xref: /linux/fs/bcachefs/inode.c (revision 71dfa617ea9f18e4585fe78364217cd32b1fc382)
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 "error.h"
12 #include "extents.h"
13 #include "extent_update.h"
14 #include "inode.h"
15 #include "str_hash.h"
16 #include "snapshot.h"
17 #include "subvolume.h"
18 #include "varint.h"
19 
20 #include <linux/random.h>
21 
22 #include <asm/unaligned.h>
23 
24 #define x(name, ...)	#name,
25 const char * const bch2_inode_opts[] = {
26 	BCH_INODE_OPTS()
27 	NULL,
28 };
29 
30 static const char * const bch2_inode_flag_strs[] = {
31 	BCH_INODE_FLAGS()
32 	NULL
33 };
34 #undef  x
35 
36 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
37 
38 static int inode_decode_field(const u8 *in, const u8 *end,
39 			      u64 out[2], unsigned *out_bits)
40 {
41 	__be64 be[2] = { 0, 0 };
42 	unsigned bytes, shift;
43 	u8 *p;
44 
45 	if (in >= end)
46 		return -1;
47 
48 	if (!*in)
49 		return -1;
50 
51 	/*
52 	 * position of highest set bit indicates number of bytes:
53 	 * shift = number of bits to remove in high byte:
54 	 */
55 	shift	= 8 - __fls(*in); /* 1 <= shift <= 8 */
56 	bytes	= byte_table[shift - 1];
57 
58 	if (in + bytes > end)
59 		return -1;
60 
61 	p = (u8 *) be + 16 - bytes;
62 	memcpy(p, in, bytes);
63 	*p ^= (1 << 8) >> shift;
64 
65 	out[0] = be64_to_cpu(be[0]);
66 	out[1] = be64_to_cpu(be[1]);
67 	*out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
68 
69 	return bytes;
70 }
71 
72 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
73 					   const struct bch_inode_unpacked *inode)
74 {
75 	struct bkey_i_inode_v3 *k = &packed->inode;
76 	u8 *out = k->v.fields;
77 	u8 *end = (void *) &packed[1];
78 	u8 *last_nonzero_field = out;
79 	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
80 	unsigned bytes;
81 	int ret;
82 
83 	bkey_inode_v3_init(&packed->inode.k_i);
84 	packed->inode.k.p.offset	= inode->bi_inum;
85 	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
86 	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
87 	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
88 	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
89 	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
90 	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
91 	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
92 	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
93 
94 
95 #define x(_name, _bits)							\
96 	nr_fields++;							\
97 									\
98 	if (inode->_name) {						\
99 		ret = bch2_varint_encode_fast(out, inode->_name);	\
100 		out += ret;						\
101 									\
102 		if (_bits > 64)						\
103 			*out++ = 0;					\
104 									\
105 		last_nonzero_field = out;				\
106 		last_nonzero_fieldnr = nr_fields;			\
107 	} else {							\
108 		*out++ = 0;						\
109 									\
110 		if (_bits > 64)						\
111 			*out++ = 0;					\
112 	}
113 
114 	BCH_INODE_FIELDS_v3()
115 #undef  x
116 	BUG_ON(out > end);
117 
118 	out = last_nonzero_field;
119 	nr_fields = last_nonzero_fieldnr;
120 
121 	bytes = out - (u8 *) &packed->inode.v;
122 	set_bkey_val_bytes(&packed->inode.k, bytes);
123 	memset_u64s_tail(&packed->inode.v, 0, bytes);
124 
125 	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
126 
127 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
128 		struct bch_inode_unpacked unpacked;
129 
130 		ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
131 		BUG_ON(ret);
132 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
133 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
134 		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
135 		BUG_ON(unpacked.bi_size		!= inode->bi_size);
136 		BUG_ON(unpacked.bi_version	!= inode->bi_version);
137 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
138 
139 #define x(_name, _bits)	if (unpacked._name != inode->_name)		\
140 			panic("unpacked %llu should be %llu",		\
141 			      (u64) unpacked._name, (u64) inode->_name);
142 		BCH_INODE_FIELDS_v3()
143 #undef  x
144 	}
145 }
146 
147 void bch2_inode_pack(struct bkey_inode_buf *packed,
148 		     const struct bch_inode_unpacked *inode)
149 {
150 	bch2_inode_pack_inlined(packed, inode);
151 }
152 
153 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
154 				struct bch_inode_unpacked *unpacked)
155 {
156 	const u8 *in = inode.v->fields;
157 	const u8 *end = bkey_val_end(inode);
158 	u64 field[2];
159 	unsigned fieldnr = 0, field_bits;
160 	int ret;
161 
162 #define x(_name, _bits)					\
163 	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
164 		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
165 		memset((void *) unpacked + offset, 0,			\
166 		       sizeof(*unpacked) - offset);			\
167 		return 0;						\
168 	}								\
169 									\
170 	ret = inode_decode_field(in, end, field, &field_bits);		\
171 	if (ret < 0)							\
172 		return ret;						\
173 									\
174 	if (field_bits > sizeof(unpacked->_name) * 8)			\
175 		return -1;						\
176 									\
177 	unpacked->_name = field[1];					\
178 	in += ret;
179 
180 	BCH_INODE_FIELDS_v2()
181 #undef  x
182 
183 	/* XXX: signal if there were more fields than expected? */
184 	return 0;
185 }
186 
187 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
188 				const u8 *in, const u8 *end,
189 				unsigned nr_fields)
190 {
191 	unsigned fieldnr = 0;
192 	int ret;
193 	u64 v[2];
194 
195 #define x(_name, _bits)							\
196 	if (fieldnr < nr_fields) {					\
197 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
198 		if (ret < 0)						\
199 			return ret;					\
200 		in += ret;						\
201 									\
202 		if (_bits > 64) {					\
203 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
204 			if (ret < 0)					\
205 				return ret;				\
206 			in += ret;					\
207 		} else {						\
208 			v[1] = 0;					\
209 		}							\
210 	} else {							\
211 		v[0] = v[1] = 0;					\
212 	}								\
213 									\
214 	unpacked->_name = v[0];						\
215 	if (v[1] || v[0] != unpacked->_name)				\
216 		return -1;						\
217 	fieldnr++;
218 
219 	BCH_INODE_FIELDS_v2()
220 #undef  x
221 
222 	/* XXX: signal if there were more fields than expected? */
223 	return 0;
224 }
225 
226 static int bch2_inode_unpack_v3(struct bkey_s_c k,
227 				struct bch_inode_unpacked *unpacked)
228 {
229 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
230 	const u8 *in = inode.v->fields;
231 	const u8 *end = bkey_val_end(inode);
232 	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
233 	unsigned fieldnr = 0;
234 	int ret;
235 	u64 v[2];
236 
237 	unpacked->bi_inum	= inode.k->p.offset;
238 	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
239 	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
240 	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
241 	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
242 	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
243 	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
244 	unpacked->bi_mode	= INODEv3_MODE(inode.v);
245 
246 #define x(_name, _bits)							\
247 	if (fieldnr < nr_fields) {					\
248 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
249 		if (ret < 0)						\
250 			return ret;					\
251 		in += ret;						\
252 									\
253 		if (_bits > 64) {					\
254 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
255 			if (ret < 0)					\
256 				return ret;				\
257 			in += ret;					\
258 		} else {						\
259 			v[1] = 0;					\
260 		}							\
261 	} else {							\
262 		v[0] = v[1] = 0;					\
263 	}								\
264 									\
265 	unpacked->_name = v[0];						\
266 	if (v[1] || v[0] != unpacked->_name)				\
267 		return -1;						\
268 	fieldnr++;
269 
270 	BCH_INODE_FIELDS_v3()
271 #undef  x
272 
273 	/* XXX: signal if there were more fields than expected? */
274 	return 0;
275 }
276 
277 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
278 					       struct bch_inode_unpacked *unpacked)
279 {
280 	memset(unpacked, 0, sizeof(*unpacked));
281 
282 	switch (k.k->type) {
283 	case KEY_TYPE_inode: {
284 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
285 
286 		unpacked->bi_inum	= inode.k->p.offset;
287 		unpacked->bi_journal_seq= 0;
288 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
289 		unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
290 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
291 
292 		if (INODE_NEW_VARINT(inode.v)) {
293 			return bch2_inode_unpack_v2(unpacked, inode.v->fields,
294 						    bkey_val_end(inode),
295 						    INODE_NR_FIELDS(inode.v));
296 		} else {
297 			return bch2_inode_unpack_v1(inode, unpacked);
298 		}
299 		break;
300 	}
301 	case KEY_TYPE_inode_v2: {
302 		struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
303 
304 		unpacked->bi_inum	= inode.k->p.offset;
305 		unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
306 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
307 		unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
308 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
309 
310 		return bch2_inode_unpack_v2(unpacked, inode.v->fields,
311 					    bkey_val_end(inode),
312 					    INODEv2_NR_FIELDS(inode.v));
313 	}
314 	default:
315 		BUG();
316 	}
317 }
318 
319 int bch2_inode_unpack(struct bkey_s_c k,
320 		      struct bch_inode_unpacked *unpacked)
321 {
322 	if (likely(k.k->type == KEY_TYPE_inode_v3))
323 		return bch2_inode_unpack_v3(k, unpacked);
324 	return bch2_inode_unpack_slowpath(k, unpacked);
325 }
326 
327 int bch2_inode_peek_nowarn(struct btree_trans *trans,
328 		    struct btree_iter *iter,
329 		    struct bch_inode_unpacked *inode,
330 		    subvol_inum inum, unsigned flags)
331 {
332 	struct bkey_s_c k;
333 	u32 snapshot;
334 	int ret;
335 
336 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
337 	if (ret)
338 		return ret;
339 
340 	k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
341 			       SPOS(0, inum.inum, snapshot),
342 			       flags|BTREE_ITER_CACHED);
343 	ret = bkey_err(k);
344 	if (ret)
345 		return ret;
346 
347 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
348 	if (ret)
349 		goto err;
350 
351 	ret = bch2_inode_unpack(k, inode);
352 	if (ret)
353 		goto err;
354 
355 	return 0;
356 err:
357 	bch2_trans_iter_exit(trans, iter);
358 	return ret;
359 }
360 
361 int bch2_inode_peek(struct btree_trans *trans,
362 		    struct btree_iter *iter,
363 		    struct bch_inode_unpacked *inode,
364 		    subvol_inum inum, unsigned flags)
365 {
366 	int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
367 	bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
368 	return ret;
369 }
370 
371 int bch2_inode_write_flags(struct btree_trans *trans,
372 		     struct btree_iter *iter,
373 		     struct bch_inode_unpacked *inode,
374 		     enum btree_update_flags flags)
375 {
376 	struct bkey_inode_buf *inode_p;
377 
378 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
379 	if (IS_ERR(inode_p))
380 		return PTR_ERR(inode_p);
381 
382 	bch2_inode_pack_inlined(inode_p, inode);
383 	inode_p->inode.k.p.snapshot = iter->snapshot;
384 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
385 }
386 
387 int __bch2_fsck_write_inode(struct btree_trans *trans,
388 			 struct bch_inode_unpacked *inode,
389 			 u32 snapshot)
390 {
391 	struct bkey_inode_buf *inode_p =
392 		bch2_trans_kmalloc(trans, sizeof(*inode_p));
393 
394 	if (IS_ERR(inode_p))
395 		return PTR_ERR(inode_p);
396 
397 	bch2_inode_pack(inode_p, inode);
398 	inode_p->inode.k.p.snapshot = snapshot;
399 
400 	return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
401 				&inode_p->inode.k_i,
402 				BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
403 }
404 
405 int bch2_fsck_write_inode(struct btree_trans *trans,
406 			    struct bch_inode_unpacked *inode,
407 			    u32 snapshot)
408 {
409 	int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
410 			    __bch2_fsck_write_inode(trans, inode, snapshot));
411 	bch_err_fn(trans->c, ret);
412 	return ret;
413 }
414 
415 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
416 {
417 	struct bch_inode_unpacked u;
418 	struct bkey_inode_buf *inode_p;
419 	int ret;
420 
421 	if (!bkey_is_inode(&k->k))
422 		return ERR_PTR(-ENOENT);
423 
424 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
425 	if (IS_ERR(inode_p))
426 		return ERR_CAST(inode_p);
427 
428 	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
429 	if (ret)
430 		return ERR_PTR(ret);
431 
432 	bch2_inode_pack(inode_p, &u);
433 	return &inode_p->inode.k_i;
434 }
435 
436 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err)
437 {
438 	struct bch_inode_unpacked unpacked;
439 	int ret = 0;
440 
441 	bkey_fsck_err_on(k.k->p.inode, c, err,
442 			 inode_pos_inode_nonzero,
443 			 "nonzero k.p.inode");
444 
445 	bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err,
446 			 inode_pos_blockdev_range,
447 			 "fs inode in blockdev range");
448 
449 	bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err,
450 			 inode_unpack_error,
451 			 "invalid variable length fields");
452 
453 	bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err,
454 			 inode_checksum_type_invalid,
455 			 "invalid data checksum type (%u >= %u",
456 			 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
457 
458 	bkey_fsck_err_on(unpacked.bi_compression &&
459 			 !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err,
460 			 inode_compression_type_invalid,
461 			 "invalid compression opt %u", unpacked.bi_compression - 1);
462 
463 	bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
464 			 unpacked.bi_nlink != 0, c, err,
465 			 inode_unlinked_but_nlink_nonzero,
466 			 "flagged as unlinked but bi_nlink != 0");
467 
468 	bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err,
469 			 inode_subvol_root_but_not_dir,
470 			 "subvolume root but not a directory");
471 fsck_err:
472 	return ret;
473 }
474 
475 int bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k,
476 		       enum bkey_invalid_flags flags,
477 		       struct printbuf *err)
478 {
479 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
480 	int ret = 0;
481 
482 	bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
483 			 inode_str_hash_invalid,
484 			 "invalid str hash type (%llu >= %u)",
485 			 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
486 
487 	ret = __bch2_inode_invalid(c, k, err);
488 fsck_err:
489 	return ret;
490 }
491 
492 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
493 			  enum bkey_invalid_flags flags,
494 			  struct printbuf *err)
495 {
496 	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
497 	int ret = 0;
498 
499 	bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
500 			 inode_str_hash_invalid,
501 			 "invalid str hash type (%llu >= %u)",
502 			 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
503 
504 	ret = __bch2_inode_invalid(c, k, err);
505 fsck_err:
506 	return ret;
507 }
508 
509 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
510 			  enum bkey_invalid_flags flags,
511 			  struct printbuf *err)
512 {
513 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
514 	int ret = 0;
515 
516 	bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
517 			 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err,
518 			 inode_v3_fields_start_bad,
519 			 "invalid fields_start (got %llu, min %u max %zu)",
520 			 INODEv3_FIELDS_START(inode.v),
521 			 INODEv3_FIELDS_START_INITIAL,
522 			 bkey_val_u64s(inode.k));
523 
524 	bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
525 			 inode_str_hash_invalid,
526 			 "invalid str hash type (%llu >= %u)",
527 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
528 
529 	ret = __bch2_inode_invalid(c, k, err);
530 fsck_err:
531 	return ret;
532 }
533 
534 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
535 					  struct bch_inode_unpacked *inode)
536 {
537 	printbuf_indent_add(out, 2);
538 	prt_printf(out, "mode=%o", inode->bi_mode);
539 	prt_newline(out);
540 
541 	prt_str(out, "flags=");
542 	prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
543 	prt_printf(out, " (%x)", inode->bi_flags);
544 	prt_newline(out);
545 
546 	prt_printf(out, "journal_seq=%llu", inode->bi_journal_seq);
547 	prt_newline(out);
548 
549 	prt_printf(out, "bi_size=%llu", inode->bi_size);
550 	prt_newline(out);
551 
552 	prt_printf(out, "bi_sectors=%llu", inode->bi_sectors);
553 	prt_newline(out);
554 
555 	prt_printf(out, "bi_version=%llu", inode->bi_version);
556 	prt_newline(out);
557 
558 #define x(_name, _bits)						\
559 	prt_printf(out, #_name "=%llu", (u64) inode->_name);	\
560 	prt_newline(out);
561 	BCH_INODE_FIELDS_v3()
562 #undef  x
563 	printbuf_indent_sub(out, 2);
564 }
565 
566 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
567 {
568 	prt_printf(out, "inum: %llu ", inode->bi_inum);
569 	__bch2_inode_unpacked_to_text(out, inode);
570 }
571 
572 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
573 {
574 	struct bch_inode_unpacked inode;
575 
576 	if (bch2_inode_unpack(k, &inode)) {
577 		prt_printf(out, "(unpack error)");
578 		return;
579 	}
580 
581 	__bch2_inode_unpacked_to_text(out, &inode);
582 }
583 
584 static inline u64 bkey_inode_flags(struct bkey_s_c k)
585 {
586 	switch (k.k->type) {
587 	case KEY_TYPE_inode:
588 		return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
589 	case KEY_TYPE_inode_v2:
590 		return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
591 	case KEY_TYPE_inode_v3:
592 		return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
593 	default:
594 		return 0;
595 	}
596 }
597 
598 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
599 {
600 	return bkey_inode_flags(k) & BCH_INODE_unlinked;
601 }
602 
603 int bch2_trigger_inode(struct btree_trans *trans,
604 		       enum btree_id btree_id, unsigned level,
605 		       struct bkey_s_c old,
606 		       struct bkey_s new,
607 		       unsigned flags)
608 {
609 	s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
610 
611 	if (flags & BTREE_TRIGGER_TRANSACTIONAL) {
612 		if (nr) {
613 			int ret = bch2_replicas_deltas_realloc(trans, 0);
614 			if (ret)
615 				return ret;
616 
617 			trans->fs_usage_deltas->nr_inodes += nr;
618 		}
619 
620 		bool old_deleted = bkey_is_deleted_inode(old);
621 		bool new_deleted = bkey_is_deleted_inode(new.s_c);
622 		if (old_deleted != new_deleted) {
623 			int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes,
624 							      new.k->p, new_deleted);
625 			if (ret)
626 				return ret;
627 		}
628 	}
629 
630 	if ((flags & BTREE_TRIGGER_ATOMIC) && (flags & BTREE_TRIGGER_INSERT)) {
631 		BUG_ON(!trans->journal_res.seq);
632 
633 		bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
634 	}
635 
636 	if (flags & BTREE_TRIGGER_GC) {
637 		struct bch_fs *c = trans->c;
638 
639 		percpu_down_read(&c->mark_lock);
640 		this_cpu_add(c->usage_gc->b.nr_inodes, nr);
641 		percpu_up_read(&c->mark_lock);
642 	}
643 
644 	return 0;
645 }
646 
647 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k,
648 				  enum bkey_invalid_flags flags,
649 				  struct printbuf *err)
650 {
651 	int ret = 0;
652 
653 	bkey_fsck_err_on(k.k->p.inode, c, err,
654 			 inode_pos_inode_nonzero,
655 			 "nonzero k.p.inode");
656 fsck_err:
657 	return ret;
658 }
659 
660 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
661 				   struct bkey_s_c k)
662 {
663 	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
664 
665 	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
666 }
667 
668 void bch2_inode_init_early(struct bch_fs *c,
669 			   struct bch_inode_unpacked *inode_u)
670 {
671 	enum bch_str_hash_type str_hash =
672 		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
673 
674 	memset(inode_u, 0, sizeof(*inode_u));
675 
676 	/* ick */
677 	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
678 	get_random_bytes(&inode_u->bi_hash_seed,
679 			 sizeof(inode_u->bi_hash_seed));
680 }
681 
682 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
683 			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
684 			  struct bch_inode_unpacked *parent)
685 {
686 	inode_u->bi_mode	= mode;
687 	inode_u->bi_uid		= uid;
688 	inode_u->bi_gid		= gid;
689 	inode_u->bi_dev		= rdev;
690 	inode_u->bi_atime	= now;
691 	inode_u->bi_mtime	= now;
692 	inode_u->bi_ctime	= now;
693 	inode_u->bi_otime	= now;
694 
695 	if (parent && parent->bi_mode & S_ISGID) {
696 		inode_u->bi_gid = parent->bi_gid;
697 		if (S_ISDIR(mode))
698 			inode_u->bi_mode |= S_ISGID;
699 	}
700 
701 	if (parent) {
702 #define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
703 		BCH_INODE_OPTS()
704 #undef x
705 	}
706 }
707 
708 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
709 		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
710 		     struct bch_inode_unpacked *parent)
711 {
712 	bch2_inode_init_early(c, inode_u);
713 	bch2_inode_init_late(inode_u, bch2_current_time(c),
714 			     uid, gid, mode, rdev, parent);
715 }
716 
717 static inline u32 bkey_generation(struct bkey_s_c k)
718 {
719 	switch (k.k->type) {
720 	case KEY_TYPE_inode:
721 	case KEY_TYPE_inode_v2:
722 		BUG();
723 	case KEY_TYPE_inode_generation:
724 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
725 	default:
726 		return 0;
727 	}
728 }
729 
730 /*
731  * This just finds an empty slot:
732  */
733 int bch2_inode_create(struct btree_trans *trans,
734 		      struct btree_iter *iter,
735 		      struct bch_inode_unpacked *inode_u,
736 		      u32 snapshot, u64 cpu)
737 {
738 	struct bch_fs *c = trans->c;
739 	struct bkey_s_c k;
740 	u64 min, max, start, pos, *hint;
741 	int ret = 0;
742 	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
743 
744 	if (c->opts.shard_inode_numbers) {
745 		bits -= c->inode_shard_bits;
746 
747 		min = (cpu << bits);
748 		max = (cpu << bits) | ~(ULLONG_MAX << bits);
749 
750 		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
751 		hint = c->unused_inode_hints + cpu;
752 	} else {
753 		min = BLOCKDEV_INODE_MAX;
754 		max = ~(ULLONG_MAX << bits);
755 		hint = c->unused_inode_hints;
756 	}
757 
758 	start = READ_ONCE(*hint);
759 
760 	if (start >= max || start < min)
761 		start = min;
762 
763 	pos = start;
764 	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
765 			     BTREE_ITER_ALL_SNAPSHOTS|
766 			     BTREE_ITER_INTENT);
767 again:
768 	while ((k = bch2_btree_iter_peek(iter)).k &&
769 	       !(ret = bkey_err(k)) &&
770 	       bkey_lt(k.k->p, POS(0, max))) {
771 		if (pos < iter->pos.offset)
772 			goto found_slot;
773 
774 		/*
775 		 * We don't need to iterate over keys in every snapshot once
776 		 * we've found just one:
777 		 */
778 		pos = iter->pos.offset + 1;
779 		bch2_btree_iter_set_pos(iter, POS(0, pos));
780 	}
781 
782 	if (!ret && pos < max)
783 		goto found_slot;
784 
785 	if (!ret && start == min)
786 		ret = -BCH_ERR_ENOSPC_inode_create;
787 
788 	if (ret) {
789 		bch2_trans_iter_exit(trans, iter);
790 		return ret;
791 	}
792 
793 	/* Retry from start */
794 	pos = start = min;
795 	bch2_btree_iter_set_pos(iter, POS(0, pos));
796 	goto again;
797 found_slot:
798 	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
799 	k = bch2_btree_iter_peek_slot(iter);
800 	ret = bkey_err(k);
801 	if (ret) {
802 		bch2_trans_iter_exit(trans, iter);
803 		return ret;
804 	}
805 
806 	*hint			= k.k->p.offset;
807 	inode_u->bi_inum	= k.k->p.offset;
808 	inode_u->bi_generation	= bkey_generation(k);
809 	return 0;
810 }
811 
812 static int bch2_inode_delete_keys(struct btree_trans *trans,
813 				  subvol_inum inum, enum btree_id id)
814 {
815 	struct btree_iter iter;
816 	struct bkey_s_c k;
817 	struct bkey_i delete;
818 	struct bpos end = POS(inum.inum, U64_MAX);
819 	u32 snapshot;
820 	int ret = 0;
821 
822 	/*
823 	 * We're never going to be deleting partial extents, no need to use an
824 	 * extent iterator:
825 	 */
826 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
827 			     BTREE_ITER_INTENT);
828 
829 	while (1) {
830 		bch2_trans_begin(trans);
831 
832 		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
833 		if (ret)
834 			goto err;
835 
836 		bch2_btree_iter_set_snapshot(&iter, snapshot);
837 
838 		k = bch2_btree_iter_peek_upto(&iter, end);
839 		ret = bkey_err(k);
840 		if (ret)
841 			goto err;
842 
843 		if (!k.k)
844 			break;
845 
846 		bkey_init(&delete.k);
847 		delete.k.p = iter.pos;
848 
849 		if (iter.flags & BTREE_ITER_IS_EXTENTS)
850 			bch2_key_resize(&delete.k,
851 					bpos_min(end, k.k->p).offset -
852 					iter.pos.offset);
853 
854 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
855 		      bch2_trans_commit(trans, NULL, NULL,
856 					BCH_TRANS_COMMIT_no_enospc);
857 err:
858 		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
859 			break;
860 	}
861 
862 	bch2_trans_iter_exit(trans, &iter);
863 	return ret;
864 }
865 
866 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
867 {
868 	struct btree_trans *trans = bch2_trans_get(c);
869 	struct btree_iter iter = { NULL };
870 	struct bkey_i_inode_generation delete;
871 	struct bch_inode_unpacked inode_u;
872 	struct bkey_s_c k;
873 	u32 snapshot;
874 	int ret;
875 
876 	/*
877 	 * If this was a directory, there shouldn't be any real dirents left -
878 	 * but there could be whiteouts (from hash collisions) that we should
879 	 * delete:
880 	 *
881 	 * XXX: the dirent could ideally would delete whiteouts when they're no
882 	 * longer needed
883 	 */
884 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
885 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
886 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
887 	if (ret)
888 		goto err;
889 retry:
890 	bch2_trans_begin(trans);
891 
892 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
893 	if (ret)
894 		goto err;
895 
896 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
897 			       SPOS(0, inum.inum, snapshot),
898 			       BTREE_ITER_INTENT|BTREE_ITER_CACHED);
899 	ret = bkey_err(k);
900 	if (ret)
901 		goto err;
902 
903 	if (!bkey_is_inode(k.k)) {
904 		bch2_fs_inconsistent(c,
905 				     "inode %llu:%u not found when deleting",
906 				     inum.inum, snapshot);
907 		ret = -EIO;
908 		goto err;
909 	}
910 
911 	bch2_inode_unpack(k, &inode_u);
912 
913 	bkey_inode_generation_init(&delete.k_i);
914 	delete.k.p = iter.pos;
915 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
916 
917 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
918 		bch2_trans_commit(trans, NULL, NULL,
919 				BCH_TRANS_COMMIT_no_enospc);
920 err:
921 	bch2_trans_iter_exit(trans, &iter);
922 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
923 		goto retry;
924 
925 	bch2_trans_put(trans);
926 	return ret;
927 }
928 
929 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
930 				  subvol_inum inum,
931 				  struct bch_inode_unpacked *inode)
932 {
933 	struct btree_iter iter;
934 	int ret;
935 
936 	ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
937 	if (!ret)
938 		bch2_trans_iter_exit(trans, &iter);
939 	return ret;
940 }
941 
942 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
943 				  subvol_inum inum,
944 				  struct bch_inode_unpacked *inode)
945 {
946 	struct btree_iter iter;
947 	int ret;
948 
949 	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
950 	if (!ret)
951 		bch2_trans_iter_exit(trans, &iter);
952 	return ret;
953 }
954 
955 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
956 			    struct bch_inode_unpacked *inode)
957 {
958 	return bch2_trans_do(c, NULL, NULL, 0,
959 		bch2_inode_find_by_inum_trans(trans, inum, inode));
960 }
961 
962 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
963 {
964 	if (bi->bi_flags & BCH_INODE_unlinked)
965 		bi->bi_flags &= ~BCH_INODE_unlinked;
966 	else {
967 		if (bi->bi_nlink == U32_MAX)
968 			return -EINVAL;
969 
970 		bi->bi_nlink++;
971 	}
972 
973 	return 0;
974 }
975 
976 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
977 {
978 	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
979 		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
980 					bi->bi_inum);
981 		return;
982 	}
983 
984 	if (bi->bi_flags & BCH_INODE_unlinked) {
985 		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
986 		return;
987 	}
988 
989 	if (bi->bi_nlink)
990 		bi->bi_nlink--;
991 	else
992 		bi->bi_flags |= BCH_INODE_unlinked;
993 }
994 
995 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
996 {
997 	struct bch_opts ret = { 0 };
998 #define x(_name, _bits)							\
999 	if (inode->bi_##_name)						\
1000 		opt_set(ret, _name, inode->bi_##_name - 1);
1001 	BCH_INODE_OPTS()
1002 #undef x
1003 	return ret;
1004 }
1005 
1006 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
1007 			 struct bch_inode_unpacked *inode)
1008 {
1009 #define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
1010 	BCH_INODE_OPTS()
1011 #undef x
1012 
1013 	if (opts->nocow)
1014 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
1015 }
1016 
1017 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
1018 {
1019 	struct bch_inode_unpacked inode;
1020 	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
1021 
1022 	if (ret)
1023 		return ret;
1024 
1025 	bch2_inode_opts_get(opts, trans->c, &inode);
1026 	return 0;
1027 }
1028 
1029 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1030 {
1031 	struct bch_fs *c = trans->c;
1032 	struct btree_iter iter = { NULL };
1033 	struct bkey_i_inode_generation delete;
1034 	struct bch_inode_unpacked inode_u;
1035 	struct bkey_s_c k;
1036 	int ret;
1037 
1038 	do {
1039 		ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1040 						      SPOS(inum, 0, snapshot),
1041 						      SPOS(inum, U64_MAX, snapshot),
1042 						      0, NULL) ?:
1043 			bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1044 						      SPOS(inum, 0, snapshot),
1045 						      SPOS(inum, U64_MAX, snapshot),
1046 						      0, NULL) ?:
1047 			bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1048 						      SPOS(inum, 0, snapshot),
1049 						      SPOS(inum, U64_MAX, snapshot),
1050 						      0, NULL);
1051 	} while (ret == -BCH_ERR_transaction_restart_nested);
1052 	if (ret)
1053 		goto err;
1054 retry:
1055 	bch2_trans_begin(trans);
1056 
1057 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1058 			       SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1059 	ret = bkey_err(k);
1060 	if (ret)
1061 		goto err;
1062 
1063 	if (!bkey_is_inode(k.k)) {
1064 		bch2_fs_inconsistent(c,
1065 				     "inode %llu:%u not found when deleting",
1066 				     inum, snapshot);
1067 		ret = -EIO;
1068 		goto err;
1069 	}
1070 
1071 	bch2_inode_unpack(k, &inode_u);
1072 
1073 	/* Subvolume root? */
1074 	if (inode_u.bi_subvol)
1075 		bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1076 
1077 	bkey_inode_generation_init(&delete.k_i);
1078 	delete.k.p = iter.pos;
1079 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1080 
1081 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1082 		bch2_trans_commit(trans, NULL, NULL,
1083 				BCH_TRANS_COMMIT_no_enospc);
1084 err:
1085 	bch2_trans_iter_exit(trans, &iter);
1086 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1087 		goto retry;
1088 
1089 	return ret ?: -BCH_ERR_transaction_restart_nested;
1090 }
1091 
1092 static int may_delete_deleted_inode(struct btree_trans *trans,
1093 				    struct btree_iter *iter,
1094 				    struct bpos pos,
1095 				    bool *need_another_pass)
1096 {
1097 	struct bch_fs *c = trans->c;
1098 	struct btree_iter inode_iter;
1099 	struct bkey_s_c k;
1100 	struct bch_inode_unpacked inode;
1101 	int ret;
1102 
1103 	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1104 	ret = bkey_err(k);
1105 	if (ret)
1106 		return ret;
1107 
1108 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1109 	if (fsck_err_on(!bkey_is_inode(k.k), c,
1110 			deleted_inode_missing,
1111 			"nonexistent inode %llu:%u in deleted_inodes btree",
1112 			pos.offset, pos.snapshot))
1113 		goto delete;
1114 
1115 	ret = bch2_inode_unpack(k, &inode);
1116 	if (ret)
1117 		goto out;
1118 
1119 	if (S_ISDIR(inode.bi_mode)) {
1120 		ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot);
1121 		if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY),
1122 				c, deleted_inode_is_dir,
1123 				"non empty directory %llu:%u in deleted_inodes btree",
1124 				pos.offset, pos.snapshot))
1125 			goto delete;
1126 		if (ret)
1127 			goto out;
1128 	}
1129 
1130 	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c,
1131 			deleted_inode_not_unlinked,
1132 			"non-deleted inode %llu:%u in deleted_inodes btree",
1133 			pos.offset, pos.snapshot))
1134 		goto delete;
1135 
1136 	if (c->sb.clean &&
1137 	    !fsck_err(c,
1138 		      deleted_inode_but_clean,
1139 		      "filesystem marked as clean but have deleted inode %llu:%u",
1140 		      pos.offset, pos.snapshot)) {
1141 		ret = 0;
1142 		goto out;
1143 	}
1144 
1145 	if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1146 		struct bpos new_min_pos;
1147 
1148 		ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1149 		if (ret)
1150 			goto out;
1151 
1152 		inode.bi_flags &= ~BCH_INODE_unlinked;
1153 
1154 		ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1155 					     BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1156 		bch_err_msg(c, ret, "clearing inode unlinked flag");
1157 		if (ret)
1158 			goto out;
1159 
1160 		/*
1161 		 * We'll need another write buffer flush to pick up the new
1162 		 * unlinked inodes in the snapshot leaves:
1163 		 */
1164 		*need_another_pass = true;
1165 		goto out;
1166 	}
1167 
1168 	ret = 1;
1169 out:
1170 fsck_err:
1171 	bch2_trans_iter_exit(trans, &inode_iter);
1172 	return ret;
1173 delete:
1174 	ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false);
1175 	goto out;
1176 }
1177 
1178 int bch2_delete_dead_inodes(struct bch_fs *c)
1179 {
1180 	struct btree_trans *trans = bch2_trans_get(c);
1181 	bool need_another_pass;
1182 	int ret;
1183 again:
1184 	/*
1185 	 * if we ran check_inodes() unlinked inodes will have already been
1186 	 * cleaned up but the write buffer will be out of sync; therefore we
1187 	 * alway need a write buffer flush
1188 	 */
1189 	ret = bch2_btree_write_buffer_flush_sync(trans);
1190 	if (ret)
1191 		goto err;
1192 
1193 	need_another_pass = false;
1194 
1195 	/*
1196 	 * Weird transaction restart handling here because on successful delete,
1197 	 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1198 	 * but we can't retry because the btree write buffer won't have been
1199 	 * flushed and we'd spin:
1200 	 */
1201 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1202 					BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k,
1203 					NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1204 		ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1205 		if (ret > 0) {
1206 			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1207 
1208 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1209 			/*
1210 			 * We don't want to loop here: a transaction restart
1211 			 * error here means we handled a transaction restart and
1212 			 * we're actually done, but if we loop we'll retry the
1213 			 * same key because the write buffer hasn't been flushed
1214 			 * yet
1215 			 */
1216 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1217 				ret = 0;
1218 				continue;
1219 			}
1220 		}
1221 
1222 		ret;
1223 	}));
1224 
1225 	if (!ret && need_another_pass)
1226 		goto again;
1227 err:
1228 	bch2_trans_put(trans);
1229 	return ret;
1230 }
1231