xref: /linux/fs/bcachefs/inode.c (revision ebc733e54a1a79ea2dde2ba5121ae73a188e20d4)
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 "error.h"
11 #include "extents.h"
12 #include "extent_update.h"
13 #include "inode.h"
14 #include "str_hash.h"
15 #include "snapshot.h"
16 #include "subvolume.h"
17 #include "varint.h"
18 
19 #include <linux/random.h>
20 
21 #include <asm/unaligned.h>
22 
23 #define x(name, ...)	#name,
24 const char * const bch2_inode_opts[] = {
25 	BCH_INODE_OPTS()
26 	NULL,
27 };
28 
29 static const char * const bch2_inode_flag_strs[] = {
30 	BCH_INODE_FLAGS()
31 	NULL
32 };
33 #undef  x
34 
35 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
36 
37 static int inode_decode_field(const u8 *in, const u8 *end,
38 			      u64 out[2], unsigned *out_bits)
39 {
40 	__be64 be[2] = { 0, 0 };
41 	unsigned bytes, shift;
42 	u8 *p;
43 
44 	if (in >= end)
45 		return -1;
46 
47 	if (!*in)
48 		return -1;
49 
50 	/*
51 	 * position of highest set bit indicates number of bytes:
52 	 * shift = number of bits to remove in high byte:
53 	 */
54 	shift	= 8 - __fls(*in); /* 1 <= shift <= 8 */
55 	bytes	= byte_table[shift - 1];
56 
57 	if (in + bytes > end)
58 		return -1;
59 
60 	p = (u8 *) be + 16 - bytes;
61 	memcpy(p, in, bytes);
62 	*p ^= (1 << 8) >> shift;
63 
64 	out[0] = be64_to_cpu(be[0]);
65 	out[1] = be64_to_cpu(be[1]);
66 	*out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
67 
68 	return bytes;
69 }
70 
71 static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
72 					   const struct bch_inode_unpacked *inode)
73 {
74 	struct bkey_i_inode_v3 *k = &packed->inode;
75 	u8 *out = k->v.fields;
76 	u8 *end = (void *) &packed[1];
77 	u8 *last_nonzero_field = out;
78 	unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
79 	unsigned bytes;
80 	int ret;
81 
82 	bkey_inode_v3_init(&packed->inode.k_i);
83 	packed->inode.k.p.offset	= inode->bi_inum;
84 	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
85 	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
86 	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
87 	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
88 	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
89 	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
90 	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
91 	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
92 
93 
94 #define x(_name, _bits)							\
95 	nr_fields++;							\
96 									\
97 	if (inode->_name) {						\
98 		ret = bch2_varint_encode_fast(out, inode->_name);	\
99 		out += ret;						\
100 									\
101 		if (_bits > 64)						\
102 			*out++ = 0;					\
103 									\
104 		last_nonzero_field = out;				\
105 		last_nonzero_fieldnr = nr_fields;			\
106 	} else {							\
107 		*out++ = 0;						\
108 									\
109 		if (_bits > 64)						\
110 			*out++ = 0;					\
111 	}
112 
113 	BCH_INODE_FIELDS_v3()
114 #undef  x
115 	BUG_ON(out > end);
116 
117 	out = last_nonzero_field;
118 	nr_fields = last_nonzero_fieldnr;
119 
120 	bytes = out - (u8 *) &packed->inode.v;
121 	set_bkey_val_bytes(&packed->inode.k, bytes);
122 	memset_u64s_tail(&packed->inode.v, 0, bytes);
123 
124 	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
125 
126 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
127 		struct bch_inode_unpacked unpacked;
128 
129 		ret = bch2_inode_unpack(bkey_i_to_s_c(&packed->inode.k_i), &unpacked);
130 		BUG_ON(ret);
131 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
132 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
133 		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
134 		BUG_ON(unpacked.bi_size		!= inode->bi_size);
135 		BUG_ON(unpacked.bi_version	!= inode->bi_version);
136 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
137 
138 #define x(_name, _bits)	if (unpacked._name != inode->_name)		\
139 			panic("unpacked %llu should be %llu",		\
140 			      (u64) unpacked._name, (u64) inode->_name);
141 		BCH_INODE_FIELDS_v3()
142 #undef  x
143 	}
144 }
145 
146 void bch2_inode_pack(struct bkey_inode_buf *packed,
147 		     const struct bch_inode_unpacked *inode)
148 {
149 	bch2_inode_pack_inlined(packed, inode);
150 }
151 
152 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
153 				struct bch_inode_unpacked *unpacked)
154 {
155 	const u8 *in = inode.v->fields;
156 	const u8 *end = bkey_val_end(inode);
157 	u64 field[2];
158 	unsigned fieldnr = 0, field_bits;
159 	int ret;
160 
161 #define x(_name, _bits)					\
162 	if (fieldnr++ == INODE_NR_FIELDS(inode.v)) {			\
163 		unsigned offset = offsetof(struct bch_inode_unpacked, _name);\
164 		memset((void *) unpacked + offset, 0,			\
165 		       sizeof(*unpacked) - offset);			\
166 		return 0;						\
167 	}								\
168 									\
169 	ret = inode_decode_field(in, end, field, &field_bits);		\
170 	if (ret < 0)							\
171 		return ret;						\
172 									\
173 	if (field_bits > sizeof(unpacked->_name) * 8)			\
174 		return -1;						\
175 									\
176 	unpacked->_name = field[1];					\
177 	in += ret;
178 
179 	BCH_INODE_FIELDS_v2()
180 #undef  x
181 
182 	/* XXX: signal if there were more fields than expected? */
183 	return 0;
184 }
185 
186 static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
187 				const u8 *in, const u8 *end,
188 				unsigned nr_fields)
189 {
190 	unsigned fieldnr = 0;
191 	int ret;
192 	u64 v[2];
193 
194 #define x(_name, _bits)							\
195 	if (fieldnr < nr_fields) {					\
196 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
197 		if (ret < 0)						\
198 			return ret;					\
199 		in += ret;						\
200 									\
201 		if (_bits > 64) {					\
202 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
203 			if (ret < 0)					\
204 				return ret;				\
205 			in += ret;					\
206 		} else {						\
207 			v[1] = 0;					\
208 		}							\
209 	} else {							\
210 		v[0] = v[1] = 0;					\
211 	}								\
212 									\
213 	unpacked->_name = v[0];						\
214 	if (v[1] || v[0] != unpacked->_name)				\
215 		return -1;						\
216 	fieldnr++;
217 
218 	BCH_INODE_FIELDS_v2()
219 #undef  x
220 
221 	/* XXX: signal if there were more fields than expected? */
222 	return 0;
223 }
224 
225 static int bch2_inode_unpack_v3(struct bkey_s_c k,
226 				struct bch_inode_unpacked *unpacked)
227 {
228 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
229 	const u8 *in = inode.v->fields;
230 	const u8 *end = bkey_val_end(inode);
231 	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
232 	unsigned fieldnr = 0;
233 	int ret;
234 	u64 v[2];
235 
236 	unpacked->bi_inum	= inode.k->p.offset;
237 	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
238 	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
239 	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
240 	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
241 	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
242 	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
243 	unpacked->bi_mode	= INODEv3_MODE(inode.v);
244 
245 #define x(_name, _bits)							\
246 	if (fieldnr < nr_fields) {					\
247 		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
248 		if (ret < 0)						\
249 			return ret;					\
250 		in += ret;						\
251 									\
252 		if (_bits > 64) {					\
253 			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
254 			if (ret < 0)					\
255 				return ret;				\
256 			in += ret;					\
257 		} else {						\
258 			v[1] = 0;					\
259 		}							\
260 	} else {							\
261 		v[0] = v[1] = 0;					\
262 	}								\
263 									\
264 	unpacked->_name = v[0];						\
265 	if (v[1] || v[0] != unpacked->_name)				\
266 		return -1;						\
267 	fieldnr++;
268 
269 	BCH_INODE_FIELDS_v3()
270 #undef  x
271 
272 	/* XXX: signal if there were more fields than expected? */
273 	return 0;
274 }
275 
276 static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
277 					       struct bch_inode_unpacked *unpacked)
278 {
279 	memset(unpacked, 0, sizeof(*unpacked));
280 
281 	switch (k.k->type) {
282 	case KEY_TYPE_inode: {
283 		struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
284 
285 		unpacked->bi_inum	= inode.k->p.offset;
286 		unpacked->bi_journal_seq= 0;
287 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
288 		unpacked->bi_flags	= le32_to_cpu(inode.v->bi_flags);
289 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
290 
291 		if (INODE_NEW_VARINT(inode.v)) {
292 			return bch2_inode_unpack_v2(unpacked, inode.v->fields,
293 						    bkey_val_end(inode),
294 						    INODE_NR_FIELDS(inode.v));
295 		} else {
296 			return bch2_inode_unpack_v1(inode, unpacked);
297 		}
298 		break;
299 	}
300 	case KEY_TYPE_inode_v2: {
301 		struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
302 
303 		unpacked->bi_inum	= inode.k->p.offset;
304 		unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
305 		unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
306 		unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
307 		unpacked->bi_mode	= le16_to_cpu(inode.v->bi_mode);
308 
309 		return bch2_inode_unpack_v2(unpacked, inode.v->fields,
310 					    bkey_val_end(inode),
311 					    INODEv2_NR_FIELDS(inode.v));
312 	}
313 	default:
314 		BUG();
315 	}
316 }
317 
318 int bch2_inode_unpack(struct bkey_s_c k,
319 		      struct bch_inode_unpacked *unpacked)
320 {
321 	if (likely(k.k->type == KEY_TYPE_inode_v3))
322 		return bch2_inode_unpack_v3(k, unpacked);
323 	return bch2_inode_unpack_slowpath(k, unpacked);
324 }
325 
326 static int bch2_inode_peek_nowarn(struct btree_trans *trans,
327 		    struct btree_iter *iter,
328 		    struct bch_inode_unpacked *inode,
329 		    subvol_inum inum, unsigned flags)
330 {
331 	struct bkey_s_c k;
332 	u32 snapshot;
333 	int ret;
334 
335 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
336 	if (ret)
337 		return ret;
338 
339 	k = bch2_bkey_get_iter(trans, iter, BTREE_ID_inodes,
340 			       SPOS(0, inum.inum, snapshot),
341 			       flags|BTREE_ITER_CACHED);
342 	ret = bkey_err(k);
343 	if (ret)
344 		return ret;
345 
346 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
347 	if (ret)
348 		goto err;
349 
350 	ret = bch2_inode_unpack(k, inode);
351 	if (ret)
352 		goto err;
353 
354 	return 0;
355 err:
356 	bch2_trans_iter_exit(trans, iter);
357 	return ret;
358 }
359 
360 int bch2_inode_peek(struct btree_trans *trans,
361 		    struct btree_iter *iter,
362 		    struct bch_inode_unpacked *inode,
363 		    subvol_inum inum, unsigned flags)
364 {
365 	int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
366 	bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
367 	return ret;
368 }
369 
370 int bch2_inode_write_flags(struct btree_trans *trans,
371 		     struct btree_iter *iter,
372 		     struct bch_inode_unpacked *inode,
373 		     enum btree_update_flags flags)
374 {
375 	struct bkey_inode_buf *inode_p;
376 
377 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
378 	if (IS_ERR(inode_p))
379 		return PTR_ERR(inode_p);
380 
381 	bch2_inode_pack_inlined(inode_p, inode);
382 	inode_p->inode.k.p.snapshot = iter->snapshot;
383 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
384 }
385 
386 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
387 {
388 	struct bch_inode_unpacked u;
389 	struct bkey_inode_buf *inode_p;
390 	int ret;
391 
392 	if (!bkey_is_inode(&k->k))
393 		return ERR_PTR(-ENOENT);
394 
395 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
396 	if (IS_ERR(inode_p))
397 		return ERR_CAST(inode_p);
398 
399 	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
400 	if (ret)
401 		return ERR_PTR(ret);
402 
403 	bch2_inode_pack(inode_p, &u);
404 	return &inode_p->inode.k_i;
405 }
406 
407 static int __bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k, struct printbuf *err)
408 {
409 	struct bch_inode_unpacked unpacked;
410 	int ret = 0;
411 
412 	bkey_fsck_err_on(k.k->p.inode, c, err,
413 			 inode_pos_inode_nonzero,
414 			 "nonzero k.p.inode");
415 
416 	bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX, c, err,
417 			 inode_pos_blockdev_range,
418 			 "fs inode in blockdev range");
419 
420 	bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked), c, err,
421 			 inode_unpack_error,
422 			 "invalid variable length fields");
423 
424 	bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1, c, err,
425 			 inode_checksum_type_invalid,
426 			 "invalid data checksum type (%u >= %u",
427 			 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
428 
429 	bkey_fsck_err_on(unpacked.bi_compression &&
430 			 !bch2_compression_opt_valid(unpacked.bi_compression - 1), c, err,
431 			 inode_compression_type_invalid,
432 			 "invalid compression opt %u", unpacked.bi_compression - 1);
433 
434 	bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
435 			 unpacked.bi_nlink != 0, c, err,
436 			 inode_unlinked_but_nlink_nonzero,
437 			 "flagged as unlinked but bi_nlink != 0");
438 
439 	bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode), c, err,
440 			 inode_subvol_root_but_not_dir,
441 			 "subvolume root but not a directory");
442 fsck_err:
443 	return ret;
444 }
445 
446 int bch2_inode_invalid(struct bch_fs *c, struct bkey_s_c k,
447 		       enum bkey_invalid_flags flags,
448 		       struct printbuf *err)
449 {
450 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
451 	int ret = 0;
452 
453 	bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
454 			 inode_str_hash_invalid,
455 			 "invalid str hash type (%llu >= %u)",
456 			 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
457 
458 	ret = __bch2_inode_invalid(c, k, err);
459 fsck_err:
460 	return ret;
461 }
462 
463 int bch2_inode_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
464 			  enum bkey_invalid_flags flags,
465 			  struct printbuf *err)
466 {
467 	struct bkey_s_c_inode_v2 inode = bkey_s_c_to_inode_v2(k);
468 	int ret = 0;
469 
470 	bkey_fsck_err_on(INODEv2_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
471 			 inode_str_hash_invalid,
472 			 "invalid str hash type (%llu >= %u)",
473 			 INODEv2_STR_HASH(inode.v), BCH_STR_HASH_NR);
474 
475 	ret = __bch2_inode_invalid(c, k, err);
476 fsck_err:
477 	return ret;
478 }
479 
480 int bch2_inode_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
481 			  enum bkey_invalid_flags flags,
482 			  struct printbuf *err)
483 {
484 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
485 	int ret = 0;
486 
487 	bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
488 			 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k), c, err,
489 			 inode_v3_fields_start_bad,
490 			 "invalid fields_start (got %llu, min %u max %zu)",
491 			 INODEv3_FIELDS_START(inode.v),
492 			 INODEv3_FIELDS_START_INITIAL,
493 			 bkey_val_u64s(inode.k));
494 
495 	bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR, c, err,
496 			 inode_str_hash_invalid,
497 			 "invalid str hash type (%llu >= %u)",
498 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
499 
500 	ret = __bch2_inode_invalid(c, k, err);
501 fsck_err:
502 	return ret;
503 }
504 
505 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
506 					  struct bch_inode_unpacked *inode)
507 {
508 	prt_printf(out, "mode=%o ", inode->bi_mode);
509 
510 	prt_str(out, "flags=");
511 	prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
512 	prt_printf(out, " (%x)", inode->bi_flags);
513 
514 	prt_printf(out, " journal_seq=%llu bi_size=%llu bi_sectors=%llu bi_version=%llu",
515 	       inode->bi_journal_seq,
516 	       inode->bi_size,
517 	       inode->bi_sectors,
518 	       inode->bi_version);
519 
520 #define x(_name, _bits)						\
521 	prt_printf(out, " "#_name "=%llu", (u64) inode->_name);
522 	BCH_INODE_FIELDS_v3()
523 #undef  x
524 }
525 
526 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
527 {
528 	prt_printf(out, "inum: %llu ", inode->bi_inum);
529 	__bch2_inode_unpacked_to_text(out, inode);
530 }
531 
532 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
533 {
534 	struct bch_inode_unpacked inode;
535 
536 	if (bch2_inode_unpack(k, &inode)) {
537 		prt_printf(out, "(unpack error)");
538 		return;
539 	}
540 
541 	__bch2_inode_unpacked_to_text(out, &inode);
542 }
543 
544 static inline u64 bkey_inode_flags(struct bkey_s_c k)
545 {
546 	switch (k.k->type) {
547 	case KEY_TYPE_inode:
548 		return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
549 	case KEY_TYPE_inode_v2:
550 		return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
551 	case KEY_TYPE_inode_v3:
552 		return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
553 	default:
554 		return 0;
555 	}
556 }
557 
558 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
559 {
560 	return bkey_inode_flags(k) & BCH_INODE_unlinked;
561 }
562 
563 int bch2_trans_mark_inode(struct btree_trans *trans,
564 			  enum btree_id btree_id, unsigned level,
565 			  struct bkey_s_c old,
566 			  struct bkey_i *new,
567 			  unsigned flags)
568 {
569 	int nr = bkey_is_inode(&new->k) - bkey_is_inode(old.k);
570 	bool old_deleted = bkey_is_deleted_inode(old);
571 	bool new_deleted = bkey_is_deleted_inode(bkey_i_to_s_c(new));
572 
573 	if (nr) {
574 		int ret = bch2_replicas_deltas_realloc(trans, 0);
575 		struct replicas_delta_list *d = trans->fs_usage_deltas;
576 
577 		if (ret)
578 			return ret;
579 
580 		d->nr_inodes += nr;
581 	}
582 
583 	if (old_deleted != new_deleted) {
584 		int ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, new->k.p, new_deleted);
585 		if (ret)
586 			return ret;
587 	}
588 
589 	return 0;
590 }
591 
592 int bch2_mark_inode(struct btree_trans *trans,
593 		    enum btree_id btree_id, unsigned level,
594 		    struct bkey_s_c old, struct bkey_s_c new,
595 		    unsigned flags)
596 {
597 	struct bch_fs *c = trans->c;
598 	struct bch_fs_usage *fs_usage;
599 	u64 journal_seq = trans->journal_res.seq;
600 
601 	if (flags & BTREE_TRIGGER_INSERT) {
602 		struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
603 
604 		BUG_ON(!journal_seq);
605 		BUG_ON(new.k->type != KEY_TYPE_inode_v3);
606 
607 		v->bi_journal_seq = cpu_to_le64(journal_seq);
608 	}
609 
610 	if (flags & BTREE_TRIGGER_GC) {
611 		percpu_down_read(&c->mark_lock);
612 		preempt_disable();
613 
614 		fs_usage = fs_usage_ptr(c, journal_seq, flags & BTREE_TRIGGER_GC);
615 		fs_usage->nr_inodes += bkey_is_inode(new.k);
616 		fs_usage->nr_inodes -= bkey_is_inode(old.k);
617 
618 		preempt_enable();
619 		percpu_up_read(&c->mark_lock);
620 	}
621 	return 0;
622 }
623 
624 int bch2_inode_generation_invalid(struct bch_fs *c, struct bkey_s_c k,
625 				  enum bkey_invalid_flags flags,
626 				  struct printbuf *err)
627 {
628 	int ret = 0;
629 
630 	bkey_fsck_err_on(k.k->p.inode, c, err,
631 			 inode_pos_inode_nonzero,
632 			 "nonzero k.p.inode");
633 fsck_err:
634 	return ret;
635 }
636 
637 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
638 				   struct bkey_s_c k)
639 {
640 	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
641 
642 	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
643 }
644 
645 void bch2_inode_init_early(struct bch_fs *c,
646 			   struct bch_inode_unpacked *inode_u)
647 {
648 	enum bch_str_hash_type str_hash =
649 		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
650 
651 	memset(inode_u, 0, sizeof(*inode_u));
652 
653 	/* ick */
654 	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
655 	get_random_bytes(&inode_u->bi_hash_seed,
656 			 sizeof(inode_u->bi_hash_seed));
657 }
658 
659 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
660 			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
661 			  struct bch_inode_unpacked *parent)
662 {
663 	inode_u->bi_mode	= mode;
664 	inode_u->bi_uid		= uid;
665 	inode_u->bi_gid		= gid;
666 	inode_u->bi_dev		= rdev;
667 	inode_u->bi_atime	= now;
668 	inode_u->bi_mtime	= now;
669 	inode_u->bi_ctime	= now;
670 	inode_u->bi_otime	= now;
671 
672 	if (parent && parent->bi_mode & S_ISGID) {
673 		inode_u->bi_gid = parent->bi_gid;
674 		if (S_ISDIR(mode))
675 			inode_u->bi_mode |= S_ISGID;
676 	}
677 
678 	if (parent) {
679 #define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
680 		BCH_INODE_OPTS()
681 #undef x
682 	}
683 }
684 
685 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
686 		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
687 		     struct bch_inode_unpacked *parent)
688 {
689 	bch2_inode_init_early(c, inode_u);
690 	bch2_inode_init_late(inode_u, bch2_current_time(c),
691 			     uid, gid, mode, rdev, parent);
692 }
693 
694 static inline u32 bkey_generation(struct bkey_s_c k)
695 {
696 	switch (k.k->type) {
697 	case KEY_TYPE_inode:
698 	case KEY_TYPE_inode_v2:
699 		BUG();
700 	case KEY_TYPE_inode_generation:
701 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
702 	default:
703 		return 0;
704 	}
705 }
706 
707 /*
708  * This just finds an empty slot:
709  */
710 int bch2_inode_create(struct btree_trans *trans,
711 		      struct btree_iter *iter,
712 		      struct bch_inode_unpacked *inode_u,
713 		      u32 snapshot, u64 cpu)
714 {
715 	struct bch_fs *c = trans->c;
716 	struct bkey_s_c k;
717 	u64 min, max, start, pos, *hint;
718 	int ret = 0;
719 	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
720 
721 	if (c->opts.shard_inode_numbers) {
722 		bits -= c->inode_shard_bits;
723 
724 		min = (cpu << bits);
725 		max = (cpu << bits) | ~(ULLONG_MAX << bits);
726 
727 		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
728 		hint = c->unused_inode_hints + cpu;
729 	} else {
730 		min = BLOCKDEV_INODE_MAX;
731 		max = ~(ULLONG_MAX << bits);
732 		hint = c->unused_inode_hints;
733 	}
734 
735 	start = READ_ONCE(*hint);
736 
737 	if (start >= max || start < min)
738 		start = min;
739 
740 	pos = start;
741 	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
742 			     BTREE_ITER_ALL_SNAPSHOTS|
743 			     BTREE_ITER_INTENT);
744 again:
745 	while ((k = bch2_btree_iter_peek(iter)).k &&
746 	       !(ret = bkey_err(k)) &&
747 	       bkey_lt(k.k->p, POS(0, max))) {
748 		if (pos < iter->pos.offset)
749 			goto found_slot;
750 
751 		/*
752 		 * We don't need to iterate over keys in every snapshot once
753 		 * we've found just one:
754 		 */
755 		pos = iter->pos.offset + 1;
756 		bch2_btree_iter_set_pos(iter, POS(0, pos));
757 	}
758 
759 	if (!ret && pos < max)
760 		goto found_slot;
761 
762 	if (!ret && start == min)
763 		ret = -BCH_ERR_ENOSPC_inode_create;
764 
765 	if (ret) {
766 		bch2_trans_iter_exit(trans, iter);
767 		return ret;
768 	}
769 
770 	/* Retry from start */
771 	pos = start = min;
772 	bch2_btree_iter_set_pos(iter, POS(0, pos));
773 	goto again;
774 found_slot:
775 	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
776 	k = bch2_btree_iter_peek_slot(iter);
777 	ret = bkey_err(k);
778 	if (ret) {
779 		bch2_trans_iter_exit(trans, iter);
780 		return ret;
781 	}
782 
783 	*hint			= k.k->p.offset;
784 	inode_u->bi_inum	= k.k->p.offset;
785 	inode_u->bi_generation	= bkey_generation(k);
786 	return 0;
787 }
788 
789 static int bch2_inode_delete_keys(struct btree_trans *trans,
790 				  subvol_inum inum, enum btree_id id)
791 {
792 	struct btree_iter iter;
793 	struct bkey_s_c k;
794 	struct bkey_i delete;
795 	struct bpos end = POS(inum.inum, U64_MAX);
796 	u32 snapshot;
797 	int ret = 0;
798 
799 	/*
800 	 * We're never going to be deleting partial extents, no need to use an
801 	 * extent iterator:
802 	 */
803 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
804 			     BTREE_ITER_INTENT);
805 
806 	while (1) {
807 		bch2_trans_begin(trans);
808 
809 		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
810 		if (ret)
811 			goto err;
812 
813 		bch2_btree_iter_set_snapshot(&iter, snapshot);
814 
815 		k = bch2_btree_iter_peek_upto(&iter, end);
816 		ret = bkey_err(k);
817 		if (ret)
818 			goto err;
819 
820 		if (!k.k)
821 			break;
822 
823 		bkey_init(&delete.k);
824 		delete.k.p = iter.pos;
825 
826 		if (iter.flags & BTREE_ITER_IS_EXTENTS)
827 			bch2_key_resize(&delete.k,
828 					bpos_min(end, k.k->p).offset -
829 					iter.pos.offset);
830 
831 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
832 		      bch2_trans_commit(trans, NULL, NULL,
833 					BTREE_INSERT_NOFAIL);
834 err:
835 		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
836 			break;
837 	}
838 
839 	bch2_trans_iter_exit(trans, &iter);
840 	return ret;
841 }
842 
843 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
844 {
845 	struct btree_trans *trans = bch2_trans_get(c);
846 	struct btree_iter iter = { NULL };
847 	struct bkey_i_inode_generation delete;
848 	struct bch_inode_unpacked inode_u;
849 	struct bkey_s_c k;
850 	u32 snapshot;
851 	int ret;
852 
853 	/*
854 	 * If this was a directory, there shouldn't be any real dirents left -
855 	 * but there could be whiteouts (from hash collisions) that we should
856 	 * delete:
857 	 *
858 	 * XXX: the dirent could ideally would delete whiteouts when they're no
859 	 * longer needed
860 	 */
861 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
862 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
863 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
864 	if (ret)
865 		goto err;
866 retry:
867 	bch2_trans_begin(trans);
868 
869 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
870 	if (ret)
871 		goto err;
872 
873 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
874 			       SPOS(0, inum.inum, snapshot),
875 			       BTREE_ITER_INTENT|BTREE_ITER_CACHED);
876 	ret = bkey_err(k);
877 	if (ret)
878 		goto err;
879 
880 	if (!bkey_is_inode(k.k)) {
881 		bch2_fs_inconsistent(c,
882 				     "inode %llu:%u not found when deleting",
883 				     inum.inum, snapshot);
884 		ret = -EIO;
885 		goto err;
886 	}
887 
888 	bch2_inode_unpack(k, &inode_u);
889 
890 	bkey_inode_generation_init(&delete.k_i);
891 	delete.k.p = iter.pos;
892 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
893 
894 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
895 		bch2_trans_commit(trans, NULL, NULL,
896 				BTREE_INSERT_NOFAIL);
897 err:
898 	bch2_trans_iter_exit(trans, &iter);
899 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
900 		goto retry;
901 
902 	bch2_trans_put(trans);
903 	return ret;
904 }
905 
906 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
907 				  subvol_inum inum,
908 				  struct bch_inode_unpacked *inode)
909 {
910 	struct btree_iter iter;
911 	int ret;
912 
913 	ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
914 	if (!ret)
915 		bch2_trans_iter_exit(trans, &iter);
916 	return ret;
917 }
918 
919 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
920 				  subvol_inum inum,
921 				  struct bch_inode_unpacked *inode)
922 {
923 	struct btree_iter iter;
924 	int ret;
925 
926 	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
927 	if (!ret)
928 		bch2_trans_iter_exit(trans, &iter);
929 	return ret;
930 }
931 
932 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
933 			    struct bch_inode_unpacked *inode)
934 {
935 	return bch2_trans_do(c, NULL, NULL, 0,
936 		bch2_inode_find_by_inum_trans(trans, inum, inode));
937 }
938 
939 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
940 {
941 	if (bi->bi_flags & BCH_INODE_unlinked)
942 		bi->bi_flags &= ~BCH_INODE_unlinked;
943 	else {
944 		if (bi->bi_nlink == U32_MAX)
945 			return -EINVAL;
946 
947 		bi->bi_nlink++;
948 	}
949 
950 	return 0;
951 }
952 
953 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
954 {
955 	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
956 		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
957 					bi->bi_inum);
958 		return;
959 	}
960 
961 	if (bi->bi_flags & BCH_INODE_unlinked) {
962 		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
963 		return;
964 	}
965 
966 	if (bi->bi_nlink)
967 		bi->bi_nlink--;
968 	else
969 		bi->bi_flags |= BCH_INODE_unlinked;
970 }
971 
972 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
973 {
974 	struct bch_opts ret = { 0 };
975 #define x(_name, _bits)							\
976 	if (inode->bi_##_name)						\
977 		opt_set(ret, _name, inode->bi_##_name - 1);
978 	BCH_INODE_OPTS()
979 #undef x
980 	return ret;
981 }
982 
983 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
984 			 struct bch_inode_unpacked *inode)
985 {
986 #define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
987 	BCH_INODE_OPTS()
988 #undef x
989 
990 	if (opts->nocow)
991 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
992 }
993 
994 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
995 {
996 	struct bch_inode_unpacked inode;
997 	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
998 
999 	if (ret)
1000 		return ret;
1001 
1002 	bch2_inode_opts_get(opts, trans->c, &inode);
1003 	return 0;
1004 }
1005 
1006 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1007 {
1008 	struct bch_fs *c = trans->c;
1009 	struct btree_iter iter = { NULL };
1010 	struct bkey_i_inode_generation delete;
1011 	struct bch_inode_unpacked inode_u;
1012 	struct bkey_s_c k;
1013 	int ret;
1014 
1015 	do {
1016 		ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1017 						      SPOS(inum, 0, snapshot),
1018 						      SPOS(inum, U64_MAX, snapshot),
1019 						      0, NULL) ?:
1020 			bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1021 						      SPOS(inum, 0, snapshot),
1022 						      SPOS(inum, U64_MAX, snapshot),
1023 						      0, NULL) ?:
1024 			bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1025 						      SPOS(inum, 0, snapshot),
1026 						      SPOS(inum, U64_MAX, snapshot),
1027 						      0, NULL);
1028 	} while (ret == -BCH_ERR_transaction_restart_nested);
1029 	if (ret)
1030 		goto err;
1031 retry:
1032 	bch2_trans_begin(trans);
1033 
1034 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1035 			       SPOS(0, inum, snapshot), BTREE_ITER_INTENT);
1036 	ret = bkey_err(k);
1037 	if (ret)
1038 		goto err;
1039 
1040 	if (!bkey_is_inode(k.k)) {
1041 		bch2_fs_inconsistent(c,
1042 				     "inode %llu:%u not found when deleting",
1043 				     inum, snapshot);
1044 		ret = -EIO;
1045 		goto err;
1046 	}
1047 
1048 	bch2_inode_unpack(k, &inode_u);
1049 
1050 	/* Subvolume root? */
1051 	if (inode_u.bi_subvol)
1052 		bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1053 
1054 	bkey_inode_generation_init(&delete.k_i);
1055 	delete.k.p = iter.pos;
1056 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1057 
1058 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1059 		bch2_trans_commit(trans, NULL, NULL,
1060 				BTREE_INSERT_NOFAIL);
1061 err:
1062 	bch2_trans_iter_exit(trans, &iter);
1063 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1064 		goto retry;
1065 
1066 	return ret ?: -BCH_ERR_transaction_restart_nested;
1067 }
1068 
1069 static int may_delete_deleted_inode(struct btree_trans *trans,
1070 				    struct btree_iter *iter,
1071 				    struct bpos pos,
1072 				    bool *need_another_pass)
1073 {
1074 	struct bch_fs *c = trans->c;
1075 	struct btree_iter inode_iter;
1076 	struct bkey_s_c k;
1077 	struct bch_inode_unpacked inode;
1078 	int ret;
1079 
1080 	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_CACHED);
1081 	ret = bkey_err(k);
1082 	if (ret)
1083 		return ret;
1084 
1085 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1086 	if (fsck_err_on(!bkey_is_inode(k.k), c,
1087 			deleted_inode_missing,
1088 			"nonexistent inode %llu:%u in deleted_inodes btree",
1089 			pos.offset, pos.snapshot))
1090 		goto delete;
1091 
1092 	ret = bch2_inode_unpack(k, &inode);
1093 	if (ret)
1094 		goto out;
1095 
1096 	if (fsck_err_on(S_ISDIR(inode.bi_mode), c,
1097 			deleted_inode_is_dir,
1098 			"directory %llu:%u in deleted_inodes btree",
1099 			pos.offset, pos.snapshot))
1100 		goto delete;
1101 
1102 	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked), c,
1103 			deleted_inode_not_unlinked,
1104 			"non-deleted inode %llu:%u in deleted_inodes btree",
1105 			pos.offset, pos.snapshot))
1106 		goto delete;
1107 
1108 	if (c->sb.clean &&
1109 	    !fsck_err(c,
1110 		      deleted_inode_but_clean,
1111 		      "filesystem marked as clean but have deleted inode %llu:%u",
1112 		      pos.offset, pos.snapshot)) {
1113 		ret = 0;
1114 		goto out;
1115 	}
1116 
1117 	if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1118 		struct bpos new_min_pos;
1119 
1120 		ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1121 		if (ret)
1122 			goto out;
1123 
1124 		inode.bi_flags &= ~BCH_INODE_unlinked;
1125 
1126 		ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1127 					     BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
1128 		bch_err_msg(c, ret, "clearing inode unlinked flag");
1129 		if (ret)
1130 			goto out;
1131 
1132 		/*
1133 		 * We'll need another write buffer flush to pick up the new
1134 		 * unlinked inodes in the snapshot leaves:
1135 		 */
1136 		*need_another_pass = true;
1137 		goto out;
1138 	}
1139 
1140 	ret = 1;
1141 out:
1142 fsck_err:
1143 	bch2_trans_iter_exit(trans, &inode_iter);
1144 	return ret;
1145 delete:
1146 	ret = bch2_btree_bit_mod(trans, BTREE_ID_deleted_inodes, pos, false);
1147 	goto out;
1148 }
1149 
1150 int bch2_delete_dead_inodes(struct bch_fs *c)
1151 {
1152 	struct btree_trans *trans = bch2_trans_get(c);
1153 	struct btree_iter iter;
1154 	struct bkey_s_c k;
1155 	bool need_another_pass;
1156 	int ret;
1157 again:
1158 	need_another_pass = false;
1159 
1160 	ret = bch2_btree_write_buffer_flush_sync(trans);
1161 	if (ret)
1162 		goto err;
1163 
1164 	/*
1165 	 * Weird transaction restart handling here because on successful delete,
1166 	 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1167 	 * but we can't retry because the btree write buffer won't have been
1168 	 * flushed and we'd spin:
1169 	 */
1170 	for_each_btree_key(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1171 			   BTREE_ITER_PREFETCH|BTREE_ITER_ALL_SNAPSHOTS, k, ret) {
1172 		ret = commit_do(trans, NULL, NULL,
1173 				BTREE_INSERT_NOFAIL|
1174 				BTREE_INSERT_LAZY_RW,
1175 			may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass));
1176 		if (ret < 0)
1177 			break;
1178 
1179 		if (ret) {
1180 			if (!test_bit(BCH_FS_RW, &c->flags)) {
1181 				bch2_trans_unlock(trans);
1182 				bch2_fs_lazy_rw(c);
1183 			}
1184 
1185 			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1186 
1187 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1188 			if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
1189 				break;
1190 		}
1191 	}
1192 	bch2_trans_iter_exit(trans, &iter);
1193 
1194 	if (!ret && need_another_pass)
1195 		goto again;
1196 err:
1197 	bch2_trans_put(trans);
1198 
1199 	return ret;
1200 }
1201