1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * Copyright 2010, 2011 Kent Overstreet <kent. 3 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
4 * Copyright 2012 Google, Inc. 4 * Copyright 2012 Google, Inc.
5 */ 5 */
6 6
7 #include "bcachefs.h" 7 #include "bcachefs.h"
8 #include "alloc_foreground.h" 8 #include "alloc_foreground.h"
9 #include "bkey_buf.h" 9 #include "bkey_buf.h"
10 #include "bset.h" 10 #include "bset.h"
11 #include "btree_update.h" 11 #include "btree_update.h"
12 #include "buckets.h" 12 #include "buckets.h"
13 #include "checksum.h" 13 #include "checksum.h"
14 #include "clock.h" 14 #include "clock.h"
15 #include "compress.h" 15 #include "compress.h"
16 #include "debug.h" 16 #include "debug.h"
17 #include "ec.h" 17 #include "ec.h"
18 #include "error.h" 18 #include "error.h"
19 #include "extent_update.h" 19 #include "extent_update.h"
20 #include "inode.h" 20 #include "inode.h"
21 #include "io_write.h" 21 #include "io_write.h"
22 #include "journal.h" 22 #include "journal.h"
23 #include "keylist.h" 23 #include "keylist.h"
24 #include "move.h" 24 #include "move.h"
25 #include "nocow_locking.h" 25 #include "nocow_locking.h"
26 #include "rebalance.h" 26 #include "rebalance.h"
27 #include "subvolume.h" 27 #include "subvolume.h"
28 #include "super.h" 28 #include "super.h"
29 #include "super-io.h" 29 #include "super-io.h"
30 #include "trace.h" 30 #include "trace.h"
31 31
32 #include <linux/blkdev.h> 32 #include <linux/blkdev.h>
33 #include <linux/prefetch.h> 33 #include <linux/prefetch.h>
34 #include <linux/random.h> 34 #include <linux/random.h>
35 #include <linux/sched/mm.h> 35 #include <linux/sched/mm.h>
36 36
37 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT 37 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
38 38
39 static inline void bch2_congested_acct(struct 39 static inline void bch2_congested_acct(struct bch_dev *ca, u64 io_latency,
40 u64 now 40 u64 now, int rw)
41 { 41 {
42 u64 latency_capable = 42 u64 latency_capable =
43 ca->io_latency[rw].quantiles.e 43 ca->io_latency[rw].quantiles.entries[QUANTILE_IDX(1)].m;
44 /* ideally we'd be taking into account 44 /* ideally we'd be taking into account the device's variance here: */
45 u64 latency_threshold = latency_capabl 45 u64 latency_threshold = latency_capable << (rw == READ ? 2 : 3);
46 s64 latency_over = io_latency - latenc 46 s64 latency_over = io_latency - latency_threshold;
47 47
48 if (latency_threshold && latency_over 48 if (latency_threshold && latency_over > 0) {
49 /* 49 /*
50 * bump up congested by approx 50 * bump up congested by approximately latency_over * 4 /
51 * latency_threshold - we don' 51 * latency_threshold - we don't need much accuracy here so don't
52 * bother with the divide: 52 * bother with the divide:
53 */ 53 */
54 if (atomic_read(&ca->congested 54 if (atomic_read(&ca->congested) < CONGESTED_MAX)
55 atomic_add(latency_ove 55 atomic_add(latency_over >>
56 max_t(int, 56 max_t(int, ilog2(latency_threshold) - 2, 0),
57 &ca->conges 57 &ca->congested);
58 58
59 ca->congested_last = now; 59 ca->congested_last = now;
60 } else if (atomic_read(&ca->congested) 60 } else if (atomic_read(&ca->congested) > 0) {
61 atomic_dec(&ca->congested); 61 atomic_dec(&ca->congested);
62 } 62 }
63 } 63 }
64 64
65 void bch2_latency_acct(struct bch_dev *ca, u64 65 void bch2_latency_acct(struct bch_dev *ca, u64 submit_time, int rw)
66 { 66 {
67 atomic64_t *latency = &ca->cur_latency 67 atomic64_t *latency = &ca->cur_latency[rw];
68 u64 now = local_clock(); 68 u64 now = local_clock();
69 u64 io_latency = time_after64(now, sub 69 u64 io_latency = time_after64(now, submit_time)
70 ? now - submit_time 70 ? now - submit_time
71 : 0; 71 : 0;
72 u64 old, new; !! 72 u64 old, new, v = atomic64_read(latency);
73 73
74 old = atomic64_read(latency); <<
75 do { 74 do {
>> 75 old = v;
>> 76
76 /* 77 /*
77 * If the io latency was reaso 78 * If the io latency was reasonably close to the current
78 * latency, skip doing the upd 79 * latency, skip doing the update and atomic operation - most of
79 * the time: 80 * the time:
80 */ 81 */
81 if (abs((int) (old - io_latenc 82 if (abs((int) (old - io_latency)) < (old >> 1) &&
82 now & ~(~0U << 5)) 83 now & ~(~0U << 5))
83 break; 84 break;
84 85
85 new = ewma_add(old, io_latency 86 new = ewma_add(old, io_latency, 5);
86 } while (!atomic64_try_cmpxchg(latency !! 87 } while ((v = atomic64_cmpxchg(latency, old, new)) != old);
87 88
88 bch2_congested_acct(ca, io_latency, no 89 bch2_congested_acct(ca, io_latency, now, rw);
89 90
90 __bch2_time_stats_update(&ca->io_laten 91 __bch2_time_stats_update(&ca->io_latency[rw].stats, submit_time, now);
91 } 92 }
92 93
93 #endif 94 #endif
94 95
95 /* Allocate, free from mempool: */ 96 /* Allocate, free from mempool: */
96 97
97 void bch2_bio_free_pages_pool(struct bch_fs *c 98 void bch2_bio_free_pages_pool(struct bch_fs *c, struct bio *bio)
98 { 99 {
99 struct bvec_iter_all iter; 100 struct bvec_iter_all iter;
100 struct bio_vec *bv; 101 struct bio_vec *bv;
101 102
102 bio_for_each_segment_all(bv, bio, iter 103 bio_for_each_segment_all(bv, bio, iter)
103 if (bv->bv_page != ZERO_PAGE(0 104 if (bv->bv_page != ZERO_PAGE(0))
104 mempool_free(bv->bv_pa 105 mempool_free(bv->bv_page, &c->bio_bounce_pages);
105 bio->bi_vcnt = 0; 106 bio->bi_vcnt = 0;
106 } 107 }
107 108
108 static struct page *__bio_alloc_page_pool(stru 109 static struct page *__bio_alloc_page_pool(struct bch_fs *c, bool *using_mempool)
109 { 110 {
110 struct page *page; 111 struct page *page;
111 112
112 if (likely(!*using_mempool)) { 113 if (likely(!*using_mempool)) {
113 page = alloc_page(GFP_NOFS); 114 page = alloc_page(GFP_NOFS);
114 if (unlikely(!page)) { 115 if (unlikely(!page)) {
115 mutex_lock(&c->bio_bou 116 mutex_lock(&c->bio_bounce_pages_lock);
116 *using_mempool = true; 117 *using_mempool = true;
117 goto pool_alloc; 118 goto pool_alloc;
118 119
119 } 120 }
120 } else { 121 } else {
121 pool_alloc: 122 pool_alloc:
122 page = mempool_alloc(&c->bio_b 123 page = mempool_alloc(&c->bio_bounce_pages, GFP_NOFS);
123 } 124 }
124 125
125 return page; 126 return page;
126 } 127 }
127 128
128 void bch2_bio_alloc_pages_pool(struct bch_fs * 129 void bch2_bio_alloc_pages_pool(struct bch_fs *c, struct bio *bio,
129 size_t size) 130 size_t size)
130 { 131 {
131 bool using_mempool = false; 132 bool using_mempool = false;
132 133
133 while (size) { 134 while (size) {
134 struct page *page = __bio_allo 135 struct page *page = __bio_alloc_page_pool(c, &using_mempool);
135 unsigned len = min_t(size_t, P 136 unsigned len = min_t(size_t, PAGE_SIZE, size);
136 137
137 BUG_ON(!bio_add_page(bio, page 138 BUG_ON(!bio_add_page(bio, page, len, 0));
138 size -= len; 139 size -= len;
139 } 140 }
140 141
141 if (using_mempool) 142 if (using_mempool)
142 mutex_unlock(&c->bio_bounce_pa 143 mutex_unlock(&c->bio_bounce_pages_lock);
143 } 144 }
144 145
145 /* Extent update path: */ 146 /* Extent update path: */
146 147
147 int bch2_sum_sector_overwrites(struct btree_tr 148 int bch2_sum_sector_overwrites(struct btree_trans *trans,
148 struct btree_it 149 struct btree_iter *extent_iter,
149 struct bkey_i * 150 struct bkey_i *new,
150 bool *usage_inc 151 bool *usage_increasing,
151 s64 *i_sectors_ 152 s64 *i_sectors_delta,
152 s64 *disk_secto 153 s64 *disk_sectors_delta)
153 { 154 {
154 struct bch_fs *c = trans->c; 155 struct bch_fs *c = trans->c;
155 struct btree_iter iter; 156 struct btree_iter iter;
156 struct bkey_s_c old; 157 struct bkey_s_c old;
157 unsigned new_replicas = bch2_bkey_repl 158 unsigned new_replicas = bch2_bkey_replicas(c, bkey_i_to_s_c(new));
158 bool new_compressed = bch2_bkey_sector 159 bool new_compressed = bch2_bkey_sectors_compressed(bkey_i_to_s_c(new));
159 int ret = 0; 160 int ret = 0;
160 161
161 *usage_increasing = false; 162 *usage_increasing = false;
162 *i_sectors_delta = 0; 163 *i_sectors_delta = 0;
163 *disk_sectors_delta = 0; 164 *disk_sectors_delta = 0;
164 165
165 bch2_trans_copy_iter(&iter, extent_ite 166 bch2_trans_copy_iter(&iter, extent_iter);
166 167
167 for_each_btree_key_upto_continue_nores 168 for_each_btree_key_upto_continue_norestart(iter,
168 new->k.p, BTRE 169 new->k.p, BTREE_ITER_slots, old, ret) {
169 s64 sectors = min(new->k.p.off 170 s64 sectors = min(new->k.p.offset, old.k->p.offset) -
170 max(bkey_start_offset( 171 max(bkey_start_offset(&new->k),
171 bkey_start_offset( 172 bkey_start_offset(old.k));
172 173
173 *i_sectors_delta += sectors * 174 *i_sectors_delta += sectors *
174 (bkey_extent_is_alloca 175 (bkey_extent_is_allocation(&new->k) -
175 bkey_extent_is_alloca 176 bkey_extent_is_allocation(old.k));
176 177
177 *disk_sectors_delta += sectors 178 *disk_sectors_delta += sectors * bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(new));
178 *disk_sectors_delta -= new->k. 179 *disk_sectors_delta -= new->k.p.snapshot == old.k->p.snapshot
179 ? sectors * bch2_bkey_ 180 ? sectors * bch2_bkey_nr_ptrs_fully_allocated(old)
180 : 0; 181 : 0;
181 182
182 if (!*usage_increasing && 183 if (!*usage_increasing &&
183 (new->k.p.snapshot != old. 184 (new->k.p.snapshot != old.k->p.snapshot ||
184 new_replicas > bch2_bkey_ 185 new_replicas > bch2_bkey_replicas(c, old) ||
185 (!new_compressed && bch2_ 186 (!new_compressed && bch2_bkey_sectors_compressed(old))))
186 *usage_increasing = tr 187 *usage_increasing = true;
187 188
188 if (bkey_ge(old.k->p, new->k.p 189 if (bkey_ge(old.k->p, new->k.p))
189 break; 190 break;
190 } 191 }
191 192
192 bch2_trans_iter_exit(trans, &iter); 193 bch2_trans_iter_exit(trans, &iter);
193 return ret; 194 return ret;
194 } 195 }
195 196
196 static inline int bch2_extent_update_i_size_se 197 static inline int bch2_extent_update_i_size_sectors(struct btree_trans *trans,
197 198 struct btree_iter *extent_iter,
198 199 u64 new_i_size,
199 200 s64 i_sectors_delta)
200 { 201 {
201 /* 202 /*
202 * Crazy performance optimization: 203 * Crazy performance optimization:
203 * Every extent update needs to also u 204 * Every extent update needs to also update the inode: the inode trigger
204 * will set bi->journal_seq to the jou 205 * will set bi->journal_seq to the journal sequence number of this
205 * transaction - for fsync. 206 * transaction - for fsync.
206 * 207 *
207 * But if that's the only reason we're 208 * But if that's the only reason we're updating the inode (we're not
208 * updating bi_size or bi_sectors), th 209 * updating bi_size or bi_sectors), then we don't need the inode update
209 * to be journalled - if we crash, the 210 * to be journalled - if we crash, the bi_journal_seq update will be
210 * lost, but that's fine. 211 * lost, but that's fine.
211 */ 212 */
212 unsigned inode_update_flags = BTREE_UP 213 unsigned inode_update_flags = BTREE_UPDATE_nojournal;
213 214
214 struct btree_iter iter; 215 struct btree_iter iter;
215 struct bkey_s_c k = bch2_bkey_get_iter 216 struct bkey_s_c k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_inodes,
216 SPOS(0, 217 SPOS(0,
217 extent_iter 218 extent_iter->pos.inode,
218 extent_iter 219 extent_iter->snapshot),
219 BTREE_ITER_cache 220 BTREE_ITER_cached);
220 int ret = bkey_err(k); 221 int ret = bkey_err(k);
221 if (unlikely(ret)) 222 if (unlikely(ret))
222 return ret; 223 return ret;
223 224
224 /* 225 /*
225 * varint_decode_fast(), in the inode 226 * varint_decode_fast(), in the inode .invalid method, reads up to 7
226 * bytes past the end of the buffer: 227 * bytes past the end of the buffer:
227 */ 228 */
228 struct bkey_i *k_mut = bch2_trans_kmal 229 struct bkey_i *k_mut = bch2_trans_kmalloc_nomemzero(trans, bkey_bytes(k.k) + 8);
229 ret = PTR_ERR_OR_ZERO(k_mut); 230 ret = PTR_ERR_OR_ZERO(k_mut);
230 if (unlikely(ret)) 231 if (unlikely(ret))
231 goto err; 232 goto err;
232 233
233 bkey_reassemble(k_mut, k); 234 bkey_reassemble(k_mut, k);
234 235
235 if (unlikely(k_mut->k.type != KEY_TYPE 236 if (unlikely(k_mut->k.type != KEY_TYPE_inode_v3)) {
236 k_mut = bch2_inode_to_v3(trans 237 k_mut = bch2_inode_to_v3(trans, k_mut);
237 ret = PTR_ERR_OR_ZERO(k_mut); 238 ret = PTR_ERR_OR_ZERO(k_mut);
238 if (unlikely(ret)) 239 if (unlikely(ret))
239 goto err; 240 goto err;
240 } 241 }
241 242
242 struct bkey_i_inode_v3 *inode = bkey_i 243 struct bkey_i_inode_v3 *inode = bkey_i_to_inode_v3(k_mut);
243 244
244 if (!(le64_to_cpu(inode->v.bi_flags) & 245 if (!(le64_to_cpu(inode->v.bi_flags) & BCH_INODE_i_size_dirty) &&
245 new_i_size > le64_to_cpu(inode->v. 246 new_i_size > le64_to_cpu(inode->v.bi_size)) {
246 inode->v.bi_size = cpu_to_le64 247 inode->v.bi_size = cpu_to_le64(new_i_size);
247 inode_update_flags = 0; 248 inode_update_flags = 0;
248 } 249 }
249 250
250 if (i_sectors_delta) { 251 if (i_sectors_delta) {
251 le64_add_cpu(&inode->v.bi_sect 252 le64_add_cpu(&inode->v.bi_sectors, i_sectors_delta);
252 inode_update_flags = 0; 253 inode_update_flags = 0;
253 } 254 }
254 255
255 if (inode->k.p.snapshot != iter.snapsh 256 if (inode->k.p.snapshot != iter.snapshot) {
256 inode->k.p.snapshot = iter.sna 257 inode->k.p.snapshot = iter.snapshot;
257 inode_update_flags = 0; 258 inode_update_flags = 0;
258 } 259 }
259 260
260 ret = bch2_trans_update(trans, &iter, 261 ret = bch2_trans_update(trans, &iter, &inode->k_i,
261 BTREE_UPDATE_i 262 BTREE_UPDATE_internal_snapshot_node|
262 inode_update_f 263 inode_update_flags);
263 err: 264 err:
264 bch2_trans_iter_exit(trans, &iter); 265 bch2_trans_iter_exit(trans, &iter);
265 return ret; 266 return ret;
266 } 267 }
267 268
268 int bch2_extent_update(struct btree_trans *tra 269 int bch2_extent_update(struct btree_trans *trans,
269 subvol_inum inum, 270 subvol_inum inum,
270 struct btree_iter *iter 271 struct btree_iter *iter,
271 struct bkey_i *k, 272 struct bkey_i *k,
272 struct disk_reservation 273 struct disk_reservation *disk_res,
273 u64 new_i_size, 274 u64 new_i_size,
274 s64 *i_sectors_delta_to 275 s64 *i_sectors_delta_total,
275 bool check_enospc) 276 bool check_enospc)
276 { 277 {
277 struct bpos next_pos; 278 struct bpos next_pos;
278 bool usage_increasing; 279 bool usage_increasing;
279 s64 i_sectors_delta = 0, disk_sectors_ 280 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
280 int ret; 281 int ret;
281 282
282 /* 283 /*
283 * This traverses us the iterator with 284 * This traverses us the iterator without changing iter->path->pos to
284 * search_key() (which is pos + 1 for 285 * search_key() (which is pos + 1 for extents): we want there to be a
285 * path already traversed at iter->pos 286 * path already traversed at iter->pos because
286 * bch2_trans_extent_update() will use 287 * bch2_trans_extent_update() will use it to attempt extent merging
287 */ 288 */
288 ret = __bch2_btree_iter_traverse(iter) 289 ret = __bch2_btree_iter_traverse(iter);
289 if (ret) 290 if (ret)
290 return ret; 291 return ret;
291 292
292 ret = bch2_extent_trim_atomic(trans, i 293 ret = bch2_extent_trim_atomic(trans, iter, k);
293 if (ret) 294 if (ret)
294 return ret; 295 return ret;
295 296
296 next_pos = k->k.p; 297 next_pos = k->k.p;
297 298
298 ret = bch2_sum_sector_overwrites(trans 299 ret = bch2_sum_sector_overwrites(trans, iter, k,
299 &usage_increasing, 300 &usage_increasing,
300 &i_sectors_delta, 301 &i_sectors_delta,
301 &disk_sectors_delta); 302 &disk_sectors_delta);
302 if (ret) 303 if (ret)
303 return ret; 304 return ret;
304 305
305 if (disk_res && 306 if (disk_res &&
306 disk_sectors_delta > (s64) disk_re 307 disk_sectors_delta > (s64) disk_res->sectors) {
307 ret = bch2_disk_reservation_ad 308 ret = bch2_disk_reservation_add(trans->c, disk_res,
308 disk_s 309 disk_sectors_delta - disk_res->sectors,
309 !check 310 !check_enospc || !usage_increasing
310 ? BCH_ 311 ? BCH_DISK_RESERVATION_NOFAIL : 0);
311 if (ret) 312 if (ret)
312 return ret; 313 return ret;
313 } 314 }
314 315
315 /* 316 /*
316 * Note: 317 * Note:
317 * We always have to do an inode updat 318 * We always have to do an inode update - even when i_size/i_sectors
318 * aren't changing - for fsync to work 319 * aren't changing - for fsync to work properly; fsync relies on
319 * inode->bi_journal_seq which is upda 320 * inode->bi_journal_seq which is updated by the trigger code:
320 */ 321 */
321 ret = bch2_extent_update_i_size_sect 322 ret = bch2_extent_update_i_size_sectors(trans, iter,
322 323 min(k->k.p.offset << 9, new_i_size),
323 324 i_sectors_delta) ?:
324 bch2_trans_update(trans, iter, 325 bch2_trans_update(trans, iter, k, 0) ?:
325 bch2_trans_commit(trans, disk_ 326 bch2_trans_commit(trans, disk_res, NULL,
326 BCH_TRANS_COMM 327 BCH_TRANS_COMMIT_no_check_rw|
327 BCH_TRANS_COMM 328 BCH_TRANS_COMMIT_no_enospc);
328 if (unlikely(ret)) 329 if (unlikely(ret))
329 return ret; 330 return ret;
330 331
331 if (i_sectors_delta_total) 332 if (i_sectors_delta_total)
332 *i_sectors_delta_total += i_se 333 *i_sectors_delta_total += i_sectors_delta;
333 bch2_btree_iter_set_pos(iter, next_pos 334 bch2_btree_iter_set_pos(iter, next_pos);
334 return 0; 335 return 0;
335 } 336 }
336 337
337 static int bch2_write_index_default(struct bch 338 static int bch2_write_index_default(struct bch_write_op *op)
338 { 339 {
339 struct bch_fs *c = op->c; 340 struct bch_fs *c = op->c;
340 struct bkey_buf sk; 341 struct bkey_buf sk;
341 struct keylist *keys = &op->insert_key 342 struct keylist *keys = &op->insert_keys;
342 struct bkey_i *k = bch2_keylist_front( 343 struct bkey_i *k = bch2_keylist_front(keys);
343 struct btree_trans *trans = bch2_trans 344 struct btree_trans *trans = bch2_trans_get(c);
344 struct btree_iter iter; 345 struct btree_iter iter;
345 subvol_inum inum = { 346 subvol_inum inum = {
346 .subvol = op->subvol, 347 .subvol = op->subvol,
347 .inum = k->k.p.inode, 348 .inum = k->k.p.inode,
348 }; 349 };
349 int ret; 350 int ret;
350 351
351 BUG_ON(!inum.subvol); 352 BUG_ON(!inum.subvol);
352 353
353 bch2_bkey_buf_init(&sk); 354 bch2_bkey_buf_init(&sk);
354 355
355 do { 356 do {
356 bch2_trans_begin(trans); 357 bch2_trans_begin(trans);
357 358
358 k = bch2_keylist_front(keys); 359 k = bch2_keylist_front(keys);
359 bch2_bkey_buf_copy(&sk, c, k); 360 bch2_bkey_buf_copy(&sk, c, k);
360 361
361 ret = bch2_subvolume_get_snaps 362 ret = bch2_subvolume_get_snapshot(trans, inum.subvol,
362 363 &sk.k->k.p.snapshot);
363 if (bch2_err_matches(ret, BCH_ 364 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
364 continue; 365 continue;
365 if (ret) 366 if (ret)
366 break; 367 break;
367 368
368 bch2_trans_iter_init(trans, &i 369 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
369 bkey_star 370 bkey_start_pos(&sk.k->k),
370 BTREE_ITE 371 BTREE_ITER_slots|BTREE_ITER_intent);
371 372
372 ret = bch2_bkey_set_needs_re 373 ret = bch2_bkey_set_needs_rebalance(c, sk.k, &op->opts) ?:
373 bch2_extent_update(tra 374 bch2_extent_update(trans, inum, &iter, sk.k,
374 &op->r 375 &op->res,
375 op->ne 376 op->new_i_size, &op->i_sectors_delta,
376 op->fl 377 op->flags & BCH_WRITE_CHECK_ENOSPC);
377 bch2_trans_iter_exit(trans, &i 378 bch2_trans_iter_exit(trans, &iter);
378 379
379 if (bch2_err_matches(ret, BCH_ 380 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
380 continue; 381 continue;
381 if (ret) 382 if (ret)
382 break; 383 break;
383 384
384 if (bkey_ge(iter.pos, k->k.p)) 385 if (bkey_ge(iter.pos, k->k.p))
385 bch2_keylist_pop_front 386 bch2_keylist_pop_front(&op->insert_keys);
386 else 387 else
387 bch2_cut_front(iter.po 388 bch2_cut_front(iter.pos, k);
388 } while (!bch2_keylist_empty(keys)); 389 } while (!bch2_keylist_empty(keys));
389 390
390 bch2_trans_put(trans); 391 bch2_trans_put(trans);
391 bch2_bkey_buf_exit(&sk, c); 392 bch2_bkey_buf_exit(&sk, c);
392 393
393 return ret; 394 return ret;
394 } 395 }
395 396
396 /* Writes */ 397 /* Writes */
397 398
398 void bch2_submit_wbio_replicas(struct bch_writ 399 void bch2_submit_wbio_replicas(struct bch_write_bio *wbio, struct bch_fs *c,
399 enum bch_data_t 400 enum bch_data_type type,
400 const struct bk 401 const struct bkey_i *k,
401 bool nocow) 402 bool nocow)
402 { 403 {
403 struct bkey_ptrs_c ptrs = bch2_bkey_pt 404 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(bkey_i_to_s_c(k));
404 struct bch_write_bio *n; 405 struct bch_write_bio *n;
405 406
406 BUG_ON(c->opts.nochanges); 407 BUG_ON(c->opts.nochanges);
407 408
408 bkey_for_each_ptr(ptrs, ptr) { 409 bkey_for_each_ptr(ptrs, ptr) {
409 struct bch_dev *ca = nocow 410 struct bch_dev *ca = nocow
410 ? bch2_dev_have_ref(c, 411 ? bch2_dev_have_ref(c, ptr->dev)
411 : bch2_dev_get_ioref(c 412 : bch2_dev_get_ioref(c, ptr->dev, type == BCH_DATA_btree ? READ : WRITE);
412 413
413 if (to_entry(ptr + 1) < ptrs.e 414 if (to_entry(ptr + 1) < ptrs.end) {
414 n = to_wbio(bio_alloc_ 415 n = to_wbio(bio_alloc_clone(NULL, &wbio->bio, GFP_NOFS, &c->replica_set));
415 416
416 n->bio.bi_end_io 417 n->bio.bi_end_io = wbio->bio.bi_end_io;
417 n->bio.bi_private 418 n->bio.bi_private = wbio->bio.bi_private;
418 n->parent 419 n->parent = wbio;
419 n->split 420 n->split = true;
420 n->bounce 421 n->bounce = false;
421 n->put_bio 422 n->put_bio = true;
422 n->bio.bi_opf 423 n->bio.bi_opf = wbio->bio.bi_opf;
423 bio_inc_remaining(&wbi 424 bio_inc_remaining(&wbio->bio);
424 } else { 425 } else {
425 n = wbio; 426 n = wbio;
426 n->split 427 n->split = false;
427 } 428 }
428 429
429 n->c = c; 430 n->c = c;
430 n->dev = ptr- 431 n->dev = ptr->dev;
431 n->have_ioref = ca ! 432 n->have_ioref = ca != NULL;
432 n->nocow = noco 433 n->nocow = nocow;
433 n->submit_time = loca 434 n->submit_time = local_clock();
434 n->inode_offset = bkey 435 n->inode_offset = bkey_start_offset(&k->k);
435 if (nocow) 436 if (nocow)
436 n->nocow_bucket = PTR_ 437 n->nocow_bucket = PTR_BUCKET_NR(ca, ptr);
437 n->bio.bi_iter.bi_sector = ptr 438 n->bio.bi_iter.bi_sector = ptr->offset;
438 439
439 if (likely(n->have_ioref)) { 440 if (likely(n->have_ioref)) {
440 this_cpu_add(ca->io_do 441 this_cpu_add(ca->io_done->sectors[WRITE][type],
441 bio_secto 442 bio_sectors(&n->bio));
442 443
443 bio_set_dev(&n->bio, c 444 bio_set_dev(&n->bio, ca->disk_sb.bdev);
444 445
445 if (type != BCH_DATA_b 446 if (type != BCH_DATA_btree && unlikely(c->opts.no_data_io)) {
446 bio_endio(&n-> 447 bio_endio(&n->bio);
447 continue; 448 continue;
448 } 449 }
449 450
450 submit_bio(&n->bio); 451 submit_bio(&n->bio);
451 } else { 452 } else {
452 n->bio.bi_status 453 n->bio.bi_status = BLK_STS_REMOVED;
453 bio_endio(&n->bio); 454 bio_endio(&n->bio);
454 } 455 }
455 } 456 }
456 } 457 }
457 458
458 static void __bch2_write(struct bch_write_op * 459 static void __bch2_write(struct bch_write_op *);
459 460
460 static void bch2_write_done(struct closure *cl 461 static void bch2_write_done(struct closure *cl)
461 { 462 {
462 struct bch_write_op *op = container_of 463 struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
463 struct bch_fs *c = op->c; 464 struct bch_fs *c = op->c;
464 465
465 EBUG_ON(op->open_buckets.nr); 466 EBUG_ON(op->open_buckets.nr);
466 467
467 bch2_time_stats_update(&c->times[BCH_T 468 bch2_time_stats_update(&c->times[BCH_TIME_data_write], op->start_time);
468 bch2_disk_reservation_put(c, &op->res) 469 bch2_disk_reservation_put(c, &op->res);
469 470
470 if (!(op->flags & BCH_WRITE_MOVE)) 471 if (!(op->flags & BCH_WRITE_MOVE))
471 bch2_write_ref_put(c, BCH_WRIT 472 bch2_write_ref_put(c, BCH_WRITE_REF_write);
472 bch2_keylist_free(&op->insert_keys, op 473 bch2_keylist_free(&op->insert_keys, op->inline_keys);
473 474
474 EBUG_ON(cl->parent); 475 EBUG_ON(cl->parent);
475 closure_debug_destroy(cl); 476 closure_debug_destroy(cl);
476 if (op->end_io) 477 if (op->end_io)
477 op->end_io(op); 478 op->end_io(op);
478 } 479 }
479 480
480 static noinline int bch2_write_drop_io_error_p 481 static noinline int bch2_write_drop_io_error_ptrs(struct bch_write_op *op)
481 { 482 {
482 struct keylist *keys = &op->insert_key 483 struct keylist *keys = &op->insert_keys;
483 struct bkey_i *src, *dst = keys->keys, 484 struct bkey_i *src, *dst = keys->keys, *n;
484 485
485 for (src = keys->keys; src != keys->to 486 for (src = keys->keys; src != keys->top; src = n) {
486 n = bkey_next(src); 487 n = bkey_next(src);
487 488
488 if (bkey_extent_is_direct_data 489 if (bkey_extent_is_direct_data(&src->k)) {
489 bch2_bkey_drop_ptrs(bk 490 bch2_bkey_drop_ptrs(bkey_i_to_s(src), ptr,
490 te 491 test_bit(ptr->dev, op->failed.d));
491 492
492 if (!bch2_bkey_nr_ptrs 493 if (!bch2_bkey_nr_ptrs(bkey_i_to_s_c(src)))
493 return -EIO; 494 return -EIO;
494 } 495 }
495 496
496 if (dst != src) 497 if (dst != src)
497 memmove_u64s_down(dst, 498 memmove_u64s_down(dst, src, src->k.u64s);
498 dst = bkey_next(dst); 499 dst = bkey_next(dst);
499 } 500 }
500 501
501 keys->top = dst; 502 keys->top = dst;
502 return 0; 503 return 0;
503 } 504 }
504 505
505 /** 506 /**
506 * __bch2_write_index - after a write, update 507 * __bch2_write_index - after a write, update index to point to new data
507 * @op: bch_write_op to process 508 * @op: bch_write_op to process
508 */ 509 */
509 static void __bch2_write_index(struct bch_writ 510 static void __bch2_write_index(struct bch_write_op *op)
510 { 511 {
511 struct bch_fs *c = op->c; 512 struct bch_fs *c = op->c;
512 struct keylist *keys = &op->insert_key 513 struct keylist *keys = &op->insert_keys;
513 unsigned dev; 514 unsigned dev;
514 int ret = 0; 515 int ret = 0;
515 516
516 if (unlikely(op->flags & BCH_WRITE_IO_ 517 if (unlikely(op->flags & BCH_WRITE_IO_ERROR)) {
517 ret = bch2_write_drop_io_error 518 ret = bch2_write_drop_io_error_ptrs(op);
518 if (ret) 519 if (ret)
519 goto err; 520 goto err;
520 } 521 }
521 522
522 if (!bch2_keylist_empty(keys)) { 523 if (!bch2_keylist_empty(keys)) {
523 u64 sectors_start = keylist_se 524 u64 sectors_start = keylist_sectors(keys);
524 525
525 ret = !(op->flags & BCH_WRITE_ 526 ret = !(op->flags & BCH_WRITE_MOVE)
526 ? bch2_write_index_def 527 ? bch2_write_index_default(op)
527 : bch2_data_update_ind 528 : bch2_data_update_index_update(op);
528 529
529 BUG_ON(bch2_err_matches(ret, B 530 BUG_ON(bch2_err_matches(ret, BCH_ERR_transaction_restart));
530 BUG_ON(keylist_sectors(keys) & 531 BUG_ON(keylist_sectors(keys) && !ret);
531 532
532 op->written += sectors_start - 533 op->written += sectors_start - keylist_sectors(keys);
533 534
534 if (ret && !bch2_err_matches(r 535 if (ret && !bch2_err_matches(ret, EROFS)) {
535 struct bkey_i *insert 536 struct bkey_i *insert = bch2_keylist_front(&op->insert_keys);
536 537
537 bch_err_inum_offset_ra 538 bch_err_inum_offset_ratelimited(c,
538 insert->k.p.in 539 insert->k.p.inode, insert->k.p.offset << 9,
539 "%s write erro 540 "%s write error while doing btree update: %s",
540 op->flags & BC 541 op->flags & BCH_WRITE_MOVE ? "move" : "user",
541 bch2_err_str(r 542 bch2_err_str(ret));
542 } 543 }
543 544
544 if (ret) 545 if (ret)
545 goto err; 546 goto err;
546 } 547 }
547 out: 548 out:
548 /* If some a bucket wasn't written, we 549 /* If some a bucket wasn't written, we can't erasure code it: */
549 for_each_set_bit(dev, op->failed.d, BC 550 for_each_set_bit(dev, op->failed.d, BCH_SB_MEMBERS_MAX)
550 bch2_open_bucket_write_error(c 551 bch2_open_bucket_write_error(c, &op->open_buckets, dev);
551 552
552 bch2_open_buckets_put(c, &op->open_buc 553 bch2_open_buckets_put(c, &op->open_buckets);
553 return; 554 return;
554 err: 555 err:
555 keys->top = keys->keys; 556 keys->top = keys->keys;
556 op->error = ret; 557 op->error = ret;
557 op->flags |= BCH_WRITE_SUBMITTED; !! 558 op->flags |= BCH_WRITE_DONE;
558 goto out; 559 goto out;
559 } 560 }
560 561
561 static inline void __wp_update_state(struct wr 562 static inline void __wp_update_state(struct write_point *wp, enum write_point_state state)
562 { 563 {
563 if (state != wp->state) { 564 if (state != wp->state) {
564 u64 now = ktime_get_ns(); 565 u64 now = ktime_get_ns();
565 566
566 if (wp->last_state_change && 567 if (wp->last_state_change &&
567 time_after64(now, wp->last 568 time_after64(now, wp->last_state_change))
568 wp->time[wp->state] += 569 wp->time[wp->state] += now - wp->last_state_change;
569 wp->state = state; 570 wp->state = state;
570 wp->last_state_change = now; 571 wp->last_state_change = now;
571 } 572 }
572 } 573 }
573 574
574 static inline void wp_update_state(struct writ 575 static inline void wp_update_state(struct write_point *wp, bool running)
575 { 576 {
576 enum write_point_state state; 577 enum write_point_state state;
577 578
578 state = running ? WRI 579 state = running ? WRITE_POINT_running :
579 !list_empty(&wp->writes) ? WRI 580 !list_empty(&wp->writes) ? WRITE_POINT_waiting_io
580 : WRI 581 : WRITE_POINT_stopped;
581 582
582 __wp_update_state(wp, state); 583 __wp_update_state(wp, state);
583 } 584 }
584 585
585 static CLOSURE_CALLBACK(bch2_write_index) 586 static CLOSURE_CALLBACK(bch2_write_index)
586 { 587 {
587 closure_type(op, struct bch_write_op, 588 closure_type(op, struct bch_write_op, cl);
588 struct write_point *wp = op->wp; 589 struct write_point *wp = op->wp;
589 struct workqueue_struct *wq = index_up 590 struct workqueue_struct *wq = index_update_wq(op);
590 unsigned long flags; 591 unsigned long flags;
591 592
592 if ((op->flags & BCH_WRITE_SUBMITTED) !! 593 if ((op->flags & BCH_WRITE_DONE) &&
593 (op->flags & BCH_WRITE_MOVE)) 594 (op->flags & BCH_WRITE_MOVE))
594 bch2_bio_free_pages_pool(op->c 595 bch2_bio_free_pages_pool(op->c, &op->wbio.bio);
595 596
596 spin_lock_irqsave(&wp->writes_lock, fl 597 spin_lock_irqsave(&wp->writes_lock, flags);
597 if (wp->state == WRITE_POINT_waiting_i 598 if (wp->state == WRITE_POINT_waiting_io)
598 __wp_update_state(wp, WRITE_PO 599 __wp_update_state(wp, WRITE_POINT_waiting_work);
599 list_add_tail(&op->wp_list, &wp->write 600 list_add_tail(&op->wp_list, &wp->writes);
600 spin_unlock_irqrestore (&wp->writes_lo 601 spin_unlock_irqrestore (&wp->writes_lock, flags);
601 602
602 queue_work(wq, &wp->index_update_work) 603 queue_work(wq, &wp->index_update_work);
603 } 604 }
604 605
605 static inline void bch2_write_queue(struct bch 606 static inline void bch2_write_queue(struct bch_write_op *op, struct write_point *wp)
606 { 607 {
607 op->wp = wp; 608 op->wp = wp;
608 609
609 if (wp->state == WRITE_POINT_stopped) 610 if (wp->state == WRITE_POINT_stopped) {
610 spin_lock_irq(&wp->writes_lock 611 spin_lock_irq(&wp->writes_lock);
611 __wp_update_state(wp, WRITE_PO 612 __wp_update_state(wp, WRITE_POINT_waiting_io);
612 spin_unlock_irq(&wp->writes_lo 613 spin_unlock_irq(&wp->writes_lock);
613 } 614 }
614 } 615 }
615 616
616 void bch2_write_point_do_index_updates(struct 617 void bch2_write_point_do_index_updates(struct work_struct *work)
617 { 618 {
618 struct write_point *wp = 619 struct write_point *wp =
619 container_of(work, struct writ 620 container_of(work, struct write_point, index_update_work);
620 struct bch_write_op *op; 621 struct bch_write_op *op;
621 622
622 while (1) { 623 while (1) {
623 spin_lock_irq(&wp->writes_lock 624 spin_lock_irq(&wp->writes_lock);
624 op = list_first_entry_or_null( 625 op = list_first_entry_or_null(&wp->writes, struct bch_write_op, wp_list);
625 if (op) 626 if (op)
626 list_del(&op->wp_list) 627 list_del(&op->wp_list);
627 wp_update_state(wp, op != NULL 628 wp_update_state(wp, op != NULL);
628 spin_unlock_irq(&wp->writes_lo 629 spin_unlock_irq(&wp->writes_lock);
629 630
630 if (!op) 631 if (!op)
631 break; 632 break;
632 633
633 op->flags |= BCH_WRITE_IN_WORK 634 op->flags |= BCH_WRITE_IN_WORKER;
634 635
635 __bch2_write_index(op); 636 __bch2_write_index(op);
636 637
637 if (!(op->flags & BCH_WRITE_SU !! 638 if (!(op->flags & BCH_WRITE_DONE))
638 __bch2_write(op); 639 __bch2_write(op);
639 else 640 else
640 bch2_write_done(&op->c 641 bch2_write_done(&op->cl);
641 } 642 }
642 } 643 }
643 644
644 static void bch2_write_endio(struct bio *bio) 645 static void bch2_write_endio(struct bio *bio)
645 { 646 {
646 struct closure *cl = bio- 647 struct closure *cl = bio->bi_private;
647 struct bch_write_op *op = cont 648 struct bch_write_op *op = container_of(cl, struct bch_write_op, cl);
648 struct bch_write_bio *wbio = to_w 649 struct bch_write_bio *wbio = to_wbio(bio);
649 struct bch_write_bio *parent = wbio 650 struct bch_write_bio *parent = wbio->split ? wbio->parent : NULL;
650 struct bch_fs *c = wbio 651 struct bch_fs *c = wbio->c;
651 struct bch_dev *ca = wbio 652 struct bch_dev *ca = wbio->have_ioref
652 ? bch2_dev_have_ref(c, wbio->d 653 ? bch2_dev_have_ref(c, wbio->dev)
653 : NULL; 654 : NULL;
654 655
655 if (bch2_dev_inum_io_err_on(bio->bi_st 656 if (bch2_dev_inum_io_err_on(bio->bi_status, ca, BCH_MEMBER_ERROR_write,
656 op->pos.in 657 op->pos.inode,
657 wbio->inod 658 wbio->inode_offset << 9,
658 "data writ 659 "data write error: %s",
659 bch2_blk_s 660 bch2_blk_status_to_str(bio->bi_status))) {
660 set_bit(wbio->dev, op->failed. 661 set_bit(wbio->dev, op->failed.d);
661 op->flags |= BCH_WRITE_IO_ERRO 662 op->flags |= BCH_WRITE_IO_ERROR;
662 } 663 }
663 664
664 if (wbio->nocow) { 665 if (wbio->nocow) {
665 bch2_bucket_nocow_unlock(&c->n 666 bch2_bucket_nocow_unlock(&c->nocow_locks,
666 POS(c 667 POS(ca->dev_idx, wbio->nocow_bucket),
667 BUCKE 668 BUCKET_NOCOW_LOCK_UPDATE);
668 set_bit(wbio->dev, op->devs_ne 669 set_bit(wbio->dev, op->devs_need_flush->d);
669 } 670 }
670 671
671 if (wbio->have_ioref) { 672 if (wbio->have_ioref) {
672 bch2_latency_acct(ca, wbio->su 673 bch2_latency_acct(ca, wbio->submit_time, WRITE);
673 percpu_ref_put(&ca->io_ref); 674 percpu_ref_put(&ca->io_ref);
674 } 675 }
675 676
676 if (wbio->bounce) 677 if (wbio->bounce)
677 bch2_bio_free_pages_pool(c, bi 678 bch2_bio_free_pages_pool(c, bio);
678 679
679 if (wbio->put_bio) 680 if (wbio->put_bio)
680 bio_put(bio); 681 bio_put(bio);
681 682
682 if (parent) 683 if (parent)
683 bio_endio(&parent->bio); 684 bio_endio(&parent->bio);
684 else 685 else
685 closure_put(cl); 686 closure_put(cl);
686 } 687 }
687 688
688 static void init_append_extent(struct bch_writ 689 static void init_append_extent(struct bch_write_op *op,
689 struct write_po 690 struct write_point *wp,
690 struct bversion 691 struct bversion version,
691 struct bch_exte 692 struct bch_extent_crc_unpacked crc)
692 { 693 {
693 struct bkey_i_extent *e; 694 struct bkey_i_extent *e;
694 695
695 op->pos.offset += crc.uncompressed_siz 696 op->pos.offset += crc.uncompressed_size;
696 697
697 e = bkey_extent_init(op->insert_keys.t 698 e = bkey_extent_init(op->insert_keys.top);
698 e->k.p = op->pos; 699 e->k.p = op->pos;
699 e->k.size = crc.uncompressed_siz 700 e->k.size = crc.uncompressed_size;
700 e->k.bversion = version; !! 701 e->k.version = version;
701 702
702 if (crc.csum_type || 703 if (crc.csum_type ||
703 crc.compression_type || 704 crc.compression_type ||
704 crc.nonce) 705 crc.nonce)
705 bch2_extent_crc_append(&e->k_i 706 bch2_extent_crc_append(&e->k_i, crc);
706 707
707 bch2_alloc_sectors_append_ptrs_inlined 708 bch2_alloc_sectors_append_ptrs_inlined(op->c, wp, &e->k_i, crc.compressed_size,
708 op->fla 709 op->flags & BCH_WRITE_CACHED);
709 710
710 bch2_keylist_push(&op->insert_keys); 711 bch2_keylist_push(&op->insert_keys);
711 } 712 }
712 713
713 static struct bio *bch2_write_bio_alloc(struct 714 static struct bio *bch2_write_bio_alloc(struct bch_fs *c,
714 struct 715 struct write_point *wp,
715 struct 716 struct bio *src,
716 bool * 717 bool *page_alloc_failed,
717 void * 718 void *buf)
718 { 719 {
719 struct bch_write_bio *wbio; 720 struct bch_write_bio *wbio;
720 struct bio *bio; 721 struct bio *bio;
721 unsigned output_available = 722 unsigned output_available =
722 min(wp->sectors_free << 9, src 723 min(wp->sectors_free << 9, src->bi_iter.bi_size);
723 unsigned pages = DIV_ROUND_UP(output_a 724 unsigned pages = DIV_ROUND_UP(output_available +
724 (buf 725 (buf
725 ? ((uns 726 ? ((unsigned long) buf & (PAGE_SIZE - 1))
726 : 0), P 727 : 0), PAGE_SIZE);
727 728
728 pages = min(pages, BIO_MAX_VECS); 729 pages = min(pages, BIO_MAX_VECS);
729 730
730 bio = bio_alloc_bioset(NULL, pages, 0, 731 bio = bio_alloc_bioset(NULL, pages, 0,
731 GFP_NOFS, &c->b 732 GFP_NOFS, &c->bio_write);
732 wbio = wbio_init(bi 733 wbio = wbio_init(bio);
733 wbio->put_bio = true; 734 wbio->put_bio = true;
734 /* copy WRITE_SYNC flag */ 735 /* copy WRITE_SYNC flag */
735 wbio->bio.bi_opf = src->bi_opf; 736 wbio->bio.bi_opf = src->bi_opf;
736 737
737 if (buf) { 738 if (buf) {
738 bch2_bio_map(bio, buf, output_ 739 bch2_bio_map(bio, buf, output_available);
739 return bio; 740 return bio;
740 } 741 }
741 742
742 wbio->bounce = true; 743 wbio->bounce = true;
743 744
744 /* 745 /*
745 * We can't use mempool for more than 746 * We can't use mempool for more than c->sb.encoded_extent_max
746 * worth of pages, but we'd like to al 747 * worth of pages, but we'd like to allocate more if we can:
747 */ 748 */
748 bch2_bio_alloc_pages_pool(c, bio, 749 bch2_bio_alloc_pages_pool(c, bio,
749 min_t(unsign 750 min_t(unsigned, output_available,
750 c->opt 751 c->opts.encoded_extent_max));
751 752
752 if (bio->bi_iter.bi_size < output_avai 753 if (bio->bi_iter.bi_size < output_available)
753 *page_alloc_failed = 754 *page_alloc_failed =
754 bch2_bio_alloc_pages(b 755 bch2_bio_alloc_pages(bio,
755 o 756 output_available -
756 b 757 bio->bi_iter.bi_size,
757 G 758 GFP_NOFS) != 0;
758 759
759 return bio; 760 return bio;
760 } 761 }
761 762
762 static int bch2_write_rechecksum(struct bch_fs 763 static int bch2_write_rechecksum(struct bch_fs *c,
763 struct bch_wr 764 struct bch_write_op *op,
764 unsigned new_ 765 unsigned new_csum_type)
765 { 766 {
766 struct bio *bio = &op->wbio.bio; 767 struct bio *bio = &op->wbio.bio;
767 struct bch_extent_crc_unpacked new_crc 768 struct bch_extent_crc_unpacked new_crc;
768 int ret; 769 int ret;
769 770
770 /* bch2_rechecksum_bio() can't encrypt 771 /* bch2_rechecksum_bio() can't encrypt or decrypt data: */
771 772
772 if (bch2_csum_type_is_encryption(op->c 773 if (bch2_csum_type_is_encryption(op->crc.csum_type) !=
773 bch2_csum_type_is_encryption(new_c 774 bch2_csum_type_is_encryption(new_csum_type))
774 new_csum_type = op->crc.csum_t 775 new_csum_type = op->crc.csum_type;
775 776
776 ret = bch2_rechecksum_bio(c, bio, op-> 777 ret = bch2_rechecksum_bio(c, bio, op->version, op->crc,
777 NULL, &new_c 778 NULL, &new_crc,
778 op->crc.offs 779 op->crc.offset, op->crc.live_size,
779 new_csum_typ 780 new_csum_type);
780 if (ret) 781 if (ret)
781 return ret; 782 return ret;
782 783
783 bio_advance(bio, op->crc.offset << 9); 784 bio_advance(bio, op->crc.offset << 9);
784 bio->bi_iter.bi_size = op->crc.live_si 785 bio->bi_iter.bi_size = op->crc.live_size << 9;
785 op->crc = new_crc; 786 op->crc = new_crc;
786 return 0; 787 return 0;
787 } 788 }
788 789
789 static int bch2_write_decrypt(struct bch_write 790 static int bch2_write_decrypt(struct bch_write_op *op)
790 { 791 {
791 struct bch_fs *c = op->c; 792 struct bch_fs *c = op->c;
792 struct nonce nonce = extent_nonce(op-> 793 struct nonce nonce = extent_nonce(op->version, op->crc);
793 struct bch_csum csum; 794 struct bch_csum csum;
794 int ret; 795 int ret;
795 796
796 if (!bch2_csum_type_is_encryption(op-> 797 if (!bch2_csum_type_is_encryption(op->crc.csum_type))
797 return 0; 798 return 0;
798 799
799 /* 800 /*
800 * If we need to decrypt data in the w 801 * If we need to decrypt data in the write path, we'll no longer be able
801 * to verify the existing checksum (po 802 * to verify the existing checksum (poly1305 mac, in this case) after
802 * it's decrypted - this is the last p 803 * it's decrypted - this is the last point we'll be able to reverify the
803 * checksum: 804 * checksum:
804 */ 805 */
805 csum = bch2_checksum_bio(c, op->crc.cs 806 csum = bch2_checksum_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
806 if (bch2_crc_cmp(op->crc.csum, csum) & 807 if (bch2_crc_cmp(op->crc.csum, csum) && !c->opts.no_data_io)
807 return -EIO; 808 return -EIO;
808 809
809 ret = bch2_encrypt_bio(c, op->crc.csum 810 ret = bch2_encrypt_bio(c, op->crc.csum_type, nonce, &op->wbio.bio);
810 op->crc.csum_type = 0; 811 op->crc.csum_type = 0;
811 op->crc.csum = (struct bch_csum) { 0, 812 op->crc.csum = (struct bch_csum) { 0, 0 };
812 return ret; 813 return ret;
813 } 814 }
814 815
815 static enum prep_encoded_ret { 816 static enum prep_encoded_ret {
816 PREP_ENCODED_OK, 817 PREP_ENCODED_OK,
817 PREP_ENCODED_ERR, 818 PREP_ENCODED_ERR,
818 PREP_ENCODED_CHECKSUM_ERR, 819 PREP_ENCODED_CHECKSUM_ERR,
819 PREP_ENCODED_DO_WRITE, 820 PREP_ENCODED_DO_WRITE,
820 } bch2_write_prep_encoded_data(struct bch_writ 821 } bch2_write_prep_encoded_data(struct bch_write_op *op, struct write_point *wp)
821 { 822 {
822 struct bch_fs *c = op->c; 823 struct bch_fs *c = op->c;
823 struct bio *bio = &op->wbio.bio; 824 struct bio *bio = &op->wbio.bio;
824 825
825 if (!(op->flags & BCH_WRITE_DATA_ENCOD 826 if (!(op->flags & BCH_WRITE_DATA_ENCODED))
826 return PREP_ENCODED_OK; 827 return PREP_ENCODED_OK;
827 828
828 BUG_ON(bio_sectors(bio) != op->crc.com 829 BUG_ON(bio_sectors(bio) != op->crc.compressed_size);
829 830
830 /* Can we just write the entire extent 831 /* Can we just write the entire extent as is? */
831 if (op->crc.uncompressed_size == op->c 832 if (op->crc.uncompressed_size == op->crc.live_size &&
832 op->crc.uncompressed_size <= c->op 833 op->crc.uncompressed_size <= c->opts.encoded_extent_max >> 9 &&
833 op->crc.compressed_size <= wp->sec 834 op->crc.compressed_size <= wp->sectors_free &&
834 (op->crc.compression_type == bch2_ 835 (op->crc.compression_type == bch2_compression_opt_to_type(op->compression_opt) ||
835 op->incompressible)) { 836 op->incompressible)) {
836 if (!crc_is_compressed(op->crc 837 if (!crc_is_compressed(op->crc) &&
837 op->csum_type != op->crc.c 838 op->csum_type != op->crc.csum_type &&
838 bch2_write_rechecksum(c, o 839 bch2_write_rechecksum(c, op, op->csum_type) &&
839 !c->opts.no_data_io) 840 !c->opts.no_data_io)
840 return PREP_ENCODED_CH 841 return PREP_ENCODED_CHECKSUM_ERR;
841 842
842 return PREP_ENCODED_DO_WRITE; 843 return PREP_ENCODED_DO_WRITE;
843 } 844 }
844 845
845 /* 846 /*
846 * If the data is compressed and we co 847 * If the data is compressed and we couldn't write the entire extent as
847 * is, we have to decompress it: 848 * is, we have to decompress it:
848 */ 849 */
849 if (crc_is_compressed(op->crc)) { 850 if (crc_is_compressed(op->crc)) {
850 struct bch_csum csum; 851 struct bch_csum csum;
851 852
852 if (bch2_write_decrypt(op)) 853 if (bch2_write_decrypt(op))
853 return PREP_ENCODED_CH 854 return PREP_ENCODED_CHECKSUM_ERR;
854 855
855 /* Last point we can still ver 856 /* Last point we can still verify checksum: */
856 csum = bch2_checksum_bio(c, op 857 csum = bch2_checksum_bio(c, op->crc.csum_type,
857 exten 858 extent_nonce(op->version, op->crc),
858 bio); 859 bio);
859 if (bch2_crc_cmp(op->crc.csum, 860 if (bch2_crc_cmp(op->crc.csum, csum) && !c->opts.no_data_io)
860 return PREP_ENCODED_CH 861 return PREP_ENCODED_CHECKSUM_ERR;
861 862
862 if (bch2_bio_uncompress_inplac 863 if (bch2_bio_uncompress_inplace(c, bio, &op->crc))
863 return PREP_ENCODED_ER 864 return PREP_ENCODED_ERR;
864 } 865 }
865 866
866 /* 867 /*
867 * No longer have compressed data afte 868 * No longer have compressed data after this point - data might be
868 * encrypted: 869 * encrypted:
869 */ 870 */
870 871
871 /* 872 /*
872 * If the data is checksummed and we'r 873 * If the data is checksummed and we're only writing a subset,
873 * rechecksum and adjust bio to point 874 * rechecksum and adjust bio to point to currently live data:
874 */ 875 */
875 if ((op->crc.live_size != op->crc.unco 876 if ((op->crc.live_size != op->crc.uncompressed_size ||
876 op->crc.csum_type != op->csum_typ 877 op->crc.csum_type != op->csum_type) &&
877 bch2_write_rechecksum(c, op, op->c 878 bch2_write_rechecksum(c, op, op->csum_type) &&
878 !c->opts.no_data_io) 879 !c->opts.no_data_io)
879 return PREP_ENCODED_CHECKSUM_E 880 return PREP_ENCODED_CHECKSUM_ERR;
880 881
881 /* 882 /*
882 * If we want to compress the data, it 883 * If we want to compress the data, it has to be decrypted:
883 */ 884 */
884 if ((op->compression_opt || 885 if ((op->compression_opt ||
885 bch2_csum_type_is_encryption(op-> 886 bch2_csum_type_is_encryption(op->crc.csum_type) !=
886 bch2_csum_type_is_encryption(op-> 887 bch2_csum_type_is_encryption(op->csum_type)) &&
887 bch2_write_decrypt(op)) 888 bch2_write_decrypt(op))
888 return PREP_ENCODED_CHECKSUM_E 889 return PREP_ENCODED_CHECKSUM_ERR;
889 890
890 return PREP_ENCODED_OK; 891 return PREP_ENCODED_OK;
891 } 892 }
892 893
893 static int bch2_write_extent(struct bch_write_ 894 static int bch2_write_extent(struct bch_write_op *op, struct write_point *wp,
894 struct bio **_dst 895 struct bio **_dst)
895 { 896 {
896 struct bch_fs *c = op->c; 897 struct bch_fs *c = op->c;
897 struct bio *src = &op->wbio.bio, *dst 898 struct bio *src = &op->wbio.bio, *dst = src;
898 struct bvec_iter saved_iter; 899 struct bvec_iter saved_iter;
899 void *ec_buf; 900 void *ec_buf;
900 unsigned total_output = 0, total_input 901 unsigned total_output = 0, total_input = 0;
901 bool bounce = false; 902 bool bounce = false;
902 bool page_alloc_failed = false; 903 bool page_alloc_failed = false;
903 int ret, more = 0; 904 int ret, more = 0;
904 905
905 BUG_ON(!bio_sectors(src)); 906 BUG_ON(!bio_sectors(src));
906 907
907 ec_buf = bch2_writepoint_ec_buf(c, wp) 908 ec_buf = bch2_writepoint_ec_buf(c, wp);
908 909
909 switch (bch2_write_prep_encoded_data(o 910 switch (bch2_write_prep_encoded_data(op, wp)) {
910 case PREP_ENCODED_OK: 911 case PREP_ENCODED_OK:
911 break; 912 break;
912 case PREP_ENCODED_ERR: 913 case PREP_ENCODED_ERR:
913 ret = -EIO; 914 ret = -EIO;
914 goto err; 915 goto err;
915 case PREP_ENCODED_CHECKSUM_ERR: 916 case PREP_ENCODED_CHECKSUM_ERR:
916 goto csum_err; 917 goto csum_err;
917 case PREP_ENCODED_DO_WRITE: 918 case PREP_ENCODED_DO_WRITE:
918 /* XXX look for bug here */ 919 /* XXX look for bug here */
919 if (ec_buf) { 920 if (ec_buf) {
920 dst = bch2_write_bio_a 921 dst = bch2_write_bio_alloc(c, wp, src,
921 922 &page_alloc_failed,
922 923 ec_buf);
923 bio_copy_data(dst, src 924 bio_copy_data(dst, src);
924 bounce = true; 925 bounce = true;
925 } 926 }
926 init_append_extent(op, wp, op- 927 init_append_extent(op, wp, op->version, op->crc);
927 goto do_write; 928 goto do_write;
928 } 929 }
929 930
930 if (ec_buf || 931 if (ec_buf ||
931 op->compression_opt || 932 op->compression_opt ||
932 (op->csum_type && 933 (op->csum_type &&
933 !(op->flags & BCH_WRITE_PAGES_STA 934 !(op->flags & BCH_WRITE_PAGES_STABLE)) ||
934 (bch2_csum_type_is_encryption(op-> 935 (bch2_csum_type_is_encryption(op->csum_type) &&
935 !(op->flags & BCH_WRITE_PAGES_OWN 936 !(op->flags & BCH_WRITE_PAGES_OWNED))) {
936 dst = bch2_write_bio_alloc(c, 937 dst = bch2_write_bio_alloc(c, wp, src,
937 &pa 938 &page_alloc_failed,
938 ec_ 939 ec_buf);
939 bounce = true; 940 bounce = true;
940 } 941 }
941 942
942 saved_iter = dst->bi_iter; 943 saved_iter = dst->bi_iter;
943 944
944 do { 945 do {
945 struct bch_extent_crc_unpacked 946 struct bch_extent_crc_unpacked crc = { 0 };
946 struct bversion version = op-> 947 struct bversion version = op->version;
947 size_t dst_len = 0, src_len = 948 size_t dst_len = 0, src_len = 0;
948 949
949 if (page_alloc_failed && 950 if (page_alloc_failed &&
950 dst->bi_iter.bi_size < (w 951 dst->bi_iter.bi_size < (wp->sectors_free << 9) &&
951 dst->bi_iter.bi_size < c-> 952 dst->bi_iter.bi_size < c->opts.encoded_extent_max)
952 break; 953 break;
953 954
954 BUG_ON(op->compression_opt && 955 BUG_ON(op->compression_opt &&
955 (op->flags & BCH_WRITE_ 956 (op->flags & BCH_WRITE_DATA_ENCODED) &&
956 bch2_csum_type_is_encry 957 bch2_csum_type_is_encryption(op->crc.csum_type));
957 BUG_ON(op->compression_opt && 958 BUG_ON(op->compression_opt && !bounce);
958 959
959 crc.compression_type = op->inc 960 crc.compression_type = op->incompressible
960 ? BCH_COMPRESSION_TYPE 961 ? BCH_COMPRESSION_TYPE_incompressible
961 : op->compression_opt 962 : op->compression_opt
962 ? bch2_bio_compress(c, 963 ? bch2_bio_compress(c, dst, &dst_len, src, &src_len,
963 op 964 op->compression_opt)
964 : 0; 965 : 0;
965 if (!crc_is_compressed(crc)) { 966 if (!crc_is_compressed(crc)) {
966 dst_len = min(dst->bi_ 967 dst_len = min(dst->bi_iter.bi_size, src->bi_iter.bi_size);
967 dst_len = min_t(unsign 968 dst_len = min_t(unsigned, dst_len, wp->sectors_free << 9);
968 969
969 if (op->csum_type) 970 if (op->csum_type)
970 dst_len = min_ 971 dst_len = min_t(unsigned, dst_len,
971 972 c->opts.encoded_extent_max);
972 973
973 if (bounce) { 974 if (bounce) {
974 swap(dst->bi_i 975 swap(dst->bi_iter.bi_size, dst_len);
975 bio_copy_data( 976 bio_copy_data(dst, src);
976 swap(dst->bi_i 977 swap(dst->bi_iter.bi_size, dst_len);
977 } 978 }
978 979
979 src_len = dst_len; 980 src_len = dst_len;
980 } 981 }
981 982
982 BUG_ON(!src_len || !dst_len); 983 BUG_ON(!src_len || !dst_len);
983 984
984 if (bch2_csum_type_is_encrypti 985 if (bch2_csum_type_is_encryption(op->csum_type)) {
985 if (bversion_zero(vers 986 if (bversion_zero(version)) {
986 version.lo = a 987 version.lo = atomic64_inc_return(&c->key_version);
987 } else { 988 } else {
988 crc.nonce = op 989 crc.nonce = op->nonce;
989 op->nonce += s 990 op->nonce += src_len >> 9;
990 } 991 }
991 } 992 }
992 993
993 if ((op->flags & BCH_WRITE_DAT 994 if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
994 !crc_is_compressed(crc) && 995 !crc_is_compressed(crc) &&
995 bch2_csum_type_is_encrypti 996 bch2_csum_type_is_encryption(op->crc.csum_type) ==
996 bch2_csum_type_is_encrypti 997 bch2_csum_type_is_encryption(op->csum_type)) {
997 u8 compression_type = 998 u8 compression_type = crc.compression_type;
998 u16 nonce = crc.nonce; 999 u16 nonce = crc.nonce;
999 /* 1000 /*
1000 * Note: when we're u 1001 * Note: when we're using rechecksum(), we need to be
1001 * checksumming @src 1002 * checksumming @src because it has all the data our
1002 * existing checksum 1003 * existing checksum covers - if we bounced (because we
1003 * were trying to com 1004 * were trying to compress), @dst will only have the
1004 * part of the data t 1005 * part of the data the new checksum will cover.
1005 * 1006 *
1006 * But normally we wa 1007 * But normally we want to be checksumming post bounce,
1007 * because part of th 1008 * because part of the reason for bouncing is so the
1008 * data can't be modi 1009 * data can't be modified (by userspace) while it's in
1009 * flight. 1010 * flight.
1010 */ 1011 */
1011 if (bch2_rechecksum_b 1012 if (bch2_rechecksum_bio(c, src, version, op->crc,
1012 &crc, 1013 &crc, &op->crc,
1013 src_l 1014 src_len >> 9,
1014 bio_s 1015 bio_sectors(src) - (src_len >> 9),
1015 op->c 1016 op->csum_type))
1016 goto csum_err 1017 goto csum_err;
1017 /* 1018 /*
1018 * rchecksum_bio sets 1019 * rchecksum_bio sets compression_type on crc from op->crc,
1019 * this isn't always 1020 * this isn't always correct as sometimes we're changing
1020 * an extent from unc 1021 * an extent from uncompressed to incompressible.
1021 */ 1022 */
1022 crc.compression_type 1023 crc.compression_type = compression_type;
1023 crc.nonce = nonce; 1024 crc.nonce = nonce;
1024 } else { 1025 } else {
1025 if ((op->flags & BCH_ 1026 if ((op->flags & BCH_WRITE_DATA_ENCODED) &&
1026 bch2_rechecksum_b 1027 bch2_rechecksum_bio(c, src, version, op->crc,
1027 NULL, 1028 NULL, &op->crc,
1028 src_l 1029 src_len >> 9,
1029 bio_s 1030 bio_sectors(src) - (src_len >> 9),
1030 op->c 1031 op->crc.csum_type))
1031 goto csum_err 1032 goto csum_err;
1032 1033
1033 crc.compressed_size 1034 crc.compressed_size = dst_len >> 9;
1034 crc.uncompressed_size 1035 crc.uncompressed_size = src_len >> 9;
1035 crc.live_size 1036 crc.live_size = src_len >> 9;
1036 1037
1037 swap(dst->bi_iter.bi_ 1038 swap(dst->bi_iter.bi_size, dst_len);
1038 ret = bch2_encrypt_bi 1039 ret = bch2_encrypt_bio(c, op->csum_type,
1039 1040 extent_nonce(version, crc), dst);
1040 if (ret) 1041 if (ret)
1041 goto err; 1042 goto err;
1042 1043
1043 crc.csum = bch2_check 1044 crc.csum = bch2_checksum_bio(c, op->csum_type,
1044 exte 1045 extent_nonce(version, crc), dst);
1045 crc.csum_type = op->c 1046 crc.csum_type = op->csum_type;
1046 swap(dst->bi_iter.bi_ 1047 swap(dst->bi_iter.bi_size, dst_len);
1047 } 1048 }
1048 1049
1049 init_append_extent(op, wp, ve 1050 init_append_extent(op, wp, version, crc);
1050 1051
1051 if (dst != src) 1052 if (dst != src)
1052 bio_advance(dst, dst_ 1053 bio_advance(dst, dst_len);
1053 bio_advance(src, src_len); 1054 bio_advance(src, src_len);
1054 total_output += dst_len; 1055 total_output += dst_len;
1055 total_input += src_len; 1056 total_input += src_len;
1056 } while (dst->bi_iter.bi_size && 1057 } while (dst->bi_iter.bi_size &&
1057 src->bi_iter.bi_size && 1058 src->bi_iter.bi_size &&
1058 wp->sectors_free && 1059 wp->sectors_free &&
1059 !bch2_keylist_realloc(&op->i 1060 !bch2_keylist_realloc(&op->insert_keys,
1060 op->inl 1061 op->inline_keys,
1061 ARRAY_S 1062 ARRAY_SIZE(op->inline_keys),
1062 BKEY_EX 1063 BKEY_EXTENT_U64s_MAX));
1063 1064
1064 more = src->bi_iter.bi_size != 0; 1065 more = src->bi_iter.bi_size != 0;
1065 1066
1066 dst->bi_iter = saved_iter; 1067 dst->bi_iter = saved_iter;
1067 1068
1068 if (dst == src && more) { 1069 if (dst == src && more) {
1069 BUG_ON(total_output != total_ 1070 BUG_ON(total_output != total_input);
1070 1071
1071 dst = bio_split(src, total_in 1072 dst = bio_split(src, total_input >> 9,
1072 GFP_NOFS, &c- 1073 GFP_NOFS, &c->bio_write);
1073 wbio_init(dst)->put_bio = tru 1074 wbio_init(dst)->put_bio = true;
1074 /* copy WRITE_SYNC flag */ 1075 /* copy WRITE_SYNC flag */
1075 dst->bi_opf = src 1076 dst->bi_opf = src->bi_opf;
1076 } 1077 }
1077 1078
1078 dst->bi_iter.bi_size = total_output; 1079 dst->bi_iter.bi_size = total_output;
1079 do_write: 1080 do_write:
1080 *_dst = dst; 1081 *_dst = dst;
1081 return more; 1082 return more;
1082 csum_err: 1083 csum_err:
1083 bch_err_inum_offset_ratelimited(c, !! 1084 bch_err(c, "%s writ error: error verifying existing checksum while rewriting existing data (memory corruption?)",
1084 op->pos.inode, <<
1085 op->pos.offset << 9, <<
1086 "%s write error: error verify <<
1087 op->flags & BCH_WRITE_MOVE ? 1085 op->flags & BCH_WRITE_MOVE ? "move" : "user");
1088 ret = -EIO; 1086 ret = -EIO;
1089 err: 1087 err:
1090 if (to_wbio(dst)->bounce) 1088 if (to_wbio(dst)->bounce)
1091 bch2_bio_free_pages_pool(c, d 1089 bch2_bio_free_pages_pool(c, dst);
1092 if (to_wbio(dst)->put_bio) 1090 if (to_wbio(dst)->put_bio)
1093 bio_put(dst); 1091 bio_put(dst);
1094 1092
1095 return ret; 1093 return ret;
1096 } 1094 }
1097 1095
1098 static bool bch2_extent_is_writeable(struct b 1096 static bool bch2_extent_is_writeable(struct bch_write_op *op,
1099 struct b 1097 struct bkey_s_c k)
1100 { 1098 {
1101 struct bch_fs *c = op->c; 1099 struct bch_fs *c = op->c;
1102 struct bkey_s_c_extent e; 1100 struct bkey_s_c_extent e;
1103 struct extent_ptr_decoded p; 1101 struct extent_ptr_decoded p;
1104 const union bch_extent_entry *entry; 1102 const union bch_extent_entry *entry;
1105 unsigned replicas = 0; 1103 unsigned replicas = 0;
1106 1104
1107 if (k.k->type != KEY_TYPE_extent) 1105 if (k.k->type != KEY_TYPE_extent)
1108 return false; 1106 return false;
1109 1107
1110 e = bkey_s_c_to_extent(k); 1108 e = bkey_s_c_to_extent(k);
1111 1109
1112 rcu_read_lock(); 1110 rcu_read_lock();
1113 extent_for_each_ptr_decode(e, p, entr 1111 extent_for_each_ptr_decode(e, p, entry) {
1114 if (crc_is_encoded(p.crc) || 1112 if (crc_is_encoded(p.crc) || p.has_ec) {
1115 rcu_read_unlock(); 1113 rcu_read_unlock();
1116 return false; 1114 return false;
1117 } 1115 }
1118 1116
1119 replicas += bch2_extent_ptr_d 1117 replicas += bch2_extent_ptr_durability(c, &p);
1120 } 1118 }
1121 rcu_read_unlock(); 1119 rcu_read_unlock();
1122 1120
1123 return replicas >= op->opts.data_repl 1121 return replicas >= op->opts.data_replicas;
1124 } 1122 }
1125 1123
1126 static int bch2_nocow_write_convert_one_unwri 1124 static int bch2_nocow_write_convert_one_unwritten(struct btree_trans *trans,
1127 1125 struct btree_iter *iter,
1128 1126 struct bkey_i *orig,
1129 1127 struct bkey_s_c k,
1130 1128 u64 new_i_size)
1131 { 1129 {
1132 if (!bch2_extents_match(bkey_i_to_s_c 1130 if (!bch2_extents_match(bkey_i_to_s_c(orig), k)) {
1133 /* trace this */ 1131 /* trace this */
1134 return 0; 1132 return 0;
1135 } 1133 }
1136 1134
1137 struct bkey_i *new = bch2_bkey_make_m 1135 struct bkey_i *new = bch2_bkey_make_mut_noupdate(trans, k);
1138 int ret = PTR_ERR_OR_ZERO(new); 1136 int ret = PTR_ERR_OR_ZERO(new);
1139 if (ret) 1137 if (ret)
1140 return ret; 1138 return ret;
1141 1139
1142 bch2_cut_front(bkey_start_pos(&orig-> 1140 bch2_cut_front(bkey_start_pos(&orig->k), new);
1143 bch2_cut_back(orig->k.p, new); 1141 bch2_cut_back(orig->k.p, new);
1144 1142
1145 struct bkey_ptrs ptrs = bch2_bkey_ptr 1143 struct bkey_ptrs ptrs = bch2_bkey_ptrs(bkey_i_to_s(new));
1146 bkey_for_each_ptr(ptrs, ptr) 1144 bkey_for_each_ptr(ptrs, ptr)
1147 ptr->unwritten = 0; 1145 ptr->unwritten = 0;
1148 1146
1149 /* 1147 /*
1150 * Note that we're not calling bch2_s 1148 * Note that we're not calling bch2_subvol_get_snapshot() in this path -
1151 * that was done when we kicked off t 1149 * that was done when we kicked off the write, and here it's important
1152 * that we update the extent that we 1150 * that we update the extent that we wrote to - even if a snapshot has
1153 * since been created. The write is s 1151 * since been created. The write is still outstanding, so we're ok
1154 * w.r.t. snapshot atomicity: 1152 * w.r.t. snapshot atomicity:
1155 */ 1153 */
1156 return bch2_extent_update_i_size_sec 1154 return bch2_extent_update_i_size_sectors(trans, iter,
1157 min(n 1155 min(new->k.p.offset << 9, new_i_size), 0) ?:
1158 bch2_trans_update(trans, iter 1156 bch2_trans_update(trans, iter, new,
1159 BTREE_UPDAT 1157 BTREE_UPDATE_internal_snapshot_node);
1160 } 1158 }
1161 1159
1162 static void bch2_nocow_write_convert_unwritte 1160 static void bch2_nocow_write_convert_unwritten(struct bch_write_op *op)
1163 { 1161 {
1164 struct bch_fs *c = op->c; 1162 struct bch_fs *c = op->c;
1165 struct btree_trans *trans = bch2_tran 1163 struct btree_trans *trans = bch2_trans_get(c);
1166 1164
1167 for_each_keylist_key(&op->insert_keys 1165 for_each_keylist_key(&op->insert_keys, orig) {
1168 int ret = for_each_btree_key_ 1166 int ret = for_each_btree_key_upto_commit(trans, iter, BTREE_ID_extents,
1169 bkey_sta 1167 bkey_start_pos(&orig->k), orig->k.p,
1170 BTREE_IT 1168 BTREE_ITER_intent, k,
1171 NULL, NU 1169 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, ({
1172 bch2_nocow_write_conv 1170 bch2_nocow_write_convert_one_unwritten(trans, &iter, orig, k, op->new_i_size);
1173 })); 1171 }));
1174 1172
1175 if (ret && !bch2_err_matches( 1173 if (ret && !bch2_err_matches(ret, EROFS)) {
1176 struct bkey_i *insert 1174 struct bkey_i *insert = bch2_keylist_front(&op->insert_keys);
1177 1175
1178 bch_err_inum_offset_r 1176 bch_err_inum_offset_ratelimited(c,
1179 insert->k.p.i 1177 insert->k.p.inode, insert->k.p.offset << 9,
1180 "%s write err 1178 "%s write error while doing btree update: %s",
1181 op->flags & B 1179 op->flags & BCH_WRITE_MOVE ? "move" : "user",
1182 bch2_err_str( 1180 bch2_err_str(ret));
1183 } 1181 }
1184 1182
1185 if (ret) { 1183 if (ret) {
1186 op->error = ret; 1184 op->error = ret;
1187 break; 1185 break;
1188 } 1186 }
1189 } 1187 }
1190 1188
1191 bch2_trans_put(trans); 1189 bch2_trans_put(trans);
1192 } 1190 }
1193 1191
1194 static void __bch2_nocow_write_done(struct bc 1192 static void __bch2_nocow_write_done(struct bch_write_op *op)
1195 { 1193 {
1196 if (unlikely(op->flags & BCH_WRITE_IO 1194 if (unlikely(op->flags & BCH_WRITE_IO_ERROR)) {
1197 op->error = -EIO; 1195 op->error = -EIO;
1198 } else if (unlikely(op->flags & BCH_W 1196 } else if (unlikely(op->flags & BCH_WRITE_CONVERT_UNWRITTEN))
1199 bch2_nocow_write_convert_unwr 1197 bch2_nocow_write_convert_unwritten(op);
1200 } 1198 }
1201 1199
1202 static CLOSURE_CALLBACK(bch2_nocow_write_done 1200 static CLOSURE_CALLBACK(bch2_nocow_write_done)
1203 { 1201 {
1204 closure_type(op, struct bch_write_op, 1202 closure_type(op, struct bch_write_op, cl);
1205 1203
1206 __bch2_nocow_write_done(op); 1204 __bch2_nocow_write_done(op);
1207 bch2_write_done(cl); 1205 bch2_write_done(cl);
1208 } 1206 }
1209 1207
1210 struct bucket_to_lock { 1208 struct bucket_to_lock {
1211 struct bpos b; 1209 struct bpos b;
1212 unsigned gen; 1210 unsigned gen;
1213 struct nocow_lock_bucket *l; 1211 struct nocow_lock_bucket *l;
1214 }; 1212 };
1215 1213
1216 static void bch2_nocow_write(struct bch_write 1214 static void bch2_nocow_write(struct bch_write_op *op)
1217 { 1215 {
1218 struct bch_fs *c = op->c; 1216 struct bch_fs *c = op->c;
1219 struct btree_trans *trans; 1217 struct btree_trans *trans;
1220 struct btree_iter iter; 1218 struct btree_iter iter;
1221 struct bkey_s_c k; 1219 struct bkey_s_c k;
1222 DARRAY_PREALLOCATED(struct bucket_to_ 1220 DARRAY_PREALLOCATED(struct bucket_to_lock, 3) buckets;
1223 u32 snapshot; 1221 u32 snapshot;
1224 struct bucket_to_lock *stale_at; 1222 struct bucket_to_lock *stale_at;
1225 int stale, ret; 1223 int stale, ret;
1226 1224
1227 if (op->flags & BCH_WRITE_MOVE) 1225 if (op->flags & BCH_WRITE_MOVE)
1228 return; 1226 return;
1229 1227
1230 darray_init(&buckets); 1228 darray_init(&buckets);
1231 trans = bch2_trans_get(c); 1229 trans = bch2_trans_get(c);
1232 retry: 1230 retry:
1233 bch2_trans_begin(trans); 1231 bch2_trans_begin(trans);
1234 1232
1235 ret = bch2_subvolume_get_snapshot(tra 1233 ret = bch2_subvolume_get_snapshot(trans, op->subvol, &snapshot);
1236 if (unlikely(ret)) 1234 if (unlikely(ret))
1237 goto err; 1235 goto err;
1238 1236
1239 bch2_trans_iter_init(trans, &iter, BT 1237 bch2_trans_iter_init(trans, &iter, BTREE_ID_extents,
1240 SPOS(op->pos.ino 1238 SPOS(op->pos.inode, op->pos.offset, snapshot),
1241 BTREE_ITER_slots 1239 BTREE_ITER_slots);
1242 while (1) { 1240 while (1) {
1243 struct bio *bio = &op->wbio.b 1241 struct bio *bio = &op->wbio.bio;
1244 1242
1245 buckets.nr = 0; 1243 buckets.nr = 0;
1246 1244
1247 ret = bch2_trans_relock(trans 1245 ret = bch2_trans_relock(trans);
1248 if (ret) 1246 if (ret)
1249 break; 1247 break;
1250 1248
1251 k = bch2_btree_iter_peek_slot 1249 k = bch2_btree_iter_peek_slot(&iter);
1252 ret = bkey_err(k); 1250 ret = bkey_err(k);
1253 if (ret) 1251 if (ret)
1254 break; 1252 break;
1255 1253
1256 /* fall back to normal cow wr 1254 /* fall back to normal cow write path? */
1257 if (unlikely(k.k->p.snapshot 1255 if (unlikely(k.k->p.snapshot != snapshot ||
1258 !bch2_extent_is_ 1256 !bch2_extent_is_writeable(op, k)))
1259 break; 1257 break;
1260 1258
1261 if (bch2_keylist_realloc(&op- 1259 if (bch2_keylist_realloc(&op->insert_keys,
1262 op-> 1260 op->inline_keys,
1263 ARRA 1261 ARRAY_SIZE(op->inline_keys),
1264 k.k- 1262 k.k->u64s))
1265 break; 1263 break;
1266 1264
1267 /* Get iorefs before dropping 1265 /* Get iorefs before dropping btree locks: */
1268 struct bkey_ptrs_c ptrs = bch 1266 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
1269 bkey_for_each_ptr(ptrs, ptr) 1267 bkey_for_each_ptr(ptrs, ptr) {
1270 struct bch_dev *ca = 1268 struct bch_dev *ca = bch2_dev_get_ioref(c, ptr->dev, WRITE);
1271 if (unlikely(!ca)) 1269 if (unlikely(!ca))
1272 goto err_get_ 1270 goto err_get_ioref;
1273 1271
1274 struct bpos b = PTR_B 1272 struct bpos b = PTR_BUCKET_POS(ca, ptr);
1275 struct nocow_lock_buc 1273 struct nocow_lock_bucket *l =
1276 bucket_nocow_ 1274 bucket_nocow_lock(&c->nocow_locks, bucket_to_u64(b));
1277 prefetch(l); 1275 prefetch(l);
1278 1276
1279 /* XXX allocating mem 1277 /* XXX allocating memory with btree locks held - rare */
1280 darray_push_gfp(&buck 1278 darray_push_gfp(&buckets, ((struct bucket_to_lock) {
1281 1279 .b = b, .gen = ptr->gen, .l = l,
1282 1280 }), GFP_KERNEL|__GFP_NOFAIL);
1283 1281
1284 if (ptr->unwritten) 1282 if (ptr->unwritten)
1285 op->flags |= 1283 op->flags |= BCH_WRITE_CONVERT_UNWRITTEN;
1286 } 1284 }
1287 1285
1288 /* Unlock before taking nocow 1286 /* Unlock before taking nocow locks, doing IO: */
1289 bkey_reassemble(op->insert_ke 1287 bkey_reassemble(op->insert_keys.top, k);
1290 bch2_trans_unlock(trans); 1288 bch2_trans_unlock(trans);
1291 1289
1292 bch2_cut_front(op->pos, op->i 1290 bch2_cut_front(op->pos, op->insert_keys.top);
1293 if (op->flags & BCH_WRITE_CON 1291 if (op->flags & BCH_WRITE_CONVERT_UNWRITTEN)
1294 bch2_cut_back(POS(op- 1292 bch2_cut_back(POS(op->pos.inode, op->pos.offset + bio_sectors(bio)), op->insert_keys.top);
1295 1293
1296 darray_for_each(buckets, i) { 1294 darray_for_each(buckets, i) {
1297 struct bch_dev *ca = 1295 struct bch_dev *ca = bch2_dev_have_ref(c, i->b.inode);
1298 1296
1299 __bch2_bucket_nocow_l 1297 __bch2_bucket_nocow_lock(&c->nocow_locks, i->l,
1300 1298 bucket_to_u64(i->b),
1301 1299 BUCKET_NOCOW_LOCK_UPDATE);
1302 1300
1303 int gen = bucket_gen_ !! 1301 rcu_read_lock();
1304 stale = gen < 0 ? gen !! 1302 u8 *gen = bucket_gen(ca, i->b.offset);
>> 1303 stale = !gen ? -1 : gen_after(*gen, i->gen);
>> 1304 rcu_read_unlock();
>> 1305
1305 if (unlikely(stale)) 1306 if (unlikely(stale)) {
1306 stale_at = i; 1307 stale_at = i;
1307 goto err_buck 1308 goto err_bucket_stale;
1308 } 1309 }
1309 } 1310 }
1310 1311
1311 bio = &op->wbio.bio; 1312 bio = &op->wbio.bio;
1312 if (k.k->p.offset < op->pos.o 1313 if (k.k->p.offset < op->pos.offset + bio_sectors(bio)) {
1313 bio = bio_split(bio, 1314 bio = bio_split(bio, k.k->p.offset - op->pos.offset,
1314 GFP_K 1315 GFP_KERNEL, &c->bio_write);
1315 wbio_init(bio)->put_b 1316 wbio_init(bio)->put_bio = true;
1316 bio->bi_opf = op->wbi 1317 bio->bi_opf = op->wbio.bio.bi_opf;
1317 } else { 1318 } else {
1318 op->flags |= BCH_WRIT !! 1319 op->flags |= BCH_WRITE_DONE;
1319 } 1320 }
1320 1321
1321 op->pos.offset += bio_sectors 1322 op->pos.offset += bio_sectors(bio);
1322 op->written += bio_sectors(bi 1323 op->written += bio_sectors(bio);
1323 1324
1324 bio->bi_end_io = bch2_write_ 1325 bio->bi_end_io = bch2_write_endio;
1325 bio->bi_private = &op->cl; 1326 bio->bi_private = &op->cl;
1326 bio->bi_opf |= REQ_OP_WRITE; 1327 bio->bi_opf |= REQ_OP_WRITE;
1327 closure_get(&op->cl); 1328 closure_get(&op->cl);
1328 bch2_submit_wbio_replicas(to_ 1329 bch2_submit_wbio_replicas(to_wbio(bio), c, BCH_DATA_user,
1329 op- 1330 op->insert_keys.top, true);
1330 1331
1331 bch2_keylist_push(&op->insert 1332 bch2_keylist_push(&op->insert_keys);
1332 if (op->flags & BCH_WRITE_SUB !! 1333 if (op->flags & BCH_WRITE_DONE)
1333 break; 1334 break;
1334 bch2_btree_iter_advance(&iter 1335 bch2_btree_iter_advance(&iter);
1335 } 1336 }
1336 out: 1337 out:
1337 bch2_trans_iter_exit(trans, &iter); 1338 bch2_trans_iter_exit(trans, &iter);
1338 err: 1339 err:
1339 if (bch2_err_matches(ret, BCH_ERR_tra 1340 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1340 goto retry; 1341 goto retry;
1341 1342
1342 if (ret) { 1343 if (ret) {
1343 bch_err_inum_offset_ratelimit 1344 bch_err_inum_offset_ratelimited(c,
1344 op->pos.inode, op->po 1345 op->pos.inode, op->pos.offset << 9,
1345 "%s: btree lookup err 1346 "%s: btree lookup error %s", __func__, bch2_err_str(ret));
1346 op->error = ret; 1347 op->error = ret;
1347 op->flags |= BCH_WRITE_SUBMIT !! 1348 op->flags |= BCH_WRITE_DONE;
1348 } 1349 }
1349 1350
1350 bch2_trans_put(trans); 1351 bch2_trans_put(trans);
1351 darray_exit(&buckets); 1352 darray_exit(&buckets);
1352 1353
1353 /* fallback to cow write path? */ 1354 /* fallback to cow write path? */
1354 if (!(op->flags & BCH_WRITE_SUBMITTED !! 1355 if (!(op->flags & BCH_WRITE_DONE)) {
1355 closure_sync(&op->cl); 1356 closure_sync(&op->cl);
1356 __bch2_nocow_write_done(op); 1357 __bch2_nocow_write_done(op);
1357 op->insert_keys.top = op->ins 1358 op->insert_keys.top = op->insert_keys.keys;
1358 } else if (op->flags & BCH_WRITE_SYNC 1359 } else if (op->flags & BCH_WRITE_SYNC) {
1359 closure_sync(&op->cl); 1360 closure_sync(&op->cl);
1360 bch2_nocow_write_done(&op->cl 1361 bch2_nocow_write_done(&op->cl.work);
1361 } else { 1362 } else {
1362 /* 1363 /*
1363 * XXX 1364 * XXX
1364 * needs to run out of proces 1365 * needs to run out of process context because ei_quota_lock is
1365 * a mutex 1366 * a mutex
1366 */ 1367 */
1367 continue_at(&op->cl, bch2_noc 1368 continue_at(&op->cl, bch2_nocow_write_done, index_update_wq(op));
1368 } 1369 }
1369 return; 1370 return;
1370 err_get_ioref: 1371 err_get_ioref:
1371 darray_for_each(buckets, i) 1372 darray_for_each(buckets, i)
1372 percpu_ref_put(&bch2_dev_have 1373 percpu_ref_put(&bch2_dev_have_ref(c, i->b.inode)->io_ref);
1373 1374
1374 /* Fall back to COW path: */ 1375 /* Fall back to COW path: */
1375 goto out; 1376 goto out;
1376 err_bucket_stale: 1377 err_bucket_stale:
1377 darray_for_each(buckets, i) { 1378 darray_for_each(buckets, i) {
1378 bch2_bucket_nocow_unlock(&c-> 1379 bch2_bucket_nocow_unlock(&c->nocow_locks, i->b, BUCKET_NOCOW_LOCK_UPDATE);
1379 if (i == stale_at) 1380 if (i == stale_at)
1380 break; 1381 break;
1381 } 1382 }
1382 1383
1383 struct printbuf buf = PRINTBUF; 1384 struct printbuf buf = PRINTBUF;
1384 if (bch2_fs_inconsistent_on(stale < 0 1385 if (bch2_fs_inconsistent_on(stale < 0, c,
1385 "pointer 1386 "pointer to invalid bucket in nocow path on device %llu\n %s",
1386 stale_at- 1387 stale_at->b.inode,
1387 (bch2_bke 1388 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1388 ret = -EIO; 1389 ret = -EIO;
1389 } else { 1390 } else {
1390 /* We can retry this: */ 1391 /* We can retry this: */
1391 ret = -BCH_ERR_transaction_re 1392 ret = -BCH_ERR_transaction_restart;
1392 } 1393 }
1393 printbuf_exit(&buf); 1394 printbuf_exit(&buf);
1394 1395
1395 goto err_get_ioref; 1396 goto err_get_ioref;
1396 } 1397 }
1397 1398
1398 static void __bch2_write(struct bch_write_op 1399 static void __bch2_write(struct bch_write_op *op)
1399 { 1400 {
1400 struct bch_fs *c = op->c; 1401 struct bch_fs *c = op->c;
1401 struct write_point *wp = NULL; 1402 struct write_point *wp = NULL;
1402 struct bio *bio = NULL; 1403 struct bio *bio = NULL;
1403 unsigned nofs_flags; 1404 unsigned nofs_flags;
1404 int ret; 1405 int ret;
1405 1406
1406 nofs_flags = memalloc_nofs_save(); 1407 nofs_flags = memalloc_nofs_save();
1407 1408
1408 if (unlikely(op->opts.nocow && c->opt 1409 if (unlikely(op->opts.nocow && c->opts.nocow_enabled)) {
1409 bch2_nocow_write(op); 1410 bch2_nocow_write(op);
1410 if (op->flags & BCH_WRITE_SUB !! 1411 if (op->flags & BCH_WRITE_DONE)
1411 goto out_nofs_restore 1412 goto out_nofs_restore;
1412 } 1413 }
1413 again: 1414 again:
1414 memset(&op->failed, 0, sizeof(op->fai 1415 memset(&op->failed, 0, sizeof(op->failed));
1415 1416
1416 do { 1417 do {
1417 struct bkey_i *key_to_write; 1418 struct bkey_i *key_to_write;
1418 unsigned key_to_write_offset 1419 unsigned key_to_write_offset = op->insert_keys.top_p -
1419 op->insert_keys.keys_ 1420 op->insert_keys.keys_p;
1420 1421
1421 /* +1 for possible cache devi 1422 /* +1 for possible cache device: */
1422 if (op->open_buckets.nr + op- 1423 if (op->open_buckets.nr + op->nr_replicas + 1 >
1423 ARRAY_SIZE(op->open_bucke 1424 ARRAY_SIZE(op->open_buckets.v))
1424 break; 1425 break;
1425 1426
1426 if (bch2_keylist_realloc(&op- 1427 if (bch2_keylist_realloc(&op->insert_keys,
1427 op->i 1428 op->inline_keys,
1428 ARRAY 1429 ARRAY_SIZE(op->inline_keys),
1429 BKEY_ 1430 BKEY_EXTENT_U64s_MAX))
1430 break; 1431 break;
1431 1432
1432 /* 1433 /*
1433 * The copygc thread is now g 1434 * The copygc thread is now global, which means it's no longer
1434 * freeing up space on specif 1435 * freeing up space on specific disks, which means that
1435 * allocations for specific d 1436 * allocations for specific disks may hang arbitrarily long:
1436 */ 1437 */
1437 ret = bch2_trans_run(c, lockr !! 1438 ret = bch2_trans_do(c, NULL, NULL, 0,
1438 bch2_alloc_sectors_st 1439 bch2_alloc_sectors_start_trans(trans,
1439 op->target, 1440 op->target,
1440 op->opts.eras 1441 op->opts.erasure_code && !(op->flags & BCH_WRITE_CACHED),
1441 op->write_poi 1442 op->write_point,
1442 &op->devs_hav 1443 &op->devs_have,
1443 op->nr_replic 1444 op->nr_replicas,
1444 op->nr_replic 1445 op->nr_replicas_required,
1445 op->watermark 1446 op->watermark,
1446 op->flags, 1447 op->flags,
1447 &op->cl, &wp) !! 1448 (op->flags & (BCH_WRITE_ALLOC_NOWAIT|
>> 1449 BCH_WRITE_ONLY_SPECIFIED_DEVS))
>> 1450 ? NULL : &op->cl, &wp));
1448 if (unlikely(ret)) { 1451 if (unlikely(ret)) {
1449 if (bch2_err_matches( 1452 if (bch2_err_matches(ret, BCH_ERR_operation_blocked))
1450 break; 1453 break;
1451 1454
1452 goto err; 1455 goto err;
1453 } 1456 }
1454 1457
1455 EBUG_ON(!wp); 1458 EBUG_ON(!wp);
1456 1459
1457 bch2_open_bucket_get(c, wp, & 1460 bch2_open_bucket_get(c, wp, &op->open_buckets);
1458 ret = bch2_write_extent(op, w 1461 ret = bch2_write_extent(op, wp, &bio);
1459 1462
1460 bch2_alloc_sectors_done_inlin 1463 bch2_alloc_sectors_done_inlined(c, wp);
1461 err: 1464 err:
1462 if (ret <= 0) { 1465 if (ret <= 0) {
1463 op->flags |= BCH_WRIT !! 1466 op->flags |= BCH_WRITE_DONE;
1464 1467
1465 if (ret < 0) { 1468 if (ret < 0) {
1466 if (!(op->fla 1469 if (!(op->flags & BCH_WRITE_ALLOC_NOWAIT))
1467 bch_e 1470 bch_err_inum_offset_ratelimited(c,
1468 1471 op->pos.inode,
1469 1472 op->pos.offset << 9,
1470 1473 "%s(): %s error: %s", __func__,
1471 1474 op->flags & BCH_WRITE_MOVE ? "move" : "user",
1472 1475 bch2_err_str(ret));
1473 op->error = r 1476 op->error = ret;
1474 break; 1477 break;
1475 } 1478 }
1476 } 1479 }
1477 1480
1478 bio->bi_end_io = bch2_write_ 1481 bio->bi_end_io = bch2_write_endio;
1479 bio->bi_private = &op->cl; 1482 bio->bi_private = &op->cl;
1480 bio->bi_opf |= REQ_OP_WRITE; 1483 bio->bi_opf |= REQ_OP_WRITE;
1481 1484
1482 closure_get(bio->bi_private); 1485 closure_get(bio->bi_private);
1483 1486
1484 key_to_write = (void *) (op-> 1487 key_to_write = (void *) (op->insert_keys.keys_p +
1485 key_ 1488 key_to_write_offset);
1486 1489
1487 bch2_submit_wbio_replicas(to_ 1490 bch2_submit_wbio_replicas(to_wbio(bio), c, BCH_DATA_user,
1488 key 1491 key_to_write, false);
1489 } while (ret); 1492 } while (ret);
1490 1493
1491 /* 1494 /*
1492 * Sync or no? 1495 * Sync or no?
1493 * 1496 *
1494 * If we're running asynchronously, w 1497 * If we're running asynchronously, wne may still want to block
1495 * synchronously here if we weren't a 1498 * synchronously here if we weren't able to submit all of the IO at
1496 * once, as that signals backpressure 1499 * once, as that signals backpressure to the caller.
1497 */ 1500 */
1498 if ((op->flags & BCH_WRITE_SYNC) || 1501 if ((op->flags & BCH_WRITE_SYNC) ||
1499 (!(op->flags & BCH_WRITE_SUBMITTE !! 1502 (!(op->flags & BCH_WRITE_DONE) &&
1500 !(op->flags & BCH_WRITE_IN_WORKE 1503 !(op->flags & BCH_WRITE_IN_WORKER))) {
1501 bch2_wait_on_allocator(c, &op !! 1504 if (closure_sync_timeout(&op->cl, HZ * 10)) {
>> 1505 bch2_print_allocator_stuck(c);
>> 1506 closure_sync(&op->cl);
>> 1507 }
1502 1508
1503 __bch2_write_index(op); 1509 __bch2_write_index(op);
1504 1510
1505 if (!(op->flags & BCH_WRITE_S !! 1511 if (!(op->flags & BCH_WRITE_DONE))
1506 goto again; 1512 goto again;
1507 bch2_write_done(&op->cl); 1513 bch2_write_done(&op->cl);
1508 } else { 1514 } else {
1509 bch2_write_queue(op, wp); 1515 bch2_write_queue(op, wp);
1510 continue_at(&op->cl, bch2_wri 1516 continue_at(&op->cl, bch2_write_index, NULL);
1511 } 1517 }
1512 out_nofs_restore: 1518 out_nofs_restore:
1513 memalloc_nofs_restore(nofs_flags); 1519 memalloc_nofs_restore(nofs_flags);
1514 } 1520 }
1515 1521
1516 static void bch2_write_data_inline(struct bch 1522 static void bch2_write_data_inline(struct bch_write_op *op, unsigned data_len)
1517 { 1523 {
1518 struct bio *bio = &op->wbio.bio; 1524 struct bio *bio = &op->wbio.bio;
1519 struct bvec_iter iter; 1525 struct bvec_iter iter;
1520 struct bkey_i_inline_data *id; 1526 struct bkey_i_inline_data *id;
1521 unsigned sectors; 1527 unsigned sectors;
1522 int ret; 1528 int ret;
1523 1529
1524 memset(&op->failed, 0, sizeof(op->fai 1530 memset(&op->failed, 0, sizeof(op->failed));
1525 1531
1526 op->flags |= BCH_WRITE_WROTE_DATA_INL 1532 op->flags |= BCH_WRITE_WROTE_DATA_INLINE;
1527 op->flags |= BCH_WRITE_SUBMITTED; !! 1533 op->flags |= BCH_WRITE_DONE;
1528 1534
1529 bch2_check_set_feature(op->c, BCH_FEA 1535 bch2_check_set_feature(op->c, BCH_FEATURE_inline_data);
1530 1536
1531 ret = bch2_keylist_realloc(&op->inser 1537 ret = bch2_keylist_realloc(&op->insert_keys, op->inline_keys,
1532 ARRAY_SIZE 1538 ARRAY_SIZE(op->inline_keys),
1533 BKEY_U64s 1539 BKEY_U64s + DIV_ROUND_UP(data_len, 8));
1534 if (ret) { 1540 if (ret) {
1535 op->error = ret; 1541 op->error = ret;
1536 goto err; 1542 goto err;
1537 } 1543 }
1538 1544
1539 sectors = bio_sectors(bio); 1545 sectors = bio_sectors(bio);
1540 op->pos.offset += sectors; 1546 op->pos.offset += sectors;
1541 1547
1542 id = bkey_inline_data_init(op->insert 1548 id = bkey_inline_data_init(op->insert_keys.top);
1543 id->k.p = op->pos; 1549 id->k.p = op->pos;
1544 id->k.bversion = op->version; !! 1550 id->k.version = op->version;
1545 id->k.size = sectors; 1551 id->k.size = sectors;
1546 1552
1547 iter = bio->bi_iter; 1553 iter = bio->bi_iter;
1548 iter.bi_size = data_len; 1554 iter.bi_size = data_len;
1549 memcpy_from_bio(id->v.data, bio, iter 1555 memcpy_from_bio(id->v.data, bio, iter);
1550 1556
1551 while (data_len & 7) 1557 while (data_len & 7)
1552 id->v.data[data_len++] = '\0' 1558 id->v.data[data_len++] = '\0';
1553 set_bkey_val_bytes(&id->k, data_len); 1559 set_bkey_val_bytes(&id->k, data_len);
1554 bch2_keylist_push(&op->insert_keys); 1560 bch2_keylist_push(&op->insert_keys);
1555 1561
1556 __bch2_write_index(op); 1562 __bch2_write_index(op);
1557 err: 1563 err:
1558 bch2_write_done(&op->cl); 1564 bch2_write_done(&op->cl);
1559 } 1565 }
1560 1566
1561 /** 1567 /**
1562 * bch2_write() - handle a write to a cache d 1568 * bch2_write() - handle a write to a cache device or flash only volume
1563 * @cl: &bch_write_op->cl 1569 * @cl: &bch_write_op->cl
1564 * 1570 *
1565 * This is the starting point for any data to 1571 * This is the starting point for any data to end up in a cache device; it could
1566 * be from a normal write, or a writeback wri 1572 * be from a normal write, or a writeback write, or a write to a flash only
1567 * volume - it's also used by the moving garb 1573 * volume - it's also used by the moving garbage collector to compact data in
1568 * mostly empty buckets. 1574 * mostly empty buckets.
1569 * 1575 *
1570 * It first writes the data to the cache, cre 1576 * It first writes the data to the cache, creating a list of keys to be inserted
1571 * (if the data won't fit in a single open bu 1577 * (if the data won't fit in a single open bucket, there will be multiple keys);
1572 * after the data is written it calls bch_jou 1578 * after the data is written it calls bch_journal, and after the keys have been
1573 * added to the next journal write they're in 1579 * added to the next journal write they're inserted into the btree.
1574 * 1580 *
1575 * If op->discard is true, instead of inserti 1581 * If op->discard is true, instead of inserting the data it invalidates the
1576 * region of the cache represented by op->bio 1582 * region of the cache represented by op->bio and op->inode.
1577 */ 1583 */
1578 CLOSURE_CALLBACK(bch2_write) 1584 CLOSURE_CALLBACK(bch2_write)
1579 { 1585 {
1580 closure_type(op, struct bch_write_op, 1586 closure_type(op, struct bch_write_op, cl);
1581 struct bio *bio = &op->wbio.bio; 1587 struct bio *bio = &op->wbio.bio;
1582 struct bch_fs *c = op->c; 1588 struct bch_fs *c = op->c;
1583 unsigned data_len; 1589 unsigned data_len;
1584 1590
1585 EBUG_ON(op->cl.parent); 1591 EBUG_ON(op->cl.parent);
1586 BUG_ON(!op->nr_replicas); 1592 BUG_ON(!op->nr_replicas);
1587 BUG_ON(!op->write_point.v); 1593 BUG_ON(!op->write_point.v);
1588 BUG_ON(bkey_eq(op->pos, POS_MAX)); 1594 BUG_ON(bkey_eq(op->pos, POS_MAX));
1589 <<
1590 if (op->flags & BCH_WRITE_ONLY_SPECIF <<
1591 op->flags |= BCH_WRITE_ALLOC_ <<
1592 1595
1593 op->nr_replicas_required = min_t(unsi 1596 op->nr_replicas_required = min_t(unsigned, op->nr_replicas_required, op->nr_replicas);
1594 op->start_time = local_clock(); 1597 op->start_time = local_clock();
1595 bch2_keylist_init(&op->insert_keys, o 1598 bch2_keylist_init(&op->insert_keys, op->inline_keys);
1596 wbio_init(bio)->put_bio = false; 1599 wbio_init(bio)->put_bio = false;
1597 1600
1598 if (bio->bi_iter.bi_size & (c->opts.b 1601 if (bio->bi_iter.bi_size & (c->opts.block_size - 1)) {
1599 bch_err_inum_offset_ratelimit 1602 bch_err_inum_offset_ratelimited(c,
1600 op->pos.inode, 1603 op->pos.inode,
1601 op->pos.offset << 9, 1604 op->pos.offset << 9,
1602 "%s write error: misa 1605 "%s write error: misaligned write",
1603 op->flags & BCH_WRITE 1606 op->flags & BCH_WRITE_MOVE ? "move" : "user");
1604 op->error = -EIO; 1607 op->error = -EIO;
1605 goto err; 1608 goto err;
1606 } 1609 }
1607 1610
1608 if (c->opts.nochanges) { 1611 if (c->opts.nochanges) {
1609 op->error = -BCH_ERR_erofs_no 1612 op->error = -BCH_ERR_erofs_no_writes;
1610 goto err; 1613 goto err;
1611 } 1614 }
1612 1615
1613 if (!(op->flags & BCH_WRITE_MOVE) && 1616 if (!(op->flags & BCH_WRITE_MOVE) &&
1614 !bch2_write_ref_tryget(c, BCH_WRI 1617 !bch2_write_ref_tryget(c, BCH_WRITE_REF_write)) {
1615 op->error = -BCH_ERR_erofs_no 1618 op->error = -BCH_ERR_erofs_no_writes;
1616 goto err; 1619 goto err;
1617 } 1620 }
1618 1621
1619 this_cpu_add(c->counters[BCH_COUNTER_ 1622 this_cpu_add(c->counters[BCH_COUNTER_io_write], bio_sectors(bio));
1620 bch2_increment_clock(c, bio_sectors(b 1623 bch2_increment_clock(c, bio_sectors(bio), WRITE);
1621 1624
1622 data_len = min_t(u64, bio->bi_iter.bi 1625 data_len = min_t(u64, bio->bi_iter.bi_size,
1623 op->new_i_size - (op 1626 op->new_i_size - (op->pos.offset << 9));
1624 1627
1625 if (c->opts.inline_data && 1628 if (c->opts.inline_data &&
1626 data_len <= min(block_bytes(c) / 1629 data_len <= min(block_bytes(c) / 2, 1024U)) {
1627 bch2_write_data_inline(op, da 1630 bch2_write_data_inline(op, data_len);
1628 return; 1631 return;
1629 } 1632 }
1630 1633
1631 __bch2_write(op); 1634 __bch2_write(op);
1632 return; 1635 return;
1633 err: 1636 err:
1634 bch2_disk_reservation_put(c, &op->res 1637 bch2_disk_reservation_put(c, &op->res);
1635 1638
1636 closure_debug_destroy(&op->cl); 1639 closure_debug_destroy(&op->cl);
1637 if (op->end_io) 1640 if (op->end_io)
1638 op->end_io(op); 1641 op->end_io(op);
1639 } 1642 }
1640 1643
1641 static const char * const bch2_write_flags[] 1644 static const char * const bch2_write_flags[] = {
1642 #define x(f) #f, 1645 #define x(f) #f,
1643 BCH_WRITE_FLAGS() 1646 BCH_WRITE_FLAGS()
1644 #undef x 1647 #undef x
1645 NULL 1648 NULL
1646 }; 1649 };
1647 1650
1648 void bch2_write_op_to_text(struct printbuf *o 1651 void bch2_write_op_to_text(struct printbuf *out, struct bch_write_op *op)
1649 { 1652 {
1650 prt_str(out, "pos: "); 1653 prt_str(out, "pos: ");
1651 bch2_bpos_to_text(out, op->pos); 1654 bch2_bpos_to_text(out, op->pos);
1652 prt_newline(out); 1655 prt_newline(out);
1653 printbuf_indent_add(out, 2); 1656 printbuf_indent_add(out, 2);
1654 1657
1655 prt_str(out, "started: "); 1658 prt_str(out, "started: ");
1656 bch2_pr_time_units(out, local_clock() 1659 bch2_pr_time_units(out, local_clock() - op->start_time);
1657 prt_newline(out); 1660 prt_newline(out);
1658 1661
1659 prt_str(out, "flags: "); 1662 prt_str(out, "flags: ");
1660 prt_bitflags(out, bch2_write_flags, o 1663 prt_bitflags(out, bch2_write_flags, op->flags);
1661 prt_newline(out); 1664 prt_newline(out);
1662 1665
1663 prt_printf(out, "ref: %u\n", closure_ 1666 prt_printf(out, "ref: %u\n", closure_nr_remaining(&op->cl));
1664 1667
1665 printbuf_indent_sub(out, 2); 1668 printbuf_indent_sub(out, 2);
1666 } 1669 }
1667 1670
1668 void bch2_fs_io_write_exit(struct bch_fs *c) 1671 void bch2_fs_io_write_exit(struct bch_fs *c)
1669 { 1672 {
1670 mempool_exit(&c->bio_bounce_pages); 1673 mempool_exit(&c->bio_bounce_pages);
1671 bioset_exit(&c->replica_set); 1674 bioset_exit(&c->replica_set);
1672 bioset_exit(&c->bio_write); 1675 bioset_exit(&c->bio_write);
1673 } 1676 }
1674 1677
1675 int bch2_fs_io_write_init(struct bch_fs *c) 1678 int bch2_fs_io_write_init(struct bch_fs *c)
1676 { 1679 {
1677 if (bioset_init(&c->bio_write, 1, o 1680 if (bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio), BIOSET_NEED_BVECS) ||
1678 bioset_init(&c->replica_set, 4, o 1681 bioset_init(&c->replica_set, 4, offsetof(struct bch_write_bio, bio), 0))
1679 return -BCH_ERR_ENOMEM_bio_wr 1682 return -BCH_ERR_ENOMEM_bio_write_init;
1680 1683
1681 if (mempool_init_page_pool(&c->bio_bo 1684 if (mempool_init_page_pool(&c->bio_bounce_pages,
1682 max_t(unsi 1685 max_t(unsigned,
1683 c->o 1686 c->opts.btree_node_size,
1684 c->o 1687 c->opts.encoded_extent_max) /
1685 PAGE_SIZE, 1688 PAGE_SIZE, 0))
1686 return -BCH_ERR_ENOMEM_bio_bo 1689 return -BCH_ERR_ENOMEM_bio_bounce_pages_init;
1687 1690
1688 return 0; 1691 return 0;
1689 } 1692 }
1690 1693
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.