xref: /linux/fs/bcachefs/inode.c (revision 566ab427f827b0256d3e8ce0235d088e6a9c28bd)
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 "inode.h"
16 #include "str_hash.h"
17 #include "snapshot.h"
18 #include "subvolume.h"
19 #include "varint.h"
20 
21 #include <linux/random.h>
22 
23 #include <linux/unaligned.h>
24 
25 #define x(name, ...)	#name,
26 const char * const bch2_inode_opts[] = {
27 	BCH_INODE_OPTS()
28 	NULL,
29 };
30 
31 static const char * const bch2_inode_flag_strs[] = {
32 	BCH_INODE_FLAGS()
33 	NULL
34 };
35 #undef  x
36 
37 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
38 
39 static int inode_decode_field(const u8 *in, const u8 *end,
40 			      u64 out[2], unsigned *out_bits)
41 {
42 	__be64 be[2] = { 0, 0 };
43 	unsigned bytes, shift;
44 	u8 *p;
45 
46 	if (in >= end)
47 		return -1;
48 
49 	if (!*in)
50 		return -1;
51 
52 	/*
53 	 * position of highest set bit indicates number of bytes:
54 	 * shift = number of bits to remove in high byte:
55 	 */
56 	shift	= 8 - __fls(*in); /* 1 <= shift <= 8 */
57 	bytes	= byte_table[shift - 1];
58 
59 	if (in + bytes > end)
60 		return -1;
61 
62 	p = (u8 *) be + 16 - bytes;
63 	memcpy(p, in, bytes);
64 	*p ^= (1 << 8) >> shift;
65 
66 	out[0] = be64_to_cpu(be[0]);
67 	out[1] = be64_to_cpu(be[1]);
68 	*out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
69 
70 	return bytes;
71 }
72 
73 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
74 					   const struct bch_inode_unpacked *inode)
75 {
76 	struct bkey_i_inode_v3 *k = &packed->inode;
77 	u8 *out = k->v.fields;
78 	u8 *end = (void *) &packed[1];
79 	u8 *last_nonzero_field = out;
80 	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
81 	unsigned bytes;
82 	int ret;
83 
84 	bkey_inode_v3_init(&packed->inode.k_i);
85 	packed->inode.k.p.offset	= inode->bi_inum;
86 	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
87 	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
88 	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
89 	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
90 	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
91 	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
92 	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
93 	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
94 
95 
96 #define x(_name, _bits)							\
97 	nr_fields++;							\
98 									\
99 	if (inode->_name) {						\
100 		ret = bch2_varint_encode_fast(out, inode->_name);	\
101 		out += ret;						\
102 									\
103 		if (_bits > 64)						\
104 			*out++ = 0;					\
105 									\
106 		last_nonzero_field = out;				\
107 		last_nonzero_fieldnr = nr_fields;			\
108 	} else {							\
109 		*out++ = 0;						\
110 									\
111 		if (_bits > 64)						\
112 			*out++ = 0;					\
113 	}
114 
115 	BCH_INODE_FIELDS_v3()
116 #undef  x
117 	BUG_ON(out > end);
118 
119 	out = last_nonzero_field;
120 	nr_fields = last_nonzero_fieldnr;
121 
122 	bytes = out - (u8 *) &packed->inode.v;
123 	set_bkey_val_bytes(&packed->inode.k, bytes);
124 	memset_u64s_tail(&packed->inode.v, 0, bytes);
125 
126 	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
127 
128 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
129 		struct bch_inode_unpacked unpacked;
130 
131 		ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
132 		BUG_ON(ret);
133 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
134 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
135 		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
136 		BUG_ON(unpacked.bi_size		!= inode->bi_size);
137 		BUG_ON(unpacked.bi_version	!= inode->bi_version);
138 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
139 
140 #define x(_name, _bits)	if (unpacked._name != inode->_name)		\
141 			panic("unpacked %llu should be %llu",		\
142 			      (u64) unpacked._name, (u64) inode->_name);
143 		BCH_INODE_FIELDS_v3()
144 #undef  x
145 	}
146 }
147 
148 void bch2_inode_pack(struct bkey_inode_buf *packed,
149 		     const struct bch_inode_unpacked *inode)
150 {
151 	bch2_inode_pack_inlined(packed, inode);
152 }
153 
154 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
155 				struct bch_inode_unpacked *unpacked)
156 {
157 	const u8 *in = inode.v->fields;
158 	const u8 *end = bkey_val_end(inode);
159 	u64 field[2];
160 	unsigned fieldnr = 0, field_bits;
161 	int ret;
162 
163 #define x(_name, _bits)					\
164 	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
165 		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
166 		memset((void *) unpacked + offset, 0,			\
167 		       sizeof(*unpacked) - offset);			\
168 		return 0;						\
169 	}								\
170 									\
171 	ret = inode_decode_field(in, end, field, &field_bits);		\
172 	if (ret < 0)							\
173 		return ret;						\
174 									\
175 	if (field_bits > sizeof(unpacked->_name) * 8)			\
176 		return -1;						\
177 									\
178 	unpacked->_name = field[1];					\
179 	in += ret;
180 
181 	BCH_INODE_FIELDS_v2()
182 #undef  x
183 
184 	/* XXX: signal if there were more fields than expected? */
185 	return 0;
186 }
187 
188 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
189 				const u8 *in, const u8 *end,
190 				unsigned nr_fields)
191 {
192 	unsigned fieldnr = 0;
193 	int ret;
194 	u64 v[2];
195 
196 #define x(_name, _bits)							\
197 	if (fieldnr < nr_fields) {					\
198 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
199 		if (ret < 0)						\
200 			return ret;					\
201 		in += ret;						\
202 									\
203 		if (_bits > 64) {					\
204 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
205 			if (ret < 0)					\
206 				return ret;				\
207 			in += ret;					\
208 		} else {						\
209 			v[1] = 0;					\
210 		}							\
211 	} else {							\
212 		v[0] = v[1] = 0;					\
213 	}								\
214 									\
215 	unpacked->_name = v[0];						\
216 	if (v[1] || v[0] != unpacked->_name)				\
217 		return -1;						\
218 	fieldnr++;
219 
220 	BCH_INODE_FIELDS_v2()
221 #undef  x
222 
223 	/* XXX: signal if there were more fields than expected? */
224 	return 0;
225 }
226 
227 static int bch2_inode_unpack_v3(struct bkey_s_c k,
228 				struct bch_inode_unpacked *unpacked)
229 {
230 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
231 	const u8 *in = inode.v->fields;
232 	const u8 *end = bkey_val_end(inode);
233 	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
234 	unsigned fieldnr = 0;
235 	int ret;
236 	u64 v[2];
237 
238 	unpacked->bi_inum	= inode.k->p.offset;
239 	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
240 	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
241 	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
242 	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
243 	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
244 	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
245 	unpacked->bi_mode	= INODEv3_MODE(inode.v);
246 
247 #define x(_name, _bits)							\
248 	if (fieldnr < nr_fields) {					\
249 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
250 		if (ret < 0)						\
251 			return ret;					\
252 		in += ret;						\
253 									\
254 		if (_bits > 64) {					\
255 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
256 			if (ret < 0)					\
257 				return ret;				\
258 			in += ret;					\
259 		} else {						\
260 			v[1] = 0;					\
261 		}							\
262 	} else {							\
263 		v[0] = v[1] = 0;					\
264 	}								\
265 									\
266 	unpacked->_name = v[0];						\
267 	if (v[1] || v[0] != unpacked->_name)				\
268 		return -1;						\
269 	fieldnr++;
270 
271 	BCH_INODE_FIELDS_v3()
272 #undef  x
273 
274 	/* XXX: signal if there were more fields than expected? */
275 	return 0;
276 }
277 
278 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
279 					       struct bch_inode_unpacked *unpacked)
280 {
281 	memset(unpacked, 0, sizeof(*unpacked));
282 
283 	switch (k.k->type) {
284 	case KEY_TYPE_inode: {
285 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
286 
287 		unpacked->bi_inum	= inode.k->p.offset;
288 		unpacked->bi_journal_seq= 0;
289 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
290 		unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
291 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
292 
293 		if (INODE_NEW_VARINT(inode.v)) {
294 			return bch2_inode_unpack_v2(unpacked, inode.v->fields,
295 						    bkey_val_end(inode),
296 						    INODE_NR_FIELDS(inode.v));
297 		} else {
298 			return bch2_inode_unpack_v1(inode, unpacked);
299 		}
300 		break;
301 	}
302 	case KEY_TYPE_inode_v2: {
303 		struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
304 
305 		unpacked->bi_inum	= inode.k->p.offset;
306 		unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
307 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
308 		unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
309 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
310 
311 		return bch2_inode_unpack_v2(unpacked, inode.v->fields,
312 					    bkey_val_end(inode),
313 					    INODEv2_NR_FIELDS(inode.v));
314 	}
315 	default:
316 		BUG();
317 	}
318 }
319 
320 int bch2_inode_unpack(struct bkey_s_c k,
321 		      struct bch_inode_unpacked *unpacked)
322 {
323 	unpacked->bi_snapshot = k.k->p.snapshot;
324 
325 	return likely(k.k->type == KEY_TYPE_inode_v3)
326 		? bch2_inode_unpack_v3(k, unpacked)
327 		: bch2_inode_unpack_slowpath(k, unpacked);
328 }
329 
330 int __bch2_inode_peek(struct btree_trans *trans,
331 		      struct btree_iter *iter,
332 		      struct bch_inode_unpacked *inode,
333 		      subvol_inum inum, unsigned flags,
334 		      bool warn)
335 {
336 	u32 snapshot;
337 	int ret = __bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot, warn);
338 	if (ret)
339 		return ret;
340 
341 	struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
342 					       SPOS(0, inum.inum, snapshot),
343 					       flags|BTREE_ITER_cached);
344 	ret = bkey_err(k);
345 	if (ret)
346 		return ret;
347 
348 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
349 	if (ret)
350 		goto err;
351 
352 	ret = bch2_inode_unpack(k, inode);
353 	if (ret)
354 		goto err;
355 
356 	return 0;
357 err:
358 	if (warn)
359 		bch_err_msg(trans->c, ret, "looking up inum %llu:%llu:", inum.subvol, inum.inum);
360 	bch2_trans_iter_exit(trans, iter);
361 	return ret;
362 }
363 
364 int bch2_inode_write_flags(struct btree_trans *trans,
365 		     struct btree_iter *iter,
366 		     struct bch_inode_unpacked *inode,
367 		     enum btree_iter_update_trigger_flags flags)
368 {
369 	struct bkey_inode_buf *inode_p;
370 
371 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
372 	if (IS_ERR(inode_p))
373 		return PTR_ERR(inode_p);
374 
375 	bch2_inode_pack_inlined(inode_p, inode);
376 	inode_p->inode.k.p.snapshot = iter->snapshot;
377 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
378 }
379 
380 int __bch2_fsck_write_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
381 {
382 	struct bkey_inode_buf *inode_p =
383 		bch2_trans_kmalloc(trans, sizeof(*inode_p));
384 
385 	if (IS_ERR(inode_p))
386 		return PTR_ERR(inode_p);
387 
388 	bch2_inode_pack(inode_p, inode);
389 	inode_p->inode.k.p.snapshot = inode->bi_snapshot;
390 
391 	return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
392 				&inode_p->inode.k_i,
393 				BTREE_UPDATE_internal_snapshot_node);
394 }
395 
396 int bch2_fsck_write_inode(struct btree_trans *trans, struct bch_inode_unpacked *inode)
397 {
398 	int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
399 			    __bch2_fsck_write_inode(trans, inode));
400 	bch_err_fn(trans->c, ret);
401 	return ret;
402 }
403 
404 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
405 {
406 	struct bch_inode_unpacked u;
407 	struct bkey_inode_buf *inode_p;
408 	int ret;
409 
410 	if (!bkey_is_inode(&k->k))
411 		return ERR_PTR(-ENOENT);
412 
413 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
414 	if (IS_ERR(inode_p))
415 		return ERR_CAST(inode_p);
416 
417 	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
418 	if (ret)
419 		return ERR_PTR(ret);
420 
421 	bch2_inode_pack(inode_p, &u);
422 	return &inode_p->inode.k_i;
423 }
424 
425 static int __bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
426 				 enum bch_validate_flags flags)
427 {
428 	struct bch_inode_unpacked unpacked;
429 	int ret = 0;
430 
431 	bkey_fsck_err_on(k.k->p.inode,
432 			 c, inode_pos_inode_nonzero,
433 			 "nonzero k.p.inode");
434 
435 	bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX,
436 			 c, inode_pos_blockdev_range,
437 			 "fs inode in blockdev range");
438 
439 	bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked),
440 			 c, inode_unpack_error,
441 			 "invalid variable length fields");
442 
443 	bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1,
444 			 c, inode_checksum_type_invalid,
445 			 "invalid data checksum type (%u >= %u",
446 			 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
447 
448 	bkey_fsck_err_on(unpacked.bi_compression &&
449 			 !bch2_compression_opt_valid(unpacked.bi_compression - 1),
450 			 c, inode_compression_type_invalid,
451 			 "invalid compression opt %u", unpacked.bi_compression - 1);
452 
453 	bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
454 			 unpacked.bi_nlink != 0,
455 			 c, inode_unlinked_but_nlink_nonzero,
456 			 "flagged as unlinked but bi_nlink != 0");
457 
458 	bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode),
459 			 c, inode_subvol_root_but_not_dir,
460 			 "subvolume root but not a directory");
461 fsck_err:
462 	return ret;
463 }
464 
465 int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
466 			enum bch_validate_flags flags)
467 {
468 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
469 	int ret = 0;
470 
471 	bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
472 			 c, inode_str_hash_invalid,
473 			 "invalid str hash type (%llu >= %u)",
474 			 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
475 
476 	ret = __bch2_inode_validate(c, k, flags);
477 fsck_err:
478 	return ret;
479 }
480 
481 int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k,
482 			   enum bch_validate_flags flags)
483 {
484 	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
485 	int ret = 0;
486 
487 	bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
488 			 c, inode_str_hash_invalid,
489 			 "invalid str hash type (%llu >= %u)",
490 			 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
491 
492 	ret = __bch2_inode_validate(c, k, flags);
493 fsck_err:
494 	return ret;
495 }
496 
497 int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k,
498 			   enum bch_validate_flags flags)
499 {
500 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
501 	int ret = 0;
502 
503 	bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
504 			 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k),
505 			 c, inode_v3_fields_start_bad,
506 			 "invalid fields_start (got %llu, min %u max %zu)",
507 			 INODEv3_FIELDS_START(inode.v),
508 			 INODEv3_FIELDS_START_INITIAL,
509 			 bkey_val_u64s(inode.k));
510 
511 	bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
512 			 c, inode_str_hash_invalid,
513 			 "invalid str hash type (%llu >= %u)",
514 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
515 
516 	ret = __bch2_inode_validate(c, k, flags);
517 fsck_err:
518 	return ret;
519 }
520 
521 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
522 					  struct bch_inode_unpacked *inode)
523 {
524 	prt_printf(out, "\n");
525 	printbuf_indent_add(out, 2);
526 	prt_printf(out, "mode=%o\n", inode->bi_mode);
527 
528 	prt_str(out, "flags=");
529 	prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
530 	prt_printf(out, "(%x)\n", inode->bi_flags);
531 
532 	prt_printf(out, "journal_seq=%llu\n",	inode->bi_journal_seq);
533 	prt_printf(out, "bi_size=%llu\n",	inode->bi_size);
534 	prt_printf(out, "bi_sectors=%llu\n",	inode->bi_sectors);
535 	prt_printf(out, "bi_version=%llu\n",	inode->bi_version);
536 
537 #define x(_name, _bits)						\
538 	prt_printf(out, #_name "=%llu\n", (u64) inode->_name);
539 	BCH_INODE_FIELDS_v3()
540 #undef  x
541 
542 	bch2_printbuf_strip_trailing_newline(out);
543 	printbuf_indent_sub(out, 2);
544 }
545 
546 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
547 {
548 	prt_printf(out, "inum: %llu:%u ", inode->bi_inum, inode->bi_snapshot);
549 	__bch2_inode_unpacked_to_text(out, inode);
550 }
551 
552 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
553 {
554 	struct bch_inode_unpacked inode;
555 
556 	if (bch2_inode_unpack(k, &inode)) {
557 		prt_printf(out, "(unpack error)");
558 		return;
559 	}
560 
561 	__bch2_inode_unpacked_to_text(out, &inode);
562 }
563 
564 static inline u64 bkey_inode_flags(struct bkey_s_c k)
565 {
566 	switch (k.k->type) {
567 	case KEY_TYPE_inode:
568 		return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
569 	case KEY_TYPE_inode_v2:
570 		return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
571 	case KEY_TYPE_inode_v3:
572 		return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
573 	default:
574 		return 0;
575 	}
576 }
577 
578 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
579 {
580 	return bkey_inode_flags(k) & BCH_INODE_unlinked;
581 }
582 
583 int bch2_trigger_inode(struct btree_trans *trans,
584 		       enum btree_id btree_id, unsigned level,
585 		       struct bkey_s_c old,
586 		       struct bkey_s new,
587 		       enum btree_iter_update_trigger_flags flags)
588 {
589 	if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
590 		BUG_ON(!trans->journal_res.seq);
591 		bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
592 	}
593 
594 	s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
595 	if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr) {
596 		struct disk_accounting_pos acc = { .type = BCH_DISK_ACCOUNTING_nr_inodes };
597 		int ret = bch2_disk_accounting_mod(trans, &acc, &nr, 1, flags & BTREE_TRIGGER_gc);
598 		if (ret)
599 			return ret;
600 	}
601 
602 	int deleted_delta =	(int) bkey_is_deleted_inode(new.s_c) -
603 				(int) bkey_is_deleted_inode(old);
604 	if ((flags & BTREE_TRIGGER_transactional) && deleted_delta) {
605 		int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes,
606 						      new.k->p, deleted_delta > 0);
607 		if (ret)
608 			return ret;
609 	}
610 
611 	return 0;
612 }
613 
614 int bch2_inode_generation_validate(struct bch_fs *c, struct bkey_s_c k,
615 				   enum bch_validate_flags flags)
616 {
617 	int ret = 0;
618 
619 	bkey_fsck_err_on(k.k->p.inode,
620 			 c, inode_pos_inode_nonzero,
621 			 "nonzero k.p.inode");
622 fsck_err:
623 	return ret;
624 }
625 
626 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
627 				   struct bkey_s_c k)
628 {
629 	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
630 
631 	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
632 }
633 
634 void bch2_inode_init_early(struct bch_fs *c,
635 			   struct bch_inode_unpacked *inode_u)
636 {
637 	enum bch_str_hash_type str_hash =
638 		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
639 
640 	memset(inode_u, 0, sizeof(*inode_u));
641 
642 	/* ick */
643 	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
644 	get_random_bytes(&inode_u->bi_hash_seed,
645 			 sizeof(inode_u->bi_hash_seed));
646 }
647 
648 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
649 			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
650 			  struct bch_inode_unpacked *parent)
651 {
652 	inode_u->bi_mode	= mode;
653 	inode_u->bi_uid		= uid;
654 	inode_u->bi_gid		= gid;
655 	inode_u->bi_dev		= rdev;
656 	inode_u->bi_atime	= now;
657 	inode_u->bi_mtime	= now;
658 	inode_u->bi_ctime	= now;
659 	inode_u->bi_otime	= now;
660 
661 	if (parent && parent->bi_mode & S_ISGID) {
662 		inode_u->bi_gid = parent->bi_gid;
663 		if (S_ISDIR(mode))
664 			inode_u->bi_mode |= S_ISGID;
665 	}
666 
667 	if (parent) {
668 #define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
669 		BCH_INODE_OPTS()
670 #undef x
671 	}
672 }
673 
674 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
675 		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
676 		     struct bch_inode_unpacked *parent)
677 {
678 	bch2_inode_init_early(c, inode_u);
679 	bch2_inode_init_late(inode_u, bch2_current_time(c),
680 			     uid, gid, mode, rdev, parent);
681 }
682 
683 static inline u32 bkey_generation(struct bkey_s_c k)
684 {
685 	switch (k.k->type) {
686 	case KEY_TYPE_inode:
687 	case KEY_TYPE_inode_v2:
688 		BUG();
689 	case KEY_TYPE_inode_generation:
690 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
691 	default:
692 		return 0;
693 	}
694 }
695 
696 /*
697  * This just finds an empty slot:
698  */
699 int bch2_inode_create(struct btree_trans *trans,
700 		      struct btree_iter *iter,
701 		      struct bch_inode_unpacked *inode_u,
702 		      u32 snapshot, u64 cpu)
703 {
704 	struct bch_fs *c = trans->c;
705 	struct bkey_s_c k;
706 	u64 min, max, start, pos, *hint;
707 	int ret = 0;
708 	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
709 
710 	if (c->opts.shard_inode_numbers) {
711 		bits -= c->inode_shard_bits;
712 
713 		min = (cpu << bits);
714 		max = (cpu << bits) | ~(ULLONG_MAX << bits);
715 
716 		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
717 		hint = c->unused_inode_hints + cpu;
718 	} else {
719 		min = BLOCKDEV_INODE_MAX;
720 		max = ~(ULLONG_MAX << bits);
721 		hint = c->unused_inode_hints;
722 	}
723 
724 	start = READ_ONCE(*hint);
725 
726 	if (start >= max || start < min)
727 		start = min;
728 
729 	pos = start;
730 	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
731 			     BTREE_ITER_all_snapshots|
732 			     BTREE_ITER_intent);
733 again:
734 	while ((k = bch2_btree_iter_peek(iter)).k &&
735 	       !(ret = bkey_err(k)) &&
736 	       bkey_lt(k.k->p, POS(0, max))) {
737 		if (pos < iter->pos.offset)
738 			goto found_slot;
739 
740 		/*
741 		 * We don't need to iterate over keys in every snapshot once
742 		 * we've found just one:
743 		 */
744 		pos = iter->pos.offset + 1;
745 		bch2_btree_iter_set_pos(iter, POS(0, pos));
746 	}
747 
748 	if (!ret && pos < max)
749 		goto found_slot;
750 
751 	if (!ret && start == min)
752 		ret = -BCH_ERR_ENOSPC_inode_create;
753 
754 	if (ret) {
755 		bch2_trans_iter_exit(trans, iter);
756 		return ret;
757 	}
758 
759 	/* Retry from start */
760 	pos = start = min;
761 	bch2_btree_iter_set_pos(iter, POS(0, pos));
762 	goto again;
763 found_slot:
764 	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
765 	k = bch2_btree_iter_peek_slot(iter);
766 	ret = bkey_err(k);
767 	if (ret) {
768 		bch2_trans_iter_exit(trans, iter);
769 		return ret;
770 	}
771 
772 	*hint			= k.k->p.offset;
773 	inode_u->bi_inum	= k.k->p.offset;
774 	inode_u->bi_generation	= bkey_generation(k);
775 	return 0;
776 }
777 
778 static int bch2_inode_delete_keys(struct btree_trans *trans,
779 				  subvol_inum inum, enum btree_id id)
780 {
781 	struct btree_iter iter;
782 	struct bkey_s_c k;
783 	struct bkey_i delete;
784 	struct bpos end = POS(inum.inum, U64_MAX);
785 	u32 snapshot;
786 	int ret = 0;
787 
788 	/*
789 	 * We're never going to be deleting partial extents, no need to use an
790 	 * extent iterator:
791 	 */
792 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
793 			     BTREE_ITER_intent);
794 
795 	while (1) {
796 		bch2_trans_begin(trans);
797 
798 		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
799 		if (ret)
800 			goto err;
801 
802 		bch2_btree_iter_set_snapshot(&iter, snapshot);
803 
804 		k = bch2_btree_iter_peek_upto(&iter, end);
805 		ret = bkey_err(k);
806 		if (ret)
807 			goto err;
808 
809 		if (!k.k)
810 			break;
811 
812 		bkey_init(&delete.k);
813 		delete.k.p = iter.pos;
814 
815 		if (iter.flags & BTREE_ITER_is_extents)
816 			bch2_key_resize(&delete.k,
817 					bpos_min(end, k.k->p).offset -
818 					iter.pos.offset);
819 
820 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
821 		      bch2_trans_commit(trans, NULL, NULL,
822 					BCH_TRANS_COMMIT_no_enospc);
823 err:
824 		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
825 			break;
826 	}
827 
828 	bch2_trans_iter_exit(trans, &iter);
829 	return ret;
830 }
831 
832 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
833 {
834 	struct btree_trans *trans = bch2_trans_get(c);
835 	struct btree_iter iter = { NULL };
836 	struct bkey_i_inode_generation delete;
837 	struct bch_inode_unpacked inode_u;
838 	struct bkey_s_c k;
839 	u32 snapshot;
840 	int ret;
841 
842 	/*
843 	 * If this was a directory, there shouldn't be any real dirents left -
844 	 * but there could be whiteouts (from hash collisions) that we should
845 	 * delete:
846 	 *
847 	 * XXX: the dirent could ideally would delete whiteouts when they're no
848 	 * longer needed
849 	 */
850 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
851 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
852 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
853 	if (ret)
854 		goto err;
855 retry:
856 	bch2_trans_begin(trans);
857 
858 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
859 	if (ret)
860 		goto err;
861 
862 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
863 			       SPOS(0, inum.inum, snapshot),
864 			       BTREE_ITER_intent|BTREE_ITER_cached);
865 	ret = bkey_err(k);
866 	if (ret)
867 		goto err;
868 
869 	if (!bkey_is_inode(k.k)) {
870 		bch2_fs_inconsistent(c,
871 				     "inode %llu:%u not found when deleting",
872 				     inum.inum, snapshot);
873 		ret = -EIO;
874 		goto err;
875 	}
876 
877 	bch2_inode_unpack(k, &inode_u);
878 
879 	bkey_inode_generation_init(&delete.k_i);
880 	delete.k.p = iter.pos;
881 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
882 
883 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
884 		bch2_trans_commit(trans, NULL, NULL,
885 				BCH_TRANS_COMMIT_no_enospc);
886 err:
887 	bch2_trans_iter_exit(trans, &iter);
888 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
889 		goto retry;
890 
891 	bch2_trans_put(trans);
892 	return ret;
893 }
894 
895 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
896 				  subvol_inum inum,
897 				  struct bch_inode_unpacked *inode)
898 {
899 	struct btree_iter iter;
900 	int ret;
901 
902 	ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
903 	if (!ret)
904 		bch2_trans_iter_exit(trans, &iter);
905 	return ret;
906 }
907 
908 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
909 				  subvol_inum inum,
910 				  struct bch_inode_unpacked *inode)
911 {
912 	struct btree_iter iter;
913 	int ret;
914 
915 	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
916 	if (!ret)
917 		bch2_trans_iter_exit(trans, &iter);
918 	return ret;
919 }
920 
921 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
922 			    struct bch_inode_unpacked *inode)
923 {
924 	return bch2_trans_do(c, NULL, NULL, 0,
925 		bch2_inode_find_by_inum_trans(trans, inum, inode));
926 }
927 
928 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
929 {
930 	if (bi->bi_flags & BCH_INODE_unlinked)
931 		bi->bi_flags &= ~BCH_INODE_unlinked;
932 	else {
933 		if (bi->bi_nlink == U32_MAX)
934 			return -EINVAL;
935 
936 		bi->bi_nlink++;
937 	}
938 
939 	return 0;
940 }
941 
942 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
943 {
944 	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
945 		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
946 					bi->bi_inum);
947 		return;
948 	}
949 
950 	if (bi->bi_flags & BCH_INODE_unlinked) {
951 		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
952 		return;
953 	}
954 
955 	if (bi->bi_nlink)
956 		bi->bi_nlink--;
957 	else
958 		bi->bi_flags |= BCH_INODE_unlinked;
959 }
960 
961 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
962 {
963 	struct bch_opts ret = { 0 };
964 #define x(_name, _bits)							\
965 	if (inode->bi_##_name)						\
966 		opt_set(ret, _name, inode->bi_##_name - 1);
967 	BCH_INODE_OPTS()
968 #undef x
969 	return ret;
970 }
971 
972 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
973 			 struct bch_inode_unpacked *inode)
974 {
975 #define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
976 	BCH_INODE_OPTS()
977 #undef x
978 
979 	if (opts->nocow)
980 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
981 }
982 
983 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
984 {
985 	struct bch_inode_unpacked inode;
986 	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
987 
988 	if (ret)
989 		return ret;
990 
991 	bch2_inode_opts_get(opts, trans->c, &inode);
992 	return 0;
993 }
994 
995 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
996 {
997 	struct bch_fs *c = trans->c;
998 	struct btree_iter iter = { NULL };
999 	struct bkey_i_inode_generation delete;
1000 	struct bch_inode_unpacked inode_u;
1001 	struct bkey_s_c k;
1002 	int ret;
1003 
1004 	do {
1005 		ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1006 						      SPOS(inum, 0, snapshot),
1007 						      SPOS(inum, U64_MAX, snapshot),
1008 						      0, NULL) ?:
1009 			bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1010 						      SPOS(inum, 0, snapshot),
1011 						      SPOS(inum, U64_MAX, snapshot),
1012 						      0, NULL) ?:
1013 			bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1014 						      SPOS(inum, 0, snapshot),
1015 						      SPOS(inum, U64_MAX, snapshot),
1016 						      0, NULL);
1017 	} while (ret == -BCH_ERR_transaction_restart_nested);
1018 	if (ret)
1019 		goto err;
1020 retry:
1021 	bch2_trans_begin(trans);
1022 
1023 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1024 			       SPOS(0, inum, snapshot), BTREE_ITER_intent);
1025 	ret = bkey_err(k);
1026 	if (ret)
1027 		goto err;
1028 
1029 	if (!bkey_is_inode(k.k)) {
1030 		bch2_fs_inconsistent(c,
1031 				     "inode %llu:%u not found when deleting",
1032 				     inum, snapshot);
1033 		ret = -EIO;
1034 		goto err;
1035 	}
1036 
1037 	bch2_inode_unpack(k, &inode_u);
1038 
1039 	/* Subvolume root? */
1040 	if (inode_u.bi_subvol)
1041 		bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1042 
1043 	bkey_inode_generation_init(&delete.k_i);
1044 	delete.k.p = iter.pos;
1045 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1046 
1047 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1048 		bch2_trans_commit(trans, NULL, NULL,
1049 				BCH_TRANS_COMMIT_no_enospc);
1050 err:
1051 	bch2_trans_iter_exit(trans, &iter);
1052 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1053 		goto retry;
1054 
1055 	return ret ?: -BCH_ERR_transaction_restart_nested;
1056 }
1057 
1058 static int may_delete_deleted_inode(struct btree_trans *trans,
1059 				    struct btree_iter *iter,
1060 				    struct bpos pos,
1061 				    bool *need_another_pass)
1062 {
1063 	struct bch_fs *c = trans->c;
1064 	struct btree_iter inode_iter;
1065 	struct bkey_s_c k;
1066 	struct bch_inode_unpacked inode;
1067 	int ret;
1068 
1069 	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_cached);
1070 	ret = bkey_err(k);
1071 	if (ret)
1072 		return ret;
1073 
1074 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1075 	if (fsck_err_on(!bkey_is_inode(k.k),
1076 			trans, deleted_inode_missing,
1077 			"nonexistent inode %llu:%u in deleted_inodes btree",
1078 			pos.offset, pos.snapshot))
1079 		goto delete;
1080 
1081 	ret = bch2_inode_unpack(k, &inode);
1082 	if (ret)
1083 		goto out;
1084 
1085 	if (S_ISDIR(inode.bi_mode)) {
1086 		ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot);
1087 		if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY),
1088 				trans, deleted_inode_is_dir,
1089 				"non empty directory %llu:%u in deleted_inodes btree",
1090 				pos.offset, pos.snapshot))
1091 			goto delete;
1092 		if (ret)
1093 			goto out;
1094 	}
1095 
1096 	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked),
1097 			trans, deleted_inode_not_unlinked,
1098 			"non-deleted inode %llu:%u in deleted_inodes btree",
1099 			pos.offset, pos.snapshot))
1100 		goto delete;
1101 
1102 	if (test_bit(BCH_FS_clean_recovery, &c->flags) &&
1103 	    !fsck_err(trans, deleted_inode_but_clean,
1104 		      "filesystem marked as clean but have deleted inode %llu:%u",
1105 		      pos.offset, pos.snapshot)) {
1106 		ret = 0;
1107 		goto out;
1108 	}
1109 
1110 	if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1111 		struct bpos new_min_pos;
1112 
1113 		ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1114 		if (ret)
1115 			goto out;
1116 
1117 		inode.bi_flags &= ~BCH_INODE_unlinked;
1118 
1119 		ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1120 					     BTREE_UPDATE_internal_snapshot_node);
1121 		bch_err_msg(c, ret, "clearing inode unlinked flag");
1122 		if (ret)
1123 			goto out;
1124 
1125 		/*
1126 		 * We'll need another write buffer flush to pick up the new
1127 		 * unlinked inodes in the snapshot leaves:
1128 		 */
1129 		*need_another_pass = true;
1130 		goto out;
1131 	}
1132 
1133 	ret = 1;
1134 out:
1135 fsck_err:
1136 	bch2_trans_iter_exit(trans, &inode_iter);
1137 	return ret;
1138 delete:
1139 	ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false);
1140 	goto out;
1141 }
1142 
1143 int bch2_delete_dead_inodes(struct bch_fs *c)
1144 {
1145 	struct btree_trans *trans = bch2_trans_get(c);
1146 	bool need_another_pass;
1147 	int ret;
1148 again:
1149 	/*
1150 	 * if we ran check_inodes() unlinked inodes will have already been
1151 	 * cleaned up but the write buffer will be out of sync; therefore we
1152 	 * alway need a write buffer flush
1153 	 */
1154 	ret = bch2_btree_write_buffer_flush_sync(trans);
1155 	if (ret)
1156 		goto err;
1157 
1158 	need_another_pass = false;
1159 
1160 	/*
1161 	 * Weird transaction restart handling here because on successful delete,
1162 	 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1163 	 * but we can't retry because the btree write buffer won't have been
1164 	 * flushed and we'd spin:
1165 	 */
1166 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1167 					BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1168 					NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1169 		ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1170 		if (ret > 0) {
1171 			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1172 
1173 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1174 			/*
1175 			 * We don't want to loop here: a transaction restart
1176 			 * error here means we handled a transaction restart and
1177 			 * we're actually done, but if we loop we'll retry the
1178 			 * same key because the write buffer hasn't been flushed
1179 			 * yet
1180 			 */
1181 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1182 				ret = 0;
1183 				continue;
1184 			}
1185 		}
1186 
1187 		ret;
1188 	}));
1189 
1190 	if (!ret && need_another_pass)
1191 		goto again;
1192 err:
1193 	bch2_trans_put(trans);
1194 	return ret;
1195 }
1196