Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [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 | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 19 | #include <linux/slab.h> |
Chris Mason | d6bfde8 | 2008-04-30 13:59:35 -0400 | [diff] [blame] | 20 | #include <linux/blkdev.h> |
Chris Mason | f421950 | 2008-07-22 11:18:09 -0400 | [diff] [blame] | 21 | #include <linux/writeback.h> |
| 22 | #include <linux/pagevec.h> |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 23 | #include "ctree.h" |
| 24 | #include "transaction.h" |
| 25 | #include "btrfs_inode.h" |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 26 | #include "extent_io.h" |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 27 | #include "disk-io.h" |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 28 | |
Miao Xie | 6352b91 | 2012-09-06 04:01:51 -0600 | [diff] [blame] | 29 | static struct kmem_cache *btrfs_ordered_extent_cache; |
| 30 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 31 | static u64 entry_end(struct btrfs_ordered_extent *entry) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 32 | { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 33 | if (entry->file_offset + entry->len < entry->file_offset) |
| 34 | return (u64)-1; |
| 35 | return entry->file_offset + entry->len; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 36 | } |
| 37 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 38 | /* returns NULL if the insertion worked, or it returns the node it did find |
| 39 | * in the tree |
| 40 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 41 | static struct rb_node *tree_insert(struct rb_root *root, u64 file_offset, |
| 42 | struct rb_node *node) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 43 | { |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 44 | struct rb_node **p = &root->rb_node; |
| 45 | struct rb_node *parent = NULL; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 46 | struct btrfs_ordered_extent *entry; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 47 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 48 | while (*p) { |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 49 | parent = *p; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 50 | entry = rb_entry(parent, struct btrfs_ordered_extent, rb_node); |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 51 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 52 | if (file_offset < entry->file_offset) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 53 | p = &(*p)->rb_left; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 54 | else if (file_offset >= entry_end(entry)) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 55 | p = &(*p)->rb_right; |
| 56 | else |
| 57 | return parent; |
| 58 | } |
| 59 | |
| 60 | rb_link_node(node, parent, p); |
| 61 | rb_insert_color(node, root); |
| 62 | return NULL; |
| 63 | } |
| 64 | |
Jeff Mahoney | 43c04fb | 2011-10-03 23:22:33 -0400 | [diff] [blame] | 65 | static void ordered_data_tree_panic(struct inode *inode, int errno, |
| 66 | u64 offset) |
| 67 | { |
| 68 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
| 69 | btrfs_panic(fs_info, errno, "Inconsistency in ordered tree at offset " |
David Sterba | 351fd35 | 2014-05-15 16:48:20 +0200 | [diff] [blame] | 70 | "%llu", offset); |
Jeff Mahoney | 43c04fb | 2011-10-03 23:22:33 -0400 | [diff] [blame] | 71 | } |
| 72 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 73 | /* |
| 74 | * look for a given offset in the tree, and if it can't be found return the |
| 75 | * first lesser offset |
| 76 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 77 | static struct rb_node *__tree_search(struct rb_root *root, u64 file_offset, |
| 78 | struct rb_node **prev_ret) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 79 | { |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 80 | struct rb_node *n = root->rb_node; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 81 | struct rb_node *prev = NULL; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 82 | struct rb_node *test; |
| 83 | struct btrfs_ordered_extent *entry; |
| 84 | struct btrfs_ordered_extent *prev_entry = NULL; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 85 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 86 | while (n) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 87 | entry = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 88 | prev = n; |
| 89 | prev_entry = entry; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 90 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 91 | if (file_offset < entry->file_offset) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 92 | n = n->rb_left; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 93 | else if (file_offset >= entry_end(entry)) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 94 | n = n->rb_right; |
| 95 | else |
| 96 | return n; |
| 97 | } |
| 98 | if (!prev_ret) |
| 99 | return NULL; |
| 100 | |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 101 | while (prev && file_offset >= entry_end(prev_entry)) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 102 | test = rb_next(prev); |
| 103 | if (!test) |
| 104 | break; |
| 105 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, |
| 106 | rb_node); |
| 107 | if (file_offset < entry_end(prev_entry)) |
| 108 | break; |
| 109 | |
| 110 | prev = test; |
| 111 | } |
| 112 | if (prev) |
| 113 | prev_entry = rb_entry(prev, struct btrfs_ordered_extent, |
| 114 | rb_node); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 115 | while (prev && file_offset < entry_end(prev_entry)) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 116 | test = rb_prev(prev); |
| 117 | if (!test) |
| 118 | break; |
| 119 | prev_entry = rb_entry(test, struct btrfs_ordered_extent, |
| 120 | rb_node); |
| 121 | prev = test; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 122 | } |
| 123 | *prev_ret = prev; |
| 124 | return NULL; |
| 125 | } |
| 126 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 127 | /* |
| 128 | * helper to check if a given offset is inside a given entry |
| 129 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 130 | static int offset_in_entry(struct btrfs_ordered_extent *entry, u64 file_offset) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 131 | { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 132 | if (file_offset < entry->file_offset || |
| 133 | entry->file_offset + entry->len <= file_offset) |
| 134 | return 0; |
| 135 | return 1; |
| 136 | } |
| 137 | |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 138 | static int range_overlaps(struct btrfs_ordered_extent *entry, u64 file_offset, |
| 139 | u64 len) |
| 140 | { |
| 141 | if (file_offset + len <= entry->file_offset || |
| 142 | entry->file_offset + entry->len <= file_offset) |
| 143 | return 0; |
| 144 | return 1; |
| 145 | } |
| 146 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 147 | /* |
| 148 | * look find the first ordered struct that has this offset, otherwise |
| 149 | * the first one less than this offset |
| 150 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 151 | static inline struct rb_node *tree_search(struct btrfs_ordered_inode_tree *tree, |
| 152 | u64 file_offset) |
| 153 | { |
| 154 | struct rb_root *root = &tree->tree; |
Chris Mason | c87fb6f | 2011-01-31 19:54:59 -0500 | [diff] [blame] | 155 | struct rb_node *prev = NULL; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 156 | struct rb_node *ret; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 157 | struct btrfs_ordered_extent *entry; |
| 158 | |
| 159 | if (tree->last) { |
| 160 | entry = rb_entry(tree->last, struct btrfs_ordered_extent, |
| 161 | rb_node); |
| 162 | if (offset_in_entry(entry, file_offset)) |
| 163 | return tree->last; |
| 164 | } |
| 165 | ret = __tree_search(root, file_offset, &prev); |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 166 | if (!ret) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 167 | ret = prev; |
| 168 | if (ret) |
| 169 | tree->last = ret; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 170 | return ret; |
| 171 | } |
| 172 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 173 | /* allocate and add a new ordered_extent into the per-inode tree. |
| 174 | * file_offset is the logical offset in the file |
| 175 | * |
| 176 | * start is the disk block number of an extent already reserved in the |
| 177 | * extent allocation tree |
| 178 | * |
| 179 | * len is the length of the extent |
| 180 | * |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 181 | * The tree is given a single reference on the ordered extent that was |
| 182 | * inserted. |
| 183 | */ |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 184 | static int __btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
| 185 | u64 start, u64 len, u64 disk_len, |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 186 | int type, int dio, int compress_type) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 187 | { |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 188 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 189 | struct btrfs_ordered_inode_tree *tree; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 190 | struct rb_node *node; |
| 191 | struct btrfs_ordered_extent *entry; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 192 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 193 | tree = &BTRFS_I(inode)->ordered_tree; |
Miao Xie | 6352b91 | 2012-09-06 04:01:51 -0600 | [diff] [blame] | 194 | entry = kmem_cache_zalloc(btrfs_ordered_extent_cache, GFP_NOFS); |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 195 | if (!entry) |
| 196 | return -ENOMEM; |
| 197 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 198 | entry->file_offset = file_offset; |
| 199 | entry->start = start; |
| 200 | entry->len = len; |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 201 | entry->disk_len = disk_len; |
Chris Mason | 8b62b72 | 2009-09-02 16:53:46 -0400 | [diff] [blame] | 202 | entry->bytes_left = len; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 203 | entry->inode = igrab(inode); |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 204 | entry->compress_type = compress_type; |
Josef Bacik | 77cef2e | 2013-08-29 13:57:21 -0400 | [diff] [blame] | 205 | entry->truncated_len = (u64)-1; |
Yan Zheng | d899e05 | 2008-10-30 14:25:28 -0400 | [diff] [blame] | 206 | if (type != BTRFS_ORDERED_IO_DONE && type != BTRFS_ORDERED_COMPLETE) |
Yan Zheng | 80ff385 | 2008-10-30 14:20:02 -0400 | [diff] [blame] | 207 | set_bit(type, &entry->flags); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 208 | |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 209 | if (dio) |
| 210 | set_bit(BTRFS_ORDERED_DIRECT, &entry->flags); |
| 211 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 212 | /* one ref for the tree */ |
| 213 | atomic_set(&entry->refs, 1); |
| 214 | init_waitqueue_head(&entry->wait); |
| 215 | INIT_LIST_HEAD(&entry->list); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 216 | INIT_LIST_HEAD(&entry->root_extent_list); |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 217 | INIT_LIST_HEAD(&entry->work_list); |
| 218 | init_completion(&entry->completion); |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 219 | INIT_LIST_HEAD(&entry->log_list); |
Josef Bacik | 50d9aa9 | 2014-11-21 14:52:38 -0500 | [diff] [blame] | 220 | INIT_LIST_HEAD(&entry->trans_list); |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 221 | |
liubo | 1abe9b8 | 2011-03-24 11:18:59 +0000 | [diff] [blame] | 222 | trace_btrfs_ordered_extent_add(inode, entry); |
| 223 | |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 224 | spin_lock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 225 | node = tree_insert(&tree->tree, file_offset, |
| 226 | &entry->rb_node); |
Jeff Mahoney | 43c04fb | 2011-10-03 23:22:33 -0400 | [diff] [blame] | 227 | if (node) |
| 228 | ordered_data_tree_panic(inode, -EEXIST, file_offset); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 229 | spin_unlock_irq(&tree->lock); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 230 | |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 231 | spin_lock(&root->ordered_extent_lock); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 232 | list_add_tail(&entry->root_extent_list, |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 233 | &root->ordered_extents); |
| 234 | root->nr_ordered_extents++; |
| 235 | if (root->nr_ordered_extents == 1) { |
| 236 | spin_lock(&root->fs_info->ordered_root_lock); |
| 237 | BUG_ON(!list_empty(&root->ordered_root)); |
| 238 | list_add_tail(&root->ordered_root, |
| 239 | &root->fs_info->ordered_roots); |
| 240 | spin_unlock(&root->fs_info->ordered_root_lock); |
| 241 | } |
| 242 | spin_unlock(&root->ordered_extent_lock); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 243 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 244 | return 0; |
| 245 | } |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 246 | |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 247 | int btrfs_add_ordered_extent(struct inode *inode, u64 file_offset, |
| 248 | u64 start, u64 len, u64 disk_len, int type) |
| 249 | { |
| 250 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 251 | disk_len, type, 0, |
| 252 | BTRFS_COMPRESS_NONE); |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 253 | } |
| 254 | |
| 255 | int btrfs_add_ordered_extent_dio(struct inode *inode, u64 file_offset, |
| 256 | u64 start, u64 len, u64 disk_len, int type) |
| 257 | { |
| 258 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, |
Li Zefan | 261507a0 | 2010-12-17 14:21:50 +0800 | [diff] [blame] | 259 | disk_len, type, 1, |
| 260 | BTRFS_COMPRESS_NONE); |
| 261 | } |
| 262 | |
| 263 | int btrfs_add_ordered_extent_compress(struct inode *inode, u64 file_offset, |
| 264 | u64 start, u64 len, u64 disk_len, |
| 265 | int type, int compress_type) |
| 266 | { |
| 267 | return __btrfs_add_ordered_extent(inode, file_offset, start, len, |
| 268 | disk_len, type, 0, |
| 269 | compress_type); |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 270 | } |
| 271 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 272 | /* |
| 273 | * Add a struct btrfs_ordered_sum into the list of checksums to be inserted |
Chris Mason | 3edf7d3 | 2008-07-18 06:17:13 -0400 | [diff] [blame] | 274 | * when an ordered extent is finished. If the list covers more than one |
| 275 | * ordered extent, it is split across multiples. |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 276 | */ |
Jeff Mahoney | 143bede | 2012-03-01 14:56:26 +0100 | [diff] [blame] | 277 | void btrfs_add_ordered_sum(struct inode *inode, |
| 278 | struct btrfs_ordered_extent *entry, |
| 279 | struct btrfs_ordered_sum *sum) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 280 | { |
| 281 | struct btrfs_ordered_inode_tree *tree; |
Chris Mason | 1b1e213 | 2008-06-25 16:01:31 -0400 | [diff] [blame] | 282 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 283 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 284 | spin_lock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 285 | list_add_tail(&sum->list, &entry->list); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 286 | spin_unlock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 287 | } |
| 288 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 289 | /* |
| 290 | * this is used to account for finished IO across a given range |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 291 | * of the file. The IO may span ordered extents. If |
| 292 | * a given ordered_extent is completely done, 1 is returned, otherwise |
| 293 | * 0. |
| 294 | * |
| 295 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used |
| 296 | * to make sure this function only returns 1 once for a given ordered extent. |
| 297 | * |
| 298 | * file_offset is updated to one byte past the range that is recorded as |
| 299 | * complete. This allows you to walk forward in the file. |
| 300 | */ |
| 301 | int btrfs_dec_test_first_ordered_pending(struct inode *inode, |
| 302 | struct btrfs_ordered_extent **cached, |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 303 | u64 *file_offset, u64 io_size, int uptodate) |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 304 | { |
| 305 | struct btrfs_ordered_inode_tree *tree; |
| 306 | struct rb_node *node; |
| 307 | struct btrfs_ordered_extent *entry = NULL; |
| 308 | int ret; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 309 | unsigned long flags; |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 310 | u64 dec_end; |
| 311 | u64 dec_start; |
| 312 | u64 to_dec; |
| 313 | |
| 314 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 315 | spin_lock_irqsave(&tree->lock, flags); |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 316 | node = tree_search(tree, *file_offset); |
| 317 | if (!node) { |
| 318 | ret = 1; |
| 319 | goto out; |
| 320 | } |
| 321 | |
| 322 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 323 | if (!offset_in_entry(entry, *file_offset)) { |
| 324 | ret = 1; |
| 325 | goto out; |
| 326 | } |
| 327 | |
| 328 | dec_start = max(*file_offset, entry->file_offset); |
| 329 | dec_end = min(*file_offset + io_size, entry->file_offset + |
| 330 | entry->len); |
| 331 | *file_offset = dec_end; |
| 332 | if (dec_start > dec_end) { |
Frank Holton | efe120a | 2013-12-20 11:37:06 -0500 | [diff] [blame] | 333 | btrfs_crit(BTRFS_I(inode)->root->fs_info, |
| 334 | "bad ordering dec_start %llu end %llu", dec_start, dec_end); |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 335 | } |
| 336 | to_dec = dec_end - dec_start; |
| 337 | if (to_dec > entry->bytes_left) { |
Frank Holton | efe120a | 2013-12-20 11:37:06 -0500 | [diff] [blame] | 338 | btrfs_crit(BTRFS_I(inode)->root->fs_info, |
| 339 | "bad ordered accounting left %llu size %llu", |
| 340 | entry->bytes_left, to_dec); |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 341 | } |
| 342 | entry->bytes_left -= to_dec; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 343 | if (!uptodate) |
| 344 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); |
| 345 | |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 346 | if (entry->bytes_left == 0) { |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 347 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 348 | if (waitqueue_active(&entry->wait)) |
| 349 | wake_up(&entry->wait); |
| 350 | } else { |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 351 | ret = 1; |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 352 | } |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 353 | out: |
| 354 | if (!ret && cached && entry) { |
| 355 | *cached = entry; |
| 356 | atomic_inc(&entry->refs); |
| 357 | } |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 358 | spin_unlock_irqrestore(&tree->lock, flags); |
Chris Mason | 163cf09 | 2010-11-28 19:56:33 -0500 | [diff] [blame] | 359 | return ret == 0; |
| 360 | } |
| 361 | |
| 362 | /* |
| 363 | * this is used to account for finished IO across a given range |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 364 | * of the file. The IO should not span ordered extents. If |
| 365 | * a given ordered_extent is completely done, 1 is returned, otherwise |
| 366 | * 0. |
| 367 | * |
| 368 | * test_and_set_bit on a flag in the struct btrfs_ordered_extent is used |
| 369 | * to make sure this function only returns 1 once for a given ordered extent. |
| 370 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 371 | int btrfs_dec_test_ordered_pending(struct inode *inode, |
Josef Bacik | 5a1a3df | 2010-02-02 20:51:14 +0000 | [diff] [blame] | 372 | struct btrfs_ordered_extent **cached, |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 373 | u64 file_offset, u64 io_size, int uptodate) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 374 | { |
| 375 | struct btrfs_ordered_inode_tree *tree; |
| 376 | struct rb_node *node; |
Josef Bacik | 5a1a3df | 2010-02-02 20:51:14 +0000 | [diff] [blame] | 377 | struct btrfs_ordered_extent *entry = NULL; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 378 | unsigned long flags; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 379 | int ret; |
| 380 | |
| 381 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 382 | spin_lock_irqsave(&tree->lock, flags); |
| 383 | if (cached && *cached) { |
| 384 | entry = *cached; |
| 385 | goto have_entry; |
| 386 | } |
| 387 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 388 | node = tree_search(tree, file_offset); |
| 389 | if (!node) { |
| 390 | ret = 1; |
| 391 | goto out; |
| 392 | } |
| 393 | |
| 394 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 395 | have_entry: |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 396 | if (!offset_in_entry(entry, file_offset)) { |
| 397 | ret = 1; |
| 398 | goto out; |
| 399 | } |
| 400 | |
Chris Mason | 8b62b72 | 2009-09-02 16:53:46 -0400 | [diff] [blame] | 401 | if (io_size > entry->bytes_left) { |
Frank Holton | efe120a | 2013-12-20 11:37:06 -0500 | [diff] [blame] | 402 | btrfs_crit(BTRFS_I(inode)->root->fs_info, |
| 403 | "bad ordered accounting left %llu size %llu", |
Geert Uytterhoeven | c1c9ff7 | 2013-08-20 13:20:07 +0200 | [diff] [blame] | 404 | entry->bytes_left, io_size); |
Chris Mason | 8b62b72 | 2009-09-02 16:53:46 -0400 | [diff] [blame] | 405 | } |
| 406 | entry->bytes_left -= io_size; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 407 | if (!uptodate) |
| 408 | set_bit(BTRFS_ORDERED_IOERR, &entry->flags); |
| 409 | |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 410 | if (entry->bytes_left == 0) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 411 | ret = test_and_set_bit(BTRFS_ORDERED_IO_DONE, &entry->flags); |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 412 | if (waitqueue_active(&entry->wait)) |
| 413 | wake_up(&entry->wait); |
| 414 | } else { |
Chris Mason | 8b62b72 | 2009-09-02 16:53:46 -0400 | [diff] [blame] | 415 | ret = 1; |
Miao Xie | af7a650 | 2014-03-06 13:54:56 +0800 | [diff] [blame] | 416 | } |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 417 | out: |
Josef Bacik | 5a1a3df | 2010-02-02 20:51:14 +0000 | [diff] [blame] | 418 | if (!ret && cached && entry) { |
| 419 | *cached = entry; |
| 420 | atomic_inc(&entry->refs); |
| 421 | } |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 422 | spin_unlock_irqrestore(&tree->lock, flags); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 423 | return ret == 0; |
| 424 | } |
| 425 | |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 426 | /* Needs to either be called under a log transaction or the log_mutex */ |
Miao Xie | 827463c | 2014-01-14 20:31:51 +0800 | [diff] [blame] | 427 | void btrfs_get_logged_extents(struct inode *inode, |
Filipe Manana | 0870295 | 2014-11-13 17:00:35 +0000 | [diff] [blame] | 428 | struct list_head *logged_list, |
| 429 | const loff_t start, |
| 430 | const loff_t end) |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 431 | { |
| 432 | struct btrfs_ordered_inode_tree *tree; |
| 433 | struct btrfs_ordered_extent *ordered; |
| 434 | struct rb_node *n; |
Filipe Manana | 0870295 | 2014-11-13 17:00:35 +0000 | [diff] [blame] | 435 | struct rb_node *prev; |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 436 | |
| 437 | tree = &BTRFS_I(inode)->ordered_tree; |
| 438 | spin_lock_irq(&tree->lock); |
Filipe Manana | 0870295 | 2014-11-13 17:00:35 +0000 | [diff] [blame] | 439 | n = __tree_search(&tree->tree, end, &prev); |
| 440 | if (!n) |
| 441 | n = prev; |
| 442 | for (; n; n = rb_prev(n)) { |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 443 | ordered = rb_entry(n, struct btrfs_ordered_extent, rb_node); |
Filipe Manana | 0870295 | 2014-11-13 17:00:35 +0000 | [diff] [blame] | 444 | if (ordered->file_offset > end) |
| 445 | continue; |
| 446 | if (entry_end(ordered) <= start) |
| 447 | break; |
Filipe Manana | 4d884fc | 2015-02-09 17:17:43 +0000 | [diff] [blame] | 448 | if (test_and_set_bit(BTRFS_ORDERED_LOGGED, &ordered->flags)) |
Josef Bacik | 50d9aa9 | 2014-11-21 14:52:38 -0500 | [diff] [blame] | 449 | continue; |
Filipe Manana | 0870295 | 2014-11-13 17:00:35 +0000 | [diff] [blame] | 450 | list_add(&ordered->log_list, logged_list); |
Miao Xie | 827463c | 2014-01-14 20:31:51 +0800 | [diff] [blame] | 451 | atomic_inc(&ordered->refs); |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 452 | } |
| 453 | spin_unlock_irq(&tree->lock); |
| 454 | } |
| 455 | |
Miao Xie | 827463c | 2014-01-14 20:31:51 +0800 | [diff] [blame] | 456 | void btrfs_put_logged_extents(struct list_head *logged_list) |
| 457 | { |
| 458 | struct btrfs_ordered_extent *ordered; |
| 459 | |
| 460 | while (!list_empty(logged_list)) { |
| 461 | ordered = list_first_entry(logged_list, |
| 462 | struct btrfs_ordered_extent, |
| 463 | log_list); |
| 464 | list_del_init(&ordered->log_list); |
| 465 | btrfs_put_ordered_extent(ordered); |
| 466 | } |
| 467 | } |
| 468 | |
| 469 | void btrfs_submit_logged_extents(struct list_head *logged_list, |
| 470 | struct btrfs_root *log) |
| 471 | { |
| 472 | int index = log->log_transid % 2; |
| 473 | |
| 474 | spin_lock_irq(&log->log_extents_lock[index]); |
| 475 | list_splice_tail(logged_list, &log->logged_list[index]); |
| 476 | spin_unlock_irq(&log->log_extents_lock[index]); |
| 477 | } |
| 478 | |
Josef Bacik | 50d9aa9 | 2014-11-21 14:52:38 -0500 | [diff] [blame] | 479 | void btrfs_wait_logged_extents(struct btrfs_trans_handle *trans, |
| 480 | struct btrfs_root *log, u64 transid) |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 481 | { |
| 482 | struct btrfs_ordered_extent *ordered; |
| 483 | int index = transid % 2; |
| 484 | |
| 485 | spin_lock_irq(&log->log_extents_lock[index]); |
| 486 | while (!list_empty(&log->logged_list[index])) { |
| 487 | ordered = list_first_entry(&log->logged_list[index], |
| 488 | struct btrfs_ordered_extent, |
| 489 | log_list); |
| 490 | list_del_init(&ordered->log_list); |
| 491 | spin_unlock_irq(&log->log_extents_lock[index]); |
Liu Bo | 98ce2de | 2014-07-17 16:08:36 +0800 | [diff] [blame] | 492 | |
| 493 | if (!test_bit(BTRFS_ORDERED_IO_DONE, &ordered->flags) && |
| 494 | !test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags)) { |
| 495 | struct inode *inode = ordered->inode; |
| 496 | u64 start = ordered->file_offset; |
| 497 | u64 end = ordered->file_offset + ordered->len - 1; |
| 498 | |
| 499 | WARN_ON(!inode); |
| 500 | filemap_fdatawrite_range(inode->i_mapping, start, end); |
| 501 | } |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 502 | wait_event(ordered->wait, test_bit(BTRFS_ORDERED_IO_DONE, |
| 503 | &ordered->flags)); |
Liu Bo | 98ce2de | 2014-07-17 16:08:36 +0800 | [diff] [blame] | 504 | |
Filipe Manana | 7558c8b | 2015-04-17 17:08:37 +0100 | [diff] [blame] | 505 | /* |
| 506 | * If our ordered extent completed it means it updated the |
| 507 | * fs/subvol and csum trees already, so no need to make the |
| 508 | * current transaction's commit wait for it, as we end up |
| 509 | * holding memory unnecessarily and delaying the inode's iput |
| 510 | * until the transaction commit (we schedule an iput for the |
| 511 | * inode when the ordered extent's refcount drops to 0), which |
| 512 | * prevents it from being evictable until the transaction |
| 513 | * commits. |
| 514 | */ |
| 515 | if (test_bit(BTRFS_ORDERED_COMPLETE, &ordered->flags)) |
| 516 | btrfs_put_ordered_extent(ordered); |
| 517 | else |
| 518 | list_add_tail(&ordered->trans_list, &trans->ordered); |
| 519 | |
Josef Bacik | 2ab28f3 | 2012-10-12 15:27:49 -0400 | [diff] [blame] | 520 | spin_lock_irq(&log->log_extents_lock[index]); |
| 521 | } |
| 522 | spin_unlock_irq(&log->log_extents_lock[index]); |
| 523 | } |
| 524 | |
| 525 | void btrfs_free_logged_extents(struct btrfs_root *log, u64 transid) |
| 526 | { |
| 527 | struct btrfs_ordered_extent *ordered; |
| 528 | int index = transid % 2; |
| 529 | |
| 530 | spin_lock_irq(&log->log_extents_lock[index]); |
| 531 | while (!list_empty(&log->logged_list[index])) { |
| 532 | ordered = list_first_entry(&log->logged_list[index], |
| 533 | struct btrfs_ordered_extent, |
| 534 | log_list); |
| 535 | list_del_init(&ordered->log_list); |
| 536 | spin_unlock_irq(&log->log_extents_lock[index]); |
| 537 | btrfs_put_ordered_extent(ordered); |
| 538 | spin_lock_irq(&log->log_extents_lock[index]); |
| 539 | } |
| 540 | spin_unlock_irq(&log->log_extents_lock[index]); |
| 541 | } |
| 542 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 543 | /* |
| 544 | * used to drop a reference on an ordered extent. This will free |
| 545 | * the extent if the last reference is dropped |
| 546 | */ |
Jeff Mahoney | 143bede | 2012-03-01 14:56:26 +0100 | [diff] [blame] | 547 | void btrfs_put_ordered_extent(struct btrfs_ordered_extent *entry) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 548 | { |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 549 | struct list_head *cur; |
| 550 | struct btrfs_ordered_sum *sum; |
| 551 | |
liubo | 1abe9b8 | 2011-03-24 11:18:59 +0000 | [diff] [blame] | 552 | trace_btrfs_ordered_extent_put(entry->inode, entry); |
| 553 | |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 554 | if (atomic_dec_and_test(&entry->refs)) { |
Filipe Manana | 61de718 | 2015-07-01 12:13:10 +0100 | [diff] [blame] | 555 | ASSERT(list_empty(&entry->log_list)); |
| 556 | ASSERT(list_empty(&entry->trans_list)); |
| 557 | ASSERT(list_empty(&entry->root_extent_list)); |
| 558 | ASSERT(RB_EMPTY_NODE(&entry->rb_node)); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 559 | if (entry->inode) |
| 560 | btrfs_add_delayed_iput(entry->inode); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 561 | while (!list_empty(&entry->list)) { |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 562 | cur = entry->list.next; |
| 563 | sum = list_entry(cur, struct btrfs_ordered_sum, list); |
| 564 | list_del(&sum->list); |
| 565 | kfree(sum); |
| 566 | } |
Miao Xie | 6352b91 | 2012-09-06 04:01:51 -0600 | [diff] [blame] | 567 | kmem_cache_free(btrfs_ordered_extent_cache, entry); |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 568 | } |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 569 | } |
| 570 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 571 | /* |
| 572 | * remove an ordered extent from the tree. No references are dropped |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 573 | * and waiters are woken up. |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 574 | */ |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 575 | void btrfs_remove_ordered_extent(struct inode *inode, |
| 576 | struct btrfs_ordered_extent *entry) |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 577 | { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 578 | struct btrfs_ordered_inode_tree *tree; |
Josef Bacik | 287a0ab | 2010-03-19 18:07:23 +0000 | [diff] [blame] | 579 | struct btrfs_root *root = BTRFS_I(inode)->root; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 580 | struct rb_node *node; |
| 581 | |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 582 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 583 | spin_lock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 584 | node = &entry->rb_node; |
Chris Mason | dc17ff8 | 2008-01-08 15:46:30 -0500 | [diff] [blame] | 585 | rb_erase(node, &tree->tree); |
Filipe Manana | 61de718 | 2015-07-01 12:13:10 +0100 | [diff] [blame] | 586 | RB_CLEAR_NODE(node); |
Filipe David Borba Manana | 1b8e7e4 | 2013-11-22 18:54:58 +0000 | [diff] [blame] | 587 | if (tree->last == node) |
| 588 | tree->last = NULL; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 589 | set_bit(BTRFS_ORDERED_COMPLETE, &entry->flags); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 590 | spin_unlock_irq(&tree->lock); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 591 | |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 592 | spin_lock(&root->ordered_extent_lock); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 593 | list_del_init(&entry->root_extent_list); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 594 | root->nr_ordered_extents--; |
Chris Mason | 5a3f23d | 2009-03-31 13:27:11 -0400 | [diff] [blame] | 595 | |
liubo | 1abe9b8 | 2011-03-24 11:18:59 +0000 | [diff] [blame] | 596 | trace_btrfs_ordered_extent_remove(inode, entry); |
| 597 | |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 598 | if (!root->nr_ordered_extents) { |
| 599 | spin_lock(&root->fs_info->ordered_root_lock); |
| 600 | BUG_ON(list_empty(&root->ordered_root)); |
| 601 | list_del_init(&root->ordered_root); |
| 602 | spin_unlock(&root->fs_info->ordered_root_lock); |
| 603 | } |
| 604 | spin_unlock(&root->ordered_extent_lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 605 | wake_up(&entry->wait); |
Chris Mason | 81d7ed2 | 2008-04-25 08:51:48 -0400 | [diff] [blame] | 606 | } |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 607 | |
Qu Wenruo | d458b05 | 2014-02-28 10:46:19 +0800 | [diff] [blame] | 608 | static void btrfs_run_ordered_extent_work(struct btrfs_work *work) |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 609 | { |
| 610 | struct btrfs_ordered_extent *ordered; |
| 611 | |
| 612 | ordered = container_of(work, struct btrfs_ordered_extent, flush_work); |
| 613 | btrfs_start_ordered_extent(ordered->inode, ordered, 1); |
| 614 | complete(&ordered->completion); |
| 615 | } |
| 616 | |
Chris Mason | d352ac6 | 2008-09-29 15:18:18 -0400 | [diff] [blame] | 617 | /* |
| 618 | * wait for all the ordered extents in a root. This is done when balancing |
| 619 | * space between drives. |
| 620 | */ |
Miao Xie | 31f3d25 | 2014-03-06 13:55:02 +0800 | [diff] [blame] | 621 | int btrfs_wait_ordered_extents(struct btrfs_root *root, int nr) |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 622 | { |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 623 | struct list_head splice, works; |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 624 | struct btrfs_ordered_extent *ordered, *next; |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 625 | int count = 0; |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 626 | |
| 627 | INIT_LIST_HEAD(&splice); |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 628 | INIT_LIST_HEAD(&works); |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 629 | |
Miao Xie | 31f3d25 | 2014-03-06 13:55:02 +0800 | [diff] [blame] | 630 | mutex_lock(&root->ordered_extent_mutex); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 631 | spin_lock(&root->ordered_extent_lock); |
| 632 | list_splice_init(&root->ordered_extents, &splice); |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 633 | while (!list_empty(&splice) && nr) { |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 634 | ordered = list_first_entry(&splice, struct btrfs_ordered_extent, |
| 635 | root_extent_list); |
| 636 | list_move_tail(&ordered->root_extent_list, |
| 637 | &root->ordered_extents); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 638 | atomic_inc(&ordered->refs); |
| 639 | spin_unlock(&root->ordered_extent_lock); |
| 640 | |
Qu Wenruo | a44903a | 2014-02-28 10:46:09 +0800 | [diff] [blame] | 641 | btrfs_init_work(&ordered->flush_work, |
Liu Bo | 9e0af23 | 2014-08-15 23:36:53 +0800 | [diff] [blame] | 642 | btrfs_flush_delalloc_helper, |
Qu Wenruo | a44903a | 2014-02-28 10:46:09 +0800 | [diff] [blame] | 643 | btrfs_run_ordered_extent_work, NULL, NULL); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 644 | list_add_tail(&ordered->work_list, &works); |
Qu Wenruo | a44903a | 2014-02-28 10:46:09 +0800 | [diff] [blame] | 645 | btrfs_queue_work(root->fs_info->flush_workers, |
| 646 | &ordered->flush_work); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 647 | |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 648 | cond_resched(); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 649 | spin_lock(&root->ordered_extent_lock); |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 650 | if (nr != -1) |
| 651 | nr--; |
| 652 | count++; |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 653 | } |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 654 | list_splice_tail(&splice, &root->ordered_extents); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 655 | spin_unlock(&root->ordered_extent_lock); |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 656 | |
| 657 | list_for_each_entry_safe(ordered, next, &works, work_list) { |
| 658 | list_del_init(&ordered->work_list); |
| 659 | wait_for_completion(&ordered->completion); |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 660 | btrfs_put_ordered_extent(ordered); |
Miao Xie | 9afab88 | 2012-10-25 09:41:36 +0000 | [diff] [blame] | 661 | cond_resched(); |
| 662 | } |
Miao Xie | 31f3d25 | 2014-03-06 13:55:02 +0800 | [diff] [blame] | 663 | mutex_unlock(&root->ordered_extent_mutex); |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 664 | |
| 665 | return count; |
Chris Mason | 3eaa288 | 2008-07-24 11:57:52 -0400 | [diff] [blame] | 666 | } |
| 667 | |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 668 | void btrfs_wait_ordered_roots(struct btrfs_fs_info *fs_info, int nr) |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 669 | { |
| 670 | struct btrfs_root *root; |
| 671 | struct list_head splice; |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 672 | int done; |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 673 | |
| 674 | INIT_LIST_HEAD(&splice); |
| 675 | |
Miao Xie | 8b9d83c | 2014-03-06 13:54:55 +0800 | [diff] [blame] | 676 | mutex_lock(&fs_info->ordered_operations_mutex); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 677 | spin_lock(&fs_info->ordered_root_lock); |
| 678 | list_splice_init(&fs_info->ordered_roots, &splice); |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 679 | while (!list_empty(&splice) && nr) { |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 680 | root = list_first_entry(&splice, struct btrfs_root, |
| 681 | ordered_root); |
| 682 | root = btrfs_grab_fs_root(root); |
| 683 | BUG_ON(!root); |
| 684 | list_move_tail(&root->ordered_root, |
| 685 | &fs_info->ordered_roots); |
| 686 | spin_unlock(&fs_info->ordered_root_lock); |
| 687 | |
Miao Xie | 31f3d25 | 2014-03-06 13:55:02 +0800 | [diff] [blame] | 688 | done = btrfs_wait_ordered_extents(root, nr); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 689 | btrfs_put_fs_root(root); |
| 690 | |
| 691 | spin_lock(&fs_info->ordered_root_lock); |
Miao Xie | b024419 | 2013-11-04 23:13:25 +0800 | [diff] [blame] | 692 | if (nr != -1) { |
| 693 | nr -= done; |
| 694 | WARN_ON(nr < 0); |
| 695 | } |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 696 | } |
Miao Xie | 931aa87 | 2013-11-14 17:33:21 +0800 | [diff] [blame] | 697 | list_splice_tail(&splice, &fs_info->ordered_roots); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 698 | spin_unlock(&fs_info->ordered_root_lock); |
Miao Xie | 8b9d83c | 2014-03-06 13:54:55 +0800 | [diff] [blame] | 699 | mutex_unlock(&fs_info->ordered_operations_mutex); |
Miao Xie | 199c2a9 | 2013-05-15 07:48:23 +0000 | [diff] [blame] | 700 | } |
| 701 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 702 | /* |
| 703 | * Used to start IO or wait for a given ordered extent to finish. |
| 704 | * |
| 705 | * If wait is one, this effectively waits on page writeback for all the pages |
| 706 | * in the extent, and it waits on the io completion code to insert |
| 707 | * metadata into the btree corresponding to the extent |
| 708 | */ |
| 709 | void btrfs_start_ordered_extent(struct inode *inode, |
| 710 | struct btrfs_ordered_extent *entry, |
| 711 | int wait) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 712 | { |
| 713 | u64 start = entry->file_offset; |
| 714 | u64 end = start + entry->len - 1; |
| 715 | |
liubo | 1abe9b8 | 2011-03-24 11:18:59 +0000 | [diff] [blame] | 716 | trace_btrfs_ordered_extent_start(inode, entry); |
| 717 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 718 | /* |
| 719 | * pages in the range can be dirty, clean or writeback. We |
| 720 | * start IO on any dirty ones so the wait doesn't stall waiting |
Artem Bityutskiy | b257031 | 2012-07-25 18:12:06 +0300 | [diff] [blame] | 721 | * for the flusher thread to find them |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 722 | */ |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 723 | if (!test_bit(BTRFS_ORDERED_DIRECT, &entry->flags)) |
| 724 | filemap_fdatawrite_range(inode->i_mapping, start, end); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 725 | if (wait) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 726 | wait_event(entry->wait, test_bit(BTRFS_ORDERED_COMPLETE, |
| 727 | &entry->flags)); |
Chris Mason | c8b9781 | 2008-10-29 14:49:59 -0400 | [diff] [blame] | 728 | } |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 729 | } |
| 730 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 731 | /* |
| 732 | * Used to wait on ordered extents across a large range of bytes. |
| 733 | */ |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 734 | int btrfs_wait_ordered_range(struct inode *inode, u64 start, u64 len) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 735 | { |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 736 | int ret = 0; |
Filipe Manana | 28aeeac | 2015-05-05 19:03:10 +0100 | [diff] [blame] | 737 | int ret_wb = 0; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 738 | u64 end; |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 739 | u64 orig_end; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 740 | struct btrfs_ordered_extent *ordered; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 741 | |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 742 | if (start + len < start) { |
Chris Mason | f421950 | 2008-07-22 11:18:09 -0400 | [diff] [blame] | 743 | orig_end = INT_LIMIT(loff_t); |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 744 | } else { |
| 745 | orig_end = start + len - 1; |
Chris Mason | f421950 | 2008-07-22 11:18:09 -0400 | [diff] [blame] | 746 | if (orig_end > INT_LIMIT(loff_t)) |
| 747 | orig_end = INT_LIMIT(loff_t); |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 748 | } |
Josef Bacik | 551ebb2 | 2012-04-23 14:41:09 -0400 | [diff] [blame] | 749 | |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 750 | /* start IO across the range first to instantiate any delalloc |
| 751 | * extents |
| 752 | */ |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 753 | ret = btrfs_fdatawrite_range(inode, start, orig_end); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 754 | if (ret) |
| 755 | return ret; |
Filipe Manana | 728404d | 2014-10-10 09:43:11 +0100 | [diff] [blame] | 756 | |
Filipe Manana | 28aeeac | 2015-05-05 19:03:10 +0100 | [diff] [blame] | 757 | /* |
| 758 | * If we have a writeback error don't return immediately. Wait first |
| 759 | * for any ordered extents that haven't completed yet. This is to make |
| 760 | * sure no one can dirty the same page ranges and call writepages() |
| 761 | * before the ordered extents complete - to avoid failures (-EEXIST) |
| 762 | * when adding the new ordered extents to the ordered tree. |
| 763 | */ |
| 764 | ret_wb = filemap_fdatawait_range(inode->i_mapping, start, orig_end); |
Chris Mason | f421950 | 2008-07-22 11:18:09 -0400 | [diff] [blame] | 765 | |
| 766 | end = orig_end; |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 767 | while (1) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 768 | ordered = btrfs_lookup_first_ordered_extent(inode, end); |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 769 | if (!ordered) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 770 | break; |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 771 | if (ordered->file_offset > orig_end) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 772 | btrfs_put_ordered_extent(ordered); |
| 773 | break; |
| 774 | } |
Filipe David Borba Manana | b52abf1 | 2013-11-06 15:12:40 +0000 | [diff] [blame] | 775 | if (ordered->file_offset + ordered->len <= start) { |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 776 | btrfs_put_ordered_extent(ordered); |
| 777 | break; |
| 778 | } |
Chris Mason | e5a2217 | 2008-07-18 20:42:20 -0400 | [diff] [blame] | 779 | btrfs_start_ordered_extent(inode, ordered, 1); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 780 | end = ordered->file_offset; |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 781 | if (test_bit(BTRFS_ORDERED_IOERR, &ordered->flags)) |
| 782 | ret = -EIO; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 783 | btrfs_put_ordered_extent(ordered); |
Josef Bacik | 0ef8b72 | 2013-10-25 16:13:35 -0400 | [diff] [blame] | 784 | if (ret || end == 0 || end == start) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 785 | break; |
| 786 | end--; |
| 787 | } |
Filipe Manana | 28aeeac | 2015-05-05 19:03:10 +0100 | [diff] [blame] | 788 | return ret_wb ? ret_wb : ret; |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 789 | } |
| 790 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 791 | /* |
| 792 | * find an ordered extent corresponding to file_offset. return NULL if |
| 793 | * nothing is found, otherwise take a reference on the extent and return it |
| 794 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 795 | struct btrfs_ordered_extent *btrfs_lookup_ordered_extent(struct inode *inode, |
| 796 | u64 file_offset) |
| 797 | { |
| 798 | struct btrfs_ordered_inode_tree *tree; |
| 799 | struct rb_node *node; |
| 800 | struct btrfs_ordered_extent *entry = NULL; |
| 801 | |
| 802 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 803 | spin_lock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 804 | node = tree_search(tree, file_offset); |
| 805 | if (!node) |
| 806 | goto out; |
| 807 | |
| 808 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 809 | if (!offset_in_entry(entry, file_offset)) |
| 810 | entry = NULL; |
| 811 | if (entry) |
| 812 | atomic_inc(&entry->refs); |
| 813 | out: |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 814 | spin_unlock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 815 | return entry; |
| 816 | } |
| 817 | |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 818 | /* Since the DIO code tries to lock a wide area we need to look for any ordered |
| 819 | * extents that exist in the range, rather than just the start of the range. |
| 820 | */ |
| 821 | struct btrfs_ordered_extent *btrfs_lookup_ordered_range(struct inode *inode, |
| 822 | u64 file_offset, |
| 823 | u64 len) |
| 824 | { |
| 825 | struct btrfs_ordered_inode_tree *tree; |
| 826 | struct rb_node *node; |
| 827 | struct btrfs_ordered_extent *entry = NULL; |
| 828 | |
| 829 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 830 | spin_lock_irq(&tree->lock); |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 831 | node = tree_search(tree, file_offset); |
| 832 | if (!node) { |
| 833 | node = tree_search(tree, file_offset + len); |
| 834 | if (!node) |
| 835 | goto out; |
| 836 | } |
| 837 | |
| 838 | while (1) { |
| 839 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 840 | if (range_overlaps(entry, file_offset, len)) |
| 841 | break; |
| 842 | |
| 843 | if (entry->file_offset >= file_offset + len) { |
| 844 | entry = NULL; |
| 845 | break; |
| 846 | } |
| 847 | entry = NULL; |
| 848 | node = rb_next(node); |
| 849 | if (!node) |
| 850 | break; |
| 851 | } |
| 852 | out: |
| 853 | if (entry) |
| 854 | atomic_inc(&entry->refs); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 855 | spin_unlock_irq(&tree->lock); |
Josef Bacik | 4b46fce | 2010-05-23 11:00:55 -0400 | [diff] [blame] | 856 | return entry; |
| 857 | } |
| 858 | |
Filipe Manana | b659ef0 | 2015-03-31 14:16:52 +0100 | [diff] [blame] | 859 | bool btrfs_have_ordered_extents_in_range(struct inode *inode, |
| 860 | u64 file_offset, |
| 861 | u64 len) |
| 862 | { |
| 863 | struct btrfs_ordered_extent *oe; |
| 864 | |
| 865 | oe = btrfs_lookup_ordered_range(inode, file_offset, len); |
| 866 | if (oe) { |
| 867 | btrfs_put_ordered_extent(oe); |
| 868 | return true; |
| 869 | } |
| 870 | return false; |
| 871 | } |
| 872 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 873 | /* |
| 874 | * lookup and return any extent before 'file_offset'. NULL is returned |
| 875 | * if none is found |
| 876 | */ |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 877 | struct btrfs_ordered_extent * |
Chris Mason | d397712 | 2009-01-05 21:25:51 -0500 | [diff] [blame] | 878 | btrfs_lookup_first_ordered_extent(struct inode *inode, u64 file_offset) |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 879 | { |
| 880 | struct btrfs_ordered_inode_tree *tree; |
| 881 | struct rb_node *node; |
| 882 | struct btrfs_ordered_extent *entry = NULL; |
| 883 | |
| 884 | tree = &BTRFS_I(inode)->ordered_tree; |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 885 | spin_lock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 886 | node = tree_search(tree, file_offset); |
| 887 | if (!node) |
| 888 | goto out; |
| 889 | |
| 890 | entry = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
| 891 | atomic_inc(&entry->refs); |
| 892 | out: |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 893 | spin_unlock_irq(&tree->lock); |
Chris Mason | e6dcd2d | 2008-07-17 12:53:50 -0400 | [diff] [blame] | 894 | return entry; |
| 895 | } |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 896 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 897 | /* |
| 898 | * After an extent is done, call this to conditionally update the on disk |
| 899 | * i_size. i_size is updated to cover any fully written part of the file. |
| 900 | */ |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 901 | int btrfs_ordered_update_i_size(struct inode *inode, u64 offset, |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 902 | struct btrfs_ordered_extent *ordered) |
| 903 | { |
| 904 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 905 | u64 disk_i_size; |
| 906 | u64 new_i_size; |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 907 | u64 i_size = i_size_read(inode); |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 908 | struct rb_node *node; |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 909 | struct rb_node *prev = NULL; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 910 | struct btrfs_ordered_extent *test; |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 911 | int ret = 1; |
| 912 | |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 913 | spin_lock_irq(&tree->lock); |
Josef Bacik | 77cef2e | 2013-08-29 13:57:21 -0400 | [diff] [blame] | 914 | if (ordered) { |
| 915 | offset = entry_end(ordered); |
| 916 | if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags)) |
| 917 | offset = min(offset, |
| 918 | ordered->file_offset + |
| 919 | ordered->truncated_len); |
| 920 | } else { |
| 921 | offset = ALIGN(offset, BTRFS_I(inode)->root->sectorsize); |
| 922 | } |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 923 | disk_i_size = BTRFS_I(inode)->disk_i_size; |
| 924 | |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 925 | /* truncate file */ |
| 926 | if (disk_i_size > i_size) { |
| 927 | BTRFS_I(inode)->disk_i_size = i_size; |
| 928 | ret = 0; |
| 929 | goto out; |
| 930 | } |
| 931 | |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 932 | /* |
| 933 | * if the disk i_size is already at the inode->i_size, or |
| 934 | * this ordered extent is inside the disk i_size, we're done |
| 935 | */ |
Josef Bacik | 5d1f402 | 2013-01-30 14:17:31 -0500 | [diff] [blame] | 936 | if (disk_i_size == i_size) |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 937 | goto out; |
Josef Bacik | 5d1f402 | 2013-01-30 14:17:31 -0500 | [diff] [blame] | 938 | |
| 939 | /* |
| 940 | * We still need to update disk_i_size if outstanding_isize is greater |
| 941 | * than disk_i_size. |
| 942 | */ |
| 943 | if (offset <= disk_i_size && |
| 944 | (!ordered || ordered->outstanding_isize <= disk_i_size)) |
| 945 | goto out; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 946 | |
| 947 | /* |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 948 | * walk backward from this ordered extent to disk_i_size. |
| 949 | * if we find an ordered extent then we can't update disk i_size |
| 950 | * yet |
| 951 | */ |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 952 | if (ordered) { |
| 953 | node = rb_prev(&ordered->rb_node); |
| 954 | } else { |
| 955 | prev = tree_search(tree, offset); |
| 956 | /* |
| 957 | * we insert file extents without involving ordered struct, |
| 958 | * so there should be no ordered struct cover this offset |
| 959 | */ |
| 960 | if (prev) { |
| 961 | test = rb_entry(prev, struct btrfs_ordered_extent, |
| 962 | rb_node); |
| 963 | BUG_ON(offset_in_entry(test, offset)); |
| 964 | } |
| 965 | node = prev; |
| 966 | } |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 967 | for (; node; node = rb_prev(node)) { |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 968 | test = rb_entry(node, struct btrfs_ordered_extent, rb_node); |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 969 | |
| 970 | /* We treat this entry as if it doesnt exist */ |
| 971 | if (test_bit(BTRFS_ORDERED_UPDATED_ISIZE, &test->flags)) |
| 972 | continue; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 973 | if (test->file_offset + test->len <= disk_i_size) |
| 974 | break; |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 975 | if (test->file_offset >= i_size) |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 976 | break; |
Josef Bacik | 59fe4f4 | 2013-01-30 14:31:31 -0500 | [diff] [blame] | 977 | if (entry_end(test) > disk_i_size) { |
Miao Xie | b9a8cc5 | 2012-09-06 04:01:21 -0600 | [diff] [blame] | 978 | /* |
| 979 | * we don't update disk_i_size now, so record this |
| 980 | * undealt i_size. Or we will not know the real |
| 981 | * i_size. |
| 982 | */ |
| 983 | if (test->outstanding_isize < offset) |
| 984 | test->outstanding_isize = offset; |
| 985 | if (ordered && |
| 986 | ordered->outstanding_isize > |
| 987 | test->outstanding_isize) |
| 988 | test->outstanding_isize = |
| 989 | ordered->outstanding_isize; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 990 | goto out; |
Miao Xie | b9a8cc5 | 2012-09-06 04:01:21 -0600 | [diff] [blame] | 991 | } |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 992 | } |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 993 | new_i_size = min_t(u64, offset, i_size); |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 994 | |
| 995 | /* |
Miao Xie | b9a8cc5 | 2012-09-06 04:01:21 -0600 | [diff] [blame] | 996 | * Some ordered extents may completed before the current one, and |
| 997 | * we hold the real i_size in ->outstanding_isize. |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 998 | */ |
Miao Xie | b9a8cc5 | 2012-09-06 04:01:21 -0600 | [diff] [blame] | 999 | if (ordered && ordered->outstanding_isize > new_i_size) |
| 1000 | new_i_size = min_t(u64, ordered->outstanding_isize, i_size); |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 1001 | BTRFS_I(inode)->disk_i_size = new_i_size; |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 1002 | ret = 0; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 1003 | out: |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 1004 | /* |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 1005 | * We need to do this because we can't remove ordered extents until |
| 1006 | * after the i_disk_size has been updated and then the inode has been |
| 1007 | * updated to reflect the change, so we need to tell anybody who finds |
| 1008 | * this ordered extent that we've already done all the real work, we |
| 1009 | * just haven't completed all the other work. |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 1010 | */ |
| 1011 | if (ordered) |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 1012 | set_bit(BTRFS_ORDERED_UPDATED_ISIZE, &ordered->flags); |
| 1013 | spin_unlock_irq(&tree->lock); |
Yan, Zheng | c216775 | 2009-11-12 09:34:21 +0000 | [diff] [blame] | 1014 | return ret; |
Chris Mason | dbe674a | 2008-07-17 12:54:05 -0400 | [diff] [blame] | 1015 | } |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1016 | |
Chris Mason | eb84ae0 | 2008-07-17 13:53:27 -0400 | [diff] [blame] | 1017 | /* |
| 1018 | * search the ordered extents for one corresponding to 'offset' and |
| 1019 | * try to find a checksum. This is used because we allow pages to |
| 1020 | * be reclaimed before their checksum is actually put into the btree |
| 1021 | */ |
Chris Mason | d20f704 | 2008-12-08 16:58:54 -0500 | [diff] [blame] | 1022 | int btrfs_find_ordered_sum(struct inode *inode, u64 offset, u64 disk_bytenr, |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1023 | u32 *sum, int len) |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1024 | { |
| 1025 | struct btrfs_ordered_sum *ordered_sum; |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1026 | struct btrfs_ordered_extent *ordered; |
| 1027 | struct btrfs_ordered_inode_tree *tree = &BTRFS_I(inode)->ordered_tree; |
Chris Mason | 3edf7d3 | 2008-07-18 06:17:13 -0400 | [diff] [blame] | 1028 | unsigned long num_sectors; |
| 1029 | unsigned long i; |
| 1030 | u32 sectorsize = BTRFS_I(inode)->root->sectorsize; |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1031 | int index = 0; |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1032 | |
| 1033 | ordered = btrfs_lookup_ordered_extent(inode, offset); |
| 1034 | if (!ordered) |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1035 | return 0; |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1036 | |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 1037 | spin_lock_irq(&tree->lock); |
Qinghuang Feng | c6e3087 | 2009-01-21 10:59:08 -0500 | [diff] [blame] | 1038 | list_for_each_entry_reverse(ordered_sum, &ordered->list, list) { |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1039 | if (disk_bytenr >= ordered_sum->bytenr && |
| 1040 | disk_bytenr < ordered_sum->bytenr + ordered_sum->len) { |
| 1041 | i = (disk_bytenr - ordered_sum->bytenr) >> |
| 1042 | inode->i_sb->s_blocksize_bits; |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1043 | num_sectors = ordered_sum->len >> |
| 1044 | inode->i_sb->s_blocksize_bits; |
Miao Xie | f51a4a1 | 2013-06-19 10:36:09 +0800 | [diff] [blame] | 1045 | num_sectors = min_t(int, len - index, num_sectors - i); |
| 1046 | memcpy(sum + index, ordered_sum->sums + i, |
| 1047 | num_sectors); |
| 1048 | |
| 1049 | index += (int)num_sectors; |
| 1050 | if (index == len) |
| 1051 | goto out; |
| 1052 | disk_bytenr += num_sectors * sectorsize; |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1053 | } |
| 1054 | } |
| 1055 | out: |
Josef Bacik | 5fd0204 | 2012-05-02 14:00:54 -0400 | [diff] [blame] | 1056 | spin_unlock_irq(&tree->lock); |
Chris Mason | 8964222 | 2008-07-24 09:41:53 -0400 | [diff] [blame] | 1057 | btrfs_put_ordered_extent(ordered); |
Miao Xie | e4100d9 | 2013-04-05 07:20:56 +0000 | [diff] [blame] | 1058 | return index; |
Chris Mason | ba1da2f | 2008-07-17 12:54:15 -0400 | [diff] [blame] | 1059 | } |
| 1060 | |
Miao Xie | 6352b91 | 2012-09-06 04:01:51 -0600 | [diff] [blame] | 1061 | int __init ordered_data_init(void) |
| 1062 | { |
| 1063 | btrfs_ordered_extent_cache = kmem_cache_create("btrfs_ordered_extent", |
| 1064 | sizeof(struct btrfs_ordered_extent), 0, |
| 1065 | SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, |
| 1066 | NULL); |
| 1067 | if (!btrfs_ordered_extent_cache) |
| 1068 | return -ENOMEM; |
Miao Xie | 25287e0 | 2012-10-25 09:31:03 +0000 | [diff] [blame] | 1069 | |
Miao Xie | 6352b91 | 2012-09-06 04:01:51 -0600 | [diff] [blame] | 1070 | return 0; |
| 1071 | } |
| 1072 | |
| 1073 | void ordered_data_exit(void) |
| 1074 | { |
| 1075 | if (btrfs_ordered_extent_cache) |
| 1076 | kmem_cache_destroy(btrfs_ordered_extent_cache); |
| 1077 | } |