3 #include "bkey_methods.h"
4 #include "btree_update.h"
10 #include <linux/random.h>
12 #include <asm/unaligned.h>
14 #define FIELD_BYTES() \
16 static const u8 byte_table[8] = { 1, 2, 3, 4, 6, 8, 10, 13 };
17 static const u8 bits_table[8] = {
28 static int inode_encode_field(u8 *out, u8 *end, u64 hi, u64 lo)
30 __be64 in[2] = { cpu_to_be64(hi), cpu_to_be64(lo), };
31 unsigned shift, bytes, bits = likely(!hi)
35 for (shift = 1; shift <= 8; shift++)
36 if (bits < bits_table[shift - 1])
41 bytes = byte_table[shift - 1];
43 BUG_ON(out + bytes > end);
45 memcpy(out, (u8 *) in + 16 - bytes, bytes);
46 *out |= (1 << 8) >> shift;
51 static int inode_decode_field(const u8 *in, const u8 *end,
52 u64 out[2], unsigned *out_bits)
54 __be64 be[2] = { 0, 0 };
55 unsigned bytes, shift;
65 * position of highest set bit indicates number of bytes:
66 * shift = number of bits to remove in high byte:
68 shift = 8 - __fls(*in); /* 1 <= shift <= 8 */
69 bytes = byte_table[shift - 1];
74 p = (u8 *) be + 16 - bytes;
76 *p ^= (1 << 8) >> shift;
78 out[0] = be64_to_cpu(be[0]);
79 out[1] = be64_to_cpu(be[1]);
80 *out_bits = out[0] ? 64 + fls64(out[0]) : fls64(out[1]);
85 void bch2_inode_pack(struct bkey_inode_buf *packed,
86 const struct bch_inode_unpacked *inode)
88 u8 *out = packed->inode.v.fields;
89 u8 *end = (void *) &packed[1];
90 u8 *last_nonzero_field = out;
91 unsigned nr_fields = 0, last_nonzero_fieldnr = 0;
93 bkey_inode_init(&packed->inode.k_i);
94 packed->inode.k.p.inode = inode->inum;
95 packed->inode.v.i_hash_seed = inode->i_hash_seed;
96 packed->inode.v.i_flags = cpu_to_le32(inode->i_flags);
97 packed->inode.v.i_mode = cpu_to_le16(inode->i_mode);
99 #define BCH_INODE_FIELD(_name, _bits) \
100 out += inode_encode_field(out, end, 0, inode->_name); \
103 if (inode->_name) { \
104 last_nonzero_field = out; \
105 last_nonzero_fieldnr = nr_fields; \
109 #undef BCH_INODE_FIELD
111 out = last_nonzero_field;
112 nr_fields = last_nonzero_fieldnr;
114 set_bkey_val_bytes(&packed->inode.k, out - (u8 *) &packed->inode.v);
116 (u8 *) &packed->inode.v +
117 bkey_val_bytes(&packed->inode.k) - out);
119 SET_INODE_NR_FIELDS(&packed->inode.v, nr_fields);
121 if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
122 struct bch_inode_unpacked unpacked;
124 int ret = bch2_inode_unpack(inode_i_to_s_c(&packed->inode),
127 BUG_ON(unpacked.inum != inode->inum);
128 BUG_ON(unpacked.i_hash_seed != inode->i_hash_seed);
129 BUG_ON(unpacked.i_mode != inode->i_mode);
131 #define BCH_INODE_FIELD(_name, _bits) BUG_ON(unpacked._name != inode->_name);
133 #undef BCH_INODE_FIELD
137 int bch2_inode_unpack(struct bkey_s_c_inode inode,
138 struct bch_inode_unpacked *unpacked)
140 const u8 *in = inode.v->fields;
141 const u8 *end = (void *) inode.v + bkey_val_bytes(inode.k);
143 unsigned fieldnr = 0, field_bits;
146 unpacked->inum = inode.k->p.inode;
147 unpacked->i_hash_seed = inode.v->i_hash_seed;
148 unpacked->i_flags = le32_to_cpu(inode.v->i_flags);
149 unpacked->i_mode = le16_to_cpu(inode.v->i_mode);
151 #define BCH_INODE_FIELD(_name, _bits) \
152 if (fieldnr++ == INODE_NR_FIELDS(inode.v)) { \
153 memset(&unpacked->_name, 0, \
154 sizeof(*unpacked) - \
155 offsetof(struct bch_inode_unpacked, _name)); \
159 ret = inode_decode_field(in, end, field, &field_bits); \
163 if (field_bits > sizeof(unpacked->_name) * 8) \
166 unpacked->_name = field[1]; \
170 #undef BCH_INODE_FIELD
172 /* XXX: signal if there were more fields than expected? */
177 static const char *bch2_inode_invalid(const struct bch_fs *c,
181 return "nonzero offset";
185 struct bkey_s_c_inode inode = bkey_s_c_to_inode(k);
186 struct bch_inode_unpacked unpacked;
188 if (bkey_val_bytes(k.k) < sizeof(struct bch_inode))
189 return "incorrect value size";
191 if (k.k->p.inode < BLOCKDEV_INODE_MAX)
192 return "fs inode in blockdev range";
194 if (INODE_STR_HASH(inode.v) >= BCH_STR_HASH_NR)
195 return "invalid str hash type";
197 if (bch2_inode_unpack(inode, &unpacked))
198 return "invalid variable length fields";
202 case BCH_INODE_BLOCKDEV:
203 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_blockdev))
204 return "incorrect value size";
206 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
207 return "blockdev inode in fs range";
210 case BCH_INODE_GENERATION:
211 if (bkey_val_bytes(k.k) != sizeof(struct bch_inode_generation))
212 return "incorrect value size";
216 return "invalid type";
220 static void bch2_inode_to_text(struct bch_fs *c, char *buf,
221 size_t size, struct bkey_s_c k)
223 struct bkey_s_c_inode inode;
224 struct bch_inode_unpacked unpacked;
228 inode = bkey_s_c_to_inode(k);
229 if (bch2_inode_unpack(inode, &unpacked)) {
230 scnprintf(buf, size, "(unpack error)");
234 scnprintf(buf, size, "i_size %llu", unpacked.i_size);
239 const struct bkey_ops bch2_bkey_inode_ops = {
240 .key_invalid = bch2_inode_invalid,
241 .val_to_text = bch2_inode_to_text,
244 void bch2_inode_init(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
245 uid_t uid, gid_t gid, umode_t mode, dev_t rdev)
247 s64 now = timespec_to_bch2_time(c, CURRENT_TIME);
249 memset(inode_u, 0, sizeof(*inode_u));
252 inode_u->i_flags |= c->sb.str_hash_type << INODE_STR_HASH_OFFSET;
253 get_random_bytes(&inode_u->i_hash_seed, sizeof(inode_u->i_hash_seed));
255 inode_u->i_mode = mode;
256 inode_u->i_uid = uid;
257 inode_u->i_gid = gid;
258 inode_u->i_dev = rdev;
259 inode_u->i_atime = now;
260 inode_u->i_mtime = now;
261 inode_u->i_ctime = now;
262 inode_u->i_otime = now;
265 int bch2_inode_create(struct bch_fs *c, struct bch_inode_unpacked *inode_u,
266 u64 min, u64 max, u64 *hint)
268 struct bkey_inode_buf inode_p;
269 struct btree_iter iter;
270 bool searched_from_start = false;
276 if (c->opts.inodes_32bit)
277 max = min_t(u64, max, U32_MAX);
279 if (*hint >= max || *hint < min)
283 searched_from_start = true;
285 bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(*hint, 0),
289 struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
290 u32 i_generation = 0;
292 ret = btree_iter_err(k);
294 bch2_btree_iter_unlock(&iter);
299 case BCH_INODE_BLOCKDEV:
302 if (iter.pos.inode == max)
305 bch2_btree_iter_advance_pos(&iter);
308 case BCH_INODE_GENERATION: {
309 struct bkey_s_c_inode_generation g =
310 bkey_s_c_to_inode_generation(k);
311 i_generation = le32_to_cpu(g.v->i_generation);
315 inode_u->i_generation = i_generation;
317 bch2_inode_pack(&inode_p, inode_u);
318 inode_p.inode.k.p = k.k->p;
320 ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
322 BTREE_INSERT_ENTRY(&iter,
323 &inode_p.inode.k_i));
326 bch2_btree_iter_unlock(&iter);
330 inode_p.inode.k.p.inode;
331 *hint = inode_p.inode.k.p.inode + 1;
343 bch2_btree_iter_unlock(&iter);
345 if (!searched_from_start) {
346 /* Retry from start */
348 searched_from_start = true;
355 int bch2_inode_truncate(struct bch_fs *c, u64 inode_nr, u64 new_size,
356 struct extent_insert_hook *hook, u64 *journal_seq)
358 return bch2_btree_delete_range(c, BTREE_ID_EXTENTS,
359 POS(inode_nr, new_size),
360 POS(inode_nr + 1, 0),
361 ZERO_VERSION, NULL, hook,
365 int bch2_inode_rm(struct bch_fs *c, u64 inode_nr)
367 struct btree_iter iter;
368 struct bkey_i_inode_generation delete;
371 ret = bch2_inode_truncate(c, inode_nr, 0, NULL, NULL);
375 ret = bch2_btree_delete_range(c, BTREE_ID_XATTRS,
377 POS(inode_nr + 1, 0),
378 ZERO_VERSION, NULL, NULL, NULL);
383 * If this was a directory, there shouldn't be any real dirents left -
384 * but there could be whiteouts (from hash collisions) that we should
387 * XXX: the dirent could ideally would delete whitouts when they're no
390 ret = bch2_btree_delete_range(c, BTREE_ID_DIRENTS,
392 POS(inode_nr + 1, 0),
393 ZERO_VERSION, NULL, NULL, NULL);
397 bch2_btree_iter_init(&iter, c, BTREE_ID_INODES, POS(inode_nr, 0),
400 struct bkey_s_c k = bch2_btree_iter_peek_with_holes(&iter);
401 u32 i_generation = 0;
403 ret = btree_iter_err(k);
405 bch2_btree_iter_unlock(&iter);
411 struct bch_inode_unpacked inode_u;
413 if (!bch2_inode_unpack(bkey_s_c_to_inode(k), &inode_u))
414 i_generation = cpu_to_le32(inode_u.i_generation) + 1;
417 case BCH_INODE_GENERATION: {
418 struct bkey_s_c_inode_generation g =
419 bkey_s_c_to_inode_generation(k);
420 i_generation = le32_to_cpu(g.v->i_generation);
426 bkey_init(&delete.k);
427 delete.k.p.inode = inode_nr;
429 bkey_inode_generation_init(&delete.k_i);
430 delete.k.p.inode = inode_nr;
431 delete.v.i_generation = cpu_to_le32(i_generation);
434 ret = bch2_btree_insert_at(c, NULL, NULL, NULL,
437 BTREE_INSERT_ENTRY(&iter, &delete.k_i));
438 } while (ret == -EINTR);
440 bch2_btree_iter_unlock(&iter);
444 int bch2_inode_find_by_inum(struct bch_fs *c, u64 inode_nr,
445 struct bch_inode_unpacked *inode)
447 struct btree_iter iter;
451 for_each_btree_key(&iter, c, BTREE_ID_INODES,
453 BTREE_ITER_WITH_HOLES, k) {
456 ret = bch2_inode_unpack(bkey_s_c_to_inode(k), inode);
459 /* hole, not found */
467 return bch2_btree_iter_unlock(&iter) ?: ret;
470 int bch2_cached_dev_inode_find_by_uuid(struct bch_fs *c, uuid_le *uuid,
471 struct bkey_i_inode_blockdev *ret)
473 struct btree_iter iter;
476 for_each_btree_key(&iter, c, BTREE_ID_INODES, POS(0, 0), 0, k) {
477 if (k.k->p.inode >= BLOCKDEV_INODE_MAX)
480 if (k.k->type == BCH_INODE_BLOCKDEV) {
481 struct bkey_s_c_inode_blockdev inode =
482 bkey_s_c_to_inode_blockdev(k);
484 pr_debug("found inode %llu: %pU (u64s %u)",
485 inode.k->p.inode, inode.v->i_uuid.b,
488 if (CACHED_DEV(inode.v) &&
489 !memcmp(uuid, &inode.v->i_uuid, 16)) {
490 bkey_reassemble(&ret->k_i, k);
491 bch2_btree_iter_unlock(&iter);
496 bch2_btree_iter_cond_resched(&iter);
498 bch2_btree_iter_unlock(&iter);
502 #ifdef CONFIG_BCACHEFS_DEBUG
503 void bch2_inode_pack_test(void)
505 struct bch_inode_unpacked *u, test_inodes[] = {
512 .i_sectors = U64_MAX,
516 .i_generation = U32_MAX,
521 for (u = test_inodes;
522 u < test_inodes + ARRAY_SIZE(test_inodes);
524 struct bkey_inode_buf p;
526 bch2_inode_pack(&p, u);