Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | #ifndef NO_BCACHEFS_FS |
| 3 | |
| 4 | #include "bcachefs.h" |
| 5 | #include "btree_iter.h" |
| 6 | #include "extents.h" |
| 7 | #include "fs-io.h" |
| 8 | #include "fs-io-pagecache.h" |
| 9 | #include "subvolume.h" |
| 10 | |
| 11 | #include <linux/pagevec.h> |
| 12 | #include <linux/writeback.h> |
| 13 | |
| 14 | int bch2_filemap_get_contig_folios_d(struct address_space *mapping, |
| 15 | loff_t start, u64 end, |
Kent Overstreet | c4f1f80 | 2023-11-11 12:30:19 -0500 | [diff] [blame] | 16 | fgf_t fgp_flags, gfp_t gfp, |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 17 | folios *fs) |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 18 | { |
| 19 | struct folio *f; |
| 20 | u64 pos = start; |
| 21 | int ret = 0; |
| 22 | |
| 23 | while (pos < end) { |
| 24 | if ((u64) pos >= (u64) start + (1ULL << 20)) |
| 25 | fgp_flags &= ~FGP_CREAT; |
| 26 | |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 27 | ret = darray_make_room_gfp(fs, 1, gfp & GFP_KERNEL); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 28 | if (ret) |
| 29 | break; |
| 30 | |
| 31 | f = __filemap_get_folio(mapping, pos >> PAGE_SHIFT, fgp_flags, gfp); |
| 32 | if (IS_ERR_OR_NULL(f)) |
| 33 | break; |
| 34 | |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 35 | BUG_ON(fs->nr && folio_pos(f) != pos); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 36 | |
| 37 | pos = folio_end_pos(f); |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 38 | darray_push(fs, f); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 39 | } |
| 40 | |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 41 | if (!fs->nr && !ret && (fgp_flags & FGP_CREAT)) |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 42 | ret = -ENOMEM; |
| 43 | |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 44 | return fs->nr ? 0 : ret; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 45 | } |
| 46 | |
| 47 | /* pagecache_block must be held */ |
| 48 | int bch2_write_invalidate_inode_pages_range(struct address_space *mapping, |
| 49 | loff_t start, loff_t end) |
| 50 | { |
| 51 | int ret; |
| 52 | |
| 53 | /* |
| 54 | * XXX: the way this is currently implemented, we can spin if a process |
| 55 | * is continually redirtying a specific page |
| 56 | */ |
| 57 | do { |
| 58 | if (!mapping->nrpages) |
| 59 | return 0; |
| 60 | |
| 61 | ret = filemap_write_and_wait_range(mapping, start, end); |
| 62 | if (ret) |
| 63 | break; |
| 64 | |
| 65 | if (!mapping->nrpages) |
| 66 | return 0; |
| 67 | |
| 68 | ret = invalidate_inode_pages2_range(mapping, |
| 69 | start >> PAGE_SHIFT, |
| 70 | end >> PAGE_SHIFT); |
| 71 | } while (ret == -EBUSY); |
| 72 | |
| 73 | return ret; |
| 74 | } |
| 75 | |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 76 | #if 0 |
| 77 | /* Useful for debug tracing: */ |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 78 | static const char * const bch2_folio_sector_states[] = { |
| 79 | #define x(n) #n, |
| 80 | BCH_FOLIO_SECTOR_STATE() |
| 81 | #undef x |
| 82 | NULL |
| 83 | }; |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 84 | #endif |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 85 | |
| 86 | static inline enum bch_folio_sector_state |
| 87 | folio_sector_dirty(enum bch_folio_sector_state state) |
| 88 | { |
| 89 | switch (state) { |
| 90 | case SECTOR_unallocated: |
| 91 | return SECTOR_dirty; |
| 92 | case SECTOR_reserved: |
| 93 | return SECTOR_dirty_reserved; |
| 94 | default: |
| 95 | return state; |
| 96 | } |
| 97 | } |
| 98 | |
| 99 | static inline enum bch_folio_sector_state |
| 100 | folio_sector_undirty(enum bch_folio_sector_state state) |
| 101 | { |
| 102 | switch (state) { |
| 103 | case SECTOR_dirty: |
| 104 | return SECTOR_unallocated; |
| 105 | case SECTOR_dirty_reserved: |
| 106 | return SECTOR_reserved; |
| 107 | default: |
| 108 | return state; |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | static inline enum bch_folio_sector_state |
| 113 | folio_sector_reserve(enum bch_folio_sector_state state) |
| 114 | { |
| 115 | switch (state) { |
| 116 | case SECTOR_unallocated: |
| 117 | return SECTOR_reserved; |
| 118 | case SECTOR_dirty: |
| 119 | return SECTOR_dirty_reserved; |
| 120 | default: |
| 121 | return state; |
| 122 | } |
| 123 | } |
| 124 | |
| 125 | /* for newly allocated folios: */ |
| 126 | struct bch_folio *__bch2_folio_create(struct folio *folio, gfp_t gfp) |
| 127 | { |
| 128 | struct bch_folio *s; |
| 129 | |
| 130 | s = kzalloc(sizeof(*s) + |
| 131 | sizeof(struct bch_folio_sector) * |
| 132 | folio_sectors(folio), gfp); |
| 133 | if (!s) |
| 134 | return NULL; |
| 135 | |
| 136 | spin_lock_init(&s->lock); |
| 137 | folio_attach_private(folio, s); |
| 138 | return s; |
| 139 | } |
| 140 | |
| 141 | struct bch_folio *bch2_folio_create(struct folio *folio, gfp_t gfp) |
| 142 | { |
| 143 | return bch2_folio(folio) ?: __bch2_folio_create(folio, gfp); |
| 144 | } |
| 145 | |
| 146 | static unsigned bkey_to_sector_state(struct bkey_s_c k) |
| 147 | { |
| 148 | if (bkey_extent_is_reservation(k)) |
| 149 | return SECTOR_reserved; |
| 150 | if (bkey_extent_is_allocation(k.k)) |
| 151 | return SECTOR_allocated; |
| 152 | return SECTOR_unallocated; |
| 153 | } |
| 154 | |
| 155 | static void __bch2_folio_set(struct folio *folio, |
| 156 | unsigned pg_offset, unsigned pg_len, |
| 157 | unsigned nr_ptrs, unsigned state) |
| 158 | { |
| 159 | struct bch_folio *s = bch2_folio(folio); |
| 160 | unsigned i, sectors = folio_sectors(folio); |
| 161 | |
| 162 | BUG_ON(pg_offset >= sectors); |
| 163 | BUG_ON(pg_offset + pg_len > sectors); |
| 164 | |
| 165 | spin_lock(&s->lock); |
| 166 | |
| 167 | for (i = pg_offset; i < pg_offset + pg_len; i++) { |
| 168 | s->s[i].nr_replicas = nr_ptrs; |
| 169 | bch2_folio_sector_set(folio, s, i, state); |
| 170 | } |
| 171 | |
| 172 | if (i == sectors) |
| 173 | s->uptodate = true; |
| 174 | |
| 175 | spin_unlock(&s->lock); |
| 176 | } |
| 177 | |
| 178 | /* |
| 179 | * Initialize bch_folio state (allocated/unallocated, nr_replicas) from the |
| 180 | * extents btree: |
| 181 | */ |
| 182 | int bch2_folio_set(struct bch_fs *c, subvol_inum inum, |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 183 | struct folio **fs, unsigned nr_folios) |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 184 | { |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 185 | struct btree_trans *trans; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 186 | struct btree_iter iter; |
| 187 | struct bkey_s_c k; |
| 188 | struct bch_folio *s; |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 189 | u64 offset = folio_sector(fs[0]); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 190 | unsigned folio_idx; |
| 191 | u32 snapshot; |
| 192 | bool need_set = false; |
| 193 | int ret; |
| 194 | |
| 195 | for (folio_idx = 0; folio_idx < nr_folios; folio_idx++) { |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 196 | s = bch2_folio_create(fs[folio_idx], GFP_KERNEL); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 197 | if (!s) |
| 198 | return -ENOMEM; |
| 199 | |
| 200 | need_set |= !s->uptodate; |
| 201 | } |
| 202 | |
| 203 | if (!need_set) |
| 204 | return 0; |
| 205 | |
| 206 | folio_idx = 0; |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 207 | trans = bch2_trans_get(c); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 208 | retry: |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 209 | bch2_trans_begin(trans); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 210 | |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 211 | ret = bch2_subvolume_get_snapshot(trans, inum.subvol, &snapshot); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 212 | if (ret) |
| 213 | goto err; |
| 214 | |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 215 | for_each_btree_key_norestart(trans, iter, BTREE_ID_extents, |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 216 | SPOS(inum.inum, offset, snapshot), |
Kent Overstreet | 5dd8c60 | 2024-04-07 18:05:34 -0400 | [diff] [blame] | 217 | BTREE_ITER_slots, k, ret) { |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 218 | unsigned nr_ptrs = bch2_bkey_nr_ptrs_fully_allocated(k); |
| 219 | unsigned state = bkey_to_sector_state(k); |
| 220 | |
| 221 | while (folio_idx < nr_folios) { |
Kent Overstreet | 96dea3d | 2023-09-12 18:41:22 -0400 | [diff] [blame] | 222 | struct folio *folio = fs[folio_idx]; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 223 | u64 folio_start = folio_sector(folio); |
| 224 | u64 folio_end = folio_end_sector(folio); |
| 225 | unsigned folio_offset = max(bkey_start_offset(k.k), folio_start) - |
| 226 | folio_start; |
| 227 | unsigned folio_len = min(k.k->p.offset, folio_end) - |
| 228 | folio_offset - folio_start; |
| 229 | |
| 230 | BUG_ON(k.k->p.offset < folio_start); |
| 231 | BUG_ON(bkey_start_offset(k.k) > folio_end); |
| 232 | |
| 233 | if (!bch2_folio(folio)->uptodate) |
| 234 | __bch2_folio_set(folio, folio_offset, folio_len, nr_ptrs, state); |
| 235 | |
| 236 | if (k.k->p.offset < folio_end) |
| 237 | break; |
| 238 | folio_idx++; |
| 239 | } |
| 240 | |
| 241 | if (folio_idx == nr_folios) |
| 242 | break; |
| 243 | } |
| 244 | |
| 245 | offset = iter.pos.offset; |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 246 | bch2_trans_iter_exit(trans, &iter); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 247 | err: |
| 248 | if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) |
| 249 | goto retry; |
Kent Overstreet | 6bd68ec | 2023-09-12 17:16:02 -0400 | [diff] [blame] | 250 | bch2_trans_put(trans); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 251 | |
| 252 | return ret; |
| 253 | } |
| 254 | |
| 255 | void bch2_bio_page_state_set(struct bio *bio, struct bkey_s_c k) |
| 256 | { |
| 257 | struct bvec_iter iter; |
| 258 | struct folio_vec fv; |
| 259 | unsigned nr_ptrs = k.k->type == KEY_TYPE_reflink_v |
| 260 | ? 0 : bch2_bkey_nr_ptrs_fully_allocated(k); |
| 261 | unsigned state = bkey_to_sector_state(k); |
| 262 | |
| 263 | bio_for_each_folio(fv, bio, iter) |
| 264 | __bch2_folio_set(fv.fv_folio, |
| 265 | fv.fv_offset >> 9, |
| 266 | fv.fv_len >> 9, |
| 267 | nr_ptrs, state); |
| 268 | } |
| 269 | |
| 270 | void bch2_mark_pagecache_unallocated(struct bch_inode_info *inode, |
| 271 | u64 start, u64 end) |
| 272 | { |
| 273 | pgoff_t index = start >> PAGE_SECTORS_SHIFT; |
| 274 | pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; |
| 275 | struct folio_batch fbatch; |
| 276 | unsigned i, j; |
| 277 | |
| 278 | if (end <= start) |
| 279 | return; |
| 280 | |
| 281 | folio_batch_init(&fbatch); |
| 282 | |
| 283 | while (filemap_get_folios(inode->v.i_mapping, |
| 284 | &index, end_index, &fbatch)) { |
| 285 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 286 | struct folio *folio = fbatch.folios[i]; |
| 287 | u64 folio_start = folio_sector(folio); |
| 288 | u64 folio_end = folio_end_sector(folio); |
| 289 | unsigned folio_offset = max(start, folio_start) - folio_start; |
| 290 | unsigned folio_len = min(end, folio_end) - folio_offset - folio_start; |
| 291 | struct bch_folio *s; |
| 292 | |
| 293 | BUG_ON(end <= folio_start); |
| 294 | |
| 295 | folio_lock(folio); |
| 296 | s = bch2_folio(folio); |
| 297 | |
| 298 | if (s) { |
| 299 | spin_lock(&s->lock); |
| 300 | for (j = folio_offset; j < folio_offset + folio_len; j++) |
| 301 | s->s[j].nr_replicas = 0; |
| 302 | spin_unlock(&s->lock); |
| 303 | } |
| 304 | |
| 305 | folio_unlock(folio); |
| 306 | } |
| 307 | folio_batch_release(&fbatch); |
| 308 | cond_resched(); |
| 309 | } |
| 310 | } |
| 311 | |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 312 | int bch2_mark_pagecache_reserved(struct bch_inode_info *inode, |
| 313 | u64 *start, u64 end, |
| 314 | bool nonblocking) |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 315 | { |
| 316 | struct bch_fs *c = inode->v.i_sb->s_fs_info; |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 317 | pgoff_t index = *start >> PAGE_SECTORS_SHIFT; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 318 | pgoff_t end_index = (end - 1) >> PAGE_SECTORS_SHIFT; |
| 319 | struct folio_batch fbatch; |
| 320 | s64 i_sectors_delta = 0; |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 321 | int ret = 0; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 322 | |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 323 | if (end <= *start) |
| 324 | return 0; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 325 | |
| 326 | folio_batch_init(&fbatch); |
| 327 | |
| 328 | while (filemap_get_folios(inode->v.i_mapping, |
| 329 | &index, end_index, &fbatch)) { |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 330 | for (unsigned i = 0; i < folio_batch_count(&fbatch); i++) { |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 331 | struct folio *folio = fbatch.folios[i]; |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 332 | |
| 333 | if (!nonblocking) |
| 334 | folio_lock(folio); |
| 335 | else if (!folio_trylock(folio)) { |
| 336 | folio_batch_release(&fbatch); |
| 337 | ret = -EAGAIN; |
| 338 | break; |
| 339 | } |
| 340 | |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 341 | u64 folio_start = folio_sector(folio); |
| 342 | u64 folio_end = folio_end_sector(folio); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 343 | |
| 344 | BUG_ON(end <= folio_start); |
| 345 | |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 346 | *start = min(end, folio_end); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 347 | |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 348 | struct bch_folio *s = bch2_folio(folio); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 349 | if (s) { |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 350 | unsigned folio_offset = max(*start, folio_start) - folio_start; |
| 351 | unsigned folio_len = min(end, folio_end) - folio_offset - folio_start; |
| 352 | |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 353 | spin_lock(&s->lock); |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 354 | for (unsigned j = folio_offset; j < folio_offset + folio_len; j++) { |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 355 | i_sectors_delta -= s->s[j].state == SECTOR_dirty; |
| 356 | bch2_folio_sector_set(folio, s, j, |
| 357 | folio_sector_reserve(s->s[j].state)); |
| 358 | } |
| 359 | spin_unlock(&s->lock); |
| 360 | } |
| 361 | |
| 362 | folio_unlock(folio); |
| 363 | } |
| 364 | folio_batch_release(&fbatch); |
| 365 | cond_resched(); |
| 366 | } |
| 367 | |
| 368 | bch2_i_sectors_acct(c, inode, NULL, i_sectors_delta); |
Kent Overstreet | 46bf2e9 | 2024-01-15 20:37:23 -0500 | [diff] [blame] | 369 | return ret; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 370 | } |
| 371 | |
| 372 | static inline unsigned sectors_to_reserve(struct bch_folio_sector *s, |
| 373 | unsigned nr_replicas) |
| 374 | { |
| 375 | return max(0, (int) nr_replicas - |
| 376 | s->nr_replicas - |
| 377 | s->replicas_reserved); |
| 378 | } |
| 379 | |
| 380 | int bch2_get_folio_disk_reservation(struct bch_fs *c, |
| 381 | struct bch_inode_info *inode, |
| 382 | struct folio *folio, bool check_enospc) |
| 383 | { |
| 384 | struct bch_folio *s = bch2_folio_create(folio, 0); |
| 385 | unsigned nr_replicas = inode_nr_replicas(c, inode); |
| 386 | struct disk_reservation disk_res = { 0 }; |
| 387 | unsigned i, sectors = folio_sectors(folio), disk_res_sectors = 0; |
| 388 | int ret; |
| 389 | |
| 390 | if (!s) |
| 391 | return -ENOMEM; |
| 392 | |
| 393 | for (i = 0; i < sectors; i++) |
| 394 | disk_res_sectors += sectors_to_reserve(&s->s[i], nr_replicas); |
| 395 | |
| 396 | if (!disk_res_sectors) |
| 397 | return 0; |
| 398 | |
| 399 | ret = bch2_disk_reservation_get(c, &disk_res, |
| 400 | disk_res_sectors, 1, |
| 401 | !check_enospc |
| 402 | ? BCH_DISK_RESERVATION_NOFAIL |
| 403 | : 0); |
| 404 | if (unlikely(ret)) |
| 405 | return ret; |
| 406 | |
| 407 | for (i = 0; i < sectors; i++) |
| 408 | s->s[i].replicas_reserved += |
| 409 | sectors_to_reserve(&s->s[i], nr_replicas); |
| 410 | |
| 411 | return 0; |
| 412 | } |
| 413 | |
| 414 | void bch2_folio_reservation_put(struct bch_fs *c, |
| 415 | struct bch_inode_info *inode, |
| 416 | struct bch2_folio_reservation *res) |
| 417 | { |
| 418 | bch2_disk_reservation_put(c, &res->disk); |
| 419 | bch2_quota_reservation_put(c, inode, &res->quota); |
| 420 | } |
| 421 | |
| 422 | int bch2_folio_reservation_get(struct bch_fs *c, |
| 423 | struct bch_inode_info *inode, |
| 424 | struct folio *folio, |
| 425 | struct bch2_folio_reservation *res, |
Kent Overstreet | 7554a8bb6 | 2023-09-10 17:29:39 -0400 | [diff] [blame] | 426 | size_t offset, size_t len) |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 427 | { |
| 428 | struct bch_folio *s = bch2_folio_create(folio, 0); |
| 429 | unsigned i, disk_sectors = 0, quota_sectors = 0; |
| 430 | int ret; |
| 431 | |
| 432 | if (!s) |
| 433 | return -ENOMEM; |
| 434 | |
| 435 | BUG_ON(!s->uptodate); |
| 436 | |
| 437 | for (i = round_down(offset, block_bytes(c)) >> 9; |
| 438 | i < round_up(offset + len, block_bytes(c)) >> 9; |
| 439 | i++) { |
Kent Overstreet | 7554a8bb6 | 2023-09-10 17:29:39 -0400 | [diff] [blame] | 440 | disk_sectors += sectors_to_reserve(&s->s[i], res->disk.nr_replicas); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 441 | quota_sectors += s->s[i].state == SECTOR_unallocated; |
| 442 | } |
| 443 | |
| 444 | if (disk_sectors) { |
| 445 | ret = bch2_disk_reservation_add(c, &res->disk, disk_sectors, 0); |
| 446 | if (unlikely(ret)) |
| 447 | return ret; |
| 448 | } |
| 449 | |
| 450 | if (quota_sectors) { |
Kent Overstreet | 7554a8bb6 | 2023-09-10 17:29:39 -0400 | [diff] [blame] | 451 | ret = bch2_quota_reservation_add(c, inode, &res->quota, quota_sectors, true); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 452 | if (unlikely(ret)) { |
Kent Overstreet | 7554a8bb6 | 2023-09-10 17:29:39 -0400 | [diff] [blame] | 453 | struct disk_reservation tmp = { .sectors = disk_sectors }; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 454 | |
| 455 | bch2_disk_reservation_put(c, &tmp); |
| 456 | res->disk.sectors -= disk_sectors; |
| 457 | return ret; |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | return 0; |
| 462 | } |
| 463 | |
Kent Overstreet | 7554a8bb6 | 2023-09-10 17:29:39 -0400 | [diff] [blame] | 464 | ssize_t bch2_folio_reservation_get_partial(struct bch_fs *c, |
| 465 | struct bch_inode_info *inode, |
| 466 | struct folio *folio, |
| 467 | struct bch2_folio_reservation *res, |
| 468 | size_t offset, size_t len) |
| 469 | { |
| 470 | size_t l, reserved = 0; |
| 471 | int ret; |
| 472 | |
| 473 | while ((l = len - reserved)) { |
| 474 | while ((ret = bch2_folio_reservation_get(c, inode, folio, res, offset, l))) { |
| 475 | if ((offset & (block_bytes(c) - 1)) + l <= block_bytes(c)) |
| 476 | return reserved ?: ret; |
| 477 | |
| 478 | len = reserved + l; |
| 479 | l /= 2; |
| 480 | } |
| 481 | |
| 482 | offset += l; |
| 483 | reserved += l; |
| 484 | } |
| 485 | |
| 486 | return reserved; |
| 487 | } |
| 488 | |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 489 | static void bch2_clear_folio_bits(struct folio *folio) |
| 490 | { |
| 491 | struct bch_inode_info *inode = to_bch_ei(folio->mapping->host); |
| 492 | struct bch_fs *c = inode->v.i_sb->s_fs_info; |
| 493 | struct bch_folio *s = bch2_folio(folio); |
| 494 | struct disk_reservation disk_res = { 0 }; |
| 495 | int i, sectors = folio_sectors(folio), dirty_sectors = 0; |
| 496 | |
| 497 | if (!s) |
| 498 | return; |
| 499 | |
| 500 | EBUG_ON(!folio_test_locked(folio)); |
| 501 | EBUG_ON(folio_test_writeback(folio)); |
| 502 | |
| 503 | for (i = 0; i < sectors; i++) { |
| 504 | disk_res.sectors += s->s[i].replicas_reserved; |
| 505 | s->s[i].replicas_reserved = 0; |
| 506 | |
| 507 | dirty_sectors -= s->s[i].state == SECTOR_dirty; |
| 508 | bch2_folio_sector_set(folio, s, i, folio_sector_undirty(s->s[i].state)); |
| 509 | } |
| 510 | |
| 511 | bch2_disk_reservation_put(c, &disk_res); |
| 512 | |
| 513 | bch2_i_sectors_acct(c, inode, NULL, dirty_sectors); |
| 514 | |
| 515 | bch2_folio_release(folio); |
| 516 | } |
| 517 | |
| 518 | void bch2_set_folio_dirty(struct bch_fs *c, |
| 519 | struct bch_inode_info *inode, |
| 520 | struct folio *folio, |
| 521 | struct bch2_folio_reservation *res, |
| 522 | unsigned offset, unsigned len) |
| 523 | { |
| 524 | struct bch_folio *s = bch2_folio(folio); |
| 525 | unsigned i, dirty_sectors = 0; |
| 526 | |
| 527 | WARN_ON((u64) folio_pos(folio) + offset + len > |
| 528 | round_up((u64) i_size_read(&inode->v), block_bytes(c))); |
| 529 | |
| 530 | BUG_ON(!s->uptodate); |
| 531 | |
| 532 | spin_lock(&s->lock); |
| 533 | |
| 534 | for (i = round_down(offset, block_bytes(c)) >> 9; |
| 535 | i < round_up(offset + len, block_bytes(c)) >> 9; |
| 536 | i++) { |
| 537 | unsigned sectors = sectors_to_reserve(&s->s[i], |
| 538 | res->disk.nr_replicas); |
| 539 | |
| 540 | /* |
| 541 | * This can happen if we race with the error path in |
| 542 | * bch2_writepage_io_done(): |
| 543 | */ |
| 544 | sectors = min_t(unsigned, sectors, res->disk.sectors); |
| 545 | |
| 546 | s->s[i].replicas_reserved += sectors; |
| 547 | res->disk.sectors -= sectors; |
| 548 | |
| 549 | dirty_sectors += s->s[i].state == SECTOR_unallocated; |
| 550 | |
| 551 | bch2_folio_sector_set(folio, s, i, folio_sector_dirty(s->s[i].state)); |
| 552 | } |
| 553 | |
| 554 | spin_unlock(&s->lock); |
| 555 | |
| 556 | bch2_i_sectors_acct(c, inode, &res->quota, dirty_sectors); |
| 557 | |
| 558 | if (!folio_test_dirty(folio)) |
| 559 | filemap_dirty_folio(inode->v.i_mapping, folio); |
| 560 | } |
| 561 | |
| 562 | vm_fault_t bch2_page_fault(struct vm_fault *vmf) |
| 563 | { |
| 564 | struct file *file = vmf->vma->vm_file; |
| 565 | struct address_space *mapping = file->f_mapping; |
| 566 | struct address_space *fdm = faults_disabled_mapping(); |
| 567 | struct bch_inode_info *inode = file_bch_inode(file); |
| 568 | vm_fault_t ret; |
| 569 | |
| 570 | if (fdm == mapping) |
| 571 | return VM_FAULT_SIGBUS; |
| 572 | |
| 573 | /* Lock ordering: */ |
| 574 | if (fdm > mapping) { |
| 575 | struct bch_inode_info *fdm_host = to_bch_ei(fdm->host); |
| 576 | |
| 577 | if (bch2_pagecache_add_tryget(inode)) |
| 578 | goto got_lock; |
| 579 | |
| 580 | bch2_pagecache_block_put(fdm_host); |
| 581 | |
| 582 | bch2_pagecache_add_get(inode); |
| 583 | bch2_pagecache_add_put(inode); |
| 584 | |
| 585 | bch2_pagecache_block_get(fdm_host); |
| 586 | |
| 587 | /* Signal that lock has been dropped: */ |
| 588 | set_fdm_dropped_locks(); |
| 589 | return VM_FAULT_SIGBUS; |
| 590 | } |
| 591 | |
| 592 | bch2_pagecache_add_get(inode); |
| 593 | got_lock: |
| 594 | ret = filemap_fault(vmf); |
| 595 | bch2_pagecache_add_put(inode); |
| 596 | |
| 597 | return ret; |
| 598 | } |
| 599 | |
| 600 | vm_fault_t bch2_page_mkwrite(struct vm_fault *vmf) |
| 601 | { |
| 602 | struct folio *folio = page_folio(vmf->page); |
| 603 | struct file *file = vmf->vma->vm_file; |
| 604 | struct bch_inode_info *inode = file_bch_inode(file); |
| 605 | struct address_space *mapping = file->f_mapping; |
| 606 | struct bch_fs *c = inode->v.i_sb->s_fs_info; |
| 607 | struct bch2_folio_reservation res; |
| 608 | unsigned len; |
| 609 | loff_t isize; |
| 610 | vm_fault_t ret; |
| 611 | |
| 612 | bch2_folio_reservation_init(c, inode, &res); |
| 613 | |
| 614 | sb_start_pagefault(inode->v.i_sb); |
| 615 | file_update_time(file); |
| 616 | |
| 617 | /* |
| 618 | * Not strictly necessary, but helps avoid dio writes livelocking in |
| 619 | * bch2_write_invalidate_inode_pages_range() - can drop this if/when we get |
| 620 | * a bch2_write_invalidate_inode_pages_range() that works without dropping |
| 621 | * page lock before invalidating page |
| 622 | */ |
| 623 | bch2_pagecache_add_get(inode); |
| 624 | |
| 625 | folio_lock(folio); |
| 626 | isize = i_size_read(&inode->v); |
| 627 | |
| 628 | if (folio->mapping != mapping || folio_pos(folio) >= isize) { |
| 629 | folio_unlock(folio); |
| 630 | ret = VM_FAULT_NOPAGE; |
| 631 | goto out; |
| 632 | } |
| 633 | |
| 634 | len = min_t(loff_t, folio_size(folio), isize - folio_pos(folio)); |
| 635 | |
| 636 | if (bch2_folio_set(c, inode_inum(inode), &folio, 1) ?: |
| 637 | bch2_folio_reservation_get(c, inode, folio, &res, 0, len)) { |
| 638 | folio_unlock(folio); |
| 639 | ret = VM_FAULT_SIGBUS; |
| 640 | goto out; |
| 641 | } |
| 642 | |
| 643 | bch2_set_folio_dirty(c, inode, folio, &res, 0, len); |
| 644 | bch2_folio_reservation_put(c, inode, &res); |
| 645 | |
| 646 | folio_wait_stable(folio); |
| 647 | ret = VM_FAULT_LOCKED; |
| 648 | out: |
| 649 | bch2_pagecache_add_put(inode); |
| 650 | sb_end_pagefault(inode->v.i_sb); |
| 651 | |
| 652 | return ret; |
| 653 | } |
| 654 | |
| 655 | void bch2_invalidate_folio(struct folio *folio, size_t offset, size_t length) |
| 656 | { |
| 657 | if (offset || length < folio_size(folio)) |
| 658 | return; |
| 659 | |
| 660 | bch2_clear_folio_bits(folio); |
| 661 | } |
| 662 | |
| 663 | bool bch2_release_folio(struct folio *folio, gfp_t gfp_mask) |
| 664 | { |
| 665 | if (folio_test_dirty(folio) || folio_test_writeback(folio)) |
| 666 | return false; |
| 667 | |
| 668 | bch2_clear_folio_bits(folio); |
| 669 | return true; |
| 670 | } |
| 671 | |
| 672 | /* fseek: */ |
| 673 | |
| 674 | static int folio_data_offset(struct folio *folio, loff_t pos, |
| 675 | unsigned min_replicas) |
| 676 | { |
| 677 | struct bch_folio *s = bch2_folio(folio); |
| 678 | unsigned i, sectors = folio_sectors(folio); |
| 679 | |
| 680 | if (s) |
| 681 | for (i = folio_pos_to_s(folio, pos); i < sectors; i++) |
| 682 | if (s->s[i].state >= SECTOR_dirty && |
| 683 | s->s[i].nr_replicas + s->s[i].replicas_reserved >= min_replicas) |
| 684 | return i << SECTOR_SHIFT; |
| 685 | |
| 686 | return -1; |
| 687 | } |
| 688 | |
| 689 | loff_t bch2_seek_pagecache_data(struct inode *vinode, |
| 690 | loff_t start_offset, |
| 691 | loff_t end_offset, |
| 692 | unsigned min_replicas, |
| 693 | bool nonblock) |
| 694 | { |
| 695 | struct folio_batch fbatch; |
| 696 | pgoff_t start_index = start_offset >> PAGE_SHIFT; |
| 697 | pgoff_t end_index = end_offset >> PAGE_SHIFT; |
| 698 | pgoff_t index = start_index; |
| 699 | unsigned i; |
| 700 | loff_t ret; |
| 701 | int offset; |
| 702 | |
| 703 | folio_batch_init(&fbatch); |
| 704 | |
| 705 | while (filemap_get_folios(vinode->i_mapping, |
| 706 | &index, end_index, &fbatch)) { |
| 707 | for (i = 0; i < folio_batch_count(&fbatch); i++) { |
| 708 | struct folio *folio = fbatch.folios[i]; |
| 709 | |
| 710 | if (!nonblock) { |
| 711 | folio_lock(folio); |
| 712 | } else if (!folio_trylock(folio)) { |
| 713 | folio_batch_release(&fbatch); |
| 714 | return -EAGAIN; |
| 715 | } |
| 716 | |
| 717 | offset = folio_data_offset(folio, |
| 718 | max(folio_pos(folio), start_offset), |
| 719 | min_replicas); |
| 720 | if (offset >= 0) { |
| 721 | ret = clamp(folio_pos(folio) + offset, |
| 722 | start_offset, end_offset); |
| 723 | folio_unlock(folio); |
| 724 | folio_batch_release(&fbatch); |
| 725 | return ret; |
| 726 | } |
| 727 | folio_unlock(folio); |
| 728 | } |
| 729 | folio_batch_release(&fbatch); |
| 730 | cond_resched(); |
| 731 | } |
| 732 | |
| 733 | return end_offset; |
| 734 | } |
| 735 | |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 736 | /* |
| 737 | * Search for a hole in a folio. |
| 738 | * |
| 739 | * The filemap layer returns -ENOENT if no folio exists, so reuse the same error |
| 740 | * code to indicate a pagecache hole exists at the returned offset. Otherwise |
| 741 | * return 0 if the folio is filled with data, or an error code. This function |
| 742 | * can return -EAGAIN if nonblock is specified. |
| 743 | */ |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 744 | static int folio_hole_offset(struct address_space *mapping, loff_t *offset, |
| 745 | unsigned min_replicas, bool nonblock) |
| 746 | { |
| 747 | struct folio *folio; |
| 748 | struct bch_folio *s; |
| 749 | unsigned i, sectors; |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 750 | int ret = -ENOENT; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 751 | |
| 752 | folio = __filemap_get_folio(mapping, *offset >> PAGE_SHIFT, |
| 753 | FGP_LOCK|(nonblock ? FGP_NOWAIT : 0), 0); |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 754 | if (IS_ERR(folio)) |
| 755 | return PTR_ERR(folio); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 756 | |
| 757 | s = bch2_folio(folio); |
| 758 | if (!s) |
| 759 | goto unlock; |
| 760 | |
| 761 | sectors = folio_sectors(folio); |
| 762 | for (i = folio_pos_to_s(folio, *offset); i < sectors; i++) |
| 763 | if (s->s[i].state < SECTOR_dirty || |
| 764 | s->s[i].nr_replicas + s->s[i].replicas_reserved < min_replicas) { |
| 765 | *offset = max(*offset, |
| 766 | folio_pos(folio) + (i << SECTOR_SHIFT)); |
| 767 | goto unlock; |
| 768 | } |
| 769 | |
| 770 | *offset = folio_end_pos(folio); |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 771 | ret = 0; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 772 | unlock: |
| 773 | folio_unlock(folio); |
| 774 | folio_put(folio); |
| 775 | return ret; |
| 776 | } |
| 777 | |
| 778 | loff_t bch2_seek_pagecache_hole(struct inode *vinode, |
| 779 | loff_t start_offset, |
| 780 | loff_t end_offset, |
| 781 | unsigned min_replicas, |
| 782 | bool nonblock) |
| 783 | { |
| 784 | struct address_space *mapping = vinode->i_mapping; |
| 785 | loff_t offset = start_offset; |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 786 | loff_t ret = 0; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 787 | |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 788 | while (!ret && offset < end_offset) |
| 789 | ret = folio_hole_offset(mapping, &offset, min_replicas, nonblock); |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 790 | |
Brian Foster | 8c9b0f7 | 2023-08-14 10:49:42 -0400 | [diff] [blame] | 791 | if (ret && ret != -ENOENT) |
| 792 | return ret; |
Kent Overstreet | dbbfca9 | 2023-08-03 18:18:21 -0400 | [diff] [blame] | 793 | return min(offset, end_offset); |
| 794 | } |
| 795 | |
| 796 | int bch2_clamp_data_hole(struct inode *inode, |
| 797 | u64 *hole_start, |
| 798 | u64 *hole_end, |
| 799 | unsigned min_replicas, |
| 800 | bool nonblock) |
| 801 | { |
| 802 | loff_t ret; |
| 803 | |
| 804 | ret = bch2_seek_pagecache_hole(inode, |
| 805 | *hole_start << 9, *hole_end << 9, min_replicas, nonblock) >> 9; |
| 806 | if (ret < 0) |
| 807 | return ret; |
| 808 | |
| 809 | *hole_start = ret; |
| 810 | |
| 811 | if (*hole_start == *hole_end) |
| 812 | return 0; |
| 813 | |
| 814 | ret = bch2_seek_pagecache_data(inode, |
| 815 | *hole_start << 9, *hole_end << 9, min_replicas, nonblock) >> 9; |
| 816 | if (ret < 0) |
| 817 | return ret; |
| 818 | |
| 819 | *hole_end = ret; |
| 820 | return 0; |
| 821 | } |
| 822 | |
| 823 | #endif /* NO_BCACHEFS_FS */ |