |
Josef Bacik | 5039edd | 2014-01-15 13:34:13 -0500 | [diff] [blame] | 2039 | trans->sync = true; |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2040 | |
Miao Xie | 8b050d3 | 2014-02-20 18:08:58 +0800 | [diff] [blame] | 2041 | btrfs_init_log_ctx(&ctx); |
| 2042 | |
Filipe Manana | 49dae1b | 2014-09-06 22:34:39 +0100 | [diff] [blame] | 2043 | ret = btrfs_log_dentry_safe(trans, root, dentry, start, end, &ctx); |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 2044 | if (ret < 0) { |
Filipe David Borba Manana | a0634be | 2013-09-11 20:36:44 +0100 | [diff] [blame] | 2045 | /* Fallthrough and commit/free transaction. */ |
| 2046 | ret = 1; |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 2047 | } |
Chris Mason | 49eb7e4 | 2008-09-11 15:53:12 -0400 | [diff] [blame] | 2048 | |
| 2049 | /* we've logged all the items and now have a consistent |
| 2050 | * version of the file in the log. It is possible that |
| 2051 | * someone will come in and modify the file, but that's |
| 2052 | * fine because the log is consistent on disk, and we |
| 2053 | * have references to all of the file's extents |
| 2054 | * |
| 2055 | * It is possible that someone will come in and log the |
| 2056 | * file again, but that will end up using the synchronization |
| 2057 | * inside btrfs_sync_log to keep things safe. |
| 2058 | */ |
Josef Bacik | 02c24a8 | 2011-07-16 20:44:56 -0400 | [diff] [blame] | 2059 | mutex_unlock(&inode->i_mutex); |
Chris Mason | 49eb7e4 | 2008-09-11 15:53:12 -0400 | [diff] [blame] | 2060 | |
Filipe Manana | 8407f55 | 2014-09-05 15:14:39 +0100 | [diff] [blame] | 2061 | /* |
| 2062 | * If any of the ordered extents had an error, just return it to user |
| 2063 | * space, so that the application knows some writes didn't succeed and |
| 2064 | * can take proper action (retry for e.g.). Blindly committing the |
| 2065 | * transaction in this case, would fool userspace that everything was |
| 2066 | * successful. And we also want to make sure our log doesn't contain |
| 2067 | * file extent items pointing to extents that weren't fully written to - |
| 2068 | * just like in the non fast fsync path, where we check for the ordered |
| 2069 | * operation's error flag before writing to the log tree and return -EIO |
| 2070 | * if any of them had this flag set (btrfs_wait_ordered_range) - |
| 2071 | * therefore we need to check for errors in the ordered operations, |
| 2072 | * which are indicated by ctx.io_err. |
| 2073 | */ |
| 2074 | if (ctx.io_err) { |
| 2075 | btrfs_end_transaction(trans, root); |
| 2076 | ret = ctx.io_err; |
| 2077 | goto out; |
| 2078 | } |
| 2079 | |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 2080 | if (ret != BTRFS_NO_LOG_SYNC) { |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2081 | if (!ret) { |
Miao Xie | 8b050d3 | 2014-02-20 18:08:58 +0800 | [diff] [blame] | 2082 | ret = btrfs_sync_log(trans, root, &ctx); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2083 | if (!ret) { |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 2084 | ret = btrfs_end_transaction(trans, root); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2085 | goto out; |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 2086 | } |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 2087 | } |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2088 | if (!full_sync) { |
David Sterba | 9dcbeed | 2015-11-09 11:44:45 +0100 | [diff] [blame] | 2089 | ret = btrfs_wait_ordered_range(inode, start, len); |
Filipe Manana | b05fd87 | 2014-05-29 23:31:39 +0100 | [diff] [blame] | 2090 | if (ret) { |
| 2091 | btrfs_end_transaction(trans, root); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2092 | goto out; |
Filipe Manana | b05fd87 | 2014-05-29 23:31:39 +0100 | [diff] [blame] | 2093 | } |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2094 | } |
| 2095 | ret = btrfs_commit_transaction(trans, root); |
Chris Mason | 257c62e | 2009-10-13 13:21:08 -0400 | [diff] [blame] | 2096 | } else { |
| 2097 | ret = btrfs_end_transaction(trans, root); |
Chris Mason | e02119d | 2008-09-05 16:13:11 -0400 | [diff] [blame] | 2098 | } |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2099 | out: |
Roel Kluin | 014e4ac | 2010-01-29 10:42:11 +0000 | [diff] [blame] | 2100 | return ret > 0 ? -EIO : ret; |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2101 | } |
| 2102 | |
Alexey Dobriyan | f0f37e2 | 2009-09-27 22:29:37 +0400 | [diff] [blame] | 2103 | static const struct vm_operations_struct btrfs_file_vm_ops = { |
Chris Mason | 92fee66 | 2007-07-25 12:31:35 -0400 | [diff] [blame] | 2104 | .fault = filemap_fault, |
Kirill A. Shutemov | f182036 | 2014-04-07 15:37:19 -0700 | [diff] [blame] | 2105 | .map_pages = filemap_map_pages, |
Chris Mason | 9ebefb18 | 2007-06-15 13:50:00 -0400 | [diff] [blame] | 2106 | .page_mkwrite = btrfs_page_mkwrite, |
| 2107 | }; |
| 2108 | |
| 2109 | static int btrfs_file_mmap(struct file *filp, struct vm_area_struct *vma) |
| 2110 | { |
Miao Xie | 058a457 | 2010-05-20 07:21:50 +0000 | [diff] [blame] | 2111 | struct address_space *mapping = filp->f_mapping; |
| 2112 | |
| 2113 | if (!mapping->a_ops->readpage) |
| 2114 | return -ENOEXEC; |
| 2115 | |
Chris Mason | 9ebefb18 | 2007-06-15 13:50:00 -0400 | [diff] [blame] | 2116 | file_accessed(filp); |
Miao Xie | 058a457 | 2010-05-20 07:21:50 +0000 | [diff] [blame] | 2117 | vma->vm_ops = &btrfs_file_vm_ops; |
Miao Xie | 058a457 | 2010-05-20 07:21:50 +0000 | [diff] [blame] | 2118 | |
Chris Mason | 9ebefb18 | 2007-06-15 13:50:00 -0400 | [diff] [blame] | 2119 | return 0; |
| 2120 | } |
| 2121 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2122 | static int hole_mergeable(struct inode *inode, struct extent_buffer *leaf, |
| 2123 | int slot, u64 start, u64 end) |
| 2124 | { |
| 2125 | struct btrfs_file_extent_item *fi; |
| 2126 | struct btrfs_key key; |
| 2127 | |
| 2128 | if (slot < 0 || slot >= btrfs_header_nritems(leaf)) |
| 2129 | return 0; |
| 2130 | |
| 2131 | btrfs_item_key_to_cpu(leaf, &key, slot); |
| 2132 | if (key.objectid != btrfs_ino(inode) || |
| 2133 | key.type != BTRFS_EXTENT_DATA_KEY) |
| 2134 | return 0; |
| 2135 | |
| 2136 | fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); |
| 2137 | |
| 2138 | if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG) |
| 2139 | return 0; |
| 2140 | |
| 2141 | if (btrfs_file_extent_disk_bytenr(leaf, fi)) |
| 2142 | return 0; |
| 2143 | |
| 2144 | if (key.offset == end) |
| 2145 | return 1; |
| 2146 | if (key.offset + btrfs_file_extent_num_bytes(leaf, fi) == start) |
| 2147 | return 1; |
| 2148 | return 0; |
| 2149 | } |
| 2150 | |
| 2151 | static int fill_holes(struct btrfs_trans_handle *trans, struct inode *inode, |
| 2152 | struct btrfs_path *path, u64 offset, u64 end) |
| 2153 | { |
| 2154 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 2155 | struct extent_buffer *leaf; |
| 2156 | struct btrfs_file_extent_item *fi; |
| 2157 | struct extent_map *hole_em; |
| 2158 | struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; |
| 2159 | struct btrfs_key key; |
| 2160 | int ret; |
| 2161 | |
Josef Bacik | 16e7549 | 2013-10-22 12:18:51 -0400 | [diff] [blame] | 2162 | if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) |
| 2163 | goto out; |
| 2164 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2165 | key.objectid = btrfs_ino(inode); |
| 2166 | key.type = BTRFS_EXTENT_DATA_KEY; |
| 2167 | key.offset = offset; |
| 2168 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2169 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
| 2170 | if (ret < 0) |
| 2171 | return ret; |
| 2172 | BUG_ON(!ret); |
| 2173 | |
| 2174 | leaf = path->nodes[0]; |
| 2175 | if (hole_mergeable(inode, leaf, path->slots[0]-1, offset, end)) { |
| 2176 | u64 num_bytes; |
| 2177 | |
| 2178 | path->slots[0]--; |
| 2179 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 2180 | struct btrfs_file_extent_item); |
| 2181 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + |
| 2182 | end - offset; |
| 2183 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); |
| 2184 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); |
| 2185 | btrfs_set_file_extent_offset(leaf, fi, 0); |
| 2186 | btrfs_mark_buffer_dirty(leaf); |
| 2187 | goto out; |
| 2188 | } |
| 2189 | |
chandan | 1707e26 | 2014-07-01 12:04:28 +0530 | [diff] [blame] | 2190 | if (hole_mergeable(inode, leaf, path->slots[0], offset, end)) { |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2191 | u64 num_bytes; |
| 2192 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2193 | key.offset = offset; |
Daniel Dressler | b7a0365 | 2014-11-12 13:43:09 +0900 | [diff] [blame] | 2194 | btrfs_set_item_key_safe(root->fs_info, path, &key); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2195 | fi = btrfs_item_ptr(leaf, path->slots[0], |
| 2196 | struct btrfs_file_extent_item); |
| 2197 | num_bytes = btrfs_file_extent_num_bytes(leaf, fi) + end - |
| 2198 | offset; |
| 2199 | btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes); |
| 2200 | btrfs_set_file_extent_ram_bytes(leaf, fi, num_bytes); |
| 2201 | btrfs_set_file_extent_offset(leaf, fi, 0); |
| 2202 | btrfs_mark_buffer_dirty(leaf); |
| 2203 | goto out; |
| 2204 | } |
| 2205 | btrfs_release_path(path); |
| 2206 | |
| 2207 | ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset, |
| 2208 | 0, 0, end - offset, 0, end - offset, |
| 2209 | 0, 0, 0); |
| 2210 | if (ret) |
| 2211 | return ret; |
| 2212 | |
| 2213 | out: |
| 2214 | btrfs_release_path(path); |
| 2215 | |
| 2216 | hole_em = alloc_extent_map(); |
| 2217 | if (!hole_em) { |
| 2218 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); |
| 2219 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
| 2220 | &BTRFS_I(inode)->runtime_flags); |
| 2221 | } else { |
| 2222 | hole_em->start = offset; |
| 2223 | hole_em->len = end - offset; |
Josef Bacik | cc95bef | 2013-04-04 14:31:27 -0400 | [diff] [blame] | 2224 | hole_em->ram_bytes = hole_em->len; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2225 | hole_em->orig_start = offset; |
| 2226 | |
| 2227 | hole_em->block_start = EXTENT_MAP_HOLE; |
| 2228 | hole_em->block_len = 0; |
Josef Bacik | b493968 | 2012-12-03 10:31:19 -0500 | [diff] [blame] | 2229 | hole_em->orig_block_len = 0; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2230 | hole_em->bdev = root->fs_info->fs_devices->latest_bdev; |
| 2231 | hole_em->compress_type = BTRFS_COMPRESS_NONE; |
| 2232 | hole_em->generation = trans->transid; |
| 2233 | |
| 2234 | do { |
| 2235 | btrfs_drop_extent_cache(inode, offset, end - 1, 0); |
| 2236 | write_lock(&em_tree->lock); |
Josef Bacik | 09a2a8f9 | 2013-04-05 16:51:15 -0400 | [diff] [blame] | 2237 | ret = add_extent_mapping(em_tree, hole_em, 1); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2238 | write_unlock(&em_tree->lock); |
| 2239 | } while (ret == -EEXIST); |
| 2240 | free_extent_map(hole_em); |
| 2241 | if (ret) |
| 2242 | set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, |
| 2243 | &BTRFS_I(inode)->runtime_flags); |
| 2244 | } |
| 2245 | |
| 2246 | return 0; |
| 2247 | } |
| 2248 | |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2249 | /* |
| 2250 | * Find a hole extent on given inode and change start/len to the end of hole |
| 2251 | * extent.(hole/vacuum extent whose em->start <= start && |
| 2252 | * em->start + em->len > start) |
| 2253 | * When a hole extent is found, return 1 and modify start/len. |
| 2254 | */ |
| 2255 | static int find_first_non_hole(struct inode *inode, u64 *start, u64 *len) |
| 2256 | { |
| 2257 | struct extent_map *em; |
| 2258 | int ret = 0; |
| 2259 | |
| 2260 | em = btrfs_get_extent(inode, NULL, 0, *start, *len, 0); |
| 2261 | if (IS_ERR_OR_NULL(em)) { |
| 2262 | if (!em) |
| 2263 | ret = -ENOMEM; |
| 2264 | else |
| 2265 | ret = PTR_ERR(em); |
| 2266 | return ret; |
| 2267 | } |
| 2268 | |
| 2269 | /* Hole or vacuum extent(only exists in no-hole mode) */ |
| 2270 | if (em->block_start == EXTENT_MAP_HOLE) { |
| 2271 | ret = 1; |
| 2272 | *len = em->start + em->len > *start + *len ? |
| 2273 | 0 : *start + *len - em->start - em->len; |
| 2274 | *start = em->start + em->len; |
| 2275 | } |
| 2276 | free_extent_map(em); |
| 2277 | return ret; |
| 2278 | } |
| 2279 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2280 | static int btrfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
| 2281 | { |
| 2282 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 2283 | struct extent_state *cached_state = NULL; |
| 2284 | struct btrfs_path *path; |
| 2285 | struct btrfs_block_rsv *rsv; |
| 2286 | struct btrfs_trans_handle *trans; |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2287 | u64 lockstart; |
| 2288 | u64 lockend; |
| 2289 | u64 tail_start; |
| 2290 | u64 tail_len; |
| 2291 | u64 orig_start = offset; |
| 2292 | u64 cur_offset; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2293 | u64 min_size = btrfs_calc_trunc_metadata_size(root, 1); |
| 2294 | u64 drop_end; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2295 | int ret = 0; |
| 2296 | int err = 0; |
Alexandru Moise | 6e4d6fa | 2015-09-22 21:00:07 +0000 | [diff] [blame] | 2297 | unsigned int rsv_count; |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2298 | bool same_page; |
Josef Bacik | 16e7549 | 2013-10-22 12:18:51 -0400 | [diff] [blame] | 2299 | bool no_holes = btrfs_fs_incompat(root->fs_info, NO_HOLES); |
Filipe Manana | a1a50f6 | 2014-04-26 01:35:31 +0100 | [diff] [blame] | 2300 | u64 ino_size; |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2301 | bool truncated_page = false; |
| 2302 | bool updated_inode = false; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2303 | |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2304 | ret = btrfs_wait_ordered_range(inode, offset, len); |
| 2305 | if (ret) |
| 2306 | return ret; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2307 | |
| 2308 | mutex_lock(&inode->i_mutex); |
Filipe Manana | a1a50f6 | 2014-04-26 01:35:31 +0100 | [diff] [blame] | 2309 | ino_size = round_up(inode->i_size, PAGE_CACHE_SIZE); |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2310 | ret = find_first_non_hole(inode, &offset, &len); |
| 2311 | if (ret < 0) |
| 2312 | goto out_only_mutex; |
| 2313 | if (ret && !len) { |
| 2314 | /* Already in a large hole */ |
| 2315 | ret = 0; |
| 2316 | goto out_only_mutex; |
| 2317 | } |
| 2318 | |
Qu Wenruo | 51f395a | 2014-08-08 13:06:20 +0800 | [diff] [blame] | 2319 | lockstart = round_up(offset, BTRFS_I(inode)->root->sectorsize); |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2320 | lockend = round_down(offset + len, |
| 2321 | BTRFS_I(inode)->root->sectorsize) - 1; |
| 2322 | same_page = ((offset >> PAGE_CACHE_SHIFT) == |
| 2323 | ((offset + len - 1) >> PAGE_CACHE_SHIFT)); |
| 2324 | |
Miao Xie | 7426cc0 | 2012-12-05 10:54:52 +0000 | [diff] [blame] | 2325 | /* |
| 2326 | * We needn't truncate any page which is beyond the end of the file |
| 2327 | * because we are sure there is no data there. |
| 2328 | */ |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2329 | /* |
| 2330 | * Only do this if we are in the same page and we aren't doing the |
| 2331 | * entire page. |
| 2332 | */ |
| 2333 | if (same_page && len < PAGE_CACHE_SIZE) { |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2334 | if (offset < ino_size) { |
| 2335 | truncated_page = true; |
Miao Xie | 7426cc0 | 2012-12-05 10:54:52 +0000 | [diff] [blame] | 2336 | ret = btrfs_truncate_page(inode, offset, len, 0); |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2337 | } else { |
| 2338 | ret = 0; |
| 2339 | } |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2340 | goto out_only_mutex; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2341 | } |
| 2342 | |
| 2343 | /* zero back part of the first page */ |
Filipe Manana | 12870f1 | 2014-02-15 15:55:58 +0000 | [diff] [blame] | 2344 | if (offset < ino_size) { |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2345 | truncated_page = true; |
Miao Xie | 7426cc0 | 2012-12-05 10:54:52 +0000 | [diff] [blame] | 2346 | ret = btrfs_truncate_page(inode, offset, 0, 0); |
| 2347 | if (ret) { |
| 2348 | mutex_unlock(&inode->i_mutex); |
| 2349 | return ret; |
| 2350 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2351 | } |
| 2352 | |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2353 | /* Check the aligned pages after the first unaligned page, |
| 2354 | * if offset != orig_start, which means the first unaligned page |
| 2355 | * including serveral following pages are already in holes, |
| 2356 | * the extra check can be skipped */ |
| 2357 | if (offset == orig_start) { |
| 2358 | /* after truncate page, check hole again */ |
| 2359 | len = offset + len - lockstart; |
| 2360 | offset = lockstart; |
| 2361 | ret = find_first_non_hole(inode, &offset, &len); |
| 2362 | if (ret < 0) |
| 2363 | goto out_only_mutex; |
| 2364 | if (ret && !len) { |
| 2365 | ret = 0; |
| 2366 | goto out_only_mutex; |
| 2367 | } |
| 2368 | lockstart = offset; |
| 2369 | } |
| 2370 | |
| 2371 | /* Check the tail unaligned part is in a hole */ |
| 2372 | tail_start = lockend + 1; |
| 2373 | tail_len = offset + len - tail_start; |
| 2374 | if (tail_len) { |
| 2375 | ret = find_first_non_hole(inode, &tail_start, &tail_len); |
| 2376 | if (unlikely(ret < 0)) |
| 2377 | goto out_only_mutex; |
| 2378 | if (!ret) { |
| 2379 | /* zero the front end of the last page */ |
| 2380 | if (tail_start + tail_len < ino_size) { |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2381 | truncated_page = true; |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2382 | ret = btrfs_truncate_page(inode, |
| 2383 | tail_start + tail_len, 0, 1); |
| 2384 | if (ret) |
| 2385 | goto out_only_mutex; |
Qu Wenruo | 51f395a | 2014-08-08 13:06:20 +0800 | [diff] [blame] | 2386 | } |
Miao Xie | 0061280 | 2012-12-05 10:54:12 +0000 | [diff] [blame] | 2387 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2388 | } |
| 2389 | |
| 2390 | if (lockend < lockstart) { |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2391 | ret = 0; |
| 2392 | goto out_only_mutex; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2393 | } |
| 2394 | |
| 2395 | while (1) { |
| 2396 | struct btrfs_ordered_extent *ordered; |
| 2397 | |
| 2398 | truncate_pagecache_range(inode, lockstart, lockend); |
| 2399 | |
| 2400 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
| 2401 | 0, &cached_state); |
| 2402 | ordered = btrfs_lookup_first_ordered_extent(inode, lockend); |
| 2403 | |
| 2404 | /* |
| 2405 | * We need to make sure we have no ordered extents in this range |
| 2406 | * and nobody raced in and read a page in this range, if we did |
| 2407 | * we need to try again. |
| 2408 | */ |
| 2409 | if ((!ordered || |
Filipe David Borba Manana | 6126e3c | 2013-11-19 16:19:24 +0000 | [diff] [blame] | 2410 | (ordered->file_offset + ordered->len <= lockstart || |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2411 | ordered->file_offset > lockend)) && |
Alex Gartrell | fc4adbf | 2014-05-20 13:07:56 -0700 | [diff] [blame] | 2412 | !btrfs_page_exists_in_range(inode, lockstart, lockend)) { |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2413 | if (ordered) |
| 2414 | btrfs_put_ordered_extent(ordered); |
| 2415 | break; |
| 2416 | } |
| 2417 | if (ordered) |
| 2418 | btrfs_put_ordered_extent(ordered); |
| 2419 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, |
| 2420 | lockend, &cached_state, GFP_NOFS); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2421 | ret = btrfs_wait_ordered_range(inode, lockstart, |
| 2422 | lockend - lockstart + 1); |
| 2423 | if (ret) { |
| 2424 | mutex_unlock(&inode->i_mutex); |
| 2425 | return ret; |
| 2426 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2427 | } |
| 2428 | |
| 2429 | path = btrfs_alloc_path(); |
| 2430 | if (!path) { |
| 2431 | ret = -ENOMEM; |
| 2432 | goto out; |
| 2433 | } |
| 2434 | |
Miao Xie | 66d8f3d | 2012-09-06 04:02:28 -0600 | [diff] [blame] | 2435 | rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2436 | if (!rsv) { |
| 2437 | ret = -ENOMEM; |
| 2438 | goto out_free; |
| 2439 | } |
| 2440 | rsv->size = btrfs_calc_trunc_metadata_size(root, 1); |
| 2441 | rsv->failfast = 1; |
| 2442 | |
| 2443 | /* |
| 2444 | * 1 - update the inode |
| 2445 | * 1 - removing the extents in the range |
Josef Bacik | 16e7549 | 2013-10-22 12:18:51 -0400 | [diff] [blame] | 2446 | * 1 - adding the hole extent if no_holes isn't set |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2447 | */ |
Josef Bacik | 16e7549 | 2013-10-22 12:18:51 -0400 | [diff] [blame] | 2448 | rsv_count = no_holes ? 2 : 3; |
| 2449 | trans = btrfs_start_transaction(root, rsv_count); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2450 | if (IS_ERR(trans)) { |
| 2451 | err = PTR_ERR(trans); |
| 2452 | goto out_free; |
| 2453 | } |
| 2454 | |
| 2455 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv, |
| 2456 | min_size); |
| 2457 | BUG_ON(ret); |
| 2458 | trans->block_rsv = rsv; |
| 2459 | |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2460 | cur_offset = lockstart; |
| 2461 | len = lockend - cur_offset; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2462 | while (cur_offset < lockend) { |
| 2463 | ret = __btrfs_drop_extents(trans, root, inode, path, |
| 2464 | cur_offset, lockend + 1, |
Filipe David Borba Manana | 1acae57 | 2014-01-07 11:42:27 +0000 | [diff] [blame] | 2465 | &drop_end, 1, 0, 0, NULL); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2466 | if (ret != -ENOSPC) |
| 2467 | break; |
| 2468 | |
| 2469 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
| 2470 | |
Filipe Manana | 12870f1 | 2014-02-15 15:55:58 +0000 | [diff] [blame] | 2471 | if (cur_offset < ino_size) { |
| 2472 | ret = fill_holes(trans, inode, path, cur_offset, |
| 2473 | drop_end); |
| 2474 | if (ret) { |
| 2475 | err = ret; |
| 2476 | break; |
| 2477 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2478 | } |
| 2479 | |
| 2480 | cur_offset = drop_end; |
| 2481 | |
| 2482 | ret = btrfs_update_inode(trans, root, inode); |
| 2483 | if (ret) { |
| 2484 | err = ret; |
| 2485 | break; |
| 2486 | } |
| 2487 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2488 | btrfs_end_transaction(trans, root); |
Liu Bo | b53d3f5 | 2012-11-14 14:34:34 +0000 | [diff] [blame] | 2489 | btrfs_btree_balance_dirty(root); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2490 | |
Josef Bacik | 16e7549 | 2013-10-22 12:18:51 -0400 | [diff] [blame] | 2491 | trans = btrfs_start_transaction(root, rsv_count); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2492 | if (IS_ERR(trans)) { |
| 2493 | ret = PTR_ERR(trans); |
| 2494 | trans = NULL; |
| 2495 | break; |
| 2496 | } |
| 2497 | |
| 2498 | ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, |
| 2499 | rsv, min_size); |
| 2500 | BUG_ON(ret); /* shouldn't happen */ |
| 2501 | trans->block_rsv = rsv; |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2502 | |
| 2503 | ret = find_first_non_hole(inode, &cur_offset, &len); |
| 2504 | if (unlikely(ret < 0)) |
| 2505 | break; |
| 2506 | if (ret && !len) { |
| 2507 | ret = 0; |
| 2508 | break; |
| 2509 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2510 | } |
| 2511 | |
| 2512 | if (ret) { |
| 2513 | err = ret; |
| 2514 | goto out_trans; |
| 2515 | } |
| 2516 | |
| 2517 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
Filipe Manana | fc19c5e | 2014-04-29 13:18:40 +0100 | [diff] [blame] | 2518 | /* |
Filipe Manana | 2959a32 | 2015-11-02 12:32:44 +0000 | [diff] [blame] | 2519 | * If we are using the NO_HOLES feature we might have had already an |
| 2520 | * hole that overlaps a part of the region [lockstart, lockend] and |
| 2521 | * ends at (or beyond) lockend. Since we have no file extent items to |
| 2522 | * represent holes, drop_end can be less than lockend and so we must |
| 2523 | * make sure we have an extent map representing the existing hole (the |
| 2524 | * call to __btrfs_drop_extents() might have dropped the existing extent |
| 2525 | * map representing the existing hole), otherwise the fast fsync path |
| 2526 | * will not record the existence of the hole region |
| 2527 | * [existing_hole_start, lockend]. |
| 2528 | */ |
| 2529 | if (drop_end <= lockend) |
| 2530 | drop_end = lockend + 1; |
| 2531 | /* |
Filipe Manana | fc19c5e | 2014-04-29 13:18:40 +0100 | [diff] [blame] | 2532 | * Don't insert file hole extent item if it's for a range beyond eof |
| 2533 | * (because it's useless) or if it represents a 0 bytes range (when |
| 2534 | * cur_offset == drop_end). |
| 2535 | */ |
| 2536 | if (cur_offset < ino_size && cur_offset < drop_end) { |
Filipe Manana | 12870f1 | 2014-02-15 15:55:58 +0000 | [diff] [blame] | 2537 | ret = fill_holes(trans, inode, path, cur_offset, drop_end); |
| 2538 | if (ret) { |
| 2539 | err = ret; |
| 2540 | goto out_trans; |
| 2541 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2542 | } |
| 2543 | |
| 2544 | out_trans: |
| 2545 | if (!trans) |
| 2546 | goto out_free; |
| 2547 | |
Tsutomu Itoh | e1f5790 | 2012-11-08 04:47:33 +0000 | [diff] [blame] | 2548 | inode_inc_iversion(inode); |
| 2549 | inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
| 2550 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2551 | trans->block_rsv = &root->fs_info->trans_block_rsv; |
| 2552 | ret = btrfs_update_inode(trans, root, inode); |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2553 | updated_inode = true; |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2554 | btrfs_end_transaction(trans, root); |
Liu Bo | b53d3f5 | 2012-11-14 14:34:34 +0000 | [diff] [blame] | 2555 | btrfs_btree_balance_dirty(root); |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2556 | out_free: |
| 2557 | btrfs_free_path(path); |
| 2558 | btrfs_free_block_rsv(root, rsv); |
| 2559 | out: |
| 2560 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
| 2561 | &cached_state, GFP_NOFS); |
Qu Wenruo | d778154 | 2014-05-30 15:16:10 +0800 | [diff] [blame] | 2562 | out_only_mutex: |
Filipe Manana | e8c1c76 | 2015-02-15 22:38:54 +0000 | [diff] [blame] | 2563 | if (!updated_inode && truncated_page && !ret && !err) { |
| 2564 | /* |
| 2565 | * If we only end up zeroing part of a page, we still need to |
| 2566 | * update the inode item, so that all the time fields are |
| 2567 | * updated as well as the necessary btrfs inode in memory fields |
| 2568 | * for detecting, at fsync time, if the inode isn't yet in the |
| 2569 | * log tree or it's there but not up to date. |
| 2570 | */ |
| 2571 | trans = btrfs_start_transaction(root, 1); |
| 2572 | if (IS_ERR(trans)) { |
| 2573 | err = PTR_ERR(trans); |
| 2574 | } else { |
| 2575 | err = btrfs_update_inode(trans, root, inode); |
| 2576 | ret = btrfs_end_transaction(trans, root); |
| 2577 | } |
| 2578 | } |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2579 | mutex_unlock(&inode->i_mutex); |
| 2580 | if (ret && !err) |
| 2581 | err = ret; |
| 2582 | return err; |
| 2583 | } |
| 2584 | |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2585 | /* Helper structure to record which range is already reserved */ |
| 2586 | struct falloc_range { |
| 2587 | struct list_head list; |
| 2588 | u64 start; |
| 2589 | u64 len; |
| 2590 | }; |
| 2591 | |
| 2592 | /* |
| 2593 | * Helper function to add falloc range |
| 2594 | * |
| 2595 | * Caller should have locked the larger range of extent containing |
| 2596 | * [start, len) |
| 2597 | */ |
| 2598 | static int add_falloc_range(struct list_head *head, u64 start, u64 len) |
| 2599 | { |
| 2600 | struct falloc_range *prev = NULL; |
| 2601 | struct falloc_range *range = NULL; |
| 2602 | |
| 2603 | if (list_empty(head)) |
| 2604 | goto insert; |
| 2605 | |
| 2606 | /* |
| 2607 | * As fallocate iterate by bytenr order, we only need to check |
| 2608 | * the last range. |
| 2609 | */ |
| 2610 | prev = list_entry(head->prev, struct falloc_range, list); |
| 2611 | if (prev->start + prev->len == start) { |
| 2612 | prev->len += len; |
| 2613 | return 0; |
| 2614 | } |
| 2615 | insert: |
| 2616 | range = kmalloc(sizeof(*range), GFP_NOFS); |
| 2617 | if (!range) |
| 2618 | return -ENOMEM; |
| 2619 | range->start = start; |
| 2620 | range->len = len; |
| 2621 | list_add_tail(&range->list, head); |
| 2622 | return 0; |
| 2623 | } |
| 2624 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2625 | static long btrfs_fallocate(struct file *file, int mode, |
| 2626 | loff_t offset, loff_t len) |
| 2627 | { |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 2628 | struct inode *inode = file_inode(file); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2629 | struct extent_state *cached_state = NULL; |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2630 | struct falloc_range *range; |
| 2631 | struct falloc_range *tmp; |
| 2632 | struct list_head reserve_list; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2633 | u64 cur_offset; |
| 2634 | u64 last_byte; |
| 2635 | u64 alloc_start; |
| 2636 | u64 alloc_end; |
| 2637 | u64 alloc_hint = 0; |
| 2638 | u64 locked_end; |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2639 | u64 actual_end = 0; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2640 | struct extent_map *em; |
Miao Xie | 797f427 | 2012-11-28 10:28:07 +0000 | [diff] [blame] | 2641 | int blocksize = BTRFS_I(inode)->root->sectorsize; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2642 | int ret; |
| 2643 | |
Miao Xie | 797f427 | 2012-11-28 10:28:07 +0000 | [diff] [blame] | 2644 | alloc_start = round_down(offset, blocksize); |
| 2645 | alloc_end = round_up(offset + len, blocksize); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2646 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2647 | /* Make sure we aren't being give some crap mode */ |
| 2648 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2649 | return -EOPNOTSUPP; |
| 2650 | |
Josef Bacik | 2aaa665 | 2012-08-29 14:27:18 -0400 | [diff] [blame] | 2651 | if (mode & FALLOC_FL_PUNCH_HOLE) |
| 2652 | return btrfs_punch_hole(inode, offset, len); |
| 2653 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2654 | /* |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2655 | * Only trigger disk allocation, don't trigger qgroup reserve |
| 2656 | * |
| 2657 | * For qgroup space, it will be checked later. |
Chris Mason | d98456f | 2012-01-31 20:27:41 -0500 | [diff] [blame] | 2658 | */ |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2659 | ret = btrfs_alloc_data_chunk_ondemand(inode, alloc_end - alloc_start); |
| 2660 | if (ret < 0) |
Chris Mason | d98456f | 2012-01-31 20:27:41 -0500 | [diff] [blame] | 2661 | return ret; |
| 2662 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2663 | mutex_lock(&inode->i_mutex); |
| 2664 | ret = inode_newsize_ok(inode, alloc_end); |
| 2665 | if (ret) |
| 2666 | goto out; |
| 2667 | |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2668 | /* |
| 2669 | * TODO: Move these two operations after we have checked |
| 2670 | * accurate reserved space, or fallocate can still fail but |
| 2671 | * with page truncated or size expanded. |
| 2672 | * |
| 2673 | * But that's a minor problem and won't do much harm BTW. |
| 2674 | */ |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2675 | if (alloc_start > inode->i_size) { |
Josef Bacik | a41ad39 | 2011-01-31 15:30:16 -0500 | [diff] [blame] | 2676 | ret = btrfs_cont_expand(inode, i_size_read(inode), |
| 2677 | alloc_start); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2678 | if (ret) |
| 2679 | goto out; |
Josef Bacik | a71754f | 2013-06-17 17:14:39 -0400 | [diff] [blame] | 2680 | } else { |
| 2681 | /* |
| 2682 | * If we are fallocating from the end of the file onward we |
| 2683 | * need to zero out the end of the page if i_size lands in the |
| 2684 | * middle of a page. |
| 2685 | */ |
| 2686 | ret = btrfs_truncate_page(inode, inode->i_size, 0, 0); |
| 2687 | if (ret) |
| 2688 | goto out; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2689 | } |
| 2690 | |
Josef Bacik | a71754f | 2013-06-17 17:14:39 -0400 | [diff] [blame] | 2691 | /* |
| 2692 | * wait for ordered IO before we have any locks. We'll loop again |
| 2693 | * below with the locks held. |
| 2694 | */ |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2695 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
| 2696 | alloc_end - alloc_start); |
| 2697 | if (ret) |
| 2698 | goto out; |
Josef Bacik | a71754f | 2013-06-17 17:14:39 -0400 | [diff] [blame] | 2699 | |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2700 | locked_end = alloc_end - 1; |
| 2701 | while (1) { |
| 2702 | struct btrfs_ordered_extent *ordered; |
| 2703 | |
| 2704 | /* the extent lock is ordered inside the running |
| 2705 | * transaction |
| 2706 | */ |
| 2707 | lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start, |
Jeff Mahoney | d008237 | 2012-03-01 14:57:19 +0100 | [diff] [blame] | 2708 | locked_end, 0, &cached_state); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2709 | ordered = btrfs_lookup_first_ordered_extent(inode, |
| 2710 | alloc_end - 1); |
| 2711 | if (ordered && |
| 2712 | ordered->file_offset + ordered->len > alloc_start && |
| 2713 | ordered->file_offset < alloc_end) { |
| 2714 | btrfs_put_ordered_extent(ordered); |
| 2715 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, |
| 2716 | alloc_start, locked_end, |
| 2717 | &cached_state, GFP_NOFS); |
| 2718 | /* |
| 2719 | * we can't wait on the range with the transaction |
| 2720 | * running or with the extent lock held |
| 2721 | */ |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 2722 | ret = btrfs_wait_ordered_range(inode, alloc_start, |
| 2723 | alloc_end - alloc_start); |
| 2724 | if (ret) |
| 2725 | goto out; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2726 | } else { |
| 2727 | if (ordered) |
| 2728 | btrfs_put_ordered_extent(ordered); |
| 2729 | break; |
| 2730 | } |
| 2731 | } |
| 2732 | |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2733 | /* First, check if we exceed the qgroup limit */ |
| 2734 | INIT_LIST_HEAD(&reserve_list); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2735 | cur_offset = alloc_start; |
| 2736 | while (1) { |
| 2737 | em = btrfs_get_extent(inode, NULL, 0, cur_offset, |
| 2738 | alloc_end - cur_offset, 0); |
Jeff Mahoney | 79787ea | 2012-03-12 16:03:00 +0100 | [diff] [blame] | 2739 | if (IS_ERR_OR_NULL(em)) { |
| 2740 | if (!em) |
| 2741 | ret = -ENOMEM; |
| 2742 | else |
| 2743 | ret = PTR_ERR(em); |
| 2744 | break; |
| 2745 | } |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2746 | last_byte = min(extent_map_end(em), alloc_end); |
Josef Bacik | f1e490a | 2011-08-18 10:36:39 -0400 | [diff] [blame] | 2747 | actual_end = min_t(u64, extent_map_end(em), offset + len); |
Miao Xie | 797f427 | 2012-11-28 10:28:07 +0000 | [diff] [blame] | 2748 | last_byte = ALIGN(last_byte, blocksize); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2749 | if (em->block_start == EXTENT_MAP_HOLE || |
| 2750 | (cur_offset >= inode->i_size && |
| 2751 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) { |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2752 | ret = add_falloc_range(&reserve_list, cur_offset, |
| 2753 | last_byte - cur_offset); |
| 2754 | if (ret < 0) { |
| 2755 | free_extent_map(em); |
| 2756 | break; |
Filipe Manana | 3d850dd | 2015-03-12 23:23:13 +0000 | [diff] [blame] | 2757 | } |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2758 | ret = btrfs_qgroup_reserve_data(inode, cur_offset, |
| 2759 | last_byte - cur_offset); |
| 2760 | if (ret < 0) |
| 2761 | break; |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2762 | } |
| 2763 | free_extent_map(em); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2764 | cur_offset = last_byte; |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2765 | if (cur_offset >= alloc_end) |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2766 | break; |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2767 | } |
| 2768 | |
| 2769 | /* |
| 2770 | * If ret is still 0, means we're OK to fallocate. |
| 2771 | * Or just cleanup the list and exit. |
| 2772 | */ |
| 2773 | list_for_each_entry_safe(range, tmp, &reserve_list, list) { |
| 2774 | if (!ret) |
| 2775 | ret = btrfs_prealloc_file_range(inode, mode, |
| 2776 | range->start, |
| 2777 | range->len, 1 << inode->i_blkbits, |
| 2778 | offset + len, &alloc_hint); |
| 2779 | list_del(&range->list); |
| 2780 | kfree(range); |
| 2781 | } |
| 2782 | if (ret < 0) |
| 2783 | goto out_unlock; |
| 2784 | |
| 2785 | if (actual_end > inode->i_size && |
| 2786 | !(mode & FALLOC_FL_KEEP_SIZE)) { |
| 2787 | struct btrfs_trans_handle *trans; |
| 2788 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 2789 | |
| 2790 | /* |
| 2791 | * We didn't need to allocate any more space, but we |
| 2792 | * still extended the size of the file so we need to |
| 2793 | * update i_size and the inode item. |
| 2794 | */ |
| 2795 | trans = btrfs_start_transaction(root, 1); |
| 2796 | if (IS_ERR(trans)) { |
| 2797 | ret = PTR_ERR(trans); |
| 2798 | } else { |
| 2799 | inode->i_ctime = CURRENT_TIME; |
| 2800 | i_size_write(inode, actual_end); |
| 2801 | btrfs_ordered_update_i_size(inode, actual_end, NULL); |
| 2802 | ret = btrfs_update_inode(trans, root, inode); |
| 2803 | if (ret) |
| 2804 | btrfs_end_transaction(trans, root); |
| 2805 | else |
| 2806 | ret = btrfs_end_transaction(trans, root); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2807 | } |
| 2808 | } |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2809 | out_unlock: |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2810 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end, |
| 2811 | &cached_state, GFP_NOFS); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2812 | out: |
Qu Wenruo | 14524a8 | 2015-09-08 17:22:44 +0800 | [diff] [blame] | 2813 | /* |
| 2814 | * As we waited the extent range, the data_rsv_map must be empty |
| 2815 | * in the range, as written data range will be released from it. |
| 2816 | * And for prealloacted extent, it will also be released when |
| 2817 | * its metadata is written. |
| 2818 | * So this is completely used as cleanup. |
| 2819 | */ |
| 2820 | btrfs_qgroup_free_data(inode, alloc_start, alloc_end - alloc_start); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2821 | mutex_unlock(&inode->i_mutex); |
Chris Mason | d98456f | 2012-01-31 20:27:41 -0500 | [diff] [blame] | 2822 | /* Let go of our reservation. */ |
Qu Wenruo | 7cf5b97 | 2015-09-08 17:25:55 +0800 | [diff] [blame] | 2823 | btrfs_free_reserved_data_space(inode, alloc_start, |
| 2824 | alloc_end - alloc_start); |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2825 | return ret; |
| 2826 | } |
| 2827 | |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 2828 | static int find_desired_extent(struct inode *inode, loff_t *offset, int whence) |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2829 | { |
| 2830 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2831 | struct extent_map *em = NULL; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2832 | struct extent_state *cached_state = NULL; |
Liu Bo | 4d1a40c | 2014-09-16 17:49:30 +0800 | [diff] [blame] | 2833 | u64 lockstart; |
| 2834 | u64 lockend; |
| 2835 | u64 start; |
| 2836 | u64 len; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2837 | int ret = 0; |
| 2838 | |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2839 | if (inode->i_size == 0) |
| 2840 | return -ENXIO; |
| 2841 | |
Liu Bo | 4d1a40c | 2014-09-16 17:49:30 +0800 | [diff] [blame] | 2842 | /* |
| 2843 | * *offset can be negative, in this case we start finding DATA/HOLE from |
| 2844 | * the very start of the file. |
| 2845 | */ |
| 2846 | start = max_t(loff_t, 0, *offset); |
| 2847 | |
| 2848 | lockstart = round_down(start, root->sectorsize); |
| 2849 | lockend = round_up(i_size_read(inode), root->sectorsize); |
| 2850 | if (lockend <= lockstart) |
| 2851 | lockend = lockstart + root->sectorsize; |
| 2852 | lockend--; |
| 2853 | len = lockend - lockstart + 1; |
| 2854 | |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2855 | lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend, 0, |
Jeff Mahoney | d008237 | 2012-03-01 14:57:19 +0100 | [diff] [blame] | 2856 | &cached_state); |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2857 | |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2858 | while (start < inode->i_size) { |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2859 | em = btrfs_get_extent_fiemap(inode, NULL, 0, start, len, 0); |
| 2860 | if (IS_ERR(em)) { |
Jeff Liu | 6af021d | 2012-02-09 14:25:50 +0800 | [diff] [blame] | 2861 | ret = PTR_ERR(em); |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2862 | em = NULL; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2863 | break; |
| 2864 | } |
| 2865 | |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2866 | if (whence == SEEK_HOLE && |
| 2867 | (em->block_start == EXTENT_MAP_HOLE || |
| 2868 | test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) |
| 2869 | break; |
| 2870 | else if (whence == SEEK_DATA && |
| 2871 | (em->block_start != EXTENT_MAP_HOLE && |
| 2872 | !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) |
| 2873 | break; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2874 | |
| 2875 | start = em->start + em->len; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2876 | free_extent_map(em); |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2877 | em = NULL; |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2878 | cond_resched(); |
| 2879 | } |
Josef Bacik | 7f4ca37 | 2013-10-18 11:44:46 -0400 | [diff] [blame] | 2880 | free_extent_map(em); |
| 2881 | if (!ret) { |
| 2882 | if (whence == SEEK_DATA && start >= inode->i_size) |
| 2883 | ret = -ENXIO; |
| 2884 | else |
| 2885 | *offset = min_t(loff_t, start, inode->i_size); |
| 2886 | } |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2887 | unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend, |
| 2888 | &cached_state, GFP_NOFS); |
| 2889 | return ret; |
| 2890 | } |
| 2891 | |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 2892 | static loff_t btrfs_file_llseek(struct file *file, loff_t offset, int whence) |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2893 | { |
| 2894 | struct inode *inode = file->f_mapping->host; |
| 2895 | int ret; |
| 2896 | |
| 2897 | mutex_lock(&inode->i_mutex); |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 2898 | switch (whence) { |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2899 | case SEEK_END: |
| 2900 | case SEEK_CUR: |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 2901 | offset = generic_file_llseek(file, offset, whence); |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2902 | goto out; |
| 2903 | case SEEK_DATA: |
| 2904 | case SEEK_HOLE: |
Jeff Liu | 48802c8 | 2011-09-18 10:34:02 -0400 | [diff] [blame] | 2905 | if (offset >= i_size_read(inode)) { |
| 2906 | mutex_unlock(&inode->i_mutex); |
| 2907 | return -ENXIO; |
| 2908 | } |
| 2909 | |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 2910 | ret = find_desired_extent(inode, &offset, whence); |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2911 | if (ret) { |
| 2912 | mutex_unlock(&inode->i_mutex); |
| 2913 | return ret; |
| 2914 | } |
| 2915 | } |
| 2916 | |
Jie Liu | 46a1c2c | 2013-06-25 12:02:13 +0800 | [diff] [blame] | 2917 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2918 | out: |
| 2919 | mutex_unlock(&inode->i_mutex); |
| 2920 | return offset; |
| 2921 | } |
| 2922 | |
Alexey Dobriyan | 828c095 | 2009-10-01 15:43:56 -0700 | [diff] [blame] | 2923 | const struct file_operations btrfs_file_operations = { |
Josef Bacik | b267515 | 2011-07-18 13:21:36 -0400 | [diff] [blame] | 2924 | .llseek = btrfs_file_llseek, |
Al Viro | aad4f8b | 2014-04-02 14:33:16 -0400 | [diff] [blame] | 2925 | .read_iter = generic_file_read_iter, |
Chris Mason | e9906a9 | 2007-12-14 12:56:58 -0500 | [diff] [blame] | 2926 | .splice_read = generic_file_splice_read, |
Al Viro | b30ac0f | 2014-04-03 14:29:04 -0400 | [diff] [blame] | 2927 | .write_iter = btrfs_file_write_iter, |
Chris Mason | 9ebefb18 | 2007-06-15 13:50:00 -0400 | [diff] [blame] | 2928 | .mmap = btrfs_file_mmap, |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2929 | .open = generic_file_open, |
Mingming | e1b81e6 | 2008-05-27 10:55:43 -0400 | [diff] [blame] | 2930 | .release = btrfs_release_file, |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2931 | .fsync = btrfs_sync_file, |
Christoph Hellwig | 2fe17c1 | 2011-01-14 13:07:43 +0100 | [diff] [blame] | 2932 | .fallocate = btrfs_fallocate, |
Christoph Hellwig | 34287aa | 2007-09-14 10:22:47 -0400 | [diff] [blame] | 2933 | .unlocked_ioctl = btrfs_ioctl, |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2934 | #ifdef CONFIG_COMPAT |
Christoph Hellwig | 34287aa | 2007-09-14 10:22:47 -0400 | [diff] [blame] | 2935 | .compat_ioctl = btrfs_ioctl, |
Chris Mason | 39279cc | 2007-06-12 06:35:45 -0400 | [diff] [blame] | 2936 | #endif |
| 2937 | }; |
Miao Xie | 9247f31 | 2012-11-26 09:24:43 +0000 | [diff] [blame] | 2938 | |
| 2939 | void btrfs_auto_defrag_exit(void) |
| 2940 | { |
| 2941 | if (btrfs_inode_defrag_cachep) |
| 2942 | kmem_cache_destroy(btrfs_inode_defrag_cachep); |
| 2943 | } |
| 2944 | |
| 2945 | int btrfs_auto_defrag_init(void) |
| 2946 | { |
| 2947 | btrfs_inode_defrag_cachep = kmem_cache_create("btrfs_inode_defrag", |
| 2948 | sizeof(struct inode_defrag), 0, |
| 2949 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, |
| 2950 | NULL); |
| 2951 | if (!btrfs_inode_defrag_cachep) |
| 2952 | return -ENOMEM; |
| 2953 | |
| 2954 | return 0; |
| 2955 | } |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 2956 | |
| 2957 | int btrfs_fdatawrite_range(struct inode *inode, loff_t start, loff_t end) |
| 2958 | { |
| 2959 | int ret; |
| 2960 | |
| 2961 | /* |
| 2962 | * So with compression we will find and lock a dirty page and clear the |
| 2963 | * first one as dirty, setup an async extent, and immediately return |
| 2964 | * with the entire range locked but with nobody actually marked with |
| 2965 | * writeback. So we can't just filemap_write_and_wait_range() and |
| 2966 | * expect it to work since it will just kick off a thread to do the |
| 2967 | * actual work. So we need to call filemap_fdatawrite_range _again_ |
| 2968 | * since it will wait on the page lock, which won't be unlocked until |
| 2969 | * after the pages have been marked as writeback and so we're good to go |
| 2970 | * from there. We have to do this otherwise we'll miss the ordered |
| 2971 | * extents and that results in badness. Please Josef, do not think you |
| 2972 | * know better and pull this out at some point in the future, it is |
| 2973 | * right and you are wrong. |
| 2974 | */ |
| 2975 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); |
| 2976 | if (!ret && test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT, |
| 2977 | &BTRFS_I(inode)->runtime_flags)) |
| 2978 | ret = filemap_fdatawrite_range(inode->i_mapping, start, end); |
| 2979 | |
| 2980 | return ret; |
| 2981 | } |