1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 2
3 #include <linux/blkdev.h> 3 #include <linux/blkdev.h>
4 #include <linux/iversion.h> 4 #include <linux/iversion.h>
5 #include "ctree.h" 5 #include "ctree.h"
6 #include "fs.h" 6 #include "fs.h"
7 #include "messages.h" 7 #include "messages.h"
8 #include "compression.h" 8 #include "compression.h"
9 #include "delalloc-space.h" 9 #include "delalloc-space.h"
10 #include "disk-io.h" 10 #include "disk-io.h"
11 #include "reflink.h" 11 #include "reflink.h"
12 #include "transaction.h" 12 #include "transaction.h"
13 #include "subpage.h" 13 #include "subpage.h"
14 #include "accessors.h" 14 #include "accessors.h"
15 #include "file-item.h" 15 #include "file-item.h"
16 #include "file.h" 16 #include "file.h"
17 #include "super.h" 17 #include "super.h"
18 18
19 #define BTRFS_MAX_DEDUPE_LEN SZ_16M 19 #define BTRFS_MAX_DEDUPE_LEN SZ_16M
20 20
21 static int clone_finish_inode_update(struct bt 21 static int clone_finish_inode_update(struct btrfs_trans_handle *trans,
22 struct in 22 struct inode *inode,
23 u64 endof 23 u64 endoff,
24 const u64 24 const u64 destoff,
25 const u64 25 const u64 olen,
26 int no_ti 26 int no_time_update)
27 { 27 {
>> 28 struct btrfs_root *root = BTRFS_I(inode)->root;
28 int ret; 29 int ret;
29 30
30 inode_inc_iversion(inode); 31 inode_inc_iversion(inode);
31 if (!no_time_update) { 32 if (!no_time_update) {
32 inode_set_mtime_to_ts(inode, i !! 33 inode->i_mtime = current_time(inode);
>> 34 inode->i_ctime = inode->i_mtime;
33 } 35 }
34 /* 36 /*
35 * We round up to the block size at eo 37 * We round up to the block size at eof when determining which
36 * extents to clone above, but shouldn 38 * extents to clone above, but shouldn't round up the file size.
37 */ 39 */
38 if (endoff > destoff + olen) 40 if (endoff > destoff + olen)
39 endoff = destoff + olen; 41 endoff = destoff + olen;
40 if (endoff > inode->i_size) { 42 if (endoff > inode->i_size) {
41 i_size_write(inode, endoff); 43 i_size_write(inode, endoff);
42 btrfs_inode_safe_disk_i_size_w 44 btrfs_inode_safe_disk_i_size_write(BTRFS_I(inode), 0);
43 } 45 }
44 46
45 ret = btrfs_update_inode(trans, BTRFS_ !! 47 ret = btrfs_update_inode(trans, root, BTRFS_I(inode));
46 if (ret) { 48 if (ret) {
47 btrfs_abort_transaction(trans, 49 btrfs_abort_transaction(trans, ret);
48 btrfs_end_transaction(trans); 50 btrfs_end_transaction(trans);
49 goto out; 51 goto out;
50 } 52 }
51 ret = btrfs_end_transaction(trans); 53 ret = btrfs_end_transaction(trans);
52 out: 54 out:
53 return ret; 55 return ret;
54 } 56 }
55 57
56 static int copy_inline_to_page(struct btrfs_in 58 static int copy_inline_to_page(struct btrfs_inode *inode,
57 const u64 file_ 59 const u64 file_offset,
58 char *inline_da 60 char *inline_data,
59 const u64 size, 61 const u64 size,
60 const u64 datal 62 const u64 datal,
61 const u8 comp_t 63 const u8 comp_type)
62 { 64 {
63 struct btrfs_fs_info *fs_info = inode- 65 struct btrfs_fs_info *fs_info = inode->root->fs_info;
64 const u32 block_size = fs_info->sector 66 const u32 block_size = fs_info->sectorsize;
65 const u64 range_end = file_offset + bl 67 const u64 range_end = file_offset + block_size - 1;
66 const size_t inline_size = size - btrf 68 const size_t inline_size = size - btrfs_file_extent_calc_inline_size(0);
67 char *data_start = inline_data + btrfs 69 char *data_start = inline_data + btrfs_file_extent_calc_inline_size(0);
68 struct extent_changeset *data_reserved 70 struct extent_changeset *data_reserved = NULL;
69 struct folio *folio = NULL; !! 71 struct page *page = NULL;
70 struct address_space *mapping = inode- 72 struct address_space *mapping = inode->vfs_inode.i_mapping;
71 int ret; 73 int ret;
72 74
73 ASSERT(IS_ALIGNED(file_offset, block_s 75 ASSERT(IS_ALIGNED(file_offset, block_size));
74 76
75 /* 77 /*
76 * We have flushed and locked the rang 78 * We have flushed and locked the ranges of the source and destination
77 * inodes, we also have locked the ino 79 * inodes, we also have locked the inodes, so we are safe to do a
78 * reservation here. Also we must not 80 * reservation here. Also we must not do the reservation while holding
79 * a transaction open, otherwise we wo 81 * a transaction open, otherwise we would deadlock.
80 */ 82 */
81 ret = btrfs_delalloc_reserve_space(ino 83 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, file_offset,
82 blo 84 block_size);
83 if (ret) 85 if (ret)
84 goto out; 86 goto out;
85 87
86 folio = __filemap_get_folio(mapping, f !! 88 page = find_or_create_page(mapping, file_offset >> PAGE_SHIFT,
87 FGP_LO !! 89 btrfs_alloc_write_mask(mapping));
88 btrfs_ !! 90 if (!page) {
89 if (IS_ERR(folio)) { <<
90 ret = -ENOMEM; 91 ret = -ENOMEM;
91 goto out_unlock; 92 goto out_unlock;
92 } 93 }
93 94
94 ret = set_folio_extent_mapped(folio); !! 95 ret = set_page_extent_mapped(page);
95 if (ret < 0) 96 if (ret < 0)
96 goto out_unlock; 97 goto out_unlock;
97 98
98 clear_extent_bit(&inode->io_tree, file 99 clear_extent_bit(&inode->io_tree, file_offset, range_end,
99 EXTENT_DELALLOC | EXT 100 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
100 NULL); 101 NULL);
101 ret = btrfs_set_extent_delalloc(inode, 102 ret = btrfs_set_extent_delalloc(inode, file_offset, range_end, 0, NULL);
102 if (ret) 103 if (ret)
103 goto out_unlock; 104 goto out_unlock;
104 105
105 /* 106 /*
106 * After dirtying the page our caller 107 * After dirtying the page our caller will need to start a transaction,
107 * and if we are low on metadata free 108 * and if we are low on metadata free space, that can cause flushing of
108 * delalloc for all inodes in order to 109 * delalloc for all inodes in order to get metadata space released.
109 * However we are holding the range lo 110 * However we are holding the range locked for the whole duration of
110 * the clone/dedupe operation, so we m 111 * the clone/dedupe operation, so we may deadlock if that happens and no
111 * other task releases enough space. S 112 * other task releases enough space. So mark this inode as not being
112 * possible to flush to avoid such dea 113 * possible to flush to avoid such deadlock. We will clear that flag
113 * when we finish cloning all extents, 114 * when we finish cloning all extents, since a transaction is started
114 * after finding each extent to clone. 115 * after finding each extent to clone.
115 */ 116 */
116 set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, 117 set_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &inode->runtime_flags);
117 118
118 if (comp_type == BTRFS_COMPRESS_NONE) 119 if (comp_type == BTRFS_COMPRESS_NONE) {
119 memcpy_to_folio(folio, offset_ !! 120 memcpy_to_page(page, offset_in_page(file_offset), data_start,
120 datal) !! 121 datal);
121 } else { 122 } else {
122 ret = btrfs_decompress(comp_ty !! 123 ret = btrfs_decompress(comp_type, data_start, page,
123 offset_ !! 124 offset_in_page(file_offset),
124 inline_ 125 inline_size, datal);
125 if (ret) 126 if (ret)
126 goto out_unlock; 127 goto out_unlock;
127 flush_dcache_folio(folio); !! 128 flush_dcache_page(page);
128 } 129 }
129 130
130 /* 131 /*
131 * If our inline data is smaller then 132 * If our inline data is smaller then the block/page size, then the
132 * remaining of the block/page is equi 133 * remaining of the block/page is equivalent to zeroes. We had something
133 * like the following done: 134 * like the following done:
134 * 135 *
135 * $ xfs_io -f -c "pwrite -S 0xab 0 50 136 * $ xfs_io -f -c "pwrite -S 0xab 0 500" file
136 * $ sync # (or fsync) 137 * $ sync # (or fsync)
137 * $ xfs_io -c "falloc 0 4K" file 138 * $ xfs_io -c "falloc 0 4K" file
138 * $ xfs_io -c "pwrite -S 0xcd 4K 4K" 139 * $ xfs_io -c "pwrite -S 0xcd 4K 4K"
139 * 140 *
140 * So what's in the range [500, 4095] 141 * So what's in the range [500, 4095] corresponds to zeroes.
141 */ 142 */
142 if (datal < block_size) 143 if (datal < block_size)
143 folio_zero_range(folio, datal, !! 144 memzero_page(page, datal, block_size - datal);
144 145
145 btrfs_folio_set_uptodate(fs_info, foli !! 146 btrfs_page_set_uptodate(fs_info, page, file_offset, block_size);
146 btrfs_folio_clear_checked(fs_info, fol !! 147 btrfs_page_clear_checked(fs_info, page, file_offset, block_size);
147 btrfs_folio_set_dirty(fs_info, folio, !! 148 btrfs_page_set_dirty(fs_info, page, file_offset, block_size);
148 out_unlock: 149 out_unlock:
149 if (!IS_ERR(folio)) { !! 150 if (page) {
150 folio_unlock(folio); !! 151 unlock_page(page);
151 folio_put(folio); !! 152 put_page(page);
152 } 153 }
153 if (ret) 154 if (ret)
154 btrfs_delalloc_release_space(i 155 btrfs_delalloc_release_space(inode, data_reserved, file_offset,
155 b 156 block_size, true);
156 btrfs_delalloc_release_extents(inode, 157 btrfs_delalloc_release_extents(inode, block_size);
157 out: 158 out:
158 extent_changeset_free(data_reserved); 159 extent_changeset_free(data_reserved);
159 160
160 return ret; 161 return ret;
161 } 162 }
162 163
163 /* 164 /*
164 * Deal with cloning of inline extents. We try 165 * Deal with cloning of inline extents. We try to copy the inline extent from
165 * the source inode to destination inode when 166 * the source inode to destination inode when possible. When not possible we
166 * copy the inline extent's data into the resp 167 * copy the inline extent's data into the respective page of the inode.
167 */ 168 */
168 static int clone_copy_inline_extent(struct ino 169 static int clone_copy_inline_extent(struct inode *dst,
169 struct btr 170 struct btrfs_path *path,
170 struct btr 171 struct btrfs_key *new_key,
171 const u64 172 const u64 drop_start,
172 const u64 173 const u64 datal,
173 const u64 174 const u64 size,
174 const u8 c 175 const u8 comp_type,
175 char *inli 176 char *inline_data,
176 struct btr 177 struct btrfs_trans_handle **trans_out)
177 { 178 {
178 struct btrfs_fs_info *fs_info = inode_ !! 179 struct btrfs_fs_info *fs_info = btrfs_sb(dst->i_sb);
179 struct btrfs_root *root = BTRFS_I(dst) 180 struct btrfs_root *root = BTRFS_I(dst)->root;
180 const u64 aligned_end = ALIGN(new_key- 181 const u64 aligned_end = ALIGN(new_key->offset + datal,
181 fs_info- 182 fs_info->sectorsize);
182 struct btrfs_trans_handle *trans = NUL 183 struct btrfs_trans_handle *trans = NULL;
183 struct btrfs_drop_extents_args drop_ar 184 struct btrfs_drop_extents_args drop_args = { 0 };
184 int ret; 185 int ret;
185 struct btrfs_key key; 186 struct btrfs_key key;
186 187
187 if (new_key->offset > 0) { 188 if (new_key->offset > 0) {
188 ret = copy_inline_to_page(BTRF 189 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
189 inli 190 inline_data, size, datal, comp_type);
190 goto out; 191 goto out;
191 } 192 }
192 193
193 key.objectid = btrfs_ino(BTRFS_I(dst)) 194 key.objectid = btrfs_ino(BTRFS_I(dst));
194 key.type = BTRFS_EXTENT_DATA_KEY; 195 key.type = BTRFS_EXTENT_DATA_KEY;
195 key.offset = 0; 196 key.offset = 0;
196 ret = btrfs_search_slot(NULL, root, &k 197 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
197 if (ret < 0) { 198 if (ret < 0) {
198 return ret; 199 return ret;
199 } else if (ret > 0) { 200 } else if (ret > 0) {
200 if (path->slots[0] >= btrfs_he 201 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
201 ret = btrfs_next_leaf( 202 ret = btrfs_next_leaf(root, path);
202 if (ret < 0) 203 if (ret < 0)
203 return ret; 204 return ret;
204 else if (ret > 0) 205 else if (ret > 0)
205 goto copy_inli 206 goto copy_inline_extent;
206 } 207 }
207 btrfs_item_key_to_cpu(path->no 208 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
208 if (key.objectid == btrfs_ino( 209 if (key.objectid == btrfs_ino(BTRFS_I(dst)) &&
209 key.type == BTRFS_EXTENT_D 210 key.type == BTRFS_EXTENT_DATA_KEY) {
210 /* 211 /*
211 * There's an implicit 212 * There's an implicit hole at file offset 0, copy the
212 * inline extent's dat 213 * inline extent's data to the page.
213 */ 214 */
214 ASSERT(key.offset > 0) 215 ASSERT(key.offset > 0);
215 goto copy_to_page; 216 goto copy_to_page;
216 } 217 }
217 } else if (i_size_read(dst) <= datal) 218 } else if (i_size_read(dst) <= datal) {
218 struct btrfs_file_extent_item 219 struct btrfs_file_extent_item *ei;
219 220
220 ei = btrfs_item_ptr(path->node 221 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
221 struct btr 222 struct btrfs_file_extent_item);
222 /* 223 /*
223 * If it's an inline extent re 224 * If it's an inline extent replace it with the source inline
224 * extent, otherwise copy the 225 * extent, otherwise copy the source inline extent data into
225 * the respective page at the 226 * the respective page at the destination inode.
226 */ 227 */
227 if (btrfs_file_extent_type(pat 228 if (btrfs_file_extent_type(path->nodes[0], ei) ==
228 BTRFS_FILE_EXTENT_INLINE) 229 BTRFS_FILE_EXTENT_INLINE)
229 goto copy_inline_exten 230 goto copy_inline_extent;
230 231
231 goto copy_to_page; 232 goto copy_to_page;
232 } 233 }
233 234
234 copy_inline_extent: 235 copy_inline_extent:
235 /* 236 /*
236 * We have no extent items, or we have 237 * We have no extent items, or we have an extent at offset 0 which may
237 * or may not be inlined. All these ca 238 * or may not be inlined. All these cases are dealt the same way.
238 */ 239 */
239 if (i_size_read(dst) > datal) { 240 if (i_size_read(dst) > datal) {
240 /* 241 /*
241 * At the destination offset 0 242 * At the destination offset 0 we have either a hole, a regular
242 * extent or an inline extent 243 * extent or an inline extent larger then the one we want to
243 * clone. Deal with all these 244 * clone. Deal with all these cases by copying the inline extent
244 * data into the respective pa 245 * data into the respective page at the destination inode.
245 */ 246 */
246 goto copy_to_page; 247 goto copy_to_page;
247 } 248 }
248 249
249 /* 250 /*
250 * Release path before starting a new 251 * Release path before starting a new transaction so we don't hold locks
251 * that would confuse lockdep. 252 * that would confuse lockdep.
252 */ 253 */
253 btrfs_release_path(path); 254 btrfs_release_path(path);
254 /* 255 /*
255 * If we end up here it means were cop 256 * If we end up here it means were copy the inline extent into a leaf
256 * of the destination inode. We know w 257 * of the destination inode. We know we will drop or adjust at most one
257 * extent item in the destination root 258 * extent item in the destination root.
258 * 259 *
259 * 1 unit - adjusting old extent (we m 260 * 1 unit - adjusting old extent (we may have to split it)
260 * 1 unit - add new extent 261 * 1 unit - add new extent
261 * 1 unit - inode update 262 * 1 unit - inode update
262 */ 263 */
263 trans = btrfs_start_transaction(root, 264 trans = btrfs_start_transaction(root, 3);
264 if (IS_ERR(trans)) { 265 if (IS_ERR(trans)) {
265 ret = PTR_ERR(trans); 266 ret = PTR_ERR(trans);
266 trans = NULL; 267 trans = NULL;
267 goto out; 268 goto out;
268 } 269 }
269 drop_args.path = path; 270 drop_args.path = path;
270 drop_args.start = drop_start; 271 drop_args.start = drop_start;
271 drop_args.end = aligned_end; 272 drop_args.end = aligned_end;
272 drop_args.drop_cache = true; 273 drop_args.drop_cache = true;
273 ret = btrfs_drop_extents(trans, root, 274 ret = btrfs_drop_extents(trans, root, BTRFS_I(dst), &drop_args);
274 if (ret) 275 if (ret)
275 goto out; 276 goto out;
276 ret = btrfs_insert_empty_item(trans, r 277 ret = btrfs_insert_empty_item(trans, root, path, new_key, size);
277 if (ret) 278 if (ret)
278 goto out; 279 goto out;
279 280
280 write_extent_buffer(path->nodes[0], in 281 write_extent_buffer(path->nodes[0], inline_data,
281 btrfs_item_ptr_off 282 btrfs_item_ptr_offset(path->nodes[0],
282 283 path->slots[0]),
283 size); 284 size);
284 btrfs_update_inode_bytes(BTRFS_I(dst), 285 btrfs_update_inode_bytes(BTRFS_I(dst), datal, drop_args.bytes_found);
285 btrfs_set_inode_full_sync(BTRFS_I(dst) 286 btrfs_set_inode_full_sync(BTRFS_I(dst));
286 ret = btrfs_inode_set_file_extent_rang 287 ret = btrfs_inode_set_file_extent_range(BTRFS_I(dst), 0, aligned_end);
287 out: 288 out:
288 if (!ret && !trans) { 289 if (!ret && !trans) {
289 /* 290 /*
290 * No transaction here means w 291 * No transaction here means we copied the inline extent into a
291 * page of the destination ino 292 * page of the destination inode.
292 * 293 *
293 * 1 unit to update inode item 294 * 1 unit to update inode item
294 */ 295 */
295 trans = btrfs_start_transactio 296 trans = btrfs_start_transaction(root, 1);
296 if (IS_ERR(trans)) { 297 if (IS_ERR(trans)) {
297 ret = PTR_ERR(trans); 298 ret = PTR_ERR(trans);
298 trans = NULL; 299 trans = NULL;
299 } 300 }
300 } 301 }
301 if (ret && trans) { 302 if (ret && trans) {
302 btrfs_abort_transaction(trans, 303 btrfs_abort_transaction(trans, ret);
303 btrfs_end_transaction(trans); 304 btrfs_end_transaction(trans);
304 } 305 }
305 if (!ret) 306 if (!ret)
306 *trans_out = trans; 307 *trans_out = trans;
307 308
308 return ret; 309 return ret;
309 310
310 copy_to_page: 311 copy_to_page:
311 /* 312 /*
312 * Release our path because we don't n 313 * Release our path because we don't need it anymore and also because
313 * copy_inline_to_page() needs to rese 314 * copy_inline_to_page() needs to reserve data and metadata, which may
314 * need to flush delalloc when we are 315 * need to flush delalloc when we are low on available space and
315 * therefore cause a deadlock if write 316 * therefore cause a deadlock if writeback of an inline extent needs to
316 * write to the same leaf or an ordere 317 * write to the same leaf or an ordered extent completion needs to write
317 * to the same leaf. 318 * to the same leaf.
318 */ 319 */
319 btrfs_release_path(path); 320 btrfs_release_path(path);
320 321
321 ret = copy_inline_to_page(BTRFS_I(dst) 322 ret = copy_inline_to_page(BTRFS_I(dst), new_key->offset,
322 inline_data, 323 inline_data, size, datal, comp_type);
323 goto out; 324 goto out;
324 } 325 }
325 326
326 /* 327 /*
327 * Clone a range from inode file to another. 328 * Clone a range from inode file to another.
328 * 329 *
329 * @src: Inode to clone from 330 * @src: Inode to clone from
330 * @inode: Inode to clone to 331 * @inode: Inode to clone to
331 * @off: Offset within source to s 332 * @off: Offset within source to start clone from
332 * @olen: Original length, passed b 333 * @olen: Original length, passed by user, of range to clone
333 * @olen_aligned: Block-aligned value of ol 334 * @olen_aligned: Block-aligned value of olen
334 * @destoff: Offset within @inode to s 335 * @destoff: Offset within @inode to start clone
335 * @no_time_update: Whether to update mtime/c 336 * @no_time_update: Whether to update mtime/ctime on the target inode
336 */ 337 */
337 static int btrfs_clone(struct inode *src, stru 338 static int btrfs_clone(struct inode *src, struct inode *inode,
338 const u64 off, const u6 339 const u64 off, const u64 olen, const u64 olen_aligned,
339 const u64 destoff, int 340 const u64 destoff, int no_time_update)
340 { 341 {
341 struct btrfs_fs_info *fs_info = inode_ !! 342 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
342 struct btrfs_path *path = NULL; 343 struct btrfs_path *path = NULL;
343 struct extent_buffer *leaf; 344 struct extent_buffer *leaf;
344 struct btrfs_trans_handle *trans; 345 struct btrfs_trans_handle *trans;
345 char *buf = NULL; 346 char *buf = NULL;
346 struct btrfs_key key; 347 struct btrfs_key key;
347 u32 nritems; 348 u32 nritems;
348 int slot; 349 int slot;
349 int ret; 350 int ret;
350 const u64 len = olen_aligned; 351 const u64 len = olen_aligned;
351 u64 last_dest_end = destoff; 352 u64 last_dest_end = destoff;
352 u64 prev_extent_end = off; 353 u64 prev_extent_end = off;
353 354
354 ret = -ENOMEM; 355 ret = -ENOMEM;
355 buf = kvmalloc(fs_info->nodesize, GFP_ 356 buf = kvmalloc(fs_info->nodesize, GFP_KERNEL);
356 if (!buf) 357 if (!buf)
357 return ret; 358 return ret;
358 359
359 path = btrfs_alloc_path(); 360 path = btrfs_alloc_path();
360 if (!path) { 361 if (!path) {
361 kvfree(buf); 362 kvfree(buf);
362 return ret; 363 return ret;
363 } 364 }
364 365
365 path->reada = READA_FORWARD; 366 path->reada = READA_FORWARD;
366 /* Clone data */ 367 /* Clone data */
367 key.objectid = btrfs_ino(BTRFS_I(src)) 368 key.objectid = btrfs_ino(BTRFS_I(src));
368 key.type = BTRFS_EXTENT_DATA_KEY; 369 key.type = BTRFS_EXTENT_DATA_KEY;
369 key.offset = off; 370 key.offset = off;
370 371
371 while (1) { 372 while (1) {
372 struct btrfs_file_extent_item 373 struct btrfs_file_extent_item *extent;
373 u64 extent_gen; 374 u64 extent_gen;
374 int type; 375 int type;
375 u32 size; 376 u32 size;
376 struct btrfs_key new_key; 377 struct btrfs_key new_key;
377 u64 disko = 0, diskl = 0; 378 u64 disko = 0, diskl = 0;
378 u64 datao = 0, datal = 0; 379 u64 datao = 0, datal = 0;
379 u8 comp; 380 u8 comp;
380 u64 drop_start; 381 u64 drop_start;
381 382
382 /* Note the key will change ty 383 /* Note the key will change type as we walk through the tree */
383 ret = btrfs_search_slot(NULL, 384 ret = btrfs_search_slot(NULL, BTRFS_I(src)->root, &key, path,
384 0, 0); 385 0, 0);
385 if (ret < 0) 386 if (ret < 0)
386 goto out; 387 goto out;
387 /* 388 /*
388 * First search, if no extent 389 * First search, if no extent item that starts at offset off was
389 * found but the previous item 390 * found but the previous item is an extent item, it's possible
390 * it might overlap our target 391 * it might overlap our target range, therefore process it.
391 */ 392 */
392 if (key.offset == off && ret > 393 if (key.offset == off && ret > 0 && path->slots[0] > 0) {
393 btrfs_item_key_to_cpu( 394 btrfs_item_key_to_cpu(path->nodes[0], &key,
394 395 path->slots[0] - 1);
395 if (key.type == BTRFS_ 396 if (key.type == BTRFS_EXTENT_DATA_KEY)
396 path->slots[0] 397 path->slots[0]--;
397 } 398 }
398 399
399 nritems = btrfs_header_nritems 400 nritems = btrfs_header_nritems(path->nodes[0]);
400 process_slot: 401 process_slot:
401 if (path->slots[0] >= nritems) 402 if (path->slots[0] >= nritems) {
402 ret = btrfs_next_leaf( 403 ret = btrfs_next_leaf(BTRFS_I(src)->root, path);
403 if (ret < 0) 404 if (ret < 0)
404 goto out; 405 goto out;
405 if (ret > 0) 406 if (ret > 0)
406 break; 407 break;
407 nritems = btrfs_header 408 nritems = btrfs_header_nritems(path->nodes[0]);
408 } 409 }
409 leaf = path->nodes[0]; 410 leaf = path->nodes[0];
410 slot = path->slots[0]; 411 slot = path->slots[0];
411 412
412 btrfs_item_key_to_cpu(leaf, &k 413 btrfs_item_key_to_cpu(leaf, &key, slot);
413 if (key.type > BTRFS_EXTENT_DA 414 if (key.type > BTRFS_EXTENT_DATA_KEY ||
414 key.objectid != btrfs_ino( 415 key.objectid != btrfs_ino(BTRFS_I(src)))
415 break; 416 break;
416 417
417 ASSERT(key.type == BTRFS_EXTEN 418 ASSERT(key.type == BTRFS_EXTENT_DATA_KEY);
418 419
419 extent = btrfs_item_ptr(leaf, 420 extent = btrfs_item_ptr(leaf, slot,
420 struct 421 struct btrfs_file_extent_item);
421 extent_gen = btrfs_file_extent 422 extent_gen = btrfs_file_extent_generation(leaf, extent);
422 comp = btrfs_file_extent_compr 423 comp = btrfs_file_extent_compression(leaf, extent);
423 type = btrfs_file_extent_type( 424 type = btrfs_file_extent_type(leaf, extent);
424 if (type == BTRFS_FILE_EXTENT_ 425 if (type == BTRFS_FILE_EXTENT_REG ||
425 type == BTRFS_FILE_EXTENT_ 426 type == BTRFS_FILE_EXTENT_PREALLOC) {
426 disko = btrfs_file_ext 427 disko = btrfs_file_extent_disk_bytenr(leaf, extent);
427 diskl = btrfs_file_ext 428 diskl = btrfs_file_extent_disk_num_bytes(leaf, extent);
428 datao = btrfs_file_ext 429 datao = btrfs_file_extent_offset(leaf, extent);
429 datal = btrfs_file_ext 430 datal = btrfs_file_extent_num_bytes(leaf, extent);
430 } else if (type == BTRFS_FILE_ 431 } else if (type == BTRFS_FILE_EXTENT_INLINE) {
431 /* Take upper bound, m 432 /* Take upper bound, may be compressed */
432 datal = btrfs_file_ext 433 datal = btrfs_file_extent_ram_bytes(leaf, extent);
433 } 434 }
434 435
435 /* 436 /*
436 * The first search might have 437 * The first search might have left us at an extent item that
437 * ends before our target rang 438 * ends before our target range's start, can happen if we have
438 * holes and NO_HOLES feature 439 * holes and NO_HOLES feature enabled.
439 * 440 *
440 * Subsequent searches may lea 441 * Subsequent searches may leave us on a file range we have
441 * processed before - this hap 442 * processed before - this happens due to a race with ordered
442 * extent completion for a fil 443 * extent completion for a file range that is outside our source
443 * range, but that range was p 444 * range, but that range was part of a file extent item that
444 * also covered a leading part 445 * also covered a leading part of our source range.
445 */ 446 */
446 if (key.offset + datal <= prev 447 if (key.offset + datal <= prev_extent_end) {
447 path->slots[0]++; 448 path->slots[0]++;
448 goto process_slot; 449 goto process_slot;
449 } else if (key.offset >= off + 450 } else if (key.offset >= off + len) {
450 break; 451 break;
451 } 452 }
452 453
453 prev_extent_end = key.offset + 454 prev_extent_end = key.offset + datal;
454 size = btrfs_item_size(leaf, s 455 size = btrfs_item_size(leaf, slot);
455 read_extent_buffer(leaf, buf, 456 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf, slot),
456 size); 457 size);
457 458
458 btrfs_release_path(path); 459 btrfs_release_path(path);
459 460
460 memcpy(&new_key, &key, sizeof( 461 memcpy(&new_key, &key, sizeof(new_key));
461 new_key.objectid = btrfs_ino(B 462 new_key.objectid = btrfs_ino(BTRFS_I(inode));
462 if (off <= key.offset) 463 if (off <= key.offset)
463 new_key.offset = key.o 464 new_key.offset = key.offset + destoff - off;
464 else 465 else
465 new_key.offset = desto 466 new_key.offset = destoff;
466 467
467 /* 468 /*
468 * Deal with a hole that doesn 469 * Deal with a hole that doesn't have an extent item that
469 * represents it (NO_HOLES fea 470 * represents it (NO_HOLES feature enabled).
470 * This hole is either in the 471 * This hole is either in the middle of the cloning range or at
471 * the beginning (fully overla 472 * the beginning (fully overlaps it or partially overlaps it).
472 */ 473 */
473 if (new_key.offset != last_des 474 if (new_key.offset != last_dest_end)
474 drop_start = last_dest 475 drop_start = last_dest_end;
475 else 476 else
476 drop_start = new_key.o 477 drop_start = new_key.offset;
477 478
478 if (type == BTRFS_FILE_EXTENT_ 479 if (type == BTRFS_FILE_EXTENT_REG ||
479 type == BTRFS_FILE_EXTENT_ 480 type == BTRFS_FILE_EXTENT_PREALLOC) {
480 struct btrfs_replace_e 481 struct btrfs_replace_extent_info clone_info;
481 482
482 /* 483 /*
483 * a | --- range t 484 * a | --- range to clone ---| b
484 * | ------------- ext 485 * | ------------- extent ------------- |
485 */ 486 */
486 487
487 /* Subtract range b */ 488 /* Subtract range b */
488 if (key.offset + datal 489 if (key.offset + datal > off + len)
489 datal = off + 490 datal = off + len - key.offset;
490 491
491 /* Subtract range a */ 492 /* Subtract range a */
492 if (off > key.offset) 493 if (off > key.offset) {
493 datao += off - 494 datao += off - key.offset;
494 datal -= off - 495 datal -= off - key.offset;
495 } 496 }
496 497
497 clone_info.disk_offset 498 clone_info.disk_offset = disko;
498 clone_info.disk_len = 499 clone_info.disk_len = diskl;
499 clone_info.data_offset 500 clone_info.data_offset = datao;
500 clone_info.data_len = 501 clone_info.data_len = datal;
501 clone_info.file_offset 502 clone_info.file_offset = new_key.offset;
502 clone_info.extent_buf 503 clone_info.extent_buf = buf;
503 clone_info.is_new_exte 504 clone_info.is_new_extent = false;
504 clone_info.update_time 505 clone_info.update_times = !no_time_update;
505 ret = btrfs_replace_fi 506 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
506 drop_s 507 drop_start, new_key.offset + datal - 1,
507 &clone 508 &clone_info, &trans);
508 if (ret) 509 if (ret)
509 goto out; 510 goto out;
510 } else { 511 } else {
511 ASSERT(type == BTRFS_F 512 ASSERT(type == BTRFS_FILE_EXTENT_INLINE);
512 /* 513 /*
513 * Inline extents alwa 514 * Inline extents always have to start at file offset 0
514 * and can never be bi 515 * and can never be bigger then the sector size. We can
515 * never clone only pa 516 * never clone only parts of an inline extent, since all
516 * reflink operations 517 * reflink operations must start at a sector size aligned
517 * offset, and the len 518 * offset, and the length must be aligned too or end at
518 * the i_size (which i 519 * the i_size (which implies the whole inlined data).
519 */ 520 */
520 ASSERT(key.offset == 0 521 ASSERT(key.offset == 0);
521 ASSERT(datal <= fs_inf 522 ASSERT(datal <= fs_info->sectorsize);
522 if (WARN_ON(type != BT 523 if (WARN_ON(type != BTRFS_FILE_EXTENT_INLINE) ||
523 WARN_ON(key.offset 524 WARN_ON(key.offset != 0) ||
524 WARN_ON(datal > fs 525 WARN_ON(datal > fs_info->sectorsize)) {
525 ret = -EUCLEAN 526 ret = -EUCLEAN;
526 goto out; 527 goto out;
527 } 528 }
528 529
529 ret = clone_copy_inlin 530 ret = clone_copy_inline_extent(inode, path, &new_key,
530 531 drop_start, datal, size,
531 532 comp, buf, &trans);
532 if (ret) 533 if (ret)
533 goto out; 534 goto out;
534 } 535 }
535 536
536 btrfs_release_path(path); 537 btrfs_release_path(path);
537 538
538 /* 539 /*
539 * Whenever we share an extent 540 * Whenever we share an extent we update the last_reflink_trans
540 * of each inode to the curren 541 * of each inode to the current transaction. This is needed to
541 * make sure fsync does not lo 542 * make sure fsync does not log multiple checksum items with
542 * overlapping ranges (because 543 * overlapping ranges (because some extent items might refer
543 * only to sections of the ori 544 * only to sections of the original extent). For the destination
544 * inode we do this regardless 545 * inode we do this regardless of the generation of the extents
545 * or even if they are inline 546 * or even if they are inline extents or explicit holes, to make
546 * sure a full fsync does not 547 * sure a full fsync does not skip them. For the source inode,
547 * we only need to update last 548 * we only need to update last_reflink_trans in case it's a new
548 * extent that is not a hole o 549 * extent that is not a hole or an inline extent, to deal with
549 * the checksums problem on fs 550 * the checksums problem on fsync.
550 */ 551 */
551 if (extent_gen == trans->trans 552 if (extent_gen == trans->transid && disko > 0)
552 BTRFS_I(src)->last_ref 553 BTRFS_I(src)->last_reflink_trans = trans->transid;
553 554
554 BTRFS_I(inode)->last_reflink_t 555 BTRFS_I(inode)->last_reflink_trans = trans->transid;
555 556
556 last_dest_end = ALIGN(new_key. 557 last_dest_end = ALIGN(new_key.offset + datal,
557 fs_info- 558 fs_info->sectorsize);
558 ret = clone_finish_inode_updat 559 ret = clone_finish_inode_update(trans, inode, last_dest_end,
559 560 destoff, olen, no_time_update);
560 if (ret) 561 if (ret)
561 goto out; 562 goto out;
562 if (new_key.offset + datal >= 563 if (new_key.offset + datal >= destoff + len)
563 break; 564 break;
564 565
565 btrfs_release_path(path); 566 btrfs_release_path(path);
566 key.offset = prev_extent_end; 567 key.offset = prev_extent_end;
567 568
568 if (fatal_signal_pending(curre 569 if (fatal_signal_pending(current)) {
569 ret = -EINTR; 570 ret = -EINTR;
570 goto out; 571 goto out;
571 } 572 }
572 573
573 cond_resched(); 574 cond_resched();
574 } 575 }
575 ret = 0; 576 ret = 0;
576 577
577 if (last_dest_end < destoff + len) { 578 if (last_dest_end < destoff + len) {
578 /* 579 /*
579 * We have an implicit hole th 580 * We have an implicit hole that fully or partially overlaps our
580 * cloning range at its end. T 581 * cloning range at its end. This means that we either have the
581 * NO_HOLES feature enabled or 582 * NO_HOLES feature enabled or the implicit hole happened due to
582 * mixing buffered and direct 583 * mixing buffered and direct IO writes against this file.
583 */ 584 */
584 btrfs_release_path(path); 585 btrfs_release_path(path);
585 586
586 /* 587 /*
587 * When using NO_HOLES and we 588 * When using NO_HOLES and we are cloning a range that covers
588 * only a hole (no extents) in 589 * only a hole (no extents) into a range beyond the current
589 * i_size, punching a hole in 590 * i_size, punching a hole in the target range will not create
590 * an extent map defining a ho 591 * an extent map defining a hole, because the range starts at or
591 * beyond current i_size. If t 592 * beyond current i_size. If the file previously had an i_size
592 * greater than the new i_size 593 * greater than the new i_size set by this clone operation, we
593 * need to make sure the next 594 * need to make sure the next fsync is a full fsync, so that it
594 * detects and logs a hole cov 595 * detects and logs a hole covering a range from the current
595 * i_size to the new i_size. I 596 * i_size to the new i_size. If the clone range covers extents,
596 * besides a hole, then we kno 597 * besides a hole, then we know the full sync flag was already
597 * set by previous calls to bt 598 * set by previous calls to btrfs_replace_file_extents() that
598 * replaced file extent items. 599 * replaced file extent items.
599 */ 600 */
600 if (last_dest_end >= i_size_re 601 if (last_dest_end >= i_size_read(inode))
601 btrfs_set_inode_full_s 602 btrfs_set_inode_full_sync(BTRFS_I(inode));
602 603
603 ret = btrfs_replace_file_exten 604 ret = btrfs_replace_file_extents(BTRFS_I(inode), path,
604 last_dest_end, 605 last_dest_end, destoff + len - 1, NULL, &trans);
605 if (ret) 606 if (ret)
606 goto out; 607 goto out;
607 608
608 ret = clone_finish_inode_updat 609 ret = clone_finish_inode_update(trans, inode, destoff + len,
609 610 destoff, olen, no_time_update);
610 } 611 }
611 612
612 out: 613 out:
613 btrfs_free_path(path); 614 btrfs_free_path(path);
614 kvfree(buf); 615 kvfree(buf);
615 clear_bit(BTRFS_INODE_NO_DELALLOC_FLUS 616 clear_bit(BTRFS_INODE_NO_DELALLOC_FLUSH, &BTRFS_I(inode)->runtime_flags);
616 617
617 return ret; 618 return ret;
618 } 619 }
619 620
>> 621 static void btrfs_double_extent_unlock(struct inode *inode1, u64 loff1,
>> 622 struct inode *inode2, u64 loff2, u64 len)
>> 623 {
>> 624 unlock_extent(&BTRFS_I(inode1)->io_tree, loff1, loff1 + len - 1, NULL);
>> 625 unlock_extent(&BTRFS_I(inode2)->io_tree, loff2, loff2 + len - 1, NULL);
>> 626 }
>> 627
>> 628 static void btrfs_double_extent_lock(struct inode *inode1, u64 loff1,
>> 629 struct inode *inode2, u64 loff2, u64 len)
>> 630 {
>> 631 u64 range1_end = loff1 + len - 1;
>> 632 u64 range2_end = loff2 + len - 1;
>> 633
>> 634 if (inode1 < inode2) {
>> 635 swap(inode1, inode2);
>> 636 swap(loff1, loff2);
>> 637 swap(range1_end, range2_end);
>> 638 } else if (inode1 == inode2 && loff2 < loff1) {
>> 639 swap(loff1, loff2);
>> 640 swap(range1_end, range2_end);
>> 641 }
>> 642
>> 643 lock_extent(&BTRFS_I(inode1)->io_tree, loff1, range1_end, NULL);
>> 644 lock_extent(&BTRFS_I(inode2)->io_tree, loff2, range2_end, NULL);
>> 645
>> 646 btrfs_assert_inode_range_clean(BTRFS_I(inode1), loff1, range1_end);
>> 647 btrfs_assert_inode_range_clean(BTRFS_I(inode2), loff2, range2_end);
>> 648 }
>> 649
620 static void btrfs_double_mmap_lock(struct inod 650 static void btrfs_double_mmap_lock(struct inode *inode1, struct inode *inode2)
621 { 651 {
622 if (inode1 < inode2) 652 if (inode1 < inode2)
623 swap(inode1, inode2); 653 swap(inode1, inode2);
624 down_write(&BTRFS_I(inode1)->i_mmap_lo 654 down_write(&BTRFS_I(inode1)->i_mmap_lock);
625 down_write_nested(&BTRFS_I(inode2)->i_ 655 down_write_nested(&BTRFS_I(inode2)->i_mmap_lock, SINGLE_DEPTH_NESTING);
626 } 656 }
627 657
628 static void btrfs_double_mmap_unlock(struct in 658 static void btrfs_double_mmap_unlock(struct inode *inode1, struct inode *inode2)
629 { 659 {
630 up_write(&BTRFS_I(inode1)->i_mmap_lock 660 up_write(&BTRFS_I(inode1)->i_mmap_lock);
631 up_write(&BTRFS_I(inode2)->i_mmap_lock 661 up_write(&BTRFS_I(inode2)->i_mmap_lock);
632 } 662 }
633 663
634 static int btrfs_extent_same_range(struct inod 664 static int btrfs_extent_same_range(struct inode *src, u64 loff, u64 len,
635 struct inod 665 struct inode *dst, u64 dst_loff)
636 { 666 {
637 const u64 end = dst_loff + len - 1; <<
638 struct extent_state *cached_state = NU <<
639 struct btrfs_fs_info *fs_info = BTRFS_ 667 struct btrfs_fs_info *fs_info = BTRFS_I(src)->root->fs_info;
640 const u64 bs = fs_info->sectorsize; !! 668 const u64 bs = fs_info->sb->s_blocksize;
641 int ret; 669 int ret;
642 670
643 /* 671 /*
644 * Lock destination range to serialize !! 672 * Lock destination range to serialize with concurrent readahead() and
645 * we are safe from concurrency with r !! 673 * source range to serialize with relocation.
646 * because we have already locked the <<
647 * mode. <<
648 */ 674 */
649 lock_extent(&BTRFS_I(dst)->io_tree, ds !! 675 btrfs_double_extent_lock(src, loff, dst, dst_loff, len);
650 ret = btrfs_clone(src, dst, loff, len, 676 ret = btrfs_clone(src, dst, loff, len, ALIGN(len, bs), dst_loff, 1);
651 unlock_extent(&BTRFS_I(dst)->io_tree, !! 677 btrfs_double_extent_unlock(src, loff, dst, dst_loff, len);
652 678
653 btrfs_btree_balance_dirty(fs_info); 679 btrfs_btree_balance_dirty(fs_info);
654 680
655 return ret; 681 return ret;
656 } 682 }
657 683
658 static int btrfs_extent_same(struct inode *src 684 static int btrfs_extent_same(struct inode *src, u64 loff, u64 olen,
659 struct inode *dst 685 struct inode *dst, u64 dst_loff)
660 { 686 {
661 int ret = 0; 687 int ret = 0;
662 u64 i, tail_len, chunk_count; 688 u64 i, tail_len, chunk_count;
663 struct btrfs_root *root_dst = BTRFS_I( 689 struct btrfs_root *root_dst = BTRFS_I(dst)->root;
664 690
665 spin_lock(&root_dst->root_item_lock); 691 spin_lock(&root_dst->root_item_lock);
666 if (root_dst->send_in_progress) { 692 if (root_dst->send_in_progress) {
667 btrfs_warn_rl(root_dst->fs_inf 693 btrfs_warn_rl(root_dst->fs_info,
668 "cannot deduplicate to root %llu while send op 694 "cannot deduplicate to root %llu while send operations are using it (%d in progress)",
669 btrfs_root_id(ro !! 695 root_dst->root_key.objectid,
670 root_dst->send_i 696 root_dst->send_in_progress);
671 spin_unlock(&root_dst->root_it 697 spin_unlock(&root_dst->root_item_lock);
672 return -EAGAIN; 698 return -EAGAIN;
673 } 699 }
674 root_dst->dedupe_in_progress++; 700 root_dst->dedupe_in_progress++;
675 spin_unlock(&root_dst->root_item_lock) 701 spin_unlock(&root_dst->root_item_lock);
676 702
677 tail_len = olen % BTRFS_MAX_DEDUPE_LEN 703 tail_len = olen % BTRFS_MAX_DEDUPE_LEN;
678 chunk_count = div_u64(olen, BTRFS_MAX_ 704 chunk_count = div_u64(olen, BTRFS_MAX_DEDUPE_LEN);
679 705
680 for (i = 0; i < chunk_count; i++) { 706 for (i = 0; i < chunk_count; i++) {
681 ret = btrfs_extent_same_range( 707 ret = btrfs_extent_same_range(src, loff, BTRFS_MAX_DEDUPE_LEN,
682 708 dst, dst_loff);
683 if (ret) 709 if (ret)
684 goto out; 710 goto out;
685 711
686 loff += BTRFS_MAX_DEDUPE_LEN; 712 loff += BTRFS_MAX_DEDUPE_LEN;
687 dst_loff += BTRFS_MAX_DEDUPE_L 713 dst_loff += BTRFS_MAX_DEDUPE_LEN;
688 } 714 }
689 715
690 if (tail_len > 0) 716 if (tail_len > 0)
691 ret = btrfs_extent_same_range( 717 ret = btrfs_extent_same_range(src, loff, tail_len, dst, dst_loff);
692 out: 718 out:
693 spin_lock(&root_dst->root_item_lock); 719 spin_lock(&root_dst->root_item_lock);
694 root_dst->dedupe_in_progress--; 720 root_dst->dedupe_in_progress--;
695 spin_unlock(&root_dst->root_item_lock) 721 spin_unlock(&root_dst->root_item_lock);
696 722
697 return ret; 723 return ret;
698 } 724 }
699 725
700 static noinline int btrfs_clone_files(struct f 726 static noinline int btrfs_clone_files(struct file *file, struct file *file_src,
701 u64 of 727 u64 off, u64 olen, u64 destoff)
702 { 728 {
703 struct extent_state *cached_state = NU <<
704 struct inode *inode = file_inode(file) 729 struct inode *inode = file_inode(file);
705 struct inode *src = file_inode(file_sr 730 struct inode *src = file_inode(file_src);
706 struct btrfs_fs_info *fs_info = inode_ !! 731 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
707 int ret; 732 int ret;
708 int wb_ret; 733 int wb_ret;
709 u64 len = olen; 734 u64 len = olen;
710 u64 bs = fs_info->sectorsize; !! 735 u64 bs = fs_info->sb->s_blocksize;
711 u64 end; <<
712 736
713 /* 737 /*
714 * VFS's generic_remap_file_range_prep 738 * VFS's generic_remap_file_range_prep() protects us from cloning the
715 * eof block into the middle of a file 739 * eof block into the middle of a file, which would result in corruption
716 * if the file size is not blocksize a 740 * if the file size is not blocksize aligned. So we don't need to check
717 * for that case here. 741 * for that case here.
718 */ 742 */
719 if (off + len == src->i_size) 743 if (off + len == src->i_size)
720 len = ALIGN(src->i_size, bs) - 744 len = ALIGN(src->i_size, bs) - off;
721 745
722 if (destoff > inode->i_size) { 746 if (destoff > inode->i_size) {
723 const u64 wb_start = ALIGN_DOW 747 const u64 wb_start = ALIGN_DOWN(inode->i_size, bs);
724 748
725 ret = btrfs_cont_expand(BTRFS_ 749 ret = btrfs_cont_expand(BTRFS_I(inode), inode->i_size, destoff);
726 if (ret) 750 if (ret)
727 return ret; 751 return ret;
728 /* 752 /*
729 * We may have truncated the l 753 * We may have truncated the last block if the inode's size is
730 * not sector size aligned, so 754 * not sector size aligned, so we need to wait for writeback to
731 * complete before proceeding 755 * complete before proceeding further, otherwise we can race
732 * with cloning and attempt to 756 * with cloning and attempt to increment a reference to an
733 * extent that no longer exist 757 * extent that no longer exists (writeback completed right after
734 * we found the previous exten 758 * we found the previous extent covering eof and before we
735 * attempted to increment its 759 * attempted to increment its reference count).
736 */ 760 */
737 ret = btrfs_wait_ordered_range !! 761 ret = btrfs_wait_ordered_range(inode, wb_start,
738 762 destoff - wb_start);
739 if (ret) 763 if (ret)
740 return ret; 764 return ret;
741 } 765 }
742 766
743 /* 767 /*
744 * Lock destination range to serialize !! 768 * Lock destination range to serialize with concurrent readahead() and
745 * we are safe from concurrency with r !! 769 * source range to serialize with relocation.
746 * because we have already locked the <<
747 * mode. <<
748 */ 770 */
749 end = destoff + len - 1; !! 771 btrfs_double_extent_lock(src, off, inode, destoff, len);
750 lock_extent(&BTRFS_I(inode)->io_tree, <<
751 ret = btrfs_clone(src, inode, off, ole 772 ret = btrfs_clone(src, inode, off, olen, len, destoff, 0);
752 unlock_extent(&BTRFS_I(inode)->io_tree !! 773 btrfs_double_extent_unlock(src, off, inode, destoff, len);
753 774
754 /* 775 /*
755 * We may have copied an inline extent 776 * We may have copied an inline extent into a page of the destination
756 * range, so wait for writeback to com 777 * range, so wait for writeback to complete before truncating pages
757 * from the page cache. This is a rare 778 * from the page cache. This is a rare case.
758 */ 779 */
759 wb_ret = btrfs_wait_ordered_range(BTRF !! 780 wb_ret = btrfs_wait_ordered_range(inode, destoff, len);
760 ret = ret ? ret : wb_ret; 781 ret = ret ? ret : wb_ret;
761 /* 782 /*
762 * Truncate page cache pages so that f 783 * Truncate page cache pages so that future reads will see the cloned
763 * data immediately and not the previo 784 * data immediately and not the previous data.
764 */ 785 */
765 truncate_inode_pages_range(&inode->i_d 786 truncate_inode_pages_range(&inode->i_data,
766 round_down(des 787 round_down(destoff, PAGE_SIZE),
767 round_up(desto 788 round_up(destoff + len, PAGE_SIZE) - 1);
768 789
769 btrfs_btree_balance_dirty(fs_info); 790 btrfs_btree_balance_dirty(fs_info);
770 791
771 return ret; 792 return ret;
772 } 793 }
773 794
774 static int btrfs_remap_file_range_prep(struct 795 static int btrfs_remap_file_range_prep(struct file *file_in, loff_t pos_in,
775 struct 796 struct file *file_out, loff_t pos_out,
776 loff_t 797 loff_t *len, unsigned int remap_flags)
777 { 798 {
778 struct inode *inode_in = file_inode(fi 799 struct inode *inode_in = file_inode(file_in);
779 struct inode *inode_out = file_inode(f 800 struct inode *inode_out = file_inode(file_out);
780 u64 bs = BTRFS_I(inode_out)->root->fs_ !! 801 u64 bs = BTRFS_I(inode_out)->root->fs_info->sb->s_blocksize;
781 u64 wb_len; 802 u64 wb_len;
782 int ret; 803 int ret;
783 804
784 if (!(remap_flags & REMAP_FILE_DEDUP)) 805 if (!(remap_flags & REMAP_FILE_DEDUP)) {
785 struct btrfs_root *root_out = 806 struct btrfs_root *root_out = BTRFS_I(inode_out)->root;
786 807
787 if (btrfs_root_readonly(root_o 808 if (btrfs_root_readonly(root_out))
788 return -EROFS; 809 return -EROFS;
789 810
790 ASSERT(inode_in->i_sb == inode 811 ASSERT(inode_in->i_sb == inode_out->i_sb);
791 } 812 }
792 813
793 /* Don't make the dst file partly chec 814 /* Don't make the dst file partly checksummed */
794 if ((BTRFS_I(inode_in)->flags & BTRFS_ 815 if ((BTRFS_I(inode_in)->flags & BTRFS_INODE_NODATASUM) !=
795 (BTRFS_I(inode_out)->flags & BTRFS 816 (BTRFS_I(inode_out)->flags & BTRFS_INODE_NODATASUM)) {
796 return -EINVAL; 817 return -EINVAL;
797 } 818 }
798 819
799 /* 820 /*
800 * Now that the inodes are locked, we 821 * Now that the inodes are locked, we need to start writeback ourselves
801 * and can not rely on the writeback f 822 * and can not rely on the writeback from the VFS's generic helper
802 * generic_remap_file_range_prep() bec 823 * generic_remap_file_range_prep() because:
803 * 824 *
804 * 1) For compression we must call fil 825 * 1) For compression we must call filemap_fdatawrite_range() range
805 * twice (btrfs_fdatawrite_range() 826 * twice (btrfs_fdatawrite_range() does it for us), and the generic
806 * helper only calls it once; 827 * helper only calls it once;
807 * 828 *
808 * 2) filemap_fdatawrite_range(), call 829 * 2) filemap_fdatawrite_range(), called by the generic helper only
809 * waits for the writeback to compl 830 * waits for the writeback to complete, i.e. for IO to be done, and
810 * not for the ordered extents to c 831 * not for the ordered extents to complete. We need to wait for them
811 * to complete so that new file ext 832 * to complete so that new file extent items are in the fs tree.
812 */ 833 */
813 if (*len == 0 && !(remap_flags & REMAP 834 if (*len == 0 && !(remap_flags & REMAP_FILE_DEDUP))
814 wb_len = ALIGN(inode_in->i_siz 835 wb_len = ALIGN(inode_in->i_size, bs) - ALIGN_DOWN(pos_in, bs);
815 else 836 else
816 wb_len = ALIGN(*len, bs); 837 wb_len = ALIGN(*len, bs);
817 838
818 /* 839 /*
819 * Workaround to make sure NOCOW buffe 840 * Workaround to make sure NOCOW buffered write reach disk as NOCOW.
820 * 841 *
821 * Btrfs' back references do not have 842 * Btrfs' back references do not have a block level granularity, they
822 * work at the whole extent level. 843 * work at the whole extent level.
823 * NOCOW buffered write without data s 844 * NOCOW buffered write without data space reserved may not be able
824 * to fall back to CoW due to lack of 845 * to fall back to CoW due to lack of data space, thus could cause
825 * data loss. 846 * data loss.
826 * 847 *
827 * Here we take a shortcut by flushing 848 * Here we take a shortcut by flushing the whole inode, so that all
828 * nocow write should reach disk as no 849 * nocow write should reach disk as nocow before we increase the
829 * reference of the extent. We could d 850 * reference of the extent. We could do better by only flushing NOCOW
830 * data, but that needs extra accounti 851 * data, but that needs extra accounting.
831 * 852 *
832 * Also we don't need to check ASYNC_E 853 * Also we don't need to check ASYNC_EXTENT, as async extent will be
833 * CoWed anyway, not affecting nocow p 854 * CoWed anyway, not affecting nocow part.
834 */ 855 */
835 ret = filemap_flush(inode_in->i_mappin 856 ret = filemap_flush(inode_in->i_mapping);
836 if (ret < 0) 857 if (ret < 0)
837 return ret; 858 return ret;
838 859
839 ret = btrfs_wait_ordered_range(BTRFS_I !! 860 ret = btrfs_wait_ordered_range(inode_in, ALIGN_DOWN(pos_in, bs),
840 wb_len) 861 wb_len);
841 if (ret < 0) 862 if (ret < 0)
842 return ret; 863 return ret;
843 ret = btrfs_wait_ordered_range(BTRFS_I !! 864 ret = btrfs_wait_ordered_range(inode_out, ALIGN_DOWN(pos_out, bs),
844 wb_len) 865 wb_len);
845 if (ret < 0) 866 if (ret < 0)
846 return ret; 867 return ret;
847 868
848 return generic_remap_file_range_prep(f 869 return generic_remap_file_range_prep(file_in, pos_in, file_out, pos_out,
849 le 870 len, remap_flags);
850 } 871 }
851 872
852 static bool file_sync_write(const struct file 873 static bool file_sync_write(const struct file *file)
853 { 874 {
854 if (file->f_flags & (__O_SYNC | O_DSYN 875 if (file->f_flags & (__O_SYNC | O_DSYNC))
855 return true; 876 return true;
856 if (IS_SYNC(file_inode(file))) 877 if (IS_SYNC(file_inode(file)))
857 return true; 878 return true;
858 879
859 return false; 880 return false;
860 } 881 }
861 882
862 loff_t btrfs_remap_file_range(struct file *src 883 loff_t btrfs_remap_file_range(struct file *src_file, loff_t off,
863 struct file *dst_file, loff_t 884 struct file *dst_file, loff_t destoff, loff_t len,
864 unsigned int remap_flags) 885 unsigned int remap_flags)
865 { 886 {
866 struct inode *src_inode = file_inode(s 887 struct inode *src_inode = file_inode(src_file);
867 struct inode *dst_inode = file_inode(d 888 struct inode *dst_inode = file_inode(dst_file);
868 bool same_inode = dst_inode == src_ino 889 bool same_inode = dst_inode == src_inode;
869 int ret; 890 int ret;
870 891
871 if (remap_flags & ~(REMAP_FILE_DEDUP | 892 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
872 return -EINVAL; 893 return -EINVAL;
873 894
874 if (same_inode) { 895 if (same_inode) {
875 btrfs_inode_lock(BTRFS_I(src_i 896 btrfs_inode_lock(BTRFS_I(src_inode), BTRFS_ILOCK_MMAP);
876 } else { 897 } else {
877 lock_two_nondirectories(src_in 898 lock_two_nondirectories(src_inode, dst_inode);
878 btrfs_double_mmap_lock(src_ino 899 btrfs_double_mmap_lock(src_inode, dst_inode);
879 } 900 }
880 901
881 ret = btrfs_remap_file_range_prep(src_ 902 ret = btrfs_remap_file_range_prep(src_file, off, dst_file, destoff,
882 &len 903 &len, remap_flags);
883 if (ret < 0 || len == 0) 904 if (ret < 0 || len == 0)
884 goto out_unlock; 905 goto out_unlock;
885 906
886 if (remap_flags & REMAP_FILE_DEDUP) 907 if (remap_flags & REMAP_FILE_DEDUP)
887 ret = btrfs_extent_same(src_in 908 ret = btrfs_extent_same(src_inode, off, len, dst_inode, destoff);
888 else 909 else
889 ret = btrfs_clone_files(dst_fi 910 ret = btrfs_clone_files(dst_file, src_file, off, len, destoff);
890 911
891 out_unlock: 912 out_unlock:
892 if (same_inode) { 913 if (same_inode) {
893 btrfs_inode_unlock(BTRFS_I(src 914 btrfs_inode_unlock(BTRFS_I(src_inode), BTRFS_ILOCK_MMAP);
894 } else { 915 } else {
895 btrfs_double_mmap_unlock(src_i 916 btrfs_double_mmap_unlock(src_inode, dst_inode);
896 unlock_two_nondirectories(src_ 917 unlock_two_nondirectories(src_inode, dst_inode);
897 } 918 }
898 919
899 /* 920 /*
900 * If either the source or the destina 921 * If either the source or the destination file was opened with O_SYNC,
901 * O_DSYNC or has the S_SYNC attribute 922 * O_DSYNC or has the S_SYNC attribute, fsync both the destination and
902 * source files/ranges, so that after 923 * source files/ranges, so that after a successful return (0) followed
903 * by a power failure results in the r 924 * by a power failure results in the reflinked data to be readable from
904 * both files/ranges. 925 * both files/ranges.
905 */ 926 */
906 if (ret == 0 && len > 0 && 927 if (ret == 0 && len > 0 &&
907 (file_sync_write(src_file) || file 928 (file_sync_write(src_file) || file_sync_write(dst_file))) {
908 ret = btrfs_sync_file(src_file 929 ret = btrfs_sync_file(src_file, off, off + len - 1, 0);
909 if (ret == 0) 930 if (ret == 0)
910 ret = btrfs_sync_file( 931 ret = btrfs_sync_file(dst_file, destoff,
911 932 destoff + len - 1, 0);
912 } 933 }
913 934
914 return ret < 0 ? ret : len; 935 return ret < 0 ? ret : len;
915 } 936 }
916 937
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.