xref: /linux/fs/bcachefs/inode.c (revision c1e822754cc7f28b98c6897d62e8b47b4001e422)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include "bcachefs.h"
4 #include "btree_key_cache.h"
5 #include "btree_write_buffer.h"
6 #include "bkey_methods.h"
7 #include "btree_update.h"
8 #include "buckets.h"
9 #include "compress.h"
10 #include "dirent.h"
11 #include "disk_accounting.h"
12 #include "error.h"
13 #include "extents.h"
14 #include "extent_update.h"
15 #include "fs.h"
16 #include "inode.h"
17 #include "str_hash.h"
18 #include "snapshot.h"
19 #include "subvolume.h"
20 #include "varint.h"
21 
22 #include <linux/random.h>
23 
24 #include <linux/unaligned.h>
25 
26 #define x(name, ...)	#name,
27 const char * const bch2_inode_opts[] = {
28 	BCH_INODE_OPTS()
29 	NULL,
30 };
31 
32 static const char * const bch2_inode_flag_strs[] = {
33 	BCH_INODE_FLAGS()
34 	NULL
35 };
36 #undef  x
37 
38 static int delete_ancestor_snapshot_inodes(struct btree_trans *, struct bpos);
39 
40 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
41 
inode_decode_field(const u8 * in,const u8 * end,u64 out[2],unsigned * out_bits)42 static int inode_decode_field(const u8 *in, const u8 *end,
43 			      u64 out[2], unsigned *out_bits)
44 {
45 	__be64 be[2] = { 0, 0 };
46 	unsigned bytes, shift;
47 	u8 *p;
48 
49 	if (in >= end)
50 		return -1;
51 
52 	if (!*in)
53 		return -1;
54 
55 	/*
56 	 * position of highest set bit indicates number of bytes:
57 	 * shift = number of bits to remove in high byte:
58 	 */
59 	shift	= 8 - __fls(*in); /* 1 <= shift <= 8 */
60 	bytes	= byte_table[shift - 1];
61 
62 	if (in + bytes > end)
63 		return -1;
64 
65 	p = (u8 *) be + 16 - bytes;
66 	memcpy(p, in, bytes);
67 	*p ^= (1 << 8) >> shift;
68 
69 	out[0] = be64_to_cpu(be[0]);
70 	out[1] = be64_to_cpu(be[1]);
71 	*out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
72 
73 	return bytes;
74 }
75 
bch2_inode_pack_inlined(struct bkey_inode_buf * packed,const struct bch_inode_unpacked * inode)76 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
77 					   const struct bch_inode_unpacked *inode)
78 {
79 	struct bkey_i_inode_v3 *k = &packed->inode;
80 	u8 *out = k->v.fields;
81 	u8 *end = (void *) &packed[1];
82 	u8 *last_nonzero_field = out;
83 	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
84 	unsigned bytes;
85 	int ret;
86 
87 	bkey_inode_v3_init(&packed->inode.k_i);
88 	packed->inode.k.p.offset	= inode->bi_inum;
89 	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
90 	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
91 	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
92 	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
93 	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
94 	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
95 	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
96 	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
97 
98 
99 #define x(_name, _bits)							\
100 	nr_fields++;							\
101 									\
102 	if (inode->_name) {						\
103 		ret = bch2_varint_encode_fast(out, inode->_name);	\
104 		out += ret;						\
105 									\
106 		if (_bits > 64)						\
107 			*out++ = 0;					\
108 									\
109 		last_nonzero_field = out;				\
110 		last_nonzero_fieldnr = nr_fields;			\
111 	} else {							\
112 		*out++ = 0;						\
113 									\
114 		if (_bits > 64)						\
115 			*out++ = 0;					\
116 	}
117 
118 	BCH_INODE_FIELDS_v3()
119 #undef  x
120 	BUG_ON(out > end);
121 
122 	out = last_nonzero_field;
123 	nr_fields = last_nonzero_fieldnr;
124 
125 	bytes = out - (u8 *) &packed->inode.v;
126 	set_bkey_val_bytes(&packed->inode.k, bytes);
127 	memset_u64s_tail(&packed->inode.v, 0, bytes);
128 
129 	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
130 
131 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
132 		struct bch_inode_unpacked unpacked;
133 
134 		ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
135 		BUG_ON(ret);
136 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
137 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
138 		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
139 		BUG_ON(unpacked.bi_size		!= inode->bi_size);
140 		BUG_ON(unpacked.bi_version	!= inode->bi_version);
141 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
142 
143 #define x(_name, _bits)	if (unpacked._name != inode->_name)		\
144 			panic("unpacked %llu should be %llu",		\
145 			      (u64) unpacked._name, (u64) inode->_name);
146 		BCH_INODE_FIELDS_v3()
147 #undef  x
148 	}
149 }
150 
bch2_inode_pack(struct bkey_inode_buf * packed,const struct bch_inode_unpacked * inode)151 void bch2_inode_pack(struct bkey_inode_buf *packed,
152 		     const struct bch_inode_unpacked *inode)
153 {
154 	bch2_inode_pack_inlined(packed, inode);
155 }
156 
bch2_inode_unpack_v1(struct bkey_s_c_inode inode,struct bch_inode_unpacked * unpacked)157 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
158 				struct bch_inode_unpacked *unpacked)
159 {
160 	const u8 *in = inode.v->fields;
161 	const u8 *end = bkey_val_end(inode);
162 	u64 field[2];
163 	unsigned fieldnr = 0, field_bits;
164 	int ret;
165 
166 #define x(_name, _bits)							\
167 	if (fieldnr++ == INODEv1_NR_FIELDS(inode.v)) {			\
168 		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
169 		memset((void *) unpacked + offset, 0,			\
170 		       sizeof(*unpacked) - offset);			\
171 		return 0;						\
172 	}								\
173 									\
174 	ret = inode_decode_field(in, end, field, &field_bits);		\
175 	if (ret < 0)							\
176 		return ret;						\
177 									\
178 	if (field_bits > sizeof(unpacked->_name) * 8)			\
179 		return -1;						\
180 									\
181 	unpacked->_name = field[1];					\
182 	in += ret;
183 
184 	BCH_INODE_FIELDS_v2()
185 #undef  x
186 
187 	/* XXX: signal if there were more fields than expected? */
188 	return 0;
189 }
190 
bch2_inode_unpack_v2(struct bch_inode_unpacked * unpacked,const u8 * in,const u8 * end,unsigned nr_fields)191 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
192 				const u8 *in, const u8 *end,
193 				unsigned nr_fields)
194 {
195 	unsigned fieldnr = 0;
196 	int ret;
197 	u64 v[2];
198 
199 #define x(_name, _bits)							\
200 	if (fieldnr < nr_fields) {					\
201 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
202 		if (ret < 0)						\
203 			return ret;					\
204 		in += ret;						\
205 									\
206 		if (_bits > 64) {					\
207 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
208 			if (ret < 0)					\
209 				return ret;				\
210 			in += ret;					\
211 		} else {						\
212 			v[1] = 0;					\
213 		}							\
214 	} else {							\
215 		v[0] = v[1] = 0;					\
216 	}								\
217 									\
218 	unpacked->_name = v[0];						\
219 	if (v[1] || v[0] != unpacked->_name)				\
220 		return -1;						\
221 	fieldnr++;
222 
223 	BCH_INODE_FIELDS_v2()
224 #undef  x
225 
226 	/* XXX: signal if there were more fields than expected? */
227 	return 0;
228 }
229 
bch2_inode_unpack_v3(struct bkey_s_c k,struct bch_inode_unpacked * unpacked)230 static int bch2_inode_unpack_v3(struct bkey_s_c k,
231 				struct bch_inode_unpacked *unpacked)
232 {
233 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
234 	const u8 *in = inode.v->fields;
235 	const u8 *end = bkey_val_end(inode);
236 	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
237 	unsigned fieldnr = 0;
238 	int ret;
239 	u64 v[2];
240 
241 	unpacked->bi_inum	= inode.k->p.offset;
242 	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
243 	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
244 	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
245 	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
246 	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
247 	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
248 	unpacked->bi_mode	= INODEv3_MODE(inode.v);
249 
250 #define x(_name, _bits)							\
251 	if (fieldnr < nr_fields) {					\
252 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
253 		if (ret < 0)						\
254 			return ret;					\
255 		in += ret;						\
256 									\
257 		if (_bits > 64) {					\
258 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
259 			if (ret < 0)					\
260 				return ret;				\
261 			in += ret;					\
262 		} else {						\
263 			v[1] = 0;					\
264 		}							\
265 	} else {							\
266 		v[0] = v[1] = 0;					\
267 	}								\
268 									\
269 	unpacked->_name = v[0];						\
270 	if (v[1] || v[0] != unpacked->_name)				\
271 		return -1;						\
272 	fieldnr++;
273 
274 	BCH_INODE_FIELDS_v3()
275 #undef  x
276 
277 	/* XXX: signal if there were more fields than expected? */
278 	return 0;
279 }
280 
bch2_inode_unpack_slowpath(struct bkey_s_c k,struct bch_inode_unpacked * unpacked)281 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
282 					       struct bch_inode_unpacked *unpacked)
283 {
284 	memset(unpacked, 0, sizeof(*unpacked));
285 
286 	unpacked->bi_snapshot = k.k->p.snapshot;
287 
288 	switch (k.k->type) {
289 	case KEY_TYPE_inode: {
290 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
291 
292 		unpacked->bi_inum	= inode.k->p.offset;
293 		unpacked->bi_journal_seq= 0;
294 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
295 		unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
296 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
297 
298 		if (INODEv1_NEW_VARINT(inode.v)) {
299 			return bch2_inode_unpack_v2(unpacked, inode.v->fields,
300 						    bkey_val_end(inode),
301 						    INODEv1_NR_FIELDS(inode.v));
302 		} else {
303 			return bch2_inode_unpack_v1(inode, unpacked);
304 		}
305 		break;
306 	}
307 	case KEY_TYPE_inode_v2: {
308 		struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
309 
310 		unpacked->bi_inum	= inode.k->p.offset;
311 		unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
312 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
313 		unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
314 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
315 
316 		return bch2_inode_unpack_v2(unpacked, inode.v->fields,
317 					    bkey_val_end(inode),
318 					    INODEv2_NR_FIELDS(inode.v));
319 	}
320 	default:
321 		BUG();
322 	}
323 }
324 
bch2_inode_unpack(struct bkey_s_c k,struct bch_inode_unpacked * unpacked)325 int bch2_inode_unpack(struct bkey_s_c k,
326 		      struct bch_inode_unpacked *unpacked)
327 {
328 	unpacked->bi_snapshot = k.k->p.snapshot;
329 
330 	return likely(k.k->type == KEY_TYPE_inode_v3)
331 		? bch2_inode_unpack_v3(k, unpacked)
332 		: bch2_inode_unpack_slowpath(k, unpacked);
333 }
334 
__bch2_inode_peek(struct btree_trans * trans,struct btree_iter * iter,struct bch_inode_unpacked * inode,subvol_inum inum,unsigned flags,bool warn)335 int __bch2_inode_peek(struct btree_trans *trans,
336 		      struct btree_iter *iter,
337 		      struct bch_inode_unpacked *inode,
338 		      subvol_inum inum, unsigned flags,
339 		      bool warn)
340 {
341 	u32 snapshot;
342 	int ret = __bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot, warn);
343 	if (ret)
344 		return ret;
345 
346 	struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
347 					       SPOS(0, inum.inum, snapshot),
348 					       flags|BTREE_ITER_cached);
349 	ret = bkey_err(k);
350 	if (ret)
351 		return ret;
352 
353 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
354 	if (ret)
355 		goto err;
356 
357 	ret = bch2_inode_unpack(k, inode);
358 	if (ret)
359 		goto err;
360 
361 	return 0;
362 err:
363 	if (warn)
364 		bch_err_msg(trans->c, ret, "looking up inum %llu:%llu:", inum.subvol, inum.inum);
365 	bch2_trans_iter_exit(trans, iter);
366 	return ret;
367 }
368 
bch2_inode_write_flags(struct btree_trans * trans,struct btree_iter * iter,struct bch_inode_unpacked * inode,enum btree_iter_update_trigger_flags flags)369 int bch2_inode_write_flags(struct btree_trans *trans,
370 		     struct btree_iter *iter,
371 		     struct bch_inode_unpacked *inode,
372 		     enum btree_iter_update_trigger_flags flags)
373 {
374 	struct bkey_inode_buf *inode_p;
375 
376 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
377 	if (IS_ERR(inode_p))
378 		return PTR_ERR(inode_p);
379 
380 	bch2_inode_pack_inlined(inode_p, inode);
381 	inode_p->inode.k.p.snapshot = iter->snapshot;
382 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
383 }
384 
__bch2_fsck_write_inode(struct btree_trans * trans,struct bch_inode_unpacked * inode)385 int __bch2_fsck_write_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
386 {
387 	struct bkey_inode_buf *inode_p =
388 		bch2_trans_kmalloc(trans, sizeof(*inode_p));
389 
390 	if (IS_ERR(inode_p))
391 		return PTR_ERR(inode_p);
392 
393 	bch2_inode_pack(inode_p, inode);
394 	inode_p->inode.k.p.snapshot = inode->bi_snapshot;
395 
396 	return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
397 				&inode_p->inode.k_i,
398 				BTREE_UPDATE_internal_snapshot_node);
399 }
400 
bch2_fsck_write_inode(struct btree_trans * trans,struct bch_inode_unpacked * inode)401 int bch2_fsck_write_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
402 {
403 	int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
404 			    __bch2_fsck_write_inode(trans, inode));
405 	bch_err_fn(trans->c, ret);
406 	return ret;
407 }
408 
bch2_inode_to_v3(struct btree_trans * trans,struct bkey_i * k)409 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
410 {
411 	struct bch_inode_unpacked u;
412 	struct bkey_inode_buf *inode_p;
413 	int ret;
414 
415 	if (!bkey_is_inode(&k->k))
416 		return ERR_PTR(-ENOENT);
417 
418 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
419 	if (IS_ERR(inode_p))
420 		return ERR_CAST(inode_p);
421 
422 	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
423 	if (ret)
424 		return ERR_PTR(ret);
425 
426 	bch2_inode_pack(inode_p, &u);
427 	return &inode_p->inode.k_i;
428 }
429 
__bch2_inode_validate(struct bch_fs * c,struct bkey_s_c k,enum bch_validate_flags flags)430 static int __bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
431 				 enum bch_validate_flags flags)
432 {
433 	struct bch_inode_unpacked unpacked;
434 	int ret = 0;
435 
436 	bkey_fsck_err_on(k.k->p.inode,
437 			 c, inode_pos_inode_nonzero,
438 			 "nonzero k.p.inode");
439 
440 	bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX,
441 			 c, inode_pos_blockdev_range,
442 			 "fs inode in blockdev range");
443 
444 	bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked),
445 			 c, inode_unpack_error,
446 			 "invalid variable length fields");
447 
448 	bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1,
449 			 c, inode_checksum_type_invalid,
450 			 "invalid data checksum type (%u >= %u",
451 			 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
452 
453 	bkey_fsck_err_on(unpacked.bi_compression &&
454 			 !bch2_compression_opt_valid(unpacked.bi_compression - 1),
455 			 c, inode_compression_type_invalid,
456 			 "invalid compression opt %u", unpacked.bi_compression - 1);
457 
458 	bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
459 			 unpacked.bi_nlink != 0,
460 			 c, inode_unlinked_but_nlink_nonzero,
461 			 "flagged as unlinked but bi_nlink != 0");
462 
463 	bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode),
464 			 c, inode_subvol_root_but_not_dir,
465 			 "subvolume root but not a directory");
466 fsck_err:
467 	return ret;
468 }
469 
bch2_inode_validate(struct bch_fs * c,struct bkey_s_c k,enum bch_validate_flags flags)470 int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
471 			enum bch_validate_flags flags)
472 {
473 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
474 	int ret = 0;
475 
476 	bkey_fsck_err_on(INODEv1_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
477 			 c, inode_str_hash_invalid,
478 			 "invalid str hash type (%llu >= %u)",
479 			 INODEv1_STR_HASH(inode.v), BCH_STR_HASH_NR);
480 
481 	ret = __bch2_inode_validate(c, k, flags);
482 fsck_err:
483 	return ret;
484 }
485 
bch2_inode_v2_validate(struct bch_fs * c,struct bkey_s_c k,enum bch_validate_flags flags)486 int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k,
487 			   enum bch_validate_flags flags)
488 {
489 	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
490 	int ret = 0;
491 
492 	bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
493 			 c, inode_str_hash_invalid,
494 			 "invalid str hash type (%llu >= %u)",
495 			 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
496 
497 	ret = __bch2_inode_validate(c, k, flags);
498 fsck_err:
499 	return ret;
500 }
501 
bch2_inode_v3_validate(struct bch_fs * c,struct bkey_s_c k,enum bch_validate_flags flags)502 int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k,
503 			   enum bch_validate_flags flags)
504 {
505 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
506 	int ret = 0;
507 
508 	bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
509 			 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k),
510 			 c, inode_v3_fields_start_bad,
511 			 "invalid fields_start (got %llu, min %u max %zu)",
512 			 INODEv3_FIELDS_START(inode.v),
513 			 INODEv3_FIELDS_START_INITIAL,
514 			 bkey_val_u64s(inode.k));
515 
516 	bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
517 			 c, inode_str_hash_invalid,
518 			 "invalid str hash type (%llu >= %u)",
519 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
520 
521 	ret = __bch2_inode_validate(c, k, flags);
522 fsck_err:
523 	return ret;
524 }
525 
__bch2_inode_unpacked_to_text(struct printbuf * out,struct bch_inode_unpacked * inode)526 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
527 					  struct bch_inode_unpacked *inode)
528 {
529 	prt_printf(out, "\n");
530 	printbuf_indent_add(out, 2);
531 	prt_printf(out, "mode=%o\n", inode->bi_mode);
532 
533 	prt_str(out, "flags=");
534 	prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
535 	prt_printf(out, "(%x)\n", inode->bi_flags);
536 
537 	prt_printf(out, "journal_seq=%llu\n",	inode->bi_journal_seq);
538 	prt_printf(out, "hash_seed=%llx\n",	inode->bi_hash_seed);
539 	prt_printf(out, "hash_type=");
540 	bch2_prt_str_hash_type(out, INODE_STR_HASH(inode));
541 	prt_newline(out);
542 	prt_printf(out, "bi_size=%llu\n",	inode->bi_size);
543 	prt_printf(out, "bi_sectors=%llu\n",	inode->bi_sectors);
544 	prt_printf(out, "bi_version=%llu\n",	inode->bi_version);
545 
546 #define x(_name, _bits)						\
547 	prt_printf(out, #_name "=%llu\n", (u64) inode->_name);
548 	BCH_INODE_FIELDS_v3()
549 #undef  x
550 
551 	bch2_printbuf_strip_trailing_newline(out);
552 	printbuf_indent_sub(out, 2);
553 }
554 
bch2_inode_unpacked_to_text(struct printbuf * out,struct bch_inode_unpacked * inode)555 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
556 {
557 	prt_printf(out, "inum: %llu:%u ", inode->bi_inum, inode->bi_snapshot);
558 	__bch2_inode_unpacked_to_text(out, inode);
559 }
560 
bch2_inode_to_text(struct printbuf * out,struct bch_fs * c,struct bkey_s_c k)561 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
562 {
563 	struct bch_inode_unpacked inode;
564 
565 	if (bch2_inode_unpack(k, &inode)) {
566 		prt_printf(out, "(unpack error)");
567 		return;
568 	}
569 
570 	__bch2_inode_unpacked_to_text(out, &inode);
571 }
572 
bkey_inode_flags(struct bkey_s_c k)573 static inline u64 bkey_inode_flags(struct bkey_s_c k)
574 {
575 	switch (k.k->type) {
576 	case KEY_TYPE_inode:
577 		return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
578 	case KEY_TYPE_inode_v2:
579 		return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
580 	case KEY_TYPE_inode_v3:
581 		return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
582 	default:
583 		return 0;
584 	}
585 }
586 
bkey_inode_flags_set(struct bkey_s k,u64 f)587 static inline void bkey_inode_flags_set(struct bkey_s k, u64 f)
588 {
589 	switch (k.k->type) {
590 	case KEY_TYPE_inode:
591 		bkey_s_to_inode(k).v->bi_flags = cpu_to_le32(f);
592 		return;
593 	case KEY_TYPE_inode_v2:
594 		bkey_s_to_inode_v2(k).v->bi_flags = cpu_to_le64(f);
595 		return;
596 	case KEY_TYPE_inode_v3:
597 		bkey_s_to_inode_v3(k).v->bi_flags = cpu_to_le64(f);
598 		return;
599 	default:
600 		BUG();
601 	}
602 }
603 
bkey_is_unlinked_inode(struct bkey_s_c k)604 static inline bool bkey_is_unlinked_inode(struct bkey_s_c k)
605 {
606 	unsigned f = bkey_inode_flags(k) & BCH_INODE_unlinked;
607 
608 	return (f & BCH_INODE_unlinked) && !(f & BCH_INODE_has_child_snapshot);
609 }
610 
611 static struct bkey_s_c
bch2_bkey_get_iter_snapshot_parent(struct btree_trans * trans,struct btree_iter * iter,enum btree_id btree,struct bpos pos,unsigned flags)612 bch2_bkey_get_iter_snapshot_parent(struct btree_trans *trans, struct btree_iter *iter,
613 				   enum btree_id btree, struct bpos pos,
614 				   unsigned flags)
615 {
616 	struct bch_fs *c = trans->c;
617 	struct bkey_s_c k;
618 	int ret = 0;
619 
620 	for_each_btree_key_upto_norestart(trans, *iter, btree,
621 					  bpos_successor(pos),
622 					  SPOS(pos.inode, pos.offset, U32_MAX),
623 					  flags|BTREE_ITER_all_snapshots, k, ret)
624 		if (bch2_snapshot_is_ancestor(c, pos.snapshot, k.k->p.snapshot))
625 			return k;
626 
627 	bch2_trans_iter_exit(trans, iter);
628 	return ret ? bkey_s_c_err(ret) : bkey_s_c_null;
629 }
630 
631 static struct bkey_s_c
bch2_inode_get_iter_snapshot_parent(struct btree_trans * trans,struct btree_iter * iter,struct bpos pos,unsigned flags)632 bch2_inode_get_iter_snapshot_parent(struct btree_trans *trans, struct btree_iter *iter,
633 				    struct bpos pos, unsigned flags)
634 {
635 	struct bkey_s_c k;
636 again:
637 	k = bch2_bkey_get_iter_snapshot_parent(trans, iter, BTREE_ID_inodes, pos, flags);
638 	if (!k.k ||
639 	    bkey_err(k) ||
640 	    bkey_is_inode(k.k))
641 		return k;
642 
643 	bch2_trans_iter_exit(trans, iter);
644 	pos = k.k->p;
645 	goto again;
646 }
647 
__bch2_inode_has_child_snapshots(struct btree_trans * trans,struct bpos pos)648 int __bch2_inode_has_child_snapshots(struct btree_trans *trans, struct bpos pos)
649 {
650 	struct bch_fs *c = trans->c;
651 	struct btree_iter iter;
652 	struct bkey_s_c k;
653 	int ret = 0;
654 
655 	for_each_btree_key_upto_norestart(trans, iter,
656 			BTREE_ID_inodes, POS(0, pos.offset), bpos_predecessor(pos),
657 			BTREE_ITER_all_snapshots|
658 			BTREE_ITER_with_updates, k, ret)
659 		if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, pos.snapshot) &&
660 		    bkey_is_inode(k.k)) {
661 			ret = 1;
662 			break;
663 		}
664 	bch2_trans_iter_exit(trans, &iter);
665 	return ret;
666 }
667 
update_inode_has_children(struct btree_trans * trans,struct bkey_s k,bool have_child)668 static int update_inode_has_children(struct btree_trans *trans,
669 				     struct bkey_s k,
670 				     bool have_child)
671 {
672 	if (!have_child) {
673 		int ret = bch2_inode_has_child_snapshots(trans, k.k->p);
674 		if (ret)
675 			return ret < 0 ? ret : 0;
676 	}
677 
678 	u64 f = bkey_inode_flags(k.s_c);
679 	if (have_child != !!(f & BCH_INODE_has_child_snapshot))
680 		bkey_inode_flags_set(k, f ^ BCH_INODE_has_child_snapshot);
681 
682 	return 0;
683 }
684 
update_parent_inode_has_children(struct btree_trans * trans,struct bpos pos,bool have_child)685 static int update_parent_inode_has_children(struct btree_trans *trans, struct bpos pos,
686 					    bool have_child)
687 {
688 	struct btree_iter iter;
689 	struct bkey_s_c k = bch2_inode_get_iter_snapshot_parent(trans,
690 						&iter, pos, BTREE_ITER_with_updates);
691 	int ret = bkey_err(k);
692 	if (ret)
693 		return ret;
694 	if (!k.k)
695 		return 0;
696 
697 	if (!have_child) {
698 		ret = bch2_inode_has_child_snapshots(trans, k.k->p);
699 		if (ret) {
700 			ret = ret < 0 ? ret : 0;
701 			goto err;
702 		}
703 	}
704 
705 	u64 f = bkey_inode_flags(k);
706 	if (have_child != !!(f & BCH_INODE_has_child_snapshot)) {
707 		struct bkey_i *update = bch2_bkey_make_mut(trans, &iter, &k,
708 					     BTREE_UPDATE_internal_snapshot_node);
709 		ret = PTR_ERR_OR_ZERO(update);
710 		if (ret)
711 			goto err;
712 
713 		bkey_inode_flags_set(bkey_i_to_s(update), f ^ BCH_INODE_has_child_snapshot);
714 	}
715 err:
716 	bch2_trans_iter_exit(trans, &iter);
717 	return ret;
718 }
719 
bch2_trigger_inode(struct btree_trans * trans,enum btree_id btree_id,unsigned level,struct bkey_s_c old,struct bkey_s new,enum btree_iter_update_trigger_flags flags)720 int bch2_trigger_inode(struct btree_trans *trans,
721 		       enum btree_id btree_id, unsigned level,
722 		       struct bkey_s_c old,
723 		       struct bkey_s new,
724 		       enum btree_iter_update_trigger_flags flags)
725 {
726 	struct bch_fs *c = trans->c;
727 
728 	if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
729 		BUG_ON(!trans->journal_res.seq);
730 		bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
731 	}
732 
733 	s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
734 	if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr) {
735 		struct disk_accounting_pos acc = { .type = BCH_DISK_ACCOUNTING_nr_inodes };
736 		int ret = bch2_disk_accounting_mod(trans, &acc, &nr, 1, flags & BTREE_TRIGGER_gc);
737 		if (ret)
738 			return ret;
739 	}
740 
741 	if (flags & BTREE_TRIGGER_transactional) {
742 		int unlinked_delta =	(int) bkey_is_unlinked_inode(new.s_c) -
743 					(int) bkey_is_unlinked_inode(old);
744 		if (unlinked_delta) {
745 			int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes,
746 							      new.k->p, unlinked_delta > 0);
747 			if (ret)
748 				return ret;
749 		}
750 
751 		/*
752 		 * If we're creating or deleting an inode at this snapshot ID,
753 		 * and there might be an inode in a parent snapshot ID, we might
754 		 * need to set or clear the has_child_snapshot flag on the
755 		 * parent.
756 		 */
757 		int deleted_delta = (int) bkey_is_inode(new.k) -
758 				    (int) bkey_is_inode(old.k);
759 		if (deleted_delta &&
760 		    bch2_snapshot_parent(c, new.k->p.snapshot)) {
761 			int ret = update_parent_inode_has_children(trans, new.k->p,
762 								   deleted_delta > 0);
763 			if (ret)
764 				return ret;
765 		}
766 
767 		/*
768 		 * When an inode is first updated in a new snapshot, we may need
769 		 * to clear has_child_snapshot
770 		 */
771 		if (deleted_delta > 0) {
772 			int ret = update_inode_has_children(trans, new, false);
773 			if (ret)
774 				return ret;
775 		}
776 	}
777 
778 	return 0;
779 }
780 
bch2_inode_generation_validate(struct bch_fs * c,struct bkey_s_c k,enum bch_validate_flags flags)781 int bch2_inode_generation_validate(struct bch_fs *c, struct bkey_s_c k,
782 				   enum bch_validate_flags flags)
783 {
784 	int ret = 0;
785 
786 	bkey_fsck_err_on(k.k->p.inode,
787 			 c, inode_pos_inode_nonzero,
788 			 "nonzero k.p.inode");
789 fsck_err:
790 	return ret;
791 }
792 
bch2_inode_generation_to_text(struct printbuf * out,struct bch_fs * c,struct bkey_s_c k)793 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
794 				   struct bkey_s_c k)
795 {
796 	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
797 
798 	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
799 }
800 
bch2_inode_init_early(struct bch_fs * c,struct bch_inode_unpacked * inode_u)801 void bch2_inode_init_early(struct bch_fs *c,
802 			   struct bch_inode_unpacked *inode_u)
803 {
804 	enum bch_str_hash_type str_hash =
805 		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
806 
807 	memset(inode_u, 0, sizeof(*inode_u));
808 
809 	SET_INODE_STR_HASH(inode_u, str_hash);
810 	get_random_bytes(&inode_u->bi_hash_seed, sizeof(inode_u->bi_hash_seed));
811 }
812 
bch2_inode_init_late(struct bch_inode_unpacked * inode_u,u64 now,uid_t uid,gid_t gid,umode_t mode,dev_t rdev,struct bch_inode_unpacked * parent)813 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
814 			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
815 			  struct bch_inode_unpacked *parent)
816 {
817 	inode_u->bi_mode	= mode;
818 	inode_u->bi_uid		= uid;
819 	inode_u->bi_gid		= gid;
820 	inode_u->bi_dev		= rdev;
821 	inode_u->bi_atime	= now;
822 	inode_u->bi_mtime	= now;
823 	inode_u->bi_ctime	= now;
824 	inode_u->bi_otime	= now;
825 
826 	if (parent && parent->bi_mode & S_ISGID) {
827 		inode_u->bi_gid = parent->bi_gid;
828 		if (S_ISDIR(mode))
829 			inode_u->bi_mode |= S_ISGID;
830 	}
831 
832 	if (parent) {
833 #define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
834 		BCH_INODE_OPTS()
835 #undef x
836 	}
837 }
838 
bch2_inode_init(struct bch_fs * c,struct bch_inode_unpacked * inode_u,uid_t uid,gid_t gid,umode_t mode,dev_t rdev,struct bch_inode_unpacked * parent)839 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
840 		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
841 		     struct bch_inode_unpacked *parent)
842 {
843 	bch2_inode_init_early(c, inode_u);
844 	bch2_inode_init_late(inode_u, bch2_current_time(c),
845 			     uid, gid, mode, rdev, parent);
846 }
847 
bkey_generation(struct bkey_s_c k)848 static inline u32 bkey_generation(struct bkey_s_c k)
849 {
850 	switch (k.k->type) {
851 	case KEY_TYPE_inode:
852 	case KEY_TYPE_inode_v2:
853 		BUG();
854 	case KEY_TYPE_inode_generation:
855 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
856 	default:
857 		return 0;
858 	}
859 }
860 
861 /*
862  * This just finds an empty slot:
863  */
bch2_inode_create(struct btree_trans * trans,struct btree_iter * iter,struct bch_inode_unpacked * inode_u,u32 snapshot,u64 cpu)864 int bch2_inode_create(struct btree_trans *trans,
865 		      struct btree_iter *iter,
866 		      struct bch_inode_unpacked *inode_u,
867 		      u32 snapshot, u64 cpu)
868 {
869 	struct bch_fs *c = trans->c;
870 	struct bkey_s_c k;
871 	u64 min, max, start, pos, *hint;
872 	int ret = 0;
873 	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
874 
875 	if (c->opts.shard_inode_numbers) {
876 		bits -= c->inode_shard_bits;
877 
878 		min = (cpu << bits);
879 		max = (cpu << bits) | ~(ULLONG_MAX << bits);
880 
881 		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
882 		hint = c->unused_inode_hints + cpu;
883 	} else {
884 		min = BLOCKDEV_INODE_MAX;
885 		max = ~(ULLONG_MAX << bits);
886 		hint = c->unused_inode_hints;
887 	}
888 
889 	start = READ_ONCE(*hint);
890 
891 	if (start >= max || start < min)
892 		start = min;
893 
894 	pos = start;
895 	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
896 			     BTREE_ITER_all_snapshots|
897 			     BTREE_ITER_intent);
898 again:
899 	while ((k = bch2_btree_iter_peek(iter)).k &&
900 	       !(ret = bkey_err(k)) &&
901 	       bkey_lt(k.k->p, POS(0, max))) {
902 		if (pos < iter->pos.offset)
903 			goto found_slot;
904 
905 		/*
906 		 * We don't need to iterate over keys in every snapshot once
907 		 * we've found just one:
908 		 */
909 		pos = iter->pos.offset + 1;
910 		bch2_btree_iter_set_pos(iter, POS(0, pos));
911 	}
912 
913 	if (!ret && pos < max)
914 		goto found_slot;
915 
916 	if (!ret && start == min)
917 		ret = -BCH_ERR_ENOSPC_inode_create;
918 
919 	if (ret) {
920 		bch2_trans_iter_exit(trans, iter);
921 		return ret;
922 	}
923 
924 	/* Retry from start */
925 	pos = start = min;
926 	bch2_btree_iter_set_pos(iter, POS(0, pos));
927 	goto again;
928 found_slot:
929 	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
930 	k = bch2_btree_iter_peek_slot(iter);
931 	ret = bkey_err(k);
932 	if (ret) {
933 		bch2_trans_iter_exit(trans, iter);
934 		return ret;
935 	}
936 
937 	*hint			= k.k->p.offset;
938 	inode_u->bi_inum	= k.k->p.offset;
939 	inode_u->bi_generation	= bkey_generation(k);
940 	return 0;
941 }
942 
bch2_inode_delete_keys(struct btree_trans * trans,subvol_inum inum,enum btree_id id)943 static int bch2_inode_delete_keys(struct btree_trans *trans,
944 				  subvol_inum inum, enum btree_id id)
945 {
946 	struct btree_iter iter;
947 	struct bkey_s_c k;
948 	struct bkey_i delete;
949 	struct bpos end = POS(inum.inum, U64_MAX);
950 	u32 snapshot;
951 	int ret = 0;
952 
953 	/*
954 	 * We're never going to be deleting partial extents, no need to use an
955 	 * extent iterator:
956 	 */
957 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
958 			     BTREE_ITER_intent);
959 
960 	while (1) {
961 		bch2_trans_begin(trans);
962 
963 		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
964 		if (ret)
965 			goto err;
966 
967 		bch2_btree_iter_set_snapshot(&iter, snapshot);
968 
969 		k = bch2_btree_iter_peek_upto(&iter, end);
970 		ret = bkey_err(k);
971 		if (ret)
972 			goto err;
973 
974 		if (!k.k)
975 			break;
976 
977 		bkey_init(&delete.k);
978 		delete.k.p = iter.pos;
979 
980 		if (iter.flags & BTREE_ITER_is_extents)
981 			bch2_key_resize(&delete.k,
982 					bpos_min(end, k.k->p).offset -
983 					iter.pos.offset);
984 
985 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
986 		      bch2_trans_commit(trans, NULL, NULL,
987 					BCH_TRANS_COMMIT_no_enospc);
988 err:
989 		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
990 			break;
991 	}
992 
993 	bch2_trans_iter_exit(trans, &iter);
994 	return ret;
995 }
996 
bch2_inode_rm(struct bch_fs * c,subvol_inum inum)997 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
998 {
999 	struct btree_trans *trans = bch2_trans_get(c);
1000 	struct btree_iter iter = { NULL };
1001 	struct bkey_i_inode_generation delete;
1002 	struct bch_inode_unpacked inode_u;
1003 	struct bkey_s_c k;
1004 	u32 snapshot;
1005 	int ret;
1006 
1007 	/*
1008 	 * If this was a directory, there shouldn't be any real dirents left -
1009 	 * but there could be whiteouts (from hash collisions) that we should
1010 	 * delete:
1011 	 *
1012 	 * XXX: the dirent could ideally would delete whiteouts when they're no
1013 	 * longer needed
1014 	 */
1015 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
1016 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
1017 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
1018 	if (ret)
1019 		goto err;
1020 retry:
1021 	bch2_trans_begin(trans);
1022 
1023 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
1024 	if (ret)
1025 		goto err;
1026 
1027 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1028 			       SPOS(0, inum.inum, snapshot),
1029 			       BTREE_ITER_intent|BTREE_ITER_cached);
1030 	ret = bkey_err(k);
1031 	if (ret)
1032 		goto err;
1033 
1034 	if (!bkey_is_inode(k.k)) {
1035 		bch2_fs_inconsistent(c,
1036 				     "inode %llu:%u not found when deleting",
1037 				     inum.inum, snapshot);
1038 		ret = -EIO;
1039 		goto err;
1040 	}
1041 
1042 	bch2_inode_unpack(k, &inode_u);
1043 
1044 	bkey_inode_generation_init(&delete.k_i);
1045 	delete.k.p = iter.pos;
1046 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1047 
1048 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1049 		bch2_trans_commit(trans, NULL, NULL,
1050 				BCH_TRANS_COMMIT_no_enospc);
1051 err:
1052 	bch2_trans_iter_exit(trans, &iter);
1053 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1054 		goto retry;
1055 
1056 	if (ret)
1057 		goto err2;
1058 
1059 	ret = delete_ancestor_snapshot_inodes(trans, SPOS(0, inum.inum, snapshot));
1060 err2:
1061 	bch2_trans_put(trans);
1062 	return ret;
1063 }
1064 
bch2_inode_find_by_inum_nowarn_trans(struct btree_trans * trans,subvol_inum inum,struct bch_inode_unpacked * inode)1065 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
1066 				  subvol_inum inum,
1067 				  struct bch_inode_unpacked *inode)
1068 {
1069 	struct btree_iter iter;
1070 	int ret;
1071 
1072 	ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
1073 	if (!ret)
1074 		bch2_trans_iter_exit(trans, &iter);
1075 	return ret;
1076 }
1077 
bch2_inode_find_by_inum_trans(struct btree_trans * trans,subvol_inum inum,struct bch_inode_unpacked * inode)1078 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
1079 				  subvol_inum inum,
1080 				  struct bch_inode_unpacked *inode)
1081 {
1082 	struct btree_iter iter;
1083 	int ret;
1084 
1085 	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
1086 	if (!ret)
1087 		bch2_trans_iter_exit(trans, &iter);
1088 	return ret;
1089 }
1090 
bch2_inode_find_by_inum(struct bch_fs * c,subvol_inum inum,struct bch_inode_unpacked * inode)1091 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
1092 			    struct bch_inode_unpacked *inode)
1093 {
1094 	return bch2_trans_do(c, bch2_inode_find_by_inum_trans(trans, inum, inode));
1095 }
1096 
bch2_inode_nlink_inc(struct bch_inode_unpacked * bi)1097 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
1098 {
1099 	if (bi->bi_flags & BCH_INODE_unlinked)
1100 		bi->bi_flags &= ~BCH_INODE_unlinked;
1101 	else {
1102 		if (bi->bi_nlink == U32_MAX)
1103 			return -EINVAL;
1104 
1105 		bi->bi_nlink++;
1106 	}
1107 
1108 	return 0;
1109 }
1110 
bch2_inode_nlink_dec(struct btree_trans * trans,struct bch_inode_unpacked * bi)1111 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
1112 {
1113 	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
1114 		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
1115 					bi->bi_inum);
1116 		return;
1117 	}
1118 
1119 	if (bi->bi_flags & BCH_INODE_unlinked) {
1120 		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
1121 		return;
1122 	}
1123 
1124 	if (bi->bi_nlink)
1125 		bi->bi_nlink--;
1126 	else
1127 		bi->bi_flags |= BCH_INODE_unlinked;
1128 }
1129 
bch2_inode_opts_to_opts(struct bch_inode_unpacked * inode)1130 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
1131 {
1132 	struct bch_opts ret = { 0 };
1133 #define x(_name, _bits)							\
1134 	if (inode->bi_##_name)						\
1135 		opt_set(ret, _name, inode->bi_##_name - 1);
1136 	BCH_INODE_OPTS()
1137 #undef x
1138 	return ret;
1139 }
1140 
bch2_inode_opts_get(struct bch_io_opts * opts,struct bch_fs * c,struct bch_inode_unpacked * inode)1141 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
1142 			 struct bch_inode_unpacked *inode)
1143 {
1144 #define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
1145 	BCH_INODE_OPTS()
1146 #undef x
1147 
1148 	if (opts->nocow)
1149 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
1150 }
1151 
bch2_inum_opts_get(struct btree_trans * trans,subvol_inum inum,struct bch_io_opts * opts)1152 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
1153 {
1154 	struct bch_inode_unpacked inode;
1155 	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
1156 
1157 	if (ret)
1158 		return ret;
1159 
1160 	bch2_inode_opts_get(opts, trans->c, &inode);
1161 	return 0;
1162 }
1163 
__bch2_inode_rm_snapshot(struct btree_trans * trans,u64 inum,u32 snapshot)1164 static noinline int __bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1165 {
1166 	struct bch_fs *c = trans->c;
1167 	struct btree_iter iter = { NULL };
1168 	struct bkey_i_inode_generation delete;
1169 	struct bch_inode_unpacked inode_u;
1170 	struct bkey_s_c k;
1171 	int ret;
1172 
1173 	do {
1174 		ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1175 						      SPOS(inum, 0, snapshot),
1176 						      SPOS(inum, U64_MAX, snapshot),
1177 						      0, NULL) ?:
1178 			bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1179 						      SPOS(inum, 0, snapshot),
1180 						      SPOS(inum, U64_MAX, snapshot),
1181 						      0, NULL) ?:
1182 			bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1183 						      SPOS(inum, 0, snapshot),
1184 						      SPOS(inum, U64_MAX, snapshot),
1185 						      0, NULL);
1186 	} while (ret == -BCH_ERR_transaction_restart_nested);
1187 	if (ret)
1188 		goto err;
1189 retry:
1190 	bch2_trans_begin(trans);
1191 
1192 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1193 			       SPOS(0, inum, snapshot), BTREE_ITER_intent);
1194 	ret = bkey_err(k);
1195 	if (ret)
1196 		goto err;
1197 
1198 	if (!bkey_is_inode(k.k)) {
1199 		bch2_fs_inconsistent(c,
1200 				     "inode %llu:%u not found when deleting",
1201 				     inum, snapshot);
1202 		ret = -EIO;
1203 		goto err;
1204 	}
1205 
1206 	bch2_inode_unpack(k, &inode_u);
1207 
1208 	/* Subvolume root? */
1209 	if (inode_u.bi_subvol)
1210 		bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1211 
1212 	bkey_inode_generation_init(&delete.k_i);
1213 	delete.k.p = iter.pos;
1214 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1215 
1216 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1217 		bch2_trans_commit(trans, NULL, NULL,
1218 				BCH_TRANS_COMMIT_no_enospc);
1219 err:
1220 	bch2_trans_iter_exit(trans, &iter);
1221 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1222 		goto retry;
1223 
1224 	return ret ?: -BCH_ERR_transaction_restart_nested;
1225 }
1226 
1227 /*
1228  * After deleting an inode, there may be versions in older snapshots that should
1229  * also be deleted - if they're not referenced by sibling snapshots and not open
1230  * in other subvolumes:
1231  */
delete_ancestor_snapshot_inodes(struct btree_trans * trans,struct bpos pos)1232 static int delete_ancestor_snapshot_inodes(struct btree_trans *trans, struct bpos pos)
1233 {
1234 	struct btree_iter iter;
1235 	struct bkey_s_c k;
1236 	int ret;
1237 next_parent:
1238 	ret = lockrestart_do(trans,
1239 		bkey_err(k = bch2_inode_get_iter_snapshot_parent(trans, &iter, pos, 0)));
1240 	if (ret || !k.k)
1241 		return ret;
1242 
1243 	bool unlinked = bkey_is_unlinked_inode(k);
1244 	pos = k.k->p;
1245 	bch2_trans_iter_exit(trans, &iter);
1246 
1247 	if (!unlinked)
1248 		return 0;
1249 
1250 	ret = lockrestart_do(trans, bch2_inode_or_descendents_is_open(trans, pos));
1251 	if (ret)
1252 		return ret < 0 ? ret : 0;
1253 
1254 	ret = __bch2_inode_rm_snapshot(trans, pos.offset, pos.snapshot);
1255 	if (ret)
1256 		return ret;
1257 	goto next_parent;
1258 }
1259 
bch2_inode_rm_snapshot(struct btree_trans * trans,u64 inum,u32 snapshot)1260 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1261 {
1262 	return __bch2_inode_rm_snapshot(trans, inum, snapshot) ?:
1263 		delete_ancestor_snapshot_inodes(trans, SPOS(0, inum, snapshot));
1264 }
1265 
may_delete_deleted_inode(struct btree_trans * trans,struct btree_iter * iter,struct bpos pos,bool * need_another_pass)1266 static int may_delete_deleted_inode(struct btree_trans *trans,
1267 				    struct btree_iter *iter,
1268 				    struct bpos pos,
1269 				    bool *need_another_pass)
1270 {
1271 	struct bch_fs *c = trans->c;
1272 	struct btree_iter inode_iter;
1273 	struct bkey_s_c k;
1274 	struct bch_inode_unpacked inode;
1275 	struct printbuf buf = PRINTBUF;
1276 	int ret;
1277 
1278 	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_cached);
1279 	ret = bkey_err(k);
1280 	if (ret)
1281 		return ret;
1282 
1283 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1284 	if (fsck_err_on(!bkey_is_inode(k.k),
1285 			trans, deleted_inode_missing,
1286 			"nonexistent inode %llu:%u in deleted_inodes btree",
1287 			pos.offset, pos.snapshot))
1288 		goto delete;
1289 
1290 	ret = bch2_inode_unpack(k, &inode);
1291 	if (ret)
1292 		goto out;
1293 
1294 	if (S_ISDIR(inode.bi_mode)) {
1295 		ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot);
1296 		if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY),
1297 				trans, deleted_inode_is_dir,
1298 				"non empty directory %llu:%u in deleted_inodes btree",
1299 				pos.offset, pos.snapshot))
1300 			goto delete;
1301 		if (ret)
1302 			goto out;
1303 	}
1304 
1305 	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked),
1306 			trans, deleted_inode_not_unlinked,
1307 			"non-deleted inode %llu:%u in deleted_inodes btree",
1308 			pos.offset, pos.snapshot))
1309 		goto delete;
1310 
1311 	if (fsck_err_on(inode.bi_flags & BCH_INODE_has_child_snapshot,
1312 			trans, deleted_inode_has_child_snapshots,
1313 			"inode with child snapshots %llu:%u in deleted_inodes btree",
1314 			pos.offset, pos.snapshot))
1315 		goto delete;
1316 
1317 	ret = bch2_inode_has_child_snapshots(trans, k.k->p);
1318 	if (ret < 0)
1319 		goto out;
1320 
1321 	if (ret) {
1322 		if (fsck_err(trans, inode_has_child_snapshots_wrong,
1323 			     "inode has_child_snapshots flag wrong (should be set)\n%s",
1324 			     (printbuf_reset(&buf),
1325 			      bch2_inode_unpacked_to_text(&buf, &inode),
1326 			      buf.buf))) {
1327 			inode.bi_flags |= BCH_INODE_has_child_snapshot;
1328 			ret = __bch2_fsck_write_inode(trans, &inode);
1329 			if (ret)
1330 				goto out;
1331 		}
1332 		goto delete;
1333 
1334 	}
1335 
1336 	if (test_bit(BCH_FS_clean_recovery, &c->flags) &&
1337 	    !fsck_err(trans, deleted_inode_but_clean,
1338 		      "filesystem marked as clean but have deleted inode %llu:%u",
1339 		      pos.offset, pos.snapshot)) {
1340 		ret = 0;
1341 		goto out;
1342 	}
1343 
1344 	ret = 1;
1345 out:
1346 fsck_err:
1347 	bch2_trans_iter_exit(trans, &inode_iter);
1348 	printbuf_exit(&buf);
1349 	return ret;
1350 delete:
1351 	ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false);
1352 	goto out;
1353 }
1354 
bch2_delete_dead_inodes(struct bch_fs * c)1355 int bch2_delete_dead_inodes(struct bch_fs *c)
1356 {
1357 	struct btree_trans *trans = bch2_trans_get(c);
1358 	bool need_another_pass;
1359 	int ret;
1360 again:
1361 	/*
1362 	 * if we ran check_inodes() unlinked inodes will have already been
1363 	 * cleaned up but the write buffer will be out of sync; therefore we
1364 	 * alway need a write buffer flush
1365 	 */
1366 	ret = bch2_btree_write_buffer_flush_sync(trans);
1367 	if (ret)
1368 		goto err;
1369 
1370 	need_another_pass = false;
1371 
1372 	/*
1373 	 * Weird transaction restart handling here because on successful delete,
1374 	 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1375 	 * but we can't retry because the btree write buffer won't have been
1376 	 * flushed and we'd spin:
1377 	 */
1378 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1379 					BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1380 					NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1381 		ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1382 		if (ret > 0) {
1383 			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1384 
1385 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1386 			/*
1387 			 * We don't want to loop here: a transaction restart
1388 			 * error here means we handled a transaction restart and
1389 			 * we're actually done, but if we loop we'll retry the
1390 			 * same key because the write buffer hasn't been flushed
1391 			 * yet
1392 			 */
1393 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1394 				ret = 0;
1395 				continue;
1396 			}
1397 		}
1398 
1399 		ret;
1400 	}));
1401 
1402 	if (!ret && need_another_pass)
1403 		goto again;
1404 err:
1405 	bch2_trans_put(trans);
1406 	return ret;
1407 }
1408