xref: /linux/fs/bcachefs/inode.c (revision d8b762070c3fde224f8b9ea3cf59bc41a5a3eb57)
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 <asm/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 	if (likely(k.k->type == KEY_TYPE_inode_v3))
324 		return bch2_inode_unpack_v3(k, unpacked);
325 	return bch2_inode_unpack_slowpath(k, unpacked);
326 }
327 
328 int bch2_inode_peek_nowarn(struct btree_trans *trans,
329 		    struct btree_iter *iter,
330 		    struct bch_inode_unpacked *inode,
331 		    subvol_inum inum, unsigned flags)
332 {
333 	struct bkey_s_c k;
334 	u32 snapshot;
335 	int ret;
336 
337 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
338 	if (ret)
339 		return ret;
340 
341 	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 	bch2_trans_iter_exit(trans, iter);
359 	return ret;
360 }
361 
362 int bch2_inode_peek(struct btree_trans *trans,
363 		    struct btree_iter *iter,
364 		    struct bch_inode_unpacked *inode,
365 		    subvol_inum inum, unsigned flags)
366 {
367 	int ret = bch2_inode_peek_nowarn(trans, iter, inode, inum, flags);
368 	bch_err_msg(trans->c, ret, "looking up inum %u:%llu:", inum.subvol, inum.inum);
369 	return ret;
370 }
371 
372 int bch2_inode_write_flags(struct btree_trans *trans,
373 		     struct btree_iter *iter,
374 		     struct bch_inode_unpacked *inode,
375 		     enum btree_iter_update_trigger_flags flags)
376 {
377 	struct bkey_inode_buf *inode_p;
378 
379 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
380 	if (IS_ERR(inode_p))
381 		return PTR_ERR(inode_p);
382 
383 	bch2_inode_pack_inlined(inode_p, inode);
384 	inode_p->inode.k.p.snapshot = iter->snapshot;
385 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, flags);
386 }
387 
388 int __bch2_fsck_write_inode(struct btree_trans *trans,
389 			 struct bch_inode_unpacked *inode,
390 			 u32 snapshot)
391 {
392 	struct bkey_inode_buf *inode_p =
393 		bch2_trans_kmalloc(trans, sizeof(*inode_p));
394 
395 	if (IS_ERR(inode_p))
396 		return PTR_ERR(inode_p);
397 
398 	bch2_inode_pack(inode_p, inode);
399 	inode_p->inode.k.p.snapshot = snapshot;
400 
401 	return bch2_btree_insert_nonextent(trans, BTREE_ID_inodes,
402 				&inode_p->inode.k_i,
403 				BTREE_UPDATE_internal_snapshot_node);
404 }
405 
406 int bch2_fsck_write_inode(struct btree_trans *trans,
407 			    struct bch_inode_unpacked *inode,
408 			    u32 snapshot)
409 {
410 	int ret = commit_do(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
411 			    __bch2_fsck_write_inode(trans, inode, snapshot));
412 	bch_err_fn(trans->c, ret);
413 	return ret;
414 }
415 
416 struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
417 {
418 	struct bch_inode_unpacked u;
419 	struct bkey_inode_buf *inode_p;
420 	int ret;
421 
422 	if (!bkey_is_inode(&k->k))
423 		return ERR_PTR(-ENOENT);
424 
425 	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
426 	if (IS_ERR(inode_p))
427 		return ERR_CAST(inode_p);
428 
429 	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
430 	if (ret)
431 		return ERR_PTR(ret);
432 
433 	bch2_inode_pack(inode_p, &u);
434 	return &inode_p->inode.k_i;
435 }
436 
437 static int __bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
438 				 enum bch_validate_flags flags)
439 {
440 	struct bch_inode_unpacked unpacked;
441 	int ret = 0;
442 
443 	bkey_fsck_err_on(k.k->p.inode,
444 			 c, inode_pos_inode_nonzero,
445 			 "nonzero k.p.inode");
446 
447 	bkey_fsck_err_on(k.k->p.offset < BLOCKDEV_INODE_MAX,
448 			 c, inode_pos_blockdev_range,
449 			 "fs inode in blockdev range");
450 
451 	bkey_fsck_err_on(bch2_inode_unpack(k, &unpacked),
452 			 c, inode_unpack_error,
453 			 "invalid variable length fields");
454 
455 	bkey_fsck_err_on(unpacked.bi_data_checksum >= BCH_CSUM_OPT_NR + 1,
456 			 c, inode_checksum_type_invalid,
457 			 "invalid data checksum type (%u >= %u",
458 			 unpacked.bi_data_checksum, BCH_CSUM_OPT_NR + 1);
459 
460 	bkey_fsck_err_on(unpacked.bi_compression &&
461 			 !bch2_compression_opt_valid(unpacked.bi_compression - 1),
462 			 c, inode_compression_type_invalid,
463 			 "invalid compression opt %u", unpacked.bi_compression - 1);
464 
465 	bkey_fsck_err_on((unpacked.bi_flags & BCH_INODE_unlinked) &&
466 			 unpacked.bi_nlink != 0,
467 			 c, inode_unlinked_but_nlink_nonzero,
468 			 "flagged as unlinked but bi_nlink != 0");
469 
470 	bkey_fsck_err_on(unpacked.bi_subvol && !S_ISDIR(unpacked.bi_mode),
471 			 c, inode_subvol_root_but_not_dir,
472 			 "subvolume root but not a directory");
473 fsck_err:
474 	return ret;
475 }
476 
477 int bch2_inode_validate(struct bch_fs *c, struct bkey_s_c k,
478 			enum bch_validate_flags flags)
479 {
480 	struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
481 	int ret = 0;
482 
483 	bkey_fsck_err_on(INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
484 			 c, inode_str_hash_invalid,
485 			 "invalid str hash type (%llu >= %u)",
486 			 INODE_STR_HASH(inode.v), BCH_STR_HASH_NR);
487 
488 	ret = __bch2_inode_validate(c, k, flags);
489 fsck_err:
490 	return ret;
491 }
492 
493 int bch2_inode_v2_validate(struct bch_fs *c, struct bkey_s_c k,
494 			   enum bch_validate_flags flags)
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,
500 			 c, 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_validate(c, k, flags);
505 fsck_err:
506 	return ret;
507 }
508 
509 int bch2_inode_v3_validate(struct bch_fs *c, struct bkey_s_c k,
510 			   enum bch_validate_flags flags)
511 {
512 	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
513 	int ret = 0;
514 
515 	bkey_fsck_err_on(INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
516 			 INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k),
517 			 c, inode_v3_fields_start_bad,
518 			 "invalid fields_start (got %llu, min %u max %zu)",
519 			 INODEv3_FIELDS_START(inode.v),
520 			 INODEv3_FIELDS_START_INITIAL,
521 			 bkey_val_u64s(inode.k));
522 
523 	bkey_fsck_err_on(INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR,
524 			 c, inode_str_hash_invalid,
525 			 "invalid str hash type (%llu >= %u)",
526 			 INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
527 
528 	ret = __bch2_inode_validate(c, k, flags);
529 fsck_err:
530 	return ret;
531 }
532 
533 static void __bch2_inode_unpacked_to_text(struct printbuf *out,
534 					  struct bch_inode_unpacked *inode)
535 {
536 	prt_printf(out, "\n");
537 	printbuf_indent_add(out, 2);
538 	prt_printf(out, "mode=%o\n", inode->bi_mode);
539 
540 	prt_str(out, "flags=");
541 	prt_bitflags(out, bch2_inode_flag_strs, inode->bi_flags & ((1U << 20) - 1));
542 	prt_printf(out, "(%x)\n", inode->bi_flags);
543 
544 	prt_printf(out, "journal_seq=%llu\n",	inode->bi_journal_seq);
545 	prt_printf(out, "bi_size=%llu\n",	inode->bi_size);
546 	prt_printf(out, "bi_sectors=%llu\n",	inode->bi_sectors);
547 	prt_printf(out, "bi_version=%llu\n",	inode->bi_version);
548 
549 #define x(_name, _bits)						\
550 	prt_printf(out, #_name "=%llu\n", (u64) inode->_name);
551 	BCH_INODE_FIELDS_v3()
552 #undef  x
553 
554 	bch2_printbuf_strip_trailing_newline(out);
555 	printbuf_indent_sub(out, 2);
556 }
557 
558 void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
559 {
560 	prt_printf(out, "inum: %llu ", inode->bi_inum);
561 	__bch2_inode_unpacked_to_text(out, inode);
562 }
563 
564 void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
565 {
566 	struct bch_inode_unpacked inode;
567 
568 	if (bch2_inode_unpack(k, &inode)) {
569 		prt_printf(out, "(unpack error)");
570 		return;
571 	}
572 
573 	__bch2_inode_unpacked_to_text(out, &inode);
574 }
575 
576 static inline u64 bkey_inode_flags(struct bkey_s_c k)
577 {
578 	switch (k.k->type) {
579 	case KEY_TYPE_inode:
580 		return le32_to_cpu(bkey_s_c_to_inode(k).v->bi_flags);
581 	case KEY_TYPE_inode_v2:
582 		return le64_to_cpu(bkey_s_c_to_inode_v2(k).v->bi_flags);
583 	case KEY_TYPE_inode_v3:
584 		return le64_to_cpu(bkey_s_c_to_inode_v3(k).v->bi_flags);
585 	default:
586 		return 0;
587 	}
588 }
589 
590 static inline bool bkey_is_deleted_inode(struct bkey_s_c k)
591 {
592 	return bkey_inode_flags(k) & BCH_INODE_unlinked;
593 }
594 
595 int bch2_trigger_inode(struct btree_trans *trans,
596 		       enum btree_id btree_id, unsigned level,
597 		       struct bkey_s_c old,
598 		       struct bkey_s new,
599 		       enum btree_iter_update_trigger_flags flags)
600 {
601 	if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
602 		BUG_ON(!trans->journal_res.seq);
603 		bkey_s_to_inode_v3(new).v->bi_journal_seq = cpu_to_le64(trans->journal_res.seq);
604 	}
605 
606 	s64 nr = bkey_is_inode(new.k) - bkey_is_inode(old.k);
607 	if ((flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) && nr) {
608 		struct disk_accounting_pos acc = { .type = BCH_DISK_ACCOUNTING_nr_inodes };
609 		int ret = bch2_disk_accounting_mod(trans, &acc, &nr, 1, flags & BTREE_TRIGGER_gc);
610 		if (ret)
611 			return ret;
612 	}
613 
614 	int deleted_delta =	(int) bkey_is_deleted_inode(new.s_c) -
615 				(int) bkey_is_deleted_inode(old);
616 	if ((flags & BTREE_TRIGGER_transactional) && deleted_delta) {
617 		int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes,
618 						      new.k->p, deleted_delta > 0);
619 		if (ret)
620 			return ret;
621 	}
622 
623 	return 0;
624 }
625 
626 int bch2_inode_generation_validate(struct bch_fs *c, struct bkey_s_c k,
627 				   enum bch_validate_flags flags)
628 {
629 	int ret = 0;
630 
631 	bkey_fsck_err_on(k.k->p.inode,
632 			 c, inode_pos_inode_nonzero,
633 			 "nonzero k.p.inode");
634 fsck_err:
635 	return ret;
636 }
637 
638 void bch2_inode_generation_to_text(struct printbuf *out, struct bch_fs *c,
639 				   struct bkey_s_c k)
640 {
641 	struct bkey_s_c_inode_generation gen = bkey_s_c_to_inode_generation(k);
642 
643 	prt_printf(out, "generation: %u", le32_to_cpu(gen.v->bi_generation));
644 }
645 
646 void bch2_inode_init_early(struct bch_fs *c,
647 			   struct bch_inode_unpacked *inode_u)
648 {
649 	enum bch_str_hash_type str_hash =
650 		bch2_str_hash_opt_to_type(c, c->opts.str_hash);
651 
652 	memset(inode_u, 0, sizeof(*inode_u));
653 
654 	/* ick */
655 	inode_u->bi_flags |= str_hash << INODE_STR_HASH_OFFSET;
656 	get_random_bytes(&inode_u->bi_hash_seed,
657 			 sizeof(inode_u->bi_hash_seed));
658 }
659 
660 void bch2_inode_init_late(struct bch_inode_unpacked *inode_u, u64 now,
661 			  uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
662 			  struct bch_inode_unpacked *parent)
663 {
664 	inode_u->bi_mode	= mode;
665 	inode_u->bi_uid		= uid;
666 	inode_u->bi_gid		= gid;
667 	inode_u->bi_dev		= rdev;
668 	inode_u->bi_atime	= now;
669 	inode_u->bi_mtime	= now;
670 	inode_u->bi_ctime	= now;
671 	inode_u->bi_otime	= now;
672 
673 	if (parent && parent->bi_mode & S_ISGID) {
674 		inode_u->bi_gid = parent->bi_gid;
675 		if (S_ISDIR(mode))
676 			inode_u->bi_mode |= S_ISGID;
677 	}
678 
679 	if (parent) {
680 #define x(_name, ...)	inode_u->bi_##_name = parent->bi_##_name;
681 		BCH_INODE_OPTS()
682 #undef x
683 	}
684 }
685 
686 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
687 		     uid_t uid, gid_t gid, umode_t mode, dev_t rdev,
688 		     struct bch_inode_unpacked *parent)
689 {
690 	bch2_inode_init_early(c, inode_u);
691 	bch2_inode_init_late(inode_u, bch2_current_time(c),
692 			     uid, gid, mode, rdev, parent);
693 }
694 
695 static inline u32 bkey_generation(struct bkey_s_c k)
696 {
697 	switch (k.k->type) {
698 	case KEY_TYPE_inode:
699 	case KEY_TYPE_inode_v2:
700 		BUG();
701 	case KEY_TYPE_inode_generation:
702 		return le32_to_cpu(bkey_s_c_to_inode_generation(k).v->bi_generation);
703 	default:
704 		return 0;
705 	}
706 }
707 
708 /*
709  * This just finds an empty slot:
710  */
711 int bch2_inode_create(struct btree_trans *trans,
712 		      struct btree_iter *iter,
713 		      struct bch_inode_unpacked *inode_u,
714 		      u32 snapshot, u64 cpu)
715 {
716 	struct bch_fs *c = trans->c;
717 	struct bkey_s_c k;
718 	u64 min, max, start, pos, *hint;
719 	int ret = 0;
720 	unsigned bits = (c->opts.inodes_32bit ? 31 : 63);
721 
722 	if (c->opts.shard_inode_numbers) {
723 		bits -= c->inode_shard_bits;
724 
725 		min = (cpu << bits);
726 		max = (cpu << bits) | ~(ULLONG_MAX << bits);
727 
728 		min = max_t(u64, min, BLOCKDEV_INODE_MAX);
729 		hint = c->unused_inode_hints + cpu;
730 	} else {
731 		min = BLOCKDEV_INODE_MAX;
732 		max = ~(ULLONG_MAX << bits);
733 		hint = c->unused_inode_hints;
734 	}
735 
736 	start = READ_ONCE(*hint);
737 
738 	if (start >= max || start < min)
739 		start = min;
740 
741 	pos = start;
742 	bch2_trans_iter_init(trans, iter, BTREE_ID_inodes, POS(0, pos),
743 			     BTREE_ITER_all_snapshots|
744 			     BTREE_ITER_intent);
745 again:
746 	while ((k = bch2_btree_iter_peek(iter)).k &&
747 	       !(ret = bkey_err(k)) &&
748 	       bkey_lt(k.k->p, POS(0, max))) {
749 		if (pos < iter->pos.offset)
750 			goto found_slot;
751 
752 		/*
753 		 * We don't need to iterate over keys in every snapshot once
754 		 * we've found just one:
755 		 */
756 		pos = iter->pos.offset + 1;
757 		bch2_btree_iter_set_pos(iter, POS(0, pos));
758 	}
759 
760 	if (!ret && pos < max)
761 		goto found_slot;
762 
763 	if (!ret && start == min)
764 		ret = -BCH_ERR_ENOSPC_inode_create;
765 
766 	if (ret) {
767 		bch2_trans_iter_exit(trans, iter);
768 		return ret;
769 	}
770 
771 	/* Retry from start */
772 	pos = start = min;
773 	bch2_btree_iter_set_pos(iter, POS(0, pos));
774 	goto again;
775 found_slot:
776 	bch2_btree_iter_set_pos(iter, SPOS(0, pos, snapshot));
777 	k = bch2_btree_iter_peek_slot(iter);
778 	ret = bkey_err(k);
779 	if (ret) {
780 		bch2_trans_iter_exit(trans, iter);
781 		return ret;
782 	}
783 
784 	*hint			= k.k->p.offset;
785 	inode_u->bi_inum	= k.k->p.offset;
786 	inode_u->bi_generation	= bkey_generation(k);
787 	return 0;
788 }
789 
790 static int bch2_inode_delete_keys(struct btree_trans *trans,
791 				  subvol_inum inum, enum btree_id id)
792 {
793 	struct btree_iter iter;
794 	struct bkey_s_c k;
795 	struct bkey_i delete;
796 	struct bpos end = POS(inum.inum, U64_MAX);
797 	u32 snapshot;
798 	int ret = 0;
799 
800 	/*
801 	 * We're never going to be deleting partial extents, no need to use an
802 	 * extent iterator:
803 	 */
804 	bch2_trans_iter_init(trans, &iter, id, POS(inum.inum, 0),
805 			     BTREE_ITER_intent);
806 
807 	while (1) {
808 		bch2_trans_begin(trans);
809 
810 		ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
811 		if (ret)
812 			goto err;
813 
814 		bch2_btree_iter_set_snapshot(&iter, snapshot);
815 
816 		k = bch2_btree_iter_peek_upto(&iter, end);
817 		ret = bkey_err(k);
818 		if (ret)
819 			goto err;
820 
821 		if (!k.k)
822 			break;
823 
824 		bkey_init(&delete.k);
825 		delete.k.p = iter.pos;
826 
827 		if (iter.flags & BTREE_ITER_is_extents)
828 			bch2_key_resize(&delete.k,
829 					bpos_min(end, k.k->p).offset -
830 					iter.pos.offset);
831 
832 		ret = bch2_trans_update(trans, &iter, &delete, 0) ?:
833 		      bch2_trans_commit(trans, NULL, NULL,
834 					BCH_TRANS_COMMIT_no_enospc);
835 err:
836 		if (ret && !bch2_err_matches(ret, BCH_ERR_transaction_restart))
837 			break;
838 	}
839 
840 	bch2_trans_iter_exit(trans, &iter);
841 	return ret;
842 }
843 
844 int bch2_inode_rm(struct bch_fs *c, subvol_inum inum)
845 {
846 	struct btree_trans *trans = bch2_trans_get(c);
847 	struct btree_iter iter = { NULL };
848 	struct bkey_i_inode_generation delete;
849 	struct bch_inode_unpacked inode_u;
850 	struct bkey_s_c k;
851 	u32 snapshot;
852 	int ret;
853 
854 	/*
855 	 * If this was a directory, there shouldn't be any real dirents left -
856 	 * but there could be whiteouts (from hash collisions) that we should
857 	 * delete:
858 	 *
859 	 * XXX: the dirent could ideally would delete whiteouts when they're no
860 	 * longer needed
861 	 */
862 	ret   = bch2_inode_delete_keys(trans, inum, BTREE_ID_extents) ?:
863 		bch2_inode_delete_keys(trans, inum, BTREE_ID_xattrs) ?:
864 		bch2_inode_delete_keys(trans, inum, BTREE_ID_dirents);
865 	if (ret)
866 		goto err;
867 retry:
868 	bch2_trans_begin(trans);
869 
870 	ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot);
871 	if (ret)
872 		goto err;
873 
874 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
875 			       SPOS(0, inum.inum, snapshot),
876 			       BTREE_ITER_intent|BTREE_ITER_cached);
877 	ret = bkey_err(k);
878 	if (ret)
879 		goto err;
880 
881 	if (!bkey_is_inode(k.k)) {
882 		bch2_fs_inconsistent(c,
883 				     "inode %llu:%u not found when deleting",
884 				     inum.inum, snapshot);
885 		ret = -EIO;
886 		goto err;
887 	}
888 
889 	bch2_inode_unpack(k, &inode_u);
890 
891 	bkey_inode_generation_init(&delete.k_i);
892 	delete.k.p = iter.pos;
893 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
894 
895 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
896 		bch2_trans_commit(trans, NULL, NULL,
897 				BCH_TRANS_COMMIT_no_enospc);
898 err:
899 	bch2_trans_iter_exit(trans, &iter);
900 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
901 		goto retry;
902 
903 	bch2_trans_put(trans);
904 	return ret;
905 }
906 
907 int bch2_inode_find_by_inum_nowarn_trans(struct btree_trans *trans,
908 				  subvol_inum inum,
909 				  struct bch_inode_unpacked *inode)
910 {
911 	struct btree_iter iter;
912 	int ret;
913 
914 	ret = bch2_inode_peek_nowarn(trans, &iter, inode, inum, 0);
915 	if (!ret)
916 		bch2_trans_iter_exit(trans, &iter);
917 	return ret;
918 }
919 
920 int bch2_inode_find_by_inum_trans(struct btree_trans *trans,
921 				  subvol_inum inum,
922 				  struct bch_inode_unpacked *inode)
923 {
924 	struct btree_iter iter;
925 	int ret;
926 
927 	ret = bch2_inode_peek(trans, &iter, inode, inum, 0);
928 	if (!ret)
929 		bch2_trans_iter_exit(trans, &iter);
930 	return ret;
931 }
932 
933 int bch2_inode_find_by_inum(struct bch_fs *c, subvol_inum inum,
934 			    struct bch_inode_unpacked *inode)
935 {
936 	return bch2_trans_do(c, NULL, NULL, 0,
937 		bch2_inode_find_by_inum_trans(trans, inum, inode));
938 }
939 
940 int bch2_inode_nlink_inc(struct bch_inode_unpacked *bi)
941 {
942 	if (bi->bi_flags & BCH_INODE_unlinked)
943 		bi->bi_flags &= ~BCH_INODE_unlinked;
944 	else {
945 		if (bi->bi_nlink == U32_MAX)
946 			return -EINVAL;
947 
948 		bi->bi_nlink++;
949 	}
950 
951 	return 0;
952 }
953 
954 void bch2_inode_nlink_dec(struct btree_trans *trans, struct bch_inode_unpacked *bi)
955 {
956 	if (bi->bi_nlink && (bi->bi_flags & BCH_INODE_unlinked)) {
957 		bch2_trans_inconsistent(trans, "inode %llu unlinked but link count nonzero",
958 					bi->bi_inum);
959 		return;
960 	}
961 
962 	if (bi->bi_flags & BCH_INODE_unlinked) {
963 		bch2_trans_inconsistent(trans, "inode %llu link count underflow", bi->bi_inum);
964 		return;
965 	}
966 
967 	if (bi->bi_nlink)
968 		bi->bi_nlink--;
969 	else
970 		bi->bi_flags |= BCH_INODE_unlinked;
971 }
972 
973 struct bch_opts bch2_inode_opts_to_opts(struct bch_inode_unpacked *inode)
974 {
975 	struct bch_opts ret = { 0 };
976 #define x(_name, _bits)							\
977 	if (inode->bi_##_name)						\
978 		opt_set(ret, _name, inode->bi_##_name - 1);
979 	BCH_INODE_OPTS()
980 #undef x
981 	return ret;
982 }
983 
984 void bch2_inode_opts_get(struct bch_io_opts *opts, struct bch_fs *c,
985 			 struct bch_inode_unpacked *inode)
986 {
987 #define x(_name, _bits)		opts->_name = inode_opt_get(c, inode, _name);
988 	BCH_INODE_OPTS()
989 #undef x
990 
991 	if (opts->nocow)
992 		opts->compression = opts->background_compression = opts->data_checksum = opts->erasure_code = 0;
993 }
994 
995 int bch2_inum_opts_get(struct btree_trans *trans, subvol_inum inum, struct bch_io_opts *opts)
996 {
997 	struct bch_inode_unpacked inode;
998 	int ret = lockrestart_do(trans, bch2_inode_find_by_inum_trans(trans, inum, &inode));
999 
1000 	if (ret)
1001 		return ret;
1002 
1003 	bch2_inode_opts_get(opts, trans->c, &inode);
1004 	return 0;
1005 }
1006 
1007 int bch2_inode_rm_snapshot(struct btree_trans *trans, u64 inum, u32 snapshot)
1008 {
1009 	struct bch_fs *c = trans->c;
1010 	struct btree_iter iter = { NULL };
1011 	struct bkey_i_inode_generation delete;
1012 	struct bch_inode_unpacked inode_u;
1013 	struct bkey_s_c k;
1014 	int ret;
1015 
1016 	do {
1017 		ret   = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
1018 						      SPOS(inum, 0, snapshot),
1019 						      SPOS(inum, U64_MAX, snapshot),
1020 						      0, NULL) ?:
1021 			bch2_btree_delete_range_trans(trans, BTREE_ID_dirents,
1022 						      SPOS(inum, 0, snapshot),
1023 						      SPOS(inum, U64_MAX, snapshot),
1024 						      0, NULL) ?:
1025 			bch2_btree_delete_range_trans(trans, BTREE_ID_xattrs,
1026 						      SPOS(inum, 0, snapshot),
1027 						      SPOS(inum, U64_MAX, snapshot),
1028 						      0, NULL);
1029 	} while (ret == -BCH_ERR_transaction_restart_nested);
1030 	if (ret)
1031 		goto err;
1032 retry:
1033 	bch2_trans_begin(trans);
1034 
1035 	k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
1036 			       SPOS(0, inum, snapshot), BTREE_ITER_intent);
1037 	ret = bkey_err(k);
1038 	if (ret)
1039 		goto err;
1040 
1041 	if (!bkey_is_inode(k.k)) {
1042 		bch2_fs_inconsistent(c,
1043 				     "inode %llu:%u not found when deleting",
1044 				     inum, snapshot);
1045 		ret = -EIO;
1046 		goto err;
1047 	}
1048 
1049 	bch2_inode_unpack(k, &inode_u);
1050 
1051 	/* Subvolume root? */
1052 	if (inode_u.bi_subvol)
1053 		bch_warn(c, "deleting inode %llu marked as unlinked, but also a subvolume root!?", inode_u.bi_inum);
1054 
1055 	bkey_inode_generation_init(&delete.k_i);
1056 	delete.k.p = iter.pos;
1057 	delete.v.bi_generation = cpu_to_le32(inode_u.bi_generation + 1);
1058 
1059 	ret   = bch2_trans_update(trans, &iter, &delete.k_i, 0) ?:
1060 		bch2_trans_commit(trans, NULL, NULL,
1061 				BCH_TRANS_COMMIT_no_enospc);
1062 err:
1063 	bch2_trans_iter_exit(trans, &iter);
1064 	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1065 		goto retry;
1066 
1067 	return ret ?: -BCH_ERR_transaction_restart_nested;
1068 }
1069 
1070 static int may_delete_deleted_inode(struct btree_trans *trans,
1071 				    struct btree_iter *iter,
1072 				    struct bpos pos,
1073 				    bool *need_another_pass)
1074 {
1075 	struct bch_fs *c = trans->c;
1076 	struct btree_iter inode_iter;
1077 	struct bkey_s_c k;
1078 	struct bch_inode_unpacked inode;
1079 	int ret;
1080 
1081 	k = bch2_bkey_get_iter(trans, &inode_iter, BTREE_ID_inodes, pos, BTREE_ITER_cached);
1082 	ret = bkey_err(k);
1083 	if (ret)
1084 		return ret;
1085 
1086 	ret = bkey_is_inode(k.k) ? 0 : -BCH_ERR_ENOENT_inode;
1087 	if (fsck_err_on(!bkey_is_inode(k.k),
1088 			trans, deleted_inode_missing,
1089 			"nonexistent inode %llu:%u in deleted_inodes btree",
1090 			pos.offset, pos.snapshot))
1091 		goto delete;
1092 
1093 	ret = bch2_inode_unpack(k, &inode);
1094 	if (ret)
1095 		goto out;
1096 
1097 	if (S_ISDIR(inode.bi_mode)) {
1098 		ret = bch2_empty_dir_snapshot(trans, pos.offset, 0, pos.snapshot);
1099 		if (fsck_err_on(bch2_err_matches(ret, ENOTEMPTY),
1100 				trans, deleted_inode_is_dir,
1101 				"non empty directory %llu:%u in deleted_inodes btree",
1102 				pos.offset, pos.snapshot))
1103 			goto delete;
1104 		if (ret)
1105 			goto out;
1106 	}
1107 
1108 	if (fsck_err_on(!(inode.bi_flags & BCH_INODE_unlinked),
1109 			trans, deleted_inode_not_unlinked,
1110 			"non-deleted inode %llu:%u in deleted_inodes btree",
1111 			pos.offset, pos.snapshot))
1112 		goto delete;
1113 
1114 	if (c->sb.clean &&
1115 	    !fsck_err(trans, deleted_inode_but_clean,
1116 		      "filesystem marked as clean but have deleted inode %llu:%u",
1117 		      pos.offset, pos.snapshot)) {
1118 		ret = 0;
1119 		goto out;
1120 	}
1121 
1122 	if (bch2_snapshot_is_internal_node(c, pos.snapshot)) {
1123 		struct bpos new_min_pos;
1124 
1125 		ret = bch2_propagate_key_to_snapshot_leaves(trans, inode_iter.btree_id, k, &new_min_pos);
1126 		if (ret)
1127 			goto out;
1128 
1129 		inode.bi_flags &= ~BCH_INODE_unlinked;
1130 
1131 		ret = bch2_inode_write_flags(trans, &inode_iter, &inode,
1132 					     BTREE_UPDATE_internal_snapshot_node);
1133 		bch_err_msg(c, ret, "clearing inode unlinked flag");
1134 		if (ret)
1135 			goto out;
1136 
1137 		/*
1138 		 * We'll need another write buffer flush to pick up the new
1139 		 * unlinked inodes in the snapshot leaves:
1140 		 */
1141 		*need_another_pass = true;
1142 		goto out;
1143 	}
1144 
1145 	ret = 1;
1146 out:
1147 fsck_err:
1148 	bch2_trans_iter_exit(trans, &inode_iter);
1149 	return ret;
1150 delete:
1151 	ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_deleted_inodes, pos, false);
1152 	goto out;
1153 }
1154 
1155 int bch2_delete_dead_inodes(struct bch_fs *c)
1156 {
1157 	struct btree_trans *trans = bch2_trans_get(c);
1158 	bool need_another_pass;
1159 	int ret;
1160 again:
1161 	/*
1162 	 * if we ran check_inodes() unlinked inodes will have already been
1163 	 * cleaned up but the write buffer will be out of sync; therefore we
1164 	 * alway need a write buffer flush
1165 	 */
1166 	ret = bch2_btree_write_buffer_flush_sync(trans);
1167 	if (ret)
1168 		goto err;
1169 
1170 	need_another_pass = false;
1171 
1172 	/*
1173 	 * Weird transaction restart handling here because on successful delete,
1174 	 * bch2_inode_rm_snapshot() will return a nested transaction restart,
1175 	 * but we can't retry because the btree write buffer won't have been
1176 	 * flushed and we'd spin:
1177 	 */
1178 	ret = for_each_btree_key_commit(trans, iter, BTREE_ID_deleted_inodes, POS_MIN,
1179 					BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k,
1180 					NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1181 		ret = may_delete_deleted_inode(trans, &iter, k.k->p, &need_another_pass);
1182 		if (ret > 0) {
1183 			bch_verbose(c, "deleting unlinked inode %llu:%u", k.k->p.offset, k.k->p.snapshot);
1184 
1185 			ret = bch2_inode_rm_snapshot(trans, k.k->p.offset, k.k->p.snapshot);
1186 			/*
1187 			 * We don't want to loop here: a transaction restart
1188 			 * error here means we handled a transaction restart and
1189 			 * we're actually done, but if we loop we'll retry the
1190 			 * same key because the write buffer hasn't been flushed
1191 			 * yet
1192 			 */
1193 			if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1194 				ret = 0;
1195 				continue;
1196 			}
1197 		}
1198 
1199 		ret;
1200 	}));
1201 
1202 	if (!ret && need_another_pass)
1203 		goto again;
1204 err:
1205 	bch2_trans_put(trans);
1206 	return ret;
1207 }
1208