fs/f2fs/node.c

   1 /*
   2  * fs/f2fs/node.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include <linux/mpage.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/blkdev.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/swap.h>
  18
  19 #include "f2fs.h"
  20 #include "node.h"
  21 #include "segment.h"
  22 #include <trace/events/f2fs.h>
  23
  24 static struct kmem_cache *nat_entry_slab;
  25 static struct kmem_cache *free_nid_slab;
  26
  27 static void clear_node_page_dirty(struct page *page)
  28 {
  29         struct address_space *mapping = page->mapping;
  30         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
  31         unsigned int long flags;
  32
  33         if (PageDirty(page)) {
  34                 spin_lock_irqsave(&mapping->tree_lock, flags);
  35                 radix_tree_tag_clear(&mapping->page_tree,
  36                                 page_index(page),
  37                                 PAGECACHE_TAG_DIRTY);
  38                 spin_unlock_irqrestore(&mapping->tree_lock, flags);
  39
  40                 clear_page_dirty_for_io(page);
  41                 dec_page_count(sbi, F2FS_DIRTY_NODES);
  42         }
  43         ClearPageUptodate(page);
  44 }
  45
  46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  47 {
  48         pgoff_t index = current_nat_addr(sbi, nid);
  49         return get_meta_page(sbi, index);
  50 }
  51
  52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
  53 {
  54         struct page *src_page;
  55         struct page *dst_page;
  56         pgoff_t src_off;
  57         pgoff_t dst_off;
  58         void *src_addr;
  59         void *dst_addr;
  60         struct f2fs_nm_info *nm_i = NM_I(sbi);
  61
  62         src_off = current_nat_addr(sbi, nid);
  63         dst_off = next_nat_addr(sbi, src_off);
  64
  65         /* get current nat block page with lock */
  66         src_page = get_meta_page(sbi, src_off);
  67
  68         /* Dirty src_page means that it is already the new target NAT page. */
  69         if (PageDirty(src_page))
  70                 return src_page;
  71
  72         dst_page = grab_meta_page(sbi, dst_off);
  73
  74         src_addr = page_address(src_page);
  75         dst_addr = page_address(dst_page);
  76         memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
  77         set_page_dirty(dst_page);
  78         f2fs_put_page(src_page, 1);
  79
  80         set_to_next_nat(nm_i, nid);
  81
  82         return dst_page;
  83 }
  84
  85 /*
  86  * Readahead NAT pages
  87  */
  88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
  89 {
  90         struct address_space *mapping = META_MAPPING(sbi);
  91         struct f2fs_nm_info *nm_i = NM_I(sbi);
  92         struct page *page;
  93         pgoff_t index;
  94         int i;
  95         struct f2fs_io_info fio = {
  96                 .type = META,
  97                 .rw = READ_SYNC | REQ_META | REQ_PRIO
  98         };
  99
 100
 101         for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
 102                 if (unlikely(nid >= nm_i->max_nid))
 103                         nid = 0;
 104                 index = current_nat_addr(sbi, nid);
 105
 106                 page = grab_cache_page(mapping, index);
 107                 if (!page)
 108                         continue;
 109                 if (PageUptodate(page)) {
 110                         mark_page_accessed(page);
 111                         f2fs_put_page(page, 1);
 112                         continue;
 113                 }
 114                 f2fs_submit_page_mbio(sbi, page, index, &fio);
 115                 mark_page_accessed(page);
 116                 f2fs_put_page(page, 0);
 117         }
 118         f2fs_submit_merged_bio(sbi, META, READ);
 119 }
 120
 121 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
 122 {
 123         return radix_tree_lookup(&nm_i->nat_root, n);
 124 }
 125
 126 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
 127                 nid_t start, unsigned int nr, struct nat_entry **ep)
 128 {
 129         return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
 130 }
 131
 132 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
 133 {
 134         list_del(&e->list);
 135         radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
 136         nm_i->nat_cnt--;
 137         kmem_cache_free(nat_entry_slab, e);
 138 }
 139
 140 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
 141 {
 142         struct f2fs_nm_info *nm_i = NM_I(sbi);
 143         struct nat_entry *e;
 144         int is_cp = 1;
 145
 146         read_lock(&nm_i->nat_tree_lock);
 147         e = __lookup_nat_cache(nm_i, nid);
 148         if (e && !e->checkpointed)
 149                 is_cp = 0;
 150         read_unlock(&nm_i->nat_tree_lock);
 151         return is_cp;
 152 }
 153
 154 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
 155 {
 156         struct nat_entry *new;
 157
 158         new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
 159         if (!new)
 160                 return NULL;
 161         if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
 162                 kmem_cache_free(nat_entry_slab, new);
 163                 return NULL;
 164         }
 165         memset(new, 0, sizeof(struct nat_entry));
 166         nat_set_nid(new, nid);
 167         list_add_tail(&new->list, &nm_i->nat_entries);
 168         nm_i->nat_cnt++;
 169         return new;
 170 }
 171
 172 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
 173                                                 struct f2fs_nat_entry *ne)
 174 {
 175         struct nat_entry *e;
 176 retry:
 177         write_lock(&nm_i->nat_tree_lock);
 178         e = __lookup_nat_cache(nm_i, nid);
 179         if (!e) {
 180                 e = grab_nat_entry(nm_i, nid);
 181                 if (!e) {
 182                         write_unlock(&nm_i->nat_tree_lock);
 183                         goto retry;
 184                 }
 185                 nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
 186                 nat_set_ino(e, le32_to_cpu(ne->ino));
 187                 nat_set_version(e, ne->version);
 188                 e->checkpointed = true;
 189         }
 190         write_unlock(&nm_i->nat_tree_lock);
 191 }
 192
 193 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
 194                         block_t new_blkaddr)
 195 {
 196         struct f2fs_nm_info *nm_i = NM_I(sbi);
 197         struct nat_entry *e;
 198 retry:
 199         write_lock(&nm_i->nat_tree_lock);
 200         e = __lookup_nat_cache(nm_i, ni->nid);
 201         if (!e) {
 202                 e = grab_nat_entry(nm_i, ni->nid);
 203                 if (!e) {
 204                         write_unlock(&nm_i->nat_tree_lock);
 205                         goto retry;
 206                 }
 207                 e->ni = *ni;
 208                 e->checkpointed = true;
 209                 f2fs_bug_on(ni->blk_addr == NEW_ADDR);
 210         } else if (new_blkaddr == NEW_ADDR) {
 211                 /*
 212                  * when nid is reallocated,
 213                  * previous nat entry can be remained in nat cache.
 214                  * So, reinitialize it with new information.
 215                  */
 216                 e->ni = *ni;
 217                 f2fs_bug_on(ni->blk_addr != NULL_ADDR);
 218         }
 219
 220         if (new_blkaddr == NEW_ADDR)
 221                 e->checkpointed = false;
 222
 223         /* sanity check */
 224         f2fs_bug_on(nat_get_blkaddr(e) != ni->blk_addr);
 225         f2fs_bug_on(nat_get_blkaddr(e) == NULL_ADDR &&
 226                         new_blkaddr == NULL_ADDR);
 227         f2fs_bug_on(nat_get_blkaddr(e) == NEW_ADDR &&
 228                         new_blkaddr == NEW_ADDR);
 229         f2fs_bug_on(nat_get_blkaddr(e) != NEW_ADDR &&
 230                         nat_get_blkaddr(e) != NULL_ADDR &&
 231                         new_blkaddr == NEW_ADDR);
 232
 233         /* increament version no as node is removed */
 234         if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
 235                 unsigned char version = nat_get_version(e);
 236                 nat_set_version(e, inc_node_version(version));
 237         }
 238
 239         /* change address */
 240         nat_set_blkaddr(e, new_blkaddr);
 241         __set_nat_cache_dirty(nm_i, e);
 242         write_unlock(&nm_i->nat_tree_lock);
 243 }
 244
 245 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
 246 {
 247         struct f2fs_nm_info *nm_i = NM_I(sbi);
 248
 249         if (nm_i->nat_cnt <= NM_WOUT_THRESHOLD)
 250                 return 0;
 251
 252         write_lock(&nm_i->nat_tree_lock);
 253         while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
 254                 struct nat_entry *ne;
 255                 ne = list_first_entry(&nm_i->nat_entries,
 256                                         struct nat_entry, list);
 257                 __del_from_nat_cache(nm_i, ne);
 258                 nr_shrink--;
 259         }
 260         write_unlock(&nm_i->nat_tree_lock);
 261         return nr_shrink;
 262 }
 263
 264 /*
 265  * This function returns always success
 266  */
 267 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
 268 {
 269         struct f2fs_nm_info *nm_i = NM_I(sbi);
 270         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
 271         struct f2fs_summary_block *sum = curseg->sum_blk;
 272         nid_t start_nid = START_NID(nid);
 273         struct f2fs_nat_block *nat_blk;
 274         struct page *page = NULL;
 275         struct f2fs_nat_entry ne;
 276         struct nat_entry *e;
 277         int i;
 278
 279         memset(&ne, 0, sizeof(struct f2fs_nat_entry));
 280         ni->nid = nid;
 281
 282         /* Check nat cache */
 283         read_lock(&nm_i->nat_tree_lock);
 284         e = __lookup_nat_cache(nm_i, nid);
 285         if (e) {
 286                 ni->ino = nat_get_ino(e);
 287                 ni->blk_addr = nat_get_blkaddr(e);
 288                 ni->version = nat_get_version(e);
 289         }
 290         read_unlock(&nm_i->nat_tree_lock);
 291         if (e)
 292                 return;
 293
 294         /* Check current segment summary */
 295         mutex_lock(&curseg->curseg_mutex);
 296         i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
 297         if (i >= 0) {
 298                 ne = nat_in_journal(sum, i);
 299                 node_info_from_raw_nat(ni, &ne);
 300         }
 301         mutex_unlock(&curseg->curseg_mutex);
 302         if (i >= 0)
 303                 goto cache;
 304
 305         /* Fill node_info from nat page */
 306         page = get_current_nat_page(sbi, start_nid);
 307         nat_blk = (struct f2fs_nat_block *)page_address(page);
 308         ne = nat_blk->entries[nid - start_nid];
 309         node_info_from_raw_nat(ni, &ne);
 310         f2fs_put_page(page, 1);
 311 cache:
 312         /* cache nat entry */
 313         cache_nat_entry(NM_I(sbi), nid, &ne);
 314 }
 315
 316 /*
 317  * The maximum depth is four.
 318  * Offset[0] will have raw inode offset.
 319  */
 320 static int get_node_path(struct f2fs_inode_info *fi, long block,
 321                                 int offset[4], unsigned int noffset[4])
 322 {
 323         const long direct_index = ADDRS_PER_INODE(fi);
 324         const long direct_blks = ADDRS_PER_BLOCK;
 325         const long dptrs_per_blk = NIDS_PER_BLOCK;
 326         const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
 327         const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
 328         int n = 0;
 329         int level = 0;
 330
 331         noffset[0] = 0;
 332
 333         if (block < direct_index) {
 334                 offset[n] = block;
 335                 goto got;
 336         }
 337         block -= direct_index;
 338         if (block < direct_blks) {
 339                 offset[n++] = NODE_DIR1_BLOCK;
 340                 noffset[n] = 1;
 341                 offset[n] = block;
 342                 level = 1;
 343                 goto got;
 344         }
 345         block -= direct_blks;
 346         if (block < direct_blks) {
 347                 offset[n++] = NODE_DIR2_BLOCK;
 348                 noffset[n] = 2;
 349                 offset[n] = block;
 350                 level = 1;
 351                 goto got;
 352         }
 353         block -= direct_blks;
 354         if (block < indirect_blks) {
 355                 offset[n++] = NODE_IND1_BLOCK;
 356                 noffset[n] = 3;
 357                 offset[n++] = block / direct_blks;
 358                 noffset[n] = 4 + offset[n - 1];
 359                 offset[n] = block % direct_blks;
 360                 level = 2;
 361                 goto got;
 362         }
 363         block -= indirect_blks;
 364         if (block < indirect_blks) {
 365                 offset[n++] = NODE_IND2_BLOCK;
 366                 noffset[n] = 4 + dptrs_per_blk;
 367                 offset[n++] = block / direct_blks;
 368                 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
 369                 offset[n] = block % direct_blks;
 370                 level = 2;
 371                 goto got;
 372         }
 373         block -= indirect_blks;
 374         if (block < dindirect_blks) {
 375                 offset[n++] = NODE_DIND_BLOCK;
 376                 noffset[n] = 5 + (dptrs_per_blk * 2);
 377                 offset[n++] = block / indirect_blks;
 378                 noffset[n] = 6 + (dptrs_per_blk * 2) +
 379                               offset[n - 1] * (dptrs_per_blk + 1);
 380                 offset[n++] = (block / direct_blks) % dptrs_per_blk;
 381                 noffset[n] = 7 + (dptrs_per_blk * 2) +
 382                               offset[n - 2] * (dptrs_per_blk + 1) +
 383                               offset[n - 1];
 384                 offset[n] = block % direct_blks;
 385                 level = 3;
 386                 goto got;
 387         } else {
 388                 BUG();
 389         }
 390 got:
 391         return level;
 392 }
 393
 394 /*
 395  * Caller should call f2fs_put_dnode(dn).
 396  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
 397  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
 398  * In the case of RDONLY_NODE, we don't need to care about mutex.
 399  */
 400 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
 401 {
 402         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 403         struct page *npage[4];
 404         struct page *parent;
 405         int offset[4];
 406         unsigned int noffset[4];
 407         nid_t nids[4];
 408         int level, i;
 409         int err = 0;
 410
 411         level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
 412
 413         nids[0] = dn->inode->i_ino;
 414         npage[0] = dn->inode_page;
 415
 416         if (!npage[0]) {
 417                 npage[0] = get_node_page(sbi, nids[0]);
 418                 if (IS_ERR(npage[0]))
 419                         return PTR_ERR(npage[0]);
 420         }
 421         parent = npage[0];
 422         if (level != 0)
 423                 nids[1] = get_nid(parent, offset[0], true);
 424         dn->inode_page = npage[0];
 425         dn->inode_page_locked = true;
 426
 427         /* get indirect or direct nodes */
 428         for (i = 1; i <= level; i++) {
 429                 bool done = false;
 430
 431                 if (!nids[i] && mode == ALLOC_NODE) {
 432                         /* alloc new node */
 433                         if (!alloc_nid(sbi, &(nids[i]))) {
 434                                 err = -ENOSPC;
 435                                 goto release_pages;
 436                         }
 437
 438                         dn->nid = nids[i];
 439                         npage[i] = new_node_page(dn, noffset[i], NULL);
 440                         if (IS_ERR(npage[i])) {
 441                                 alloc_nid_failed(sbi, nids[i]);
 442                                 err = PTR_ERR(npage[i]);
 443                                 goto release_pages;
 444                         }
 445
 446                         set_nid(parent, offset[i - 1], nids[i], i == 1);
 447                         alloc_nid_done(sbi, nids[i]);
 448                         done = true;
 449                 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
 450                         npage[i] = get_node_page_ra(parent, offset[i - 1]);
 451                         if (IS_ERR(npage[i])) {
 452                                 err = PTR_ERR(npage[i]);
 453                                 goto release_pages;
 454                         }
 455                         done = true;
 456                 }
 457                 if (i == 1) {
 458                         dn->inode_page_locked = false;
 459                         unlock_page(parent);
 460                 } else {
 461                         f2fs_put_page(parent, 1);
 462                 }
 463
 464                 if (!done) {
 465                         npage[i] = get_node_page(sbi, nids[i]);
 466                         if (IS_ERR(npage[i])) {
 467                                 err = PTR_ERR(npage[i]);
 468                                 f2fs_put_page(npage[0], 0);
 469                                 goto release_out;
 470                         }
 471                 }
 472                 if (i < level) {
 473                         parent = npage[i];
 474                         nids[i + 1] = get_nid(parent, offset[i], false);
 475                 }
 476         }
 477         dn->nid = nids[level];
 478         dn->ofs_in_node = offset[level];
 479         dn->node_page = npage[level];
 480         dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
 481         return 0;
 482
 483 release_pages:
 484         f2fs_put_page(parent, 1);
 485         if (i > 1)
 486                 f2fs_put_page(npage[0], 0);
 487 release_out:
 488         dn->inode_page = NULL;
 489         dn->node_page = NULL;
 490         return err;
 491 }
 492
 493 static void truncate_node(struct dnode_of_data *dn)
 494 {
 495         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 496         struct node_info ni;
 497
 498         get_node_info(sbi, dn->nid, &ni);
 499         if (dn->inode->i_blocks == 0) {
 500                 f2fs_bug_on(ni.blk_addr != NULL_ADDR);
 501                 goto invalidate;
 502         }
 503         f2fs_bug_on(ni.blk_addr == NULL_ADDR);
 504
 505         /* Deallocate node address */
 506         invalidate_blocks(sbi, ni.blk_addr);
 507         dec_valid_node_count(sbi, dn->inode);
 508         set_node_addr(sbi, &ni, NULL_ADDR);
 509
 510         if (dn->nid == dn->inode->i_ino) {
 511                 remove_orphan_inode(sbi, dn->nid);
 512                 dec_valid_inode_count(sbi);
 513         } else {
 514                 sync_inode_page(dn);
 515         }
 516 invalidate:
 517         clear_node_page_dirty(dn->node_page);
 518         F2FS_SET_SB_DIRT(sbi);
 519
 520         f2fs_put_page(dn->node_page, 1);
 521         dn->node_page = NULL;
 522         trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
 523 }
 524
 525 static int truncate_dnode(struct dnode_of_data *dn)
 526 {
 527         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 528         struct page *page;
 529
 530         if (dn->nid == 0)
 531                 return 1;
 532
 533         /* get direct node */
 534         page = get_node_page(sbi, dn->nid);
 535         if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
 536                 return 1;
 537         else if (IS_ERR(page))
 538                 return PTR_ERR(page);
 539
 540         /* Make dnode_of_data for parameter */
 541         dn->node_page = page;
 542         dn->ofs_in_node = 0;
 543         truncate_data_blocks(dn);
 544         truncate_node(dn);
 545         return 1;
 546 }
 547
 548 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
 549                                                 int ofs, int depth)
 550 {
 551         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 552         struct dnode_of_data rdn = *dn;
 553         struct page *page;
 554         struct f2fs_node *rn;
 555         nid_t child_nid;
 556         unsigned int child_nofs;
 557         int freed = 0;
 558         int i, ret;
 559
 560         if (dn->nid == 0)
 561                 return NIDS_PER_BLOCK + 1;
 562
 563         trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
 564
 565         page = get_node_page(sbi, dn->nid);
 566         if (IS_ERR(page)) {
 567                 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
 568                 return PTR_ERR(page);
 569         }
 570
 571         rn = F2FS_NODE(page);
 572         if (depth < 3) {
 573                 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
 574                         child_nid = le32_to_cpu(rn->in.nid[i]);
 575                         if (child_nid == 0)
 576                                 continue;
 577                         rdn.nid = child_nid;
 578                         ret = truncate_dnode(&rdn);
 579                         if (ret < 0)
 580                                 goto out_err;
 581                         set_nid(page, i, 0, false);
 582                 }
 583         } else {
 584                 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
 585                 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
 586                         child_nid = le32_to_cpu(rn->in.nid[i]);
 587                         if (child_nid == 0) {
 588                                 child_nofs += NIDS_PER_BLOCK + 1;
 589                                 continue;
 590                         }
 591                         rdn.nid = child_nid;
 592                         ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
 593                         if (ret == (NIDS_PER_BLOCK + 1)) {
 594                                 set_nid(page, i, 0, false);
 595                                 child_nofs += ret;
 596                         } else if (ret < 0 && ret != -ENOENT) {
 597                                 goto out_err;
 598                         }
 599                 }
 600                 freed = child_nofs;
 601         }
 602
 603         if (!ofs) {
 604                 /* remove current indirect node */
 605                 dn->node_page = page;
 606                 truncate_node(dn);
 607                 freed++;
 608         } else {
 609                 f2fs_put_page(page, 1);
 610         }
 611         trace_f2fs_truncate_nodes_exit(dn->inode, freed);
 612         return freed;
 613
 614 out_err:
 615         f2fs_put_page(page, 1);
 616         trace_f2fs_truncate_nodes_exit(dn->inode, ret);
 617         return ret;
 618 }
 619
 620 static int truncate_partial_nodes(struct dnode_of_data *dn,
 621                         struct f2fs_inode *ri, int *offset, int depth)
 622 {
 623         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 624         struct page *pages[2];
 625         nid_t nid[3];
 626         nid_t child_nid;
 627         int err = 0;
 628         int i;
 629         int idx = depth - 2;
 630
 631         nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 632         if (!nid[0])
 633                 return 0;
 634
 635         /* get indirect nodes in the path */
 636         for (i = 0; i < idx + 1; i++) {
 637                 /* refernece count'll be increased */
 638                 pages[i] = get_node_page(sbi, nid[i]);
 639                 if (IS_ERR(pages[i])) {
 640                         err = PTR_ERR(pages[i]);
 641                         idx = i - 1;
 642                         goto fail;
 643                 }
 644                 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
 645         }
 646
 647         /* free direct nodes linked to a partial indirect node */
 648         for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
 649                 child_nid = get_nid(pages[idx], i, false);
 650                 if (!child_nid)
 651                         continue;
 652                 dn->nid = child_nid;
 653                 err = truncate_dnode(dn);
 654                 if (err < 0)
 655                         goto fail;
 656                 set_nid(pages[idx], i, 0, false);
 657         }
 658
 659         if (offset[idx + 1] == 0) {
 660                 dn->node_page = pages[idx];
 661                 dn->nid = nid[idx];
 662                 truncate_node(dn);
 663         } else {
 664                 f2fs_put_page(pages[idx], 1);
 665         }
 666         offset[idx]++;
 667         offset[idx + 1] = 0;
 668         idx--;
 669 fail:
 670         for (i = idx; i >= 0; i--)
 671                 f2fs_put_page(pages[i], 1);
 672
 673         trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
 674
 675         return err;
 676 }
 677
 678 /*
 679  * All the block addresses of data and nodes should be nullified.
 680  */
 681 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
 682 {
 683         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 684         struct address_space *node_mapping = sbi->node_inode->i_mapping;
 685         int err = 0, cont = 1;
 686         int level, offset[4], noffset[4];
 687         unsigned int nofs = 0;
 688         struct f2fs_inode *ri;
 689         struct dnode_of_data dn;
 690         struct page *page;
 691
 692         trace_f2fs_truncate_inode_blocks_enter(inode, from);
 693
 694         level = get_node_path(F2FS_I(inode), from, offset, noffset);
 695 restart:
 696         page = get_node_page(sbi, inode->i_ino);
 697         if (IS_ERR(page)) {
 698                 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
 699                 return PTR_ERR(page);
 700         }
 701
 702         set_new_dnode(&dn, inode, page, NULL, 0);
 703         unlock_page(page);
 704
 705         ri = F2FS_INODE(page);
 706         switch (level) {
 707         case 0:
 708         case 1:
 709                 nofs = noffset[1];
 710                 break;
 711         case 2:
 712                 nofs = noffset[1];
 713                 if (!offset[level - 1])
 714                         goto skip_partial;
 715                 err = truncate_partial_nodes(&dn, ri, offset, level);
 716                 if (err < 0 && err != -ENOENT)
 717                         goto fail;
 718                 nofs += 1 + NIDS_PER_BLOCK;
 719                 break;
 720         case 3:
 721                 nofs = 5 + 2 * NIDS_PER_BLOCK;
 722                 if (!offset[level - 1])
 723                         goto skip_partial;
 724                 err = truncate_partial_nodes(&dn, ri, offset, level);
 725                 if (err < 0 && err != -ENOENT)
 726                         goto fail;
 727                 break;
 728         default:
 729                 BUG();
 730         }
 731
 732 skip_partial:
 733         while (cont) {
 734                 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
 735                 switch (offset[0]) {
 736                 case NODE_DIR1_BLOCK:
 737                 case NODE_DIR2_BLOCK:
 738                         err = truncate_dnode(&dn);
 739                         break;
 740
 741                 case NODE_IND1_BLOCK:
 742                 case NODE_IND2_BLOCK:
 743                         err = truncate_nodes(&dn, nofs, offset[1], 2);
 744                         break;
 745
 746                 case NODE_DIND_BLOCK:
 747                         err = truncate_nodes(&dn, nofs, offset[1], 3);
 748                         cont = 0;
 749                         break;
 750
 751                 default:
 752                         BUG();
 753                 }
 754                 if (err < 0 && err != -ENOENT)
 755                         goto fail;
 756                 if (offset[1] == 0 &&
 757                                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
 758                         lock_page(page);
 759                         if (unlikely(page->mapping != node_mapping)) {
 760                                 f2fs_put_page(page, 1);
 761                                 goto restart;
 762                         }
 763                         wait_on_page_writeback(page);
 764                         ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
 765                         set_page_dirty(page);
 766                         unlock_page(page);
 767                 }
 768                 offset[1] = 0;
 769                 offset[0]++;
 770                 nofs += err;
 771         }
 772 fail:
 773         f2fs_put_page(page, 0);
 774         trace_f2fs_truncate_inode_blocks_exit(inode, err);
 775         return err > 0 ? 0 : err;
 776 }
 777
 778 int truncate_xattr_node(struct inode *inode, struct page *page)
 779 {
 780         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 781         nid_t nid = F2FS_I(inode)->i_xattr_nid;
 782         struct dnode_of_data dn;
 783         struct page *npage;
 784
 785         if (!nid)
 786                 return 0;
 787
 788         npage = get_node_page(sbi, nid);
 789         if (IS_ERR(npage))
 790                 return PTR_ERR(npage);
 791
 792         F2FS_I(inode)->i_xattr_nid = 0;
 793
 794         /* need to do checkpoint during fsync */
 795         F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
 796
 797         set_new_dnode(&dn, inode, page, npage, nid);
 798
 799         if (page)
 800                 dn.inode_page_locked = true;
 801         truncate_node(&dn);
 802         return 0;
 803 }
 804
 805 /*
 806  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
 807  * f2fs_unlock_op().
 808  */
 809 void remove_inode_page(struct inode *inode)
 810 {
 811         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
 812         struct page *page;
 813         nid_t ino = inode->i_ino;
 814         struct dnode_of_data dn;
 815
 816         page = get_node_page(sbi, ino);
 817         if (IS_ERR(page))
 818                 return;
 819
 820         if (truncate_xattr_node(inode, page)) {
 821                 f2fs_put_page(page, 1);
 822                 return;
 823         }
 824         /* 0 is possible, after f2fs_new_inode() is failed */
 825         f2fs_bug_on(inode->i_blocks != 0 && inode->i_blocks != 1);
 826         set_new_dnode(&dn, inode, page, page, ino);
 827         truncate_node(&dn);
 828 }
 829
 830 struct page *new_inode_page(struct inode *inode, const struct qstr *name)
 831 {
 832         struct dnode_of_data dn;
 833
 834         /* allocate inode page for new inode */
 835         set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
 836
 837         /* caller should f2fs_put_page(page, 1); */
 838         return new_node_page(&dn, 0, NULL);
 839 }
 840
 841 struct page *new_node_page(struct dnode_of_data *dn,
 842                                 unsigned int ofs, struct page *ipage)
 843 {
 844         struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
 845         struct address_space *mapping = sbi->node_inode->i_mapping;
 846         struct node_info old_ni, new_ni;
 847         struct page *page;
 848         int err;
 849
 850         if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
 851                 return ERR_PTR(-EPERM);
 852
 853         page = grab_cache_page(mapping, dn->nid);
 854         if (!page)
 855                 return ERR_PTR(-ENOMEM);
 856
 857         if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
 858                 err = -ENOSPC;
 859                 goto fail;
 860         }
 861
 862         get_node_info(sbi, dn->nid, &old_ni);
 863
 864         /* Reinitialize old_ni with new node page */
 865         f2fs_bug_on(old_ni.blk_addr != NULL_ADDR);
 866         new_ni = old_ni;
 867         new_ni.ino = dn->inode->i_ino;
 868         set_node_addr(sbi, &new_ni, NEW_ADDR);
 869
 870         fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
 871         set_cold_node(dn->inode, page);
 872         SetPageUptodate(page);
 873         set_page_dirty(page);
 874
 875         if (ofs == XATTR_NODE_OFFSET)
 876                 F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
 877
 878         dn->node_page = page;
 879         if (ipage)
 880                 update_inode(dn->inode, ipage);
 881         else
 882                 sync_inode_page(dn);
 883         if (ofs == 0)
 884                 inc_valid_inode_count(sbi);
 885
 886         return page;
 887
 888 fail:
 889         clear_node_page_dirty(page);
 890         f2fs_put_page(page, 1);
 891         return ERR_PTR(err);
 892 }
 893
 894 /*
 895  * Caller should do after getting the following values.
 896  * 0: f2fs_put_page(page, 0)
 897  * LOCKED_PAGE: f2fs_put_page(page, 1)
 898  * error: nothing
 899  */
 900 static int read_node_page(struct page *page, int rw)
 901 {
 902         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
 903         struct node_info ni;
 904
 905         get_node_info(sbi, page->index, &ni);
 906
 907         if (unlikely(ni.blk_addr == NULL_ADDR)) {
 908                 f2fs_put_page(page, 1);
 909                 return -ENOENT;
 910         }
 911
 912         if (PageUptodate(page))
 913                 return LOCKED_PAGE;
 914
 915         return f2fs_submit_page_bio(sbi, page, ni.blk_addr, rw);
 916 }
 917
 918 /*
 919  * Readahead a node page
 920  */
 921 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
 922 {
 923         struct address_space *mapping = sbi->node_inode->i_mapping;
 924         struct page *apage;
 925         int err;
 926
 927         apage = find_get_page(mapping, nid);
 928         if (apage && PageUptodate(apage)) {
 929                 f2fs_put_page(apage, 0);
 930                 return;
 931         }
 932         f2fs_put_page(apage, 0);
 933
 934         apage = grab_cache_page(mapping, nid);
 935         if (!apage)
 936                 return;
 937
 938         err = read_node_page(apage, READA);
 939         if (err == 0)
 940                 f2fs_put_page(apage, 0);
 941         else if (err == LOCKED_PAGE)
 942                 f2fs_put_page(apage, 1);
 943 }
 944
 945 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
 946 {
 947         struct address_space *mapping = sbi->node_inode->i_mapping;
 948         struct page *page;
 949         int err;
 950 repeat:
 951         page = grab_cache_page(mapping, nid);
 952         if (!page)
 953                 return ERR_PTR(-ENOMEM);
 954
 955         err = read_node_page(page, READ_SYNC);
 956         if (err < 0)
 957                 return ERR_PTR(err);
 958         else if (err == LOCKED_PAGE)
 959                 goto got_it;
 960
 961         lock_page(page);
 962         if (unlikely(!PageUptodate(page))) {
 963                 f2fs_put_page(page, 1);
 964                 return ERR_PTR(-EIO);
 965         }
 966         if (unlikely(page->mapping != mapping)) {
 967                 f2fs_put_page(page, 1);
 968                 goto repeat;
 969         }
 970 got_it:
 971         f2fs_bug_on(nid != nid_of_node(page));
 972         mark_page_accessed(page);
 973         return page;
 974 }
 975
 976 /*
 977  * Return a locked page for the desired node page.
 978  * And, readahead MAX_RA_NODE number of node pages.
 979  */
 980 struct page *get_node_page_ra(struct page *parent, int start)
 981 {
 982         struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
 983         struct address_space *mapping = sbi->node_inode->i_mapping;
 984         struct blk_plug plug;
 985         struct page *page;
 986         int err, i, end;
 987         nid_t nid;
 988
 989         /* First, try getting the desired direct node. */
 990         nid = get_nid(parent, start, false);
 991         if (!nid)
 992                 return ERR_PTR(-ENOENT);
 993 repeat:
 994         page = grab_cache_page(mapping, nid);
 995         if (!page)
 996                 return ERR_PTR(-ENOMEM);
 997
 998         err = read_node_page(page, READ_SYNC);
 999         if (err < 0)
1000                 return ERR_PTR(err);
1001         else if (err == LOCKED_PAGE)
1002                 goto page_hit;
1003
1004         blk_start_plug(&plug);
1005
1006         /* Then, try readahead for siblings of the desired node */
1007         end = start + MAX_RA_NODE;
1008         end = min(end, NIDS_PER_BLOCK);
1009         for (i = start + 1; i < end; i++) {
1010                 nid = get_nid(parent, i, false);
1011                 if (!nid)
1012                         continue;
1013                 ra_node_page(sbi, nid);
1014         }
1015
1016         blk_finish_plug(&plug);
1017
1018         lock_page(page);
1019         if (unlikely(page->mapping != mapping)) {
1020                 f2fs_put_page(page, 1);
1021                 goto repeat;
1022         }
1023 page_hit:
1024         if (unlikely(!PageUptodate(page))) {
1025                 f2fs_put_page(page, 1);
1026                 return ERR_PTR(-EIO);
1027         }
1028         mark_page_accessed(page);
1029         return page;
1030 }
1031
1032 void sync_inode_page(struct dnode_of_data *dn)
1033 {
1034         if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1035                 update_inode(dn->inode, dn->node_page);
1036         } else if (dn->inode_page) {
1037                 if (!dn->inode_page_locked)
1038                         lock_page(dn->inode_page);
1039                 update_inode(dn->inode, dn->inode_page);
1040                 if (!dn->inode_page_locked)
1041                         unlock_page(dn->inode_page);
1042         } else {
1043                 update_inode_page(dn->inode);
1044         }
1045 }
1046
1047 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1048                                         struct writeback_control *wbc)
1049 {
1050         struct address_space *mapping = sbi->node_inode->i_mapping;
1051         pgoff_t index, end;
1052         struct pagevec pvec;
1053         int step = ino ? 2 : 0;
1054         int nwritten = 0, wrote = 0;
1055
1056         pagevec_init(&pvec, 0);
1057
1058 next_step:
1059         index = 0;
1060         end = LONG_MAX;
1061
1062         while (index <= end) {
1063                 int i, nr_pages;
1064                 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1065                                 PAGECACHE_TAG_DIRTY,
1066                                 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1067                 if (nr_pages == 0)
1068                         break;
1069
1070                 for (i = 0; i < nr_pages; i++) {
1071                         struct page *page = pvec.pages[i];
1072
1073                         /*
1074                          * flushing sequence with step:
1075                          * 0. indirect nodes
1076                          * 1. dentry dnodes
1077                          * 2. file dnodes
1078                          */
1079                         if (step == 0 && IS_DNODE(page))
1080                                 continue;
1081                         if (step == 1 && (!IS_DNODE(page) ||
1082                                                 is_cold_node(page)))
1083                                 continue;
1084                         if (step == 2 && (!IS_DNODE(page) ||
1085                                                 !is_cold_node(page)))
1086                                 continue;
1087
1088                         /*
1089                          * If an fsync mode,
1090                          * we should not skip writing node pages.
1091                          */
1092                         if (ino && ino_of_node(page) == ino)
1093                                 lock_page(page);
1094                         else if (!trylock_page(page))
1095                                 continue;
1096
1097                         if (unlikely(page->mapping != mapping)) {
1098 continue_unlock:
1099                                 unlock_page(page);
1100                                 continue;
1101                         }
1102                         if (ino && ino_of_node(page) != ino)
1103                                 goto continue_unlock;
1104
1105                         if (!PageDirty(page)) {
1106                                 /* someone wrote it for us */
1107                                 goto continue_unlock;
1108                         }
1109
1110                         if (!clear_page_dirty_for_io(page))
1111                                 goto continue_unlock;
1112
1113                         /* called by fsync() */
1114                         if (ino && IS_DNODE(page)) {
1115                                 int mark = !is_checkpointed_node(sbi, ino);
1116                                 set_fsync_mark(page, 1);
1117                                 if (IS_INODE(page))
1118                                         set_dentry_mark(page, mark);
1119                                 nwritten++;
1120                         } else {
1121                                 set_fsync_mark(page, 0);
1122                                 set_dentry_mark(page, 0);
1123                         }
1124                         mapping->a_ops->writepage(page, wbc);
1125                         wrote++;
1126
1127                         if (--wbc->nr_to_write == 0)
1128                                 break;
1129                 }
1130                 pagevec_release(&pvec);
1131                 cond_resched();
1132
1133                 if (wbc->nr_to_write == 0) {
1134                         step = 2;
1135                         break;
1136                 }
1137         }
1138
1139         if (step < 2) {
1140                 step++;
1141                 goto next_step;
1142         }
1143
1144         if (wrote)
1145                 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1146         return nwritten;
1147 }
1148
1149 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1150 {
1151         struct address_space *mapping = sbi->node_inode->i_mapping;
1152         pgoff_t index = 0, end = LONG_MAX;
1153         struct pagevec pvec;
1154         int nr_pages;
1155         int ret2 = 0, ret = 0;
1156
1157         pagevec_init(&pvec, 0);
1158         while ((index <= end) &&
1159                         (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1160                         PAGECACHE_TAG_WRITEBACK,
1161                         min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
1162                 unsigned i;
1163
1164                 for (i = 0; i < nr_pages; i++) {
1165                         struct page *page = pvec.pages[i];
1166
1167                         /* until radix tree lookup accepts end_index */
1168                         if (unlikely(page->index > end))
1169                                 continue;
1170
1171                         if (ino && ino_of_node(page) == ino) {
1172                                 wait_on_page_writeback(page);
1173                                 if (TestClearPageError(page))
1174                                         ret = -EIO;
1175                         }
1176                 }
1177                 pagevec_release(&pvec);
1178                 cond_resched();
1179         }
1180
1181         if (unlikely(test_and_clear_bit(AS_ENOSPC, &mapping->flags)))
1182                 ret2 = -ENOSPC;
1183         if (unlikely(test_and_clear_bit(AS_EIO, &mapping->flags)))
1184                 ret2 = -EIO;
1185         if (!ret)
1186                 ret = ret2;
1187         return ret;
1188 }
1189
1190 static int f2fs_write_node_page(struct page *page,
1191                                 struct writeback_control *wbc)
1192 {
1193         struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1194         nid_t nid;
1195         block_t new_addr;
1196         struct node_info ni;
1197         struct f2fs_io_info fio = {
1198                 .type = NODE,
1199                 .rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1200         };
1201
1202         if (unlikely(sbi->por_doing))
1203                 goto redirty_out;
1204
1205         wait_on_page_writeback(page);
1206
1207         /* get old block addr of this node page */
1208         nid = nid_of_node(page);
1209         f2fs_bug_on(page->index != nid);
1210
1211         get_node_info(sbi, nid, &ni);
1212
1213         /* This page is already truncated */
1214         if (unlikely(ni.blk_addr == NULL_ADDR)) {
1215                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1216                 unlock_page(page);
1217                 return 0;
1218         }
1219
1220         if (wbc->for_reclaim)
1221                 goto redirty_out;
1222
1223         mutex_lock(&sbi->node_write);
1224         set_page_writeback(page);
1225         write_node_page(sbi, page, &fio, nid, ni.blk_addr, &new_addr);
1226         set_node_addr(sbi, &ni, new_addr);
1227         dec_page_count(sbi, F2FS_DIRTY_NODES);
1228         mutex_unlock(&sbi->node_write);
1229         unlock_page(page);
1230         return 0;
1231
1232 redirty_out:
1233         dec_page_count(sbi, F2FS_DIRTY_NODES);
1234         wbc->pages_skipped++;
1235         set_page_dirty(page);
1236         return AOP_WRITEPAGE_ACTIVATE;
1237 }
1238
1239 /*
1240  * It is very important to gather dirty pages and write at once, so that we can
1241  * submit a big bio without interfering other data writes.
1242  * Be default, 512 pages (2MB) * 3 node types, is more reasonable.
1243  */
1244 #define COLLECT_DIRTY_NODES     1536
1245 static int f2fs_write_node_pages(struct address_space *mapping,
1246                             struct writeback_control *wbc)
1247 {
1248         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1249         long nr_to_write = wbc->nr_to_write;
1250
1251         /* balancing f2fs's metadata in background */
1252         f2fs_balance_fs_bg(sbi);
1253
1254         /* collect a number of dirty node pages and write together */
1255         if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1256                 return 0;
1257
1258         /* if mounting is failed, skip writing node pages */
1259         wbc->nr_to_write = 3 * max_hw_blocks(sbi);
1260         wbc->sync_mode = WB_SYNC_NONE;
1261         sync_node_pages(sbi, 0, wbc);
1262         wbc->nr_to_write = nr_to_write - (3 * max_hw_blocks(sbi) -
1263                                                 wbc->nr_to_write);
1264         return 0;
1265 }
1266
1267 static int f2fs_set_node_page_dirty(struct page *page)
1268 {
1269         struct address_space *mapping = page->mapping;
1270         struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1271
1272         trace_f2fs_set_page_dirty(page, NODE);
1273
1274         SetPageUptodate(page);
1275         if (!PageDirty(page)) {
1276                 __set_page_dirty_nobuffers(page);
1277                 inc_page_count(sbi, F2FS_DIRTY_NODES);
1278                 SetPagePrivate(page);
1279                 return 1;
1280         }
1281         return 0;
1282 }
1283
1284 static void f2fs_invalidate_node_page(struct page *page, unsigned int offset,
1285                                       unsigned int length)
1286 {
1287         struct inode *inode = page->mapping->host;
1288         struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1289         if (PageDirty(page))
1290                 dec_page_count(sbi, F2FS_DIRTY_NODES);
1291         ClearPagePrivate(page);
1292 }
1293
1294 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1295 {
1296         ClearPagePrivate(page);
1297         return 1;
1298 }
1299
1300 /*
1301  * Structure of the f2fs node operations
1302  */
1303 const struct address_space_operations f2fs_node_aops = {
1304         .writepage      = f2fs_write_node_page,
1305         .writepages     = f2fs_write_node_pages,
1306         .set_page_dirty = f2fs_set_node_page_dirty,
1307         .invalidatepage = f2fs_invalidate_node_page,
1308         .releasepage    = f2fs_release_node_page,
1309 };
1310
1311 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1312 {
1313         struct list_head *this;
1314         struct free_nid *i;
1315         list_for_each(this, head) {
1316                 i = list_entry(this, struct free_nid, list);
1317                 if (i->nid == n)
1318                         return i;
1319         }
1320         return NULL;
1321 }
1322
1323 static void __del_from_free_nid_list(struct free_nid *i)
1324 {
1325         list_del(&i->list);
1326         kmem_cache_free(free_nid_slab, i);
1327 }
1328
1329 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid, bool build)
1330 {
1331         struct free_nid *i;
1332         struct nat_entry *ne;
1333         bool allocated = false;
1334
1335         if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1336                 return -1;
1337
1338         /* 0 nid should not be used */
1339         if (unlikely(nid == 0))
1340                 return 0;
1341
1342         if (build) {
1343                 /* do not add allocated nids */
1344                 read_lock(&nm_i->nat_tree_lock);
1345                 ne = __lookup_nat_cache(nm_i, nid);
1346                 if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1347                         allocated = true;
1348                 read_unlock(&nm_i->nat_tree_lock);
1349                 if (allocated)
1350                         return 0;
1351         }
1352
1353         i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1354         i->nid = nid;
1355         i->state = NID_NEW;
1356
1357         spin_lock(&nm_i->free_nid_list_lock);
1358         if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1359                 spin_unlock(&nm_i->free_nid_list_lock);
1360                 kmem_cache_free(free_nid_slab, i);
1361                 return 0;
1362         }
1363         list_add_tail(&i->list, &nm_i->free_nid_list);
1364         nm_i->fcnt++;
1365         spin_unlock(&nm_i->free_nid_list_lock);
1366         return 1;
1367 }
1368
1369 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1370 {
1371         struct free_nid *i;
1372         spin_lock(&nm_i->free_nid_list_lock);
1373         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1374         if (i && i->state == NID_NEW) {
1375                 __del_from_free_nid_list(i);
1376                 nm_i->fcnt--;
1377         }
1378         spin_unlock(&nm_i->free_nid_list_lock);
1379 }
1380
1381 static void scan_nat_page(struct f2fs_nm_info *nm_i,
1382                         struct page *nat_page, nid_t start_nid)
1383 {
1384         struct f2fs_nat_block *nat_blk = page_address(nat_page);
1385         block_t blk_addr;
1386         int i;
1387
1388         i = start_nid % NAT_ENTRY_PER_BLOCK;
1389
1390         for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1391
1392                 if (unlikely(start_nid >= nm_i->max_nid))
1393                         break;
1394
1395                 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1396                 f2fs_bug_on(blk_addr == NEW_ADDR);
1397                 if (blk_addr == NULL_ADDR) {
1398                         if (add_free_nid(nm_i, start_nid, true) < 0)
1399                                 break;
1400                 }
1401         }
1402 }
1403
1404 static void build_free_nids(struct f2fs_sb_info *sbi)
1405 {
1406         struct f2fs_nm_info *nm_i = NM_I(sbi);
1407         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1408         struct f2fs_summary_block *sum = curseg->sum_blk;
1409         int i = 0;
1410         nid_t nid = nm_i->next_scan_nid;
1411
1412         /* Enough entries */
1413         if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1414                 return;
1415
1416         /* readahead nat pages to be scanned */
1417         ra_nat_pages(sbi, nid);
1418
1419         while (1) {
1420                 struct page *page = get_current_nat_page(sbi, nid);
1421
1422                 scan_nat_page(nm_i, page, nid);
1423                 f2fs_put_page(page, 1);
1424
1425                 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1426                 if (unlikely(nid >= nm_i->max_nid))
1427                         nid = 0;
1428
1429                 if (i++ == FREE_NID_PAGES)
1430                         break;
1431         }
1432
1433         /* go to the next free nat pages to find free nids abundantly */
1434         nm_i->next_scan_nid = nid;
1435
1436         /* find free nids from current sum_pages */
1437         mutex_lock(&curseg->curseg_mutex);
1438         for (i = 0; i < nats_in_cursum(sum); i++) {
1439                 block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1440                 nid = le32_to_cpu(nid_in_journal(sum, i));
1441                 if (addr == NULL_ADDR)
1442                         add_free_nid(nm_i, nid, true);
1443                 else
1444                         remove_free_nid(nm_i, nid);
1445         }
1446         mutex_unlock(&curseg->curseg_mutex);
1447 }
1448
1449 /*
1450  * If this function returns success, caller can obtain a new nid
1451  * from second parameter of this function.
1452  * The returned nid could be used ino as well as nid when inode is created.
1453  */
1454 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1455 {
1456         struct f2fs_nm_info *nm_i = NM_I(sbi);
1457         struct free_nid *i = NULL;
1458         struct list_head *this;
1459 retry:
1460         if (unlikely(sbi->total_valid_node_count + 1 >= nm_i->max_nid))
1461                 return false;
1462
1463         spin_lock(&nm_i->free_nid_list_lock);
1464
1465         /* We should not use stale free nids created by build_free_nids */
1466         if (nm_i->fcnt && !sbi->on_build_free_nids) {
1467                 f2fs_bug_on(list_empty(&nm_i->free_nid_list));
1468                 list_for_each(this, &nm_i->free_nid_list) {
1469                         i = list_entry(this, struct free_nid, list);
1470                         if (i->state == NID_NEW)
1471                                 break;
1472                 }
1473
1474                 f2fs_bug_on(i->state != NID_NEW);
1475                 *nid = i->nid;
1476                 i->state = NID_ALLOC;
1477                 nm_i->fcnt--;
1478                 spin_unlock(&nm_i->free_nid_list_lock);
1479                 return true;
1480         }
1481         spin_unlock(&nm_i->free_nid_list_lock);
1482
1483         /* Let's scan nat pages and its caches to get free nids */
1484         mutex_lock(&nm_i->build_lock);
1485         sbi->on_build_free_nids = true;
1486         build_free_nids(sbi);
1487         sbi->on_build_free_nids = false;
1488         mutex_unlock(&nm_i->build_lock);
1489         goto retry;
1490 }
1491
1492 /*
1493  * alloc_nid() should be called prior to this function.
1494  */
1495 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1496 {
1497         struct f2fs_nm_info *nm_i = NM_I(sbi);
1498         struct free_nid *i;
1499
1500         spin_lock(&nm_i->free_nid_list_lock);
1501         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1502         f2fs_bug_on(!i || i->state != NID_ALLOC);
1503         __del_from_free_nid_list(i);
1504         spin_unlock(&nm_i->free_nid_list_lock);
1505 }
1506
1507 /*
1508  * alloc_nid() should be called prior to this function.
1509  */
1510 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1511 {
1512         struct f2fs_nm_info *nm_i = NM_I(sbi);
1513         struct free_nid *i;
1514
1515         if (!nid)
1516                 return;
1517
1518         spin_lock(&nm_i->free_nid_list_lock);
1519         i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1520         f2fs_bug_on(!i || i->state != NID_ALLOC);
1521         if (nm_i->fcnt > 2 * MAX_FREE_NIDS) {
1522                 __del_from_free_nid_list(i);
1523         } else {
1524                 i->state = NID_NEW;
1525                 nm_i->fcnt++;
1526         }
1527         spin_unlock(&nm_i->free_nid_list_lock);
1528 }
1529
1530 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1531                 struct f2fs_summary *sum, struct node_info *ni,
1532                 block_t new_blkaddr)
1533 {
1534         rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1535         set_node_addr(sbi, ni, new_blkaddr);
1536         clear_node_page_dirty(page);
1537 }
1538
1539 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1540 {
1541         struct address_space *mapping = sbi->node_inode->i_mapping;
1542         struct f2fs_inode *src, *dst;
1543         nid_t ino = ino_of_node(page);
1544         struct node_info old_ni, new_ni;
1545         struct page *ipage;
1546
1547         ipage = grab_cache_page(mapping, ino);
1548         if (!ipage)
1549                 return -ENOMEM;
1550
1551         /* Should not use this inode  from free nid list */
1552         remove_free_nid(NM_I(sbi), ino);
1553
1554         get_node_info(sbi, ino, &old_ni);
1555         SetPageUptodate(ipage);
1556         fill_node_footer(ipage, ino, ino, 0, true);
1557
1558         src = F2FS_INODE(page);
1559         dst = F2FS_INODE(ipage);
1560
1561         memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1562         dst->i_size = 0;
1563         dst->i_blocks = cpu_to_le64(1);
1564         dst->i_links = cpu_to_le32(1);
1565         dst->i_xattr_nid = 0;
1566
1567         new_ni = old_ni;
1568         new_ni.ino = ino;
1569
1570         if (unlikely(!inc_valid_node_count(sbi, NULL)))
1571                 WARN_ON(1);
1572         set_node_addr(sbi, &new_ni, NEW_ADDR);
1573         inc_valid_inode_count(sbi);
1574         f2fs_put_page(ipage, 1);
1575         return 0;
1576 }
1577
1578 /*
1579  * ra_sum_pages() merge contiguous pages into one bio and submit.
1580  * these pre-readed pages are linked in pages list.
1581  */
1582 static int ra_sum_pages(struct f2fs_sb_info *sbi, struct list_head *pages,
1583                                 int start, int nrpages)
1584 {
1585         struct page *page;
1586         int page_idx = start;
1587         struct f2fs_io_info fio = {
1588                 .type = META,
1589                 .rw = READ_SYNC | REQ_META | REQ_PRIO
1590         };
1591
1592         for (; page_idx < start + nrpages; page_idx++) {
1593                 /* alloc temporal page for read node summary info*/
1594                 page = alloc_page(GFP_F2FS_ZERO);
1595                 if (!page) {
1596                         struct page *tmp;
1597                         list_for_each_entry_safe(page, tmp, pages, lru) {
1598                                 list_del(&page->lru);
1599                                 unlock_page(page);
1600                                 __free_pages(page, 0);
1601                         }
1602                         return -ENOMEM;
1603                 }
1604
1605                 lock_page(page);
1606                 page->index = page_idx;
1607                 list_add_tail(&page->lru, pages);
1608         }
1609
1610         list_for_each_entry(page, pages, lru)
1611                 f2fs_submit_page_mbio(sbi, page, page->index, &fio);
1612
1613         f2fs_submit_merged_bio(sbi, META, READ);
1614         return 0;
1615 }
1616
1617 int restore_node_summary(struct f2fs_sb_info *sbi,
1618                         unsigned int segno, struct f2fs_summary_block *sum)
1619 {
1620         struct f2fs_node *rn;
1621         struct f2fs_summary *sum_entry;
1622         struct page *page, *tmp;
1623         block_t addr;
1624         int bio_blocks = MAX_BIO_BLOCKS(max_hw_blocks(sbi));
1625         int i, last_offset, nrpages, err = 0;
1626         LIST_HEAD(page_list);
1627
1628         /* scan the node segment */
1629         last_offset = sbi->blocks_per_seg;
1630         addr = START_BLOCK(sbi, segno);
1631         sum_entry = &sum->entries[0];
1632
1633         for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1634                 nrpages = min(last_offset - i, bio_blocks);
1635
1636                 /* read ahead node pages */
1637                 err = ra_sum_pages(sbi, &page_list, addr, nrpages);
1638                 if (err)
1639                         return err;
1640
1641                 list_for_each_entry_safe(page, tmp, &page_list, lru) {
1642
1643                         lock_page(page);
1644                         if (unlikely(!PageUptodate(page))) {
1645                                 err = -EIO;
1646                         } else {
1647                                 rn = F2FS_NODE(page);
1648                                 sum_entry->nid = rn->footer.nid;
1649                                 sum_entry->version = 0;
1650                                 sum_entry->ofs_in_node = 0;
1651                                 sum_entry++;
1652                         }
1653
1654                         list_del(&page->lru);
1655                         unlock_page(page);
1656                         __free_pages(page, 0);
1657                 }
1658         }
1659         return err;
1660 }
1661
1662 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1663 {
1664         struct f2fs_nm_info *nm_i = NM_I(sbi);
1665         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1666         struct f2fs_summary_block *sum = curseg->sum_blk;
1667         int i;
1668
1669         mutex_lock(&curseg->curseg_mutex);
1670
1671         if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1672                 mutex_unlock(&curseg->curseg_mutex);
1673                 return false;
1674         }
1675
1676         for (i = 0; i < nats_in_cursum(sum); i++) {
1677                 struct nat_entry *ne;
1678                 struct f2fs_nat_entry raw_ne;
1679                 nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1680
1681                 raw_ne = nat_in_journal(sum, i);
1682 retry:
1683                 write_lock(&nm_i->nat_tree_lock);
1684                 ne = __lookup_nat_cache(nm_i, nid);
1685                 if (ne) {
1686                         __set_nat_cache_dirty(nm_i, ne);
1687                         write_unlock(&nm_i->nat_tree_lock);
1688                         continue;
1689                 }
1690                 ne = grab_nat_entry(nm_i, nid);
1691                 if (!ne) {
1692                         write_unlock(&nm_i->nat_tree_lock);
1693                         goto retry;
1694                 }
1695                 nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1696                 nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1697                 nat_set_version(ne, raw_ne.version);
1698                 __set_nat_cache_dirty(nm_i, ne);
1699                 write_unlock(&nm_i->nat_tree_lock);
1700         }
1701         update_nats_in_cursum(sum, -i);
1702         mutex_unlock(&curseg->curseg_mutex);
1703         return true;
1704 }
1705
1706 /*
1707  * This function is called during the checkpointing process.
1708  */
1709 void flush_nat_entries(struct f2fs_sb_info *sbi)
1710 {
1711         struct f2fs_nm_info *nm_i = NM_I(sbi);
1712         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1713         struct f2fs_summary_block *sum = curseg->sum_blk;
1714         struct list_head *cur, *n;
1715         struct page *page = NULL;
1716         struct f2fs_nat_block *nat_blk = NULL;
1717         nid_t start_nid = 0, end_nid = 0;
1718         bool flushed;
1719
1720         flushed = flush_nats_in_journal(sbi);
1721
1722         if (!flushed)
1723                 mutex_lock(&curseg->curseg_mutex);
1724
1725         /* 1) flush dirty nat caches */
1726         list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1727                 struct nat_entry *ne;
1728                 nid_t nid;
1729                 struct f2fs_nat_entry raw_ne;
1730                 int offset = -1;
1731                 block_t new_blkaddr;
1732
1733                 ne = list_entry(cur, struct nat_entry, list);
1734                 nid = nat_get_nid(ne);
1735
1736                 if (nat_get_blkaddr(ne) == NEW_ADDR)
1737                         continue;
1738                 if (flushed)
1739                         goto to_nat_page;
1740
1741                 /* if there is room for nat enries in curseg->sumpage */
1742                 offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1743                 if (offset >= 0) {
1744                         raw_ne = nat_in_journal(sum, offset);
1745                         goto flush_now;
1746                 }
1747 to_nat_page:
1748                 if (!page || (start_nid > nid || nid > end_nid)) {
1749                         if (page) {
1750                                 f2fs_put_page(page, 1);
1751                                 page = NULL;
1752                         }
1753                         start_nid = START_NID(nid);
1754                         end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1755
1756                         /*
1757                          * get nat block with dirty flag, increased reference
1758                          * count, mapped and lock
1759                          */
1760                         page = get_next_nat_page(sbi, start_nid);
1761                         nat_blk = page_address(page);
1762                 }
1763
1764                 f2fs_bug_on(!nat_blk);
1765                 raw_ne = nat_blk->entries[nid - start_nid];
1766 flush_now:
1767                 new_blkaddr = nat_get_blkaddr(ne);
1768
1769                 raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1770                 raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1771                 raw_ne.version = nat_get_version(ne);
1772
1773                 if (offset < 0) {
1774                         nat_blk->entries[nid - start_nid] = raw_ne;
1775                 } else {
1776                         nat_in_journal(sum, offset) = raw_ne;
1777                         nid_in_journal(sum, offset) = cpu_to_le32(nid);
1778                 }
1779
1780                 if (nat_get_blkaddr(ne) == NULL_ADDR &&
1781                                 add_free_nid(NM_I(sbi), nid, false) <= 0) {
1782                         write_lock(&nm_i->nat_tree_lock);
1783                         __del_from_nat_cache(nm_i, ne);
1784                         write_unlock(&nm_i->nat_tree_lock);
1785                 } else {
1786                         write_lock(&nm_i->nat_tree_lock);
1787                         __clear_nat_cache_dirty(nm_i, ne);
1788                         ne->checkpointed = true;
1789                         write_unlock(&nm_i->nat_tree_lock);
1790                 }
1791         }
1792         if (!flushed)
1793                 mutex_unlock(&curseg->curseg_mutex);
1794         f2fs_put_page(page, 1);
1795
1796         /* 2) shrink nat caches if necessary */
1797         try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1798 }
1799
1800 static int init_node_manager(struct f2fs_sb_info *sbi)
1801 {
1802         struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1803         struct f2fs_nm_info *nm_i = NM_I(sbi);
1804         unsigned char *version_bitmap;
1805         unsigned int nat_segs, nat_blocks;
1806
1807         nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1808
1809         /* segment_count_nat includes pair segment so divide to 2. */
1810         nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1811         nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1812         nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1813         nm_i->fcnt = 0;
1814         nm_i->nat_cnt = 0;
1815
1816         INIT_LIST_HEAD(&nm_i->free_nid_list);
1817         INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1818         INIT_LIST_HEAD(&nm_i->nat_entries);
1819         INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1820
1821         mutex_init(&nm_i->build_lock);
1822         spin_lock_init(&nm_i->free_nid_list_lock);
1823         rwlock_init(&nm_i->nat_tree_lock);
1824
1825         nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1826         nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1827         version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1828         if (!version_bitmap)
1829                 return -EFAULT;
1830
1831         nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1832                                         GFP_KERNEL);
1833         if (!nm_i->nat_bitmap)
1834                 return -ENOMEM;
1835         return 0;
1836 }
1837
1838 int build_node_manager(struct f2fs_sb_info *sbi)
1839 {
1840         int err;
1841
1842         sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1843         if (!sbi->nm_info)
1844                 return -ENOMEM;
1845
1846         err = init_node_manager(sbi);
1847         if (err)
1848                 return err;
1849
1850         build_free_nids(sbi);
1851         return 0;
1852 }
1853
1854 void destroy_node_manager(struct f2fs_sb_info *sbi)
1855 {
1856         struct f2fs_nm_info *nm_i = NM_I(sbi);
1857         struct free_nid *i, *next_i;
1858         struct nat_entry *natvec[NATVEC_SIZE];
1859         nid_t nid = 0;
1860         unsigned int found;
1861
1862         if (!nm_i)
1863                 return;
1864
1865         /* destroy free nid list */
1866         spin_lock(&nm_i->free_nid_list_lock);
1867         list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1868                 f2fs_bug_on(i->state == NID_ALLOC);
1869                 __del_from_free_nid_list(i);
1870                 nm_i->fcnt--;
1871         }
1872         f2fs_bug_on(nm_i->fcnt);
1873         spin_unlock(&nm_i->free_nid_list_lock);
1874
1875         /* destroy nat cache */
1876         write_lock(&nm_i->nat_tree_lock);
1877         while ((found = __gang_lookup_nat_cache(nm_i,
1878                                         nid, NATVEC_SIZE, natvec))) {
1879                 unsigned idx;
1880                 for (idx = 0; idx < found; idx++) {
1881                         struct nat_entry *e = natvec[idx];
1882                         nid = nat_get_nid(e) + 1;
1883                         __del_from_nat_cache(nm_i, e);
1884                 }
1885         }
1886         f2fs_bug_on(nm_i->nat_cnt);
1887         write_unlock(&nm_i->nat_tree_lock);
1888
1889         kfree(nm_i->nat_bitmap);
1890         sbi->nm_info = NULL;
1891         kfree(nm_i);
1892 }
1893
1894 int __init create_node_manager_caches(void)
1895 {
1896         nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1897                         sizeof(struct nat_entry), NULL);
1898         if (!nat_entry_slab)
1899                 return -ENOMEM;
1900
1901         free_nid_slab = f2fs_kmem_cache_create("free_nid",
1902                         sizeof(struct free_nid), NULL);
1903         if (!free_nid_slab) {
1904                 kmem_cache_destroy(nat_entry_slab);
1905                 return -ENOMEM;
1906         }
1907         return 0;
1908 }
1909
1910 void destroy_node_manager_caches(void)
1911 {
1912         kmem_cache_destroy(free_nid_slab);
1913         kmem_cache_destroy(nat_entry_slab);
1914 }