fs/f2fs/node.c

0001 // SPDX-License-Identifier: GPL-2.0
0002 /*
0003  * fs/f2fs/node.c
0004  *
0005  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
0006  *             http://www.samsung.com/
0007  */
0008 #include <linux/fs.h>
0009 #include <linux/f2fs_fs.h>
0010 #include <linux/mpage.h>
0011 #include <linux/sched/mm.h>
0012 #include <linux/blkdev.h>
0013 #include <linux/pagevec.h>
0014 #include <linux/swap.h>
0015
0016 #include "f2fs.h"
0017 #include "node.h"
0018 #include "segment.h"
0019 #include "xattr.h"
0020 #include "iostat.h"
0021 #include <trace/events/f2fs.h>
0022
0023 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
0024
0025 static struct kmem_cache *nat_entry_slab;
0026 static struct kmem_cache *free_nid_slab;
0027 static struct kmem_cache *nat_entry_set_slab;
0028 static struct kmem_cache *fsync_node_entry_slab;
0029
0030 /*
0031  * Check whether the given nid is within node id range.
0032  */
0033 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
0034 {
0035     if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
0036         set_sbi_flag(sbi, SBI_NEED_FSCK);
0037         f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
0038               __func__, nid);
0039         return -EFSCORRUPTED;
0040     }
0041     return 0;
0042 }
0043
0044 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
0045 {
0046     struct f2fs_nm_info *nm_i = NM_I(sbi);
0047     struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
0048     struct sysinfo val;
0049     unsigned long avail_ram;
0050     unsigned long mem_size = 0;
0051     bool res = false;
0052
0053     if (!nm_i)
0054         return true;
0055
0056     si_meminfo(&val);
0057
0058     /* only uses low memory */
0059     avail_ram = val.totalram - val.totalhigh;
0060
0061     /*
0062      * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
0063      */
0064     if (type == FREE_NIDS) {
0065         mem_size = (nm_i->nid_cnt[FREE_NID] *
0066                 sizeof(struct free_nid)) >> PAGE_SHIFT;
0067         res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
0068     } else if (type == NAT_ENTRIES) {
0069         mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
0070                 sizeof(struct nat_entry)) >> PAGE_SHIFT;
0071         res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
0072         if (excess_cached_nats(sbi))
0073             res = false;
0074     } else if (type == DIRTY_DENTS) {
0075         if (sbi->sb->s_bdi->wb.dirty_exceeded)
0076             return false;
0077         mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
0078         res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
0079     } else if (type == INO_ENTRIES) {
0080         int i;
0081
0082         for (i = 0; i < MAX_INO_ENTRY; i++)
0083             mem_size += sbi->im[i].ino_num *
0084                         sizeof(struct ino_entry);
0085         mem_size >>= PAGE_SHIFT;
0086         res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
0087     } else if (type == EXTENT_CACHE) {
0088         mem_size = (atomic_read(&sbi->total_ext_tree) *
0089                 sizeof(struct extent_tree) +
0090                 atomic_read(&sbi->total_ext_node) *
0091                 sizeof(struct extent_node)) >> PAGE_SHIFT;
0092         res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
0093     } else if (type == DISCARD_CACHE) {
0094         mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
0095                 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
0096         res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
0097     } else if (type == COMPRESS_PAGE) {
0098 #ifdef CONFIG_F2FS_FS_COMPRESSION
0099         unsigned long free_ram = val.freeram;
0100
0101         /*
0102          * free memory is lower than watermark or cached page count
0103          * exceed threshold, deny caching compress page.
0104          */
0105         res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
0106             (COMPRESS_MAPPING(sbi)->nrpages <
0107              free_ram * sbi->compress_percent / 100);
0108 #else
0109         res = false;
0110 #endif
0111     } else {
0112         if (!sbi->sb->s_bdi->wb.dirty_exceeded)
0113             return true;
0114     }
0115     return res;
0116 }
0117
0118 static void clear_node_page_dirty(struct page *page)
0119 {
0120     if (PageDirty(page)) {
0121         f2fs_clear_page_cache_dirty_tag(page);
0122         clear_page_dirty_for_io(page);
0123         dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
0124     }
0125     ClearPageUptodate(page);
0126 }
0127
0128 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
0129 {
0130     return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
0131 }
0132
0133 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
0134 {
0135     struct page *src_page;
0136     struct page *dst_page;
0137     pgoff_t dst_off;
0138     void *src_addr;
0139     void *dst_addr;
0140     struct f2fs_nm_info *nm_i = NM_I(sbi);
0141
0142     dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
0143
0144     /* get current nat block page with lock */
0145     src_page = get_current_nat_page(sbi, nid);
0146     if (IS_ERR(src_page))
0147         return src_page;
0148     dst_page = f2fs_grab_meta_page(sbi, dst_off);
0149     f2fs_bug_on(sbi, PageDirty(src_page));
0150
0151     src_addr = page_address(src_page);
0152     dst_addr = page_address(dst_page);
0153     memcpy(dst_addr, src_addr, PAGE_SIZE);
0154     set_page_dirty(dst_page);
0155     f2fs_put_page(src_page, 1);
0156
0157     set_to_next_nat(nm_i, nid);
0158
0159     return dst_page;
0160 }
0161
0162 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
0163                         nid_t nid, bool no_fail)
0164 {
0165     struct nat_entry *new;
0166
0167     new = f2fs_kmem_cache_alloc(nat_entry_slab,
0168                     GFP_F2FS_ZERO, no_fail, sbi);
0169     if (new) {
0170         nat_set_nid(new, nid);
0171         nat_reset_flag(new);
0172     }
0173     return new;
0174 }
0175
0176 static void __free_nat_entry(struct nat_entry *e)
0177 {
0178     kmem_cache_free(nat_entry_slab, e);
0179 }
0180
0181 /* must be locked by nat_tree_lock */
0182 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
0183     struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
0184 {
0185     if (no_fail)
0186         f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
0187     else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
0188         return NULL;
0189
0190     if (raw_ne)
0191         node_info_from_raw_nat(&ne->ni, raw_ne);
0192
0193     spin_lock(&nm_i->nat_list_lock);
0194     list_add_tail(&ne->list, &nm_i->nat_entries);
0195     spin_unlock(&nm_i->nat_list_lock);
0196
0197     nm_i->nat_cnt[TOTAL_NAT]++;
0198     nm_i->nat_cnt[RECLAIMABLE_NAT]++;
0199     return ne;
0200 }
0201
0202 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
0203 {
0204     struct nat_entry *ne;
0205
0206     ne = radix_tree_lookup(&nm_i->nat_root, n);
0207
0208     /* for recent accessed nat entry, move it to tail of lru list */
0209     if (ne && !get_nat_flag(ne, IS_DIRTY)) {
0210         spin_lock(&nm_i->nat_list_lock);
0211         if (!list_empty(&ne->list))
0212             list_move_tail(&ne->list, &nm_i->nat_entries);
0213         spin_unlock(&nm_i->nat_list_lock);
0214     }
0215
0216     return ne;
0217 }
0218
0219 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
0220         nid_t start, unsigned int nr, struct nat_entry **ep)
0221 {
0222     return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
0223 }
0224
0225 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
0226 {
0227     radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
0228     nm_i->nat_cnt[TOTAL_NAT]--;
0229     nm_i->nat_cnt[RECLAIMABLE_NAT]--;
0230     __free_nat_entry(e);
0231 }
0232
0233 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
0234                             struct nat_entry *ne)
0235 {
0236     nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
0237     struct nat_entry_set *head;
0238
0239     head = radix_tree_lookup(&nm_i->nat_set_root, set);
0240     if (!head) {
0241         head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
0242                         GFP_NOFS, true, NULL);
0243
0244         INIT_LIST_HEAD(&head->entry_list);
0245         INIT_LIST_HEAD(&head->set_list);
0246         head->set = set;
0247         head->entry_cnt = 0;
0248         f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
0249     }
0250     return head;
0251 }
0252
0253 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
0254                         struct nat_entry *ne)
0255 {
0256     struct nat_entry_set *head;
0257     bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
0258
0259     if (!new_ne)
0260         head = __grab_nat_entry_set(nm_i, ne);
0261
0262     /*
0263      * update entry_cnt in below condition:
0264      * 1. update NEW_ADDR to valid block address;
0265      * 2. update old block address to new one;
0266      */
0267     if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
0268                 !get_nat_flag(ne, IS_DIRTY)))
0269         head->entry_cnt++;
0270
0271     set_nat_flag(ne, IS_PREALLOC, new_ne);
0272
0273     if (get_nat_flag(ne, IS_DIRTY))
0274         goto refresh_list;
0275
0276     nm_i->nat_cnt[DIRTY_NAT]++;
0277     nm_i->nat_cnt[RECLAIMABLE_NAT]--;
0278     set_nat_flag(ne, IS_DIRTY, true);
0279 refresh_list:
0280     spin_lock(&nm_i->nat_list_lock);
0281     if (new_ne)
0282         list_del_init(&ne->list);
0283     else
0284         list_move_tail(&ne->list, &head->entry_list);
0285     spin_unlock(&nm_i->nat_list_lock);
0286 }
0287
0288 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
0289         struct nat_entry_set *set, struct nat_entry *ne)
0290 {
0291     spin_lock(&nm_i->nat_list_lock);
0292     list_move_tail(&ne->list, &nm_i->nat_entries);
0293     spin_unlock(&nm_i->nat_list_lock);
0294
0295     set_nat_flag(ne, IS_DIRTY, false);
0296     set->entry_cnt--;
0297     nm_i->nat_cnt[DIRTY_NAT]--;
0298     nm_i->nat_cnt[RECLAIMABLE_NAT]++;
0299 }
0300
0301 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
0302         nid_t start, unsigned int nr, struct nat_entry_set **ep)
0303 {
0304     return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
0305                             start, nr);
0306 }
0307
0308 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
0309 {
0310     return NODE_MAPPING(sbi) == page->mapping &&
0311             IS_DNODE(page) && is_cold_node(page);
0312 }
0313
0314 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
0315 {
0316     spin_lock_init(&sbi->fsync_node_lock);
0317     INIT_LIST_HEAD(&sbi->fsync_node_list);
0318     sbi->fsync_seg_id = 0;
0319     sbi->fsync_node_num = 0;
0320 }
0321
0322 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
0323                             struct page *page)
0324 {
0325     struct fsync_node_entry *fn;
0326     unsigned long flags;
0327     unsigned int seq_id;
0328
0329     fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
0330                     GFP_NOFS, true, NULL);
0331
0332     get_page(page);
0333     fn->page = page;
0334     INIT_LIST_HEAD(&fn->list);
0335
0336     spin_lock_irqsave(&sbi->fsync_node_lock, flags);
0337     list_add_tail(&fn->list, &sbi->fsync_node_list);
0338     fn->seq_id = sbi->fsync_seg_id++;
0339     seq_id = fn->seq_id;
0340     sbi->fsync_node_num++;
0341     spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
0342
0343     return seq_id;
0344 }
0345
0346 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
0347 {
0348     struct fsync_node_entry *fn;
0349     unsigned long flags;
0350
0351     spin_lock_irqsave(&sbi->fsync_node_lock, flags);
0352     list_for_each_entry(fn, &sbi->fsync_node_list, list) {
0353         if (fn->page == page) {
0354             list_del(&fn->list);
0355             sbi->fsync_node_num--;
0356             spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
0357             kmem_cache_free(fsync_node_entry_slab, fn);
0358             put_page(page);
0359             return;
0360         }
0361     }
0362     spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
0363     f2fs_bug_on(sbi, 1);
0364 }
0365
0366 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
0367 {
0368     unsigned long flags;
0369
0370     spin_lock_irqsave(&sbi->fsync_node_lock, flags);
0371     sbi->fsync_seg_id = 0;
0372     spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
0373 }
0374
0375 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
0376 {
0377     struct f2fs_nm_info *nm_i = NM_I(sbi);
0378     struct nat_entry *e;
0379     bool need = false;
0380
0381     f2fs_down_read(&nm_i->nat_tree_lock);
0382     e = __lookup_nat_cache(nm_i, nid);
0383     if (e) {
0384         if (!get_nat_flag(e, IS_CHECKPOINTED) &&
0385                 !get_nat_flag(e, HAS_FSYNCED_INODE))
0386             need = true;
0387     }
0388     f2fs_up_read(&nm_i->nat_tree_lock);
0389     return need;
0390 }
0391
0392 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
0393 {
0394     struct f2fs_nm_info *nm_i = NM_I(sbi);
0395     struct nat_entry *e;
0396     bool is_cp = true;
0397
0398     f2fs_down_read(&nm_i->nat_tree_lock);
0399     e = __lookup_nat_cache(nm_i, nid);
0400     if (e && !get_nat_flag(e, IS_CHECKPOINTED))
0401         is_cp = false;
0402     f2fs_up_read(&nm_i->nat_tree_lock);
0403     return is_cp;
0404 }
0405
0406 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
0407 {
0408     struct f2fs_nm_info *nm_i = NM_I(sbi);
0409     struct nat_entry *e;
0410     bool need_update = true;
0411
0412     f2fs_down_read(&nm_i->nat_tree_lock);
0413     e = __lookup_nat_cache(nm_i, ino);
0414     if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
0415             (get_nat_flag(e, IS_CHECKPOINTED) ||
0416              get_nat_flag(e, HAS_FSYNCED_INODE)))
0417         need_update = false;
0418     f2fs_up_read(&nm_i->nat_tree_lock);
0419     return need_update;
0420 }
0421
0422 /* must be locked by nat_tree_lock */
0423 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
0424                         struct f2fs_nat_entry *ne)
0425 {
0426     struct f2fs_nm_info *nm_i = NM_I(sbi);
0427     struct nat_entry *new, *e;
0428
0429     /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
0430     if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
0431         return;
0432
0433     new = __alloc_nat_entry(sbi, nid, false);
0434     if (!new)
0435         return;
0436
0437     f2fs_down_write(&nm_i->nat_tree_lock);
0438     e = __lookup_nat_cache(nm_i, nid);
0439     if (!e)
0440         e = __init_nat_entry(nm_i, new, ne, false);
0441     else
0442         f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
0443                 nat_get_blkaddr(e) !=
0444                     le32_to_cpu(ne->block_addr) ||
0445                 nat_get_version(e) != ne->version);
0446     f2fs_up_write(&nm_i->nat_tree_lock);
0447     if (e != new)
0448         __free_nat_entry(new);
0449 }
0450
0451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
0452             block_t new_blkaddr, bool fsync_done)
0453 {
0454     struct f2fs_nm_info *nm_i = NM_I(sbi);
0455     struct nat_entry *e;
0456     struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
0457
0458     f2fs_down_write(&nm_i->nat_tree_lock);
0459     e = __lookup_nat_cache(nm_i, ni->nid);
0460     if (!e) {
0461         e = __init_nat_entry(nm_i, new, NULL, true);
0462         copy_node_info(&e->ni, ni);
0463         f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
0464     } else if (new_blkaddr == NEW_ADDR) {
0465         /*
0466          * when nid is reallocated,
0467          * previous nat entry can be remained in nat cache.
0468          * So, reinitialize it with new information.
0469          */
0470         copy_node_info(&e->ni, ni);
0471         f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
0472     }
0473     /* let's free early to reduce memory consumption */
0474     if (e != new)
0475         __free_nat_entry(new);
0476
0477     /* sanity check */
0478     f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
0479     f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
0480             new_blkaddr == NULL_ADDR);
0481     f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
0482             new_blkaddr == NEW_ADDR);
0483     f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
0484             new_blkaddr == NEW_ADDR);
0485
0486     /* increment version no as node is removed */
0487     if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
0488         unsigned char version = nat_get_version(e);
0489
0490         nat_set_version(e, inc_node_version(version));
0491     }
0492
0493     /* change address */
0494     nat_set_blkaddr(e, new_blkaddr);
0495     if (!__is_valid_data_blkaddr(new_blkaddr))
0496         set_nat_flag(e, IS_CHECKPOINTED, false);
0497     __set_nat_cache_dirty(nm_i, e);
0498
0499     /* update fsync_mark if its inode nat entry is still alive */
0500     if (ni->nid != ni->ino)
0501         e = __lookup_nat_cache(nm_i, ni->ino);
0502     if (e) {
0503         if (fsync_done && ni->nid == ni->ino)
0504             set_nat_flag(e, HAS_FSYNCED_INODE, true);
0505         set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
0506     }
0507     f2fs_up_write(&nm_i->nat_tree_lock);
0508 }
0509
0510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
0511 {
0512     struct f2fs_nm_info *nm_i = NM_I(sbi);
0513     int nr = nr_shrink;
0514
0515     if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
0516         return 0;
0517
0518     spin_lock(&nm_i->nat_list_lock);
0519     while (nr_shrink) {
0520         struct nat_entry *ne;
0521
0522         if (list_empty(&nm_i->nat_entries))
0523             break;
0524
0525         ne = list_first_entry(&nm_i->nat_entries,
0526                     struct nat_entry, list);
0527         list_del(&ne->list);
0528         spin_unlock(&nm_i->nat_list_lock);
0529
0530         __del_from_nat_cache(nm_i, ne);
0531         nr_shrink--;
0532
0533         spin_lock(&nm_i->nat_list_lock);
0534     }
0535     spin_unlock(&nm_i->nat_list_lock);
0536
0537     f2fs_up_write(&nm_i->nat_tree_lock);
0538     return nr - nr_shrink;
0539 }
0540
0541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
0542                 struct node_info *ni, bool checkpoint_context)
0543 {
0544     struct f2fs_nm_info *nm_i = NM_I(sbi);
0545     struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
0546     struct f2fs_journal *journal = curseg->journal;
0547     nid_t start_nid = START_NID(nid);
0548     struct f2fs_nat_block *nat_blk;
0549     struct page *page = NULL;
0550     struct f2fs_nat_entry ne;
0551     struct nat_entry *e;
0552     pgoff_t index;
0553     block_t blkaddr;
0554     int i;
0555
0556     ni->nid = nid;
0557 retry:
0558     /* Check nat cache */
0559     f2fs_down_read(&nm_i->nat_tree_lock);
0560     e = __lookup_nat_cache(nm_i, nid);
0561     if (e) {
0562         ni->ino = nat_get_ino(e);
0563         ni->blk_addr = nat_get_blkaddr(e);
0564         ni->version = nat_get_version(e);
0565         f2fs_up_read(&nm_i->nat_tree_lock);
0566         return 0;
0567     }
0568
0569     /*
0570      * Check current segment summary by trying to grab journal_rwsem first.
0571      * This sem is on the critical path on the checkpoint requiring the above
0572      * nat_tree_lock. Therefore, we should retry, if we failed to grab here
0573      * while not bothering checkpoint.
0574      */
0575     if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
0576         down_read(&curseg->journal_rwsem);
0577     } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
0578                 !down_read_trylock(&curseg->journal_rwsem)) {
0579         f2fs_up_read(&nm_i->nat_tree_lock);
0580         goto retry;
0581     }
0582
0583     i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
0584     if (i >= 0) {
0585         ne = nat_in_journal(journal, i);
0586         node_info_from_raw_nat(ni, &ne);
0587     }
0588         up_read(&curseg->journal_rwsem);
0589     if (i >= 0) {
0590         f2fs_up_read(&nm_i->nat_tree_lock);
0591         goto cache;
0592     }
0593
0594     /* Fill node_info from nat page */
0595     index = current_nat_addr(sbi, nid);
0596     f2fs_up_read(&nm_i->nat_tree_lock);
0597
0598     page = f2fs_get_meta_page(sbi, index);
0599     if (IS_ERR(page))
0600         return PTR_ERR(page);
0601
0602     nat_blk = (struct f2fs_nat_block *)page_address(page);
0603     ne = nat_blk->entries[nid - start_nid];
0604     node_info_from_raw_nat(ni, &ne);
0605     f2fs_put_page(page, 1);
0606 cache:
0607     blkaddr = le32_to_cpu(ne.block_addr);
0608     if (__is_valid_data_blkaddr(blkaddr) &&
0609         !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
0610         return -EFAULT;
0611
0612     /* cache nat entry */
0613     cache_nat_entry(sbi, nid, &ne);
0614     return 0;
0615 }
0616
0617 /*
0618  * readahead MAX_RA_NODE number of node pages.
0619  */
0620 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
0621 {
0622     struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
0623     struct blk_plug plug;
0624     int i, end;
0625     nid_t nid;
0626
0627     blk_start_plug(&plug);
0628
0629     /* Then, try readahead for siblings of the desired node */
0630     end = start + n;
0631     end = min(end, NIDS_PER_BLOCK);
0632     for (i = start; i < end; i++) {
0633         nid = get_nid(parent, i, false);
0634         f2fs_ra_node_page(sbi, nid);
0635     }
0636
0637     blk_finish_plug(&plug);
0638 }
0639
0640 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
0641 {
0642     const long direct_index = ADDRS_PER_INODE(dn->inode);
0643     const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
0644     const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
0645     unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
0646     int cur_level = dn->cur_level;
0647     int max_level = dn->max_level;
0648     pgoff_t base = 0;
0649
0650     if (!dn->max_level)
0651         return pgofs + 1;
0652
0653     while (max_level-- > cur_level)
0654         skipped_unit *= NIDS_PER_BLOCK;
0655
0656     switch (dn->max_level) {
0657     case 3:
0658         base += 2 * indirect_blks;
0659         fallthrough;
0660     case 2:
0661         base += 2 * direct_blks;
0662         fallthrough;
0663     case 1:
0664         base += direct_index;
0665         break;
0666     default:
0667         f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
0668     }
0669
0670     return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
0671 }
0672
0673 /*
0674  * The maximum depth is four.
0675  * Offset[0] will have raw inode offset.
0676  */
0677 static int get_node_path(struct inode *inode, long block,
0678                 int offset[4], unsigned int noffset[4])
0679 {
0680     const long direct_index = ADDRS_PER_INODE(inode);
0681     const long direct_blks = ADDRS_PER_BLOCK(inode);
0682     const long dptrs_per_blk = NIDS_PER_BLOCK;
0683     const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
0684     const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
0685     int n = 0;
0686     int level = 0;
0687
0688     noffset[0] = 0;
0689
0690     if (block < direct_index) {
0691         offset[n] = block;
0692         goto got;
0693     }
0694     block -= direct_index;
0695     if (block < direct_blks) {
0696         offset[n++] = NODE_DIR1_BLOCK;
0697         noffset[n] = 1;
0698         offset[n] = block;
0699         level = 1;
0700         goto got;
0701     }
0702     block -= direct_blks;
0703     if (block < direct_blks) {
0704         offset[n++] = NODE_DIR2_BLOCK;
0705         noffset[n] = 2;
0706         offset[n] = block;
0707         level = 1;
0708         goto got;
0709     }
0710     block -= direct_blks;
0711     if (block < indirect_blks) {
0712         offset[n++] = NODE_IND1_BLOCK;
0713         noffset[n] = 3;
0714         offset[n++] = block / direct_blks;
0715         noffset[n] = 4 + offset[n - 1];
0716         offset[n] = block % direct_blks;
0717         level = 2;
0718         goto got;
0719     }
0720     block -= indirect_blks;
0721     if (block < indirect_blks) {
0722         offset[n++] = NODE_IND2_BLOCK;
0723         noffset[n] = 4 + dptrs_per_blk;
0724         offset[n++] = block / direct_blks;
0725         noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
0726         offset[n] = block % direct_blks;
0727         level = 2;
0728         goto got;
0729     }
0730     block -= indirect_blks;
0731     if (block < dindirect_blks) {
0732         offset[n++] = NODE_DIND_BLOCK;
0733         noffset[n] = 5 + (dptrs_per_blk * 2);
0734         offset[n++] = block / indirect_blks;
0735         noffset[n] = 6 + (dptrs_per_blk * 2) +
0736                   offset[n - 1] * (dptrs_per_blk + 1);
0737         offset[n++] = (block / direct_blks) % dptrs_per_blk;
0738         noffset[n] = 7 + (dptrs_per_blk * 2) +
0739                   offset[n - 2] * (dptrs_per_blk + 1) +
0740                   offset[n - 1];
0741         offset[n] = block % direct_blks;
0742         level = 3;
0743         goto got;
0744     } else {
0745         return -E2BIG;
0746     }
0747 got:
0748     return level;
0749 }
0750
0751 /*
0752  * Caller should call f2fs_put_dnode(dn).
0753  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
0754  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
0755  */
0756 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
0757 {
0758     struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
0759     struct page *npage[4];
0760     struct page *parent = NULL;
0761     int offset[4];
0762     unsigned int noffset[4];
0763     nid_t nids[4];
0764     int level, i = 0;
0765     int err = 0;
0766
0767     level = get_node_path(dn->inode, index, offset, noffset);
0768     if (level < 0)
0769         return level;
0770
0771     nids[0] = dn->inode->i_ino;
0772     npage[0] = dn->inode_page;
0773
0774     if (!npage[0]) {
0775         npage[0] = f2fs_get_node_page(sbi, nids[0]);
0776         if (IS_ERR(npage[0]))
0777             return PTR_ERR(npage[0]);
0778     }
0779
0780     /* if inline_data is set, should not report any block indices */
0781     if (f2fs_has_inline_data(dn->inode) && index) {
0782         err = -ENOENT;
0783         f2fs_put_page(npage[0], 1);
0784         goto release_out;
0785     }
0786
0787     parent = npage[0];
0788     if (level != 0)
0789         nids[1] = get_nid(parent, offset[0], true);
0790     dn->inode_page = npage[0];
0791     dn->inode_page_locked = true;
0792
0793     /* get indirect or direct nodes */
0794     for (i = 1; i <= level; i++) {
0795         bool done = false;
0796
0797         if (!nids[i] && mode == ALLOC_NODE) {
0798             /* alloc new node */
0799             if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
0800                 err = -ENOSPC;
0801                 goto release_pages;
0802             }
0803
0804             dn->nid = nids[i];
0805             npage[i] = f2fs_new_node_page(dn, noffset[i]);
0806             if (IS_ERR(npage[i])) {
0807                 f2fs_alloc_nid_failed(sbi, nids[i]);
0808                 err = PTR_ERR(npage[i]);
0809                 goto release_pages;
0810             }
0811
0812             set_nid(parent, offset[i - 1], nids[i], i == 1);
0813             f2fs_alloc_nid_done(sbi, nids[i]);
0814             done = true;
0815         } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
0816             npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
0817             if (IS_ERR(npage[i])) {
0818                 err = PTR_ERR(npage[i]);
0819                 goto release_pages;
0820             }
0821             done = true;
0822         }
0823         if (i == 1) {
0824             dn->inode_page_locked = false;
0825             unlock_page(parent);
0826         } else {
0827             f2fs_put_page(parent, 1);
0828         }
0829
0830         if (!done) {
0831             npage[i] = f2fs_get_node_page(sbi, nids[i]);
0832             if (IS_ERR(npage[i])) {
0833                 err = PTR_ERR(npage[i]);
0834                 f2fs_put_page(npage[0], 0);
0835                 goto release_out;
0836             }
0837         }
0838         if (i < level) {
0839             parent = npage[i];
0840             nids[i + 1] = get_nid(parent, offset[i], false);
0841         }
0842     }
0843     dn->nid = nids[level];
0844     dn->ofs_in_node = offset[level];
0845     dn->node_page = npage[level];
0846     dn->data_blkaddr = f2fs_data_blkaddr(dn);
0847
0848     if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
0849                     f2fs_sb_has_readonly(sbi)) {
0850         unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
0851         block_t blkaddr;
0852
0853         if (!c_len)
0854             goto out;
0855
0856         blkaddr = f2fs_data_blkaddr(dn);
0857         if (blkaddr == COMPRESS_ADDR)
0858             blkaddr = data_blkaddr(dn->inode, dn->node_page,
0859                         dn->ofs_in_node + 1);
0860
0861         f2fs_update_extent_tree_range_compressed(dn->inode,
0862                     index, blkaddr,
0863                     F2FS_I(dn->inode)->i_cluster_size,
0864                     c_len);
0865     }
0866 out:
0867     return 0;
0868
0869 release_pages:
0870     f2fs_put_page(parent, 1);
0871     if (i > 1)
0872         f2fs_put_page(npage[0], 0);
0873 release_out:
0874     dn->inode_page = NULL;
0875     dn->node_page = NULL;
0876     if (err == -ENOENT) {
0877         dn->cur_level = i;
0878         dn->max_level = level;
0879         dn->ofs_in_node = offset[level];
0880     }
0881     return err;
0882 }
0883
0884 static int truncate_node(struct dnode_of_data *dn)
0885 {
0886     struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
0887     struct node_info ni;
0888     int err;
0889     pgoff_t index;
0890
0891     err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
0892     if (err)
0893         return err;
0894
0895     /* Deallocate node address */
0896     f2fs_invalidate_blocks(sbi, ni.blk_addr);
0897     dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
0898     set_node_addr(sbi, &ni, NULL_ADDR, false);
0899
0900     if (dn->nid == dn->inode->i_ino) {
0901         f2fs_remove_orphan_inode(sbi, dn->nid);
0902         dec_valid_inode_count(sbi);
0903         f2fs_inode_synced(dn->inode);
0904     }
0905
0906     clear_node_page_dirty(dn->node_page);
0907     set_sbi_flag(sbi, SBI_IS_DIRTY);
0908
0909     index = dn->node_page->index;
0910     f2fs_put_page(dn->node_page, 1);
0911
0912     invalidate_mapping_pages(NODE_MAPPING(sbi),
0913             index, index);
0914
0915     dn->node_page = NULL;
0916     trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
0917
0918     return 0;
0919 }
0920
0921 static int truncate_dnode(struct dnode_of_data *dn)
0922 {
0923     struct page *page;
0924     int err;
0925
0926     if (dn->nid == 0)
0927         return 1;
0928
0929     /* get direct node */
0930     page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
0931     if (PTR_ERR(page) == -ENOENT)
0932         return 1;
0933     else if (IS_ERR(page))
0934         return PTR_ERR(page);
0935
0936     /* Make dnode_of_data for parameter */
0937     dn->node_page = page;
0938     dn->ofs_in_node = 0;
0939     f2fs_truncate_data_blocks(dn);
0940     err = truncate_node(dn);
0941     if (err)
0942         return err;
0943
0944     return 1;
0945 }
0946
0947 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
0948                         int ofs, int depth)
0949 {
0950     struct dnode_of_data rdn = *dn;
0951     struct page *page;
0952     struct f2fs_node *rn;
0953     nid_t child_nid;
0954     unsigned int child_nofs;
0955     int freed = 0;
0956     int i, ret;
0957
0958     if (dn->nid == 0)
0959         return NIDS_PER_BLOCK + 1;
0960
0961     trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
0962
0963     page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
0964     if (IS_ERR(page)) {
0965         trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
0966         return PTR_ERR(page);
0967     }
0968
0969     f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
0970
0971     rn = F2FS_NODE(page);
0972     if (depth < 3) {
0973         for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
0974             child_nid = le32_to_cpu(rn->in.nid[i]);
0975             if (child_nid == 0)
0976                 continue;
0977             rdn.nid = child_nid;
0978             ret = truncate_dnode(&rdn);
0979             if (ret < 0)
0980                 goto out_err;
0981             if (set_nid(page, i, 0, false))
0982                 dn->node_changed = true;
0983         }
0984     } else {
0985         child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
0986         for (i = ofs; i < NIDS_PER_BLOCK; i++) {
0987             child_nid = le32_to_cpu(rn->in.nid[i]);
0988             if (child_nid == 0) {
0989                 child_nofs += NIDS_PER_BLOCK + 1;
0990                 continue;
0991             }
0992             rdn.nid = child_nid;
0993             ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
0994             if (ret == (NIDS_PER_BLOCK + 1)) {
0995                 if (set_nid(page, i, 0, false))
0996                     dn->node_changed = true;
0997                 child_nofs += ret;
0998             } else if (ret < 0 && ret != -ENOENT) {
0999                 goto out_err;
1000             }
1001         }
1002         freed = child_nofs;
1003     }
1004
1005     if (!ofs) {
1006         /* remove current indirect node */
1007         dn->node_page = page;
1008         ret = truncate_node(dn);
1009         if (ret)
1010             goto out_err;
1011         freed++;
1012     } else {
1013         f2fs_put_page(page, 1);
1014     }
1015     trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1016     return freed;
1017
1018 out_err:
1019     f2fs_put_page(page, 1);
1020     trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1021     return ret;
1022 }
1023
1024 static int truncate_partial_nodes(struct dnode_of_data *dn,
1025             struct f2fs_inode *ri, int *offset, int depth)
1026 {
1027     struct page *pages[2];
1028     nid_t nid[3];
1029     nid_t child_nid;
1030     int err = 0;
1031     int i;
1032     int idx = depth - 2;
1033
1034     nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1035     if (!nid[0])
1036         return 0;
1037
1038     /* get indirect nodes in the path */
1039     for (i = 0; i < idx + 1; i++) {
1040         /* reference count'll be increased */
1041         pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1042         if (IS_ERR(pages[i])) {
1043             err = PTR_ERR(pages[i]);
1044             idx = i - 1;
1045             goto fail;
1046         }
1047         nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1048     }
1049
1050     f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1051
1052     /* free direct nodes linked to a partial indirect node */
1053     for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1054         child_nid = get_nid(pages[idx], i, false);
1055         if (!child_nid)
1056             continue;
1057         dn->nid = child_nid;
1058         err = truncate_dnode(dn);
1059         if (err < 0)
1060             goto fail;
1061         if (set_nid(pages[idx], i, 0, false))
1062             dn->node_changed = true;
1063     }
1064
1065     if (offset[idx + 1] == 0) {
1066         dn->node_page = pages[idx];
1067         dn->nid = nid[idx];
1068         err = truncate_node(dn);
1069         if (err)
1070             goto fail;
1071     } else {
1072         f2fs_put_page(pages[idx], 1);
1073     }
1074     offset[idx]++;
1075     offset[idx + 1] = 0;
1076     idx--;
1077 fail:
1078     for (i = idx; i >= 0; i--)
1079         f2fs_put_page(pages[i], 1);
1080
1081     trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1082
1083     return err;
1084 }
1085
1086 /*
1087  * All the block addresses of data and nodes should be nullified.
1088  */
1089 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1090 {
1091     struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1092     int err = 0, cont = 1;
1093     int level, offset[4], noffset[4];
1094     unsigned int nofs = 0;
1095     struct f2fs_inode *ri;
1096     struct dnode_of_data dn;
1097     struct page *page;
1098
1099     trace_f2fs_truncate_inode_blocks_enter(inode, from);
1100
1101     level = get_node_path(inode, from, offset, noffset);
1102     if (level < 0) {
1103         trace_f2fs_truncate_inode_blocks_exit(inode, level);
1104         return level;
1105     }
1106
1107     page = f2fs_get_node_page(sbi, inode->i_ino);
1108     if (IS_ERR(page)) {
1109         trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1110         return PTR_ERR(page);
1111     }
1112
1113     set_new_dnode(&dn, inode, page, NULL, 0);
1114     unlock_page(page);
1115
1116     ri = F2FS_INODE(page);
1117     switch (level) {
1118     case 0:
1119     case 1:
1120         nofs = noffset[1];
1121         break;
1122     case 2:
1123         nofs = noffset[1];
1124         if (!offset[level - 1])
1125             goto skip_partial;
1126         err = truncate_partial_nodes(&dn, ri, offset, level);
1127         if (err < 0 && err != -ENOENT)
1128             goto fail;
1129         nofs += 1 + NIDS_PER_BLOCK;
1130         break;
1131     case 3:
1132         nofs = 5 + 2 * NIDS_PER_BLOCK;
1133         if (!offset[level - 1])
1134             goto skip_partial;
1135         err = truncate_partial_nodes(&dn, ri, offset, level);
1136         if (err < 0 && err != -ENOENT)
1137             goto fail;
1138         break;
1139     default:
1140         BUG();
1141     }
1142
1143 skip_partial:
1144     while (cont) {
1145         dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1146         switch (offset[0]) {
1147         case NODE_DIR1_BLOCK:
1148         case NODE_DIR2_BLOCK:
1149             err = truncate_dnode(&dn);
1150             break;
1151
1152         case NODE_IND1_BLOCK:
1153         case NODE_IND2_BLOCK:
1154             err = truncate_nodes(&dn, nofs, offset[1], 2);
1155             break;
1156
1157         case NODE_DIND_BLOCK:
1158             err = truncate_nodes(&dn, nofs, offset[1], 3);
1159             cont = 0;
1160             break;
1161
1162         default:
1163             BUG();
1164         }
1165         if (err < 0 && err != -ENOENT)
1166             goto fail;
1167         if (offset[1] == 0 &&
1168                 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1169             lock_page(page);
1170             BUG_ON(page->mapping != NODE_MAPPING(sbi));
1171             f2fs_wait_on_page_writeback(page, NODE, true, true);
1172             ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1173             set_page_dirty(page);
1174             unlock_page(page);
1175         }
1176         offset[1] = 0;
1177         offset[0]++;
1178         nofs += err;
1179     }
1180 fail:
1181     f2fs_put_page(page, 0);
1182     trace_f2fs_truncate_inode_blocks_exit(inode, err);
1183     return err > 0 ? 0 : err;
1184 }
1185
1186 /* caller must lock inode page */
1187 int f2fs_truncate_xattr_node(struct inode *inode)
1188 {
1189     struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1190     nid_t nid = F2FS_I(inode)->i_xattr_nid;
1191     struct dnode_of_data dn;
1192     struct page *npage;
1193     int err;
1194
1195     if (!nid)
1196         return 0;
1197
1198     npage = f2fs_get_node_page(sbi, nid);
1199     if (IS_ERR(npage))
1200         return PTR_ERR(npage);
1201
1202     set_new_dnode(&dn, inode, NULL, npage, nid);
1203     err = truncate_node(&dn);
1204     if (err) {
1205         f2fs_put_page(npage, 1);
1206         return err;
1207     }
1208
1209     f2fs_i_xnid_write(inode, 0);
1210
1211     return 0;
1212 }
1213
1214 /*
1215  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1216  * f2fs_unlock_op().
1217  */
1218 int f2fs_remove_inode_page(struct inode *inode)
1219 {
1220     struct dnode_of_data dn;
1221     int err;
1222
1223     set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1224     err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1225     if (err)
1226         return err;
1227
1228     err = f2fs_truncate_xattr_node(inode);
1229     if (err) {
1230         f2fs_put_dnode(&dn);
1231         return err;
1232     }
1233
1234     /* remove potential inline_data blocks */
1235     if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1236                 S_ISLNK(inode->i_mode))
1237         f2fs_truncate_data_blocks_range(&dn, 1);
1238
1239     /* 0 is possible, after f2fs_new_inode() has failed */
1240     if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1241         f2fs_put_dnode(&dn);
1242         return -EIO;
1243     }
1244
1245     if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1246         f2fs_warn(F2FS_I_SB(inode),
1247             "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1248             inode->i_ino, (unsigned long long)inode->i_blocks);
1249         set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1250     }
1251
1252     /* will put inode & node pages */
1253     err = truncate_node(&dn);
1254     if (err) {
1255         f2fs_put_dnode(&dn);
1256         return err;
1257     }
1258     return 0;
1259 }
1260
1261 struct page *f2fs_new_inode_page(struct inode *inode)
1262 {
1263     struct dnode_of_data dn;
1264
1265     /* allocate inode page for new inode */
1266     set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1267
1268     /* caller should f2fs_put_page(page, 1); */
1269     return f2fs_new_node_page(&dn, 0);
1270 }
1271
1272 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1273 {
1274     struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1275     struct node_info new_ni;
1276     struct page *page;
1277     int err;
1278
1279     if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1280         return ERR_PTR(-EPERM);
1281
1282     page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1283     if (!page)
1284         return ERR_PTR(-ENOMEM);
1285
1286     if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1287         goto fail;
1288
1289 #ifdef CONFIG_F2FS_CHECK_FS
1290     err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1291     if (err) {
1292         dec_valid_node_count(sbi, dn->inode, !ofs);
1293         goto fail;
1294     }
1295     if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1296         err = -EFSCORRUPTED;
1297         set_sbi_flag(sbi, SBI_NEED_FSCK);
1298         goto fail;
1299     }
1300 #endif
1301     new_ni.nid = dn->nid;
1302     new_ni.ino = dn->inode->i_ino;
1303     new_ni.blk_addr = NULL_ADDR;
1304     new_ni.flag = 0;
1305     new_ni.version = 0;
1306     set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1307
1308     f2fs_wait_on_page_writeback(page, NODE, true, true);
1309     fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1310     set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1311     if (!PageUptodate(page))
1312         SetPageUptodate(page);
1313     if (set_page_dirty(page))
1314         dn->node_changed = true;
1315
1316     if (f2fs_has_xattr_block(ofs))
1317         f2fs_i_xnid_write(dn->inode, dn->nid);
1318
1319     if (ofs == 0)
1320         inc_valid_inode_count(sbi);
1321     return page;
1322
1323 fail:
1324     clear_node_page_dirty(page);
1325     f2fs_put_page(page, 1);
1326     return ERR_PTR(err);
1327 }
1328
1329 /*
1330  * Caller should do after getting the following values.
1331  * 0: f2fs_put_page(page, 0)
1332  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1333  */
1334 static int read_node_page(struct page *page, blk_opf_t op_flags)
1335 {
1336     struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1337     struct node_info ni;
1338     struct f2fs_io_info fio = {
1339         .sbi = sbi,
1340         .type = NODE,
1341         .op = REQ_OP_READ,
1342         .op_flags = op_flags,
1343         .page = page,
1344         .encrypted_page = NULL,
1345     };
1346     int err;
1347
1348     if (PageUptodate(page)) {
1349         if (!f2fs_inode_chksum_verify(sbi, page)) {
1350             ClearPageUptodate(page);
1351             return -EFSBADCRC;
1352         }
1353         return LOCKED_PAGE;
1354     }
1355
1356     err = f2fs_get_node_info(sbi, page->index, &ni, false);
1357     if (err)
1358         return err;
1359
1360     /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1361     if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1362             is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1363         ClearPageUptodate(page);
1364         return -ENOENT;
1365     }
1366
1367     fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1368
1369     err = f2fs_submit_page_bio(&fio);
1370
1371     if (!err)
1372         f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1373
1374     return err;
1375 }
1376
1377 /*
1378  * Readahead a node page
1379  */
1380 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1381 {
1382     struct page *apage;
1383     int err;
1384
1385     if (!nid)
1386         return;
1387     if (f2fs_check_nid_range(sbi, nid))
1388         return;
1389
1390     apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1391     if (apage)
1392         return;
1393
1394     apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1395     if (!apage)
1396         return;
1397
1398     err = read_node_page(apage, REQ_RAHEAD);
1399     f2fs_put_page(apage, err ? 1 : 0);
1400 }
1401
1402 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1403                     struct page *parent, int start)
1404 {
1405     struct page *page;
1406     int err;
1407
1408     if (!nid)
1409         return ERR_PTR(-ENOENT);
1410     if (f2fs_check_nid_range(sbi, nid))
1411         return ERR_PTR(-EINVAL);
1412 repeat:
1413     page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1414     if (!page)
1415         return ERR_PTR(-ENOMEM);
1416
1417     err = read_node_page(page, 0);
1418     if (err < 0) {
1419         goto out_put_err;
1420     } else if (err == LOCKED_PAGE) {
1421         err = 0;
1422         goto page_hit;
1423     }
1424
1425     if (parent)
1426         f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1427
1428     lock_page(page);
1429
1430     if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1431         f2fs_put_page(page, 1);
1432         goto repeat;
1433     }
1434
1435     if (unlikely(!PageUptodate(page))) {
1436         err = -EIO;
1437         goto out_err;
1438     }
1439
1440     if (!f2fs_inode_chksum_verify(sbi, page)) {
1441         err = -EFSBADCRC;
1442         goto out_err;
1443     }
1444 page_hit:
1445     if (likely(nid == nid_of_node(page)))
1446         return page;
1447
1448     f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1449               nid, nid_of_node(page), ino_of_node(page),
1450               ofs_of_node(page), cpver_of_node(page),
1451               next_blkaddr_of_node(page));
1452     set_sbi_flag(sbi, SBI_NEED_FSCK);
1453     err = -EINVAL;
1454 out_err:
1455     ClearPageUptodate(page);
1456 out_put_err:
1457     /* ENOENT comes from read_node_page which is not an error. */
1458     if (err != -ENOENT)
1459         f2fs_handle_page_eio(sbi, page->index, NODE);
1460     f2fs_put_page(page, 1);
1461     return ERR_PTR(err);
1462 }
1463
1464 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1465 {
1466     return __get_node_page(sbi, nid, NULL, 0);
1467 }
1468
1469 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1470 {
1471     struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1472     nid_t nid = get_nid(parent, start, false);
1473
1474     return __get_node_page(sbi, nid, parent, start);
1475 }
1476
1477 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1478 {
1479     struct inode *inode;
1480     struct page *page;
1481     int ret;
1482
1483     /* should flush inline_data before evict_inode */
1484     inode = ilookup(sbi->sb, ino);
1485     if (!inode)
1486         return;
1487
1488     page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1489                     FGP_LOCK|FGP_NOWAIT, 0);
1490     if (!page)
1491         goto iput_out;
1492
1493     if (!PageUptodate(page))
1494         goto page_out;
1495
1496     if (!PageDirty(page))
1497         goto page_out;
1498
1499     if (!clear_page_dirty_for_io(page))
1500         goto page_out;
1501
1502     ret = f2fs_write_inline_data(inode, page);
1503     inode_dec_dirty_pages(inode);
1504     f2fs_remove_dirty_inode(inode);
1505     if (ret)
1506         set_page_dirty(page);
1507 page_out:
1508     f2fs_put_page(page, 1);
1509 iput_out:
1510     iput(inode);
1511 }
1512
1513 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1514 {
1515     pgoff_t index;
1516     struct pagevec pvec;
1517     struct page *last_page = NULL;
1518     int nr_pages;
1519
1520     pagevec_init(&pvec);
1521     index = 0;
1522
1523     while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1524                 PAGECACHE_TAG_DIRTY))) {
1525         int i;
1526
1527         for (i = 0; i < nr_pages; i++) {
1528             struct page *page = pvec.pages[i];
1529
1530             if (unlikely(f2fs_cp_error(sbi))) {
1531                 f2fs_put_page(last_page, 0);
1532                 pagevec_release(&pvec);
1533                 return ERR_PTR(-EIO);
1534             }
1535
1536             if (!IS_DNODE(page) || !is_cold_node(page))
1537                 continue;
1538             if (ino_of_node(page) != ino)
1539                 continue;
1540
1541             lock_page(page);
1542
1543             if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1544 continue_unlock:
1545                 unlock_page(page);
1546                 continue;
1547             }
1548             if (ino_of_node(page) != ino)
1549                 goto continue_unlock;
1550
1551             if (!PageDirty(page)) {
1552                 /* someone wrote it for us */
1553                 goto continue_unlock;
1554             }
1555
1556             if (last_page)
1557                 f2fs_put_page(last_page, 0);
1558
1559             get_page(page);
1560             last_page = page;
1561             unlock_page(page);
1562         }
1563         pagevec_release(&pvec);
1564         cond_resched();
1565     }
1566     return last_page;
1567 }
1568
1569 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1570                 struct writeback_control *wbc, bool do_balance,
1571                 enum iostat_type io_type, unsigned int *seq_id)
1572 {
1573     struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1574     nid_t nid;
1575     struct node_info ni;
1576     struct f2fs_io_info fio = {
1577         .sbi = sbi,
1578         .ino = ino_of_node(page),
1579         .type = NODE,
1580         .op = REQ_OP_WRITE,
1581         .op_flags = wbc_to_write_flags(wbc),
1582         .page = page,
1583         .encrypted_page = NULL,
1584         .submitted = false,
1585         .io_type = io_type,
1586         .io_wbc = wbc,
1587     };
1588     unsigned int seq;
1589
1590     trace_f2fs_writepage(page, NODE);
1591
1592     if (unlikely(f2fs_cp_error(sbi))) {
1593         ClearPageUptodate(page);
1594         dec_page_count(sbi, F2FS_DIRTY_NODES);
1595         unlock_page(page);
1596         return 0;
1597     }
1598
1599     if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1600         goto redirty_out;
1601
1602     if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1603             wbc->sync_mode == WB_SYNC_NONE &&
1604             IS_DNODE(page) && is_cold_node(page))
1605         goto redirty_out;
1606
1607     /* get old block addr of this node page */
1608     nid = nid_of_node(page);
1609     f2fs_bug_on(sbi, page->index != nid);
1610
1611     if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1612         goto redirty_out;
1613
1614     if (wbc->for_reclaim) {
1615         if (!f2fs_down_read_trylock(&sbi->node_write))
1616             goto redirty_out;
1617     } else {
1618         f2fs_down_read(&sbi->node_write);
1619     }
1620
1621     /* This page is already truncated */
1622     if (unlikely(ni.blk_addr == NULL_ADDR)) {
1623         ClearPageUptodate(page);
1624         dec_page_count(sbi, F2FS_DIRTY_NODES);
1625         f2fs_up_read(&sbi->node_write);
1626         unlock_page(page);
1627         return 0;
1628     }
1629
1630     if (__is_valid_data_blkaddr(ni.blk_addr) &&
1631         !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1632                     DATA_GENERIC_ENHANCE)) {
1633         f2fs_up_read(&sbi->node_write);
1634         goto redirty_out;
1635     }
1636
1637     if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1638         fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1639
1640     /* should add to global list before clearing PAGECACHE status */
1641     if (f2fs_in_warm_node_list(sbi, page)) {
1642         seq = f2fs_add_fsync_node_entry(sbi, page);
1643         if (seq_id)
1644             *seq_id = seq;
1645     }
1646
1647     set_page_writeback(page);
1648     ClearPageError(page);
1649
1650     fio.old_blkaddr = ni.blk_addr;
1651     f2fs_do_write_node_page(nid, &fio);
1652     set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1653     dec_page_count(sbi, F2FS_DIRTY_NODES);
1654     f2fs_up_read(&sbi->node_write);
1655
1656     if (wbc->for_reclaim) {
1657         f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1658         submitted = NULL;
1659     }
1660
1661     unlock_page(page);
1662
1663     if (unlikely(f2fs_cp_error(sbi))) {
1664         f2fs_submit_merged_write(sbi, NODE);
1665         submitted = NULL;
1666     }
1667     if (submitted)
1668         *submitted = fio.submitted;
1669
1670     if (do_balance)
1671         f2fs_balance_fs(sbi, false);
1672     return 0;
1673
1674 redirty_out:
1675     redirty_page_for_writepage(wbc, page);
1676     return AOP_WRITEPAGE_ACTIVATE;
1677 }
1678
1679 int f2fs_move_node_page(struct page *node_page, int gc_type)
1680 {
1681     int err = 0;
1682
1683     if (gc_type == FG_GC) {
1684         struct writeback_control wbc = {
1685             .sync_mode = WB_SYNC_ALL,
1686             .nr_to_write = 1,
1687             .for_reclaim = 0,
1688         };
1689
1690         f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1691
1692         set_page_dirty(node_page);
1693
1694         if (!clear_page_dirty_for_io(node_page)) {
1695             err = -EAGAIN;
1696             goto out_page;
1697         }
1698
1699         if (__write_node_page(node_page, false, NULL,
1700                     &wbc, false, FS_GC_NODE_IO, NULL)) {
1701             err = -EAGAIN;
1702             unlock_page(node_page);
1703         }
1704         goto release_page;
1705     } else {
1706         /* set page dirty and write it */
1707         if (!PageWriteback(node_page))
1708             set_page_dirty(node_page);
1709     }
1710 out_page:
1711     unlock_page(node_page);
1712 release_page:
1713     f2fs_put_page(node_page, 0);
1714     return err;
1715 }
1716
1717 static int f2fs_write_node_page(struct page *page,
1718                 struct writeback_control *wbc)
1719 {
1720     return __write_node_page(page, false, NULL, wbc, false,
1721                         FS_NODE_IO, NULL);
1722 }
1723
1724 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1725             struct writeback_control *wbc, bool atomic,
1726             unsigned int *seq_id)
1727 {
1728     pgoff_t index;
1729     struct pagevec pvec;
1730     int ret = 0;
1731     struct page *last_page = NULL;
1732     bool marked = false;
1733     nid_t ino = inode->i_ino;
1734     int nr_pages;
1735     int nwritten = 0;
1736
1737     if (atomic) {
1738         last_page = last_fsync_dnode(sbi, ino);
1739         if (IS_ERR_OR_NULL(last_page))
1740             return PTR_ERR_OR_ZERO(last_page);
1741     }
1742 retry:
1743     pagevec_init(&pvec);
1744     index = 0;
1745
1746     while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1747                 PAGECACHE_TAG_DIRTY))) {
1748         int i;
1749
1750         for (i = 0; i < nr_pages; i++) {
1751             struct page *page = pvec.pages[i];
1752             bool submitted = false;
1753
1754             if (unlikely(f2fs_cp_error(sbi))) {
1755                 f2fs_put_page(last_page, 0);
1756                 pagevec_release(&pvec);
1757                 ret = -EIO;
1758                 goto out;
1759             }
1760
1761             if (!IS_DNODE(page) || !is_cold_node(page))
1762                 continue;
1763             if (ino_of_node(page) != ino)
1764                 continue;
1765
1766             lock_page(page);
1767
1768             if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1769 continue_unlock:
1770                 unlock_page(page);
1771                 continue;
1772             }
1773             if (ino_of_node(page) != ino)
1774                 goto continue_unlock;
1775
1776             if (!PageDirty(page) && page != last_page) {
1777                 /* someone wrote it for us */
1778                 goto continue_unlock;
1779             }
1780
1781             f2fs_wait_on_page_writeback(page, NODE, true, true);
1782
1783             set_fsync_mark(page, 0);
1784             set_dentry_mark(page, 0);
1785
1786             if (!atomic || page == last_page) {
1787                 set_fsync_mark(page, 1);
1788                 percpu_counter_inc(&sbi->rf_node_block_count);
1789                 if (IS_INODE(page)) {
1790                     if (is_inode_flag_set(inode,
1791                                 FI_DIRTY_INODE))
1792                         f2fs_update_inode(inode, page);
1793                     set_dentry_mark(page,
1794                         f2fs_need_dentry_mark(sbi, ino));
1795                 }
1796                 /* may be written by other thread */
1797                 if (!PageDirty(page))
1798                     set_page_dirty(page);
1799             }
1800
1801             if (!clear_page_dirty_for_io(page))
1802                 goto continue_unlock;
1803
1804             ret = __write_node_page(page, atomic &&
1805                         page == last_page,
1806                         &submitted, wbc, true,
1807                         FS_NODE_IO, seq_id);
1808             if (ret) {
1809                 unlock_page(page);
1810                 f2fs_put_page(last_page, 0);
1811                 break;
1812             } else if (submitted) {
1813                 nwritten++;
1814             }
1815
1816             if (page == last_page) {
1817                 f2fs_put_page(page, 0);
1818                 marked = true;
1819                 break;
1820             }
1821         }
1822         pagevec_release(&pvec);
1823         cond_resched();
1824
1825         if (ret || marked)
1826             break;
1827     }
1828     if (!ret && atomic && !marked) {
1829         f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1830                ino, last_page->index);
1831         lock_page(last_page);
1832         f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1833         set_page_dirty(last_page);
1834         unlock_page(last_page);
1835         goto retry;
1836     }
1837 out:
1838     if (nwritten)
1839         f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1840     return ret ? -EIO : 0;
1841 }
1842
1843 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1844 {
1845     struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1846     bool clean;
1847
1848     if (inode->i_ino != ino)
1849         return 0;
1850
1851     if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1852         return 0;
1853
1854     spin_lock(&sbi->inode_lock[DIRTY_META]);
1855     clean = list_empty(&F2FS_I(inode)->gdirty_list);
1856     spin_unlock(&sbi->inode_lock[DIRTY_META]);
1857
1858     if (clean)
1859         return 0;
1860
1861     inode = igrab(inode);
1862     if (!inode)
1863         return 0;
1864     return 1;
1865 }
1866
1867 static bool flush_dirty_inode(struct page *page)
1868 {
1869     struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1870     struct inode *inode;
1871     nid_t ino = ino_of_node(page);
1872
1873     inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1874     if (!inode)
1875         return false;
1876
1877     f2fs_update_inode(inode, page);
1878     unlock_page(page);
1879
1880     iput(inode);
1881     return true;
1882 }
1883
1884 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1885 {
1886     pgoff_t index = 0;
1887     struct pagevec pvec;
1888     int nr_pages;
1889
1890     pagevec_init(&pvec);
1891
1892     while ((nr_pages = pagevec_lookup_tag(&pvec,
1893             NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1894         int i;
1895
1896         for (i = 0; i < nr_pages; i++) {
1897             struct page *page = pvec.pages[i];
1898
1899             if (!IS_DNODE(page))
1900                 continue;
1901
1902             lock_page(page);
1903
1904             if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1905 continue_unlock:
1906                 unlock_page(page);
1907                 continue;
1908             }
1909
1910             if (!PageDirty(page)) {
1911                 /* someone wrote it for us */
1912                 goto continue_unlock;
1913             }
1914
1915             /* flush inline_data, if it's async context. */
1916             if (page_private_inline(page)) {
1917                 clear_page_private_inline(page);
1918                 unlock_page(page);
1919                 flush_inline_data(sbi, ino_of_node(page));
1920                 continue;
1921             }
1922             unlock_page(page);
1923         }
1924         pagevec_release(&pvec);
1925         cond_resched();
1926     }
1927 }
1928
1929 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1930                 struct writeback_control *wbc,
1931                 bool do_balance, enum iostat_type io_type)
1932 {
1933     pgoff_t index;
1934     struct pagevec pvec;
1935     int step = 0;
1936     int nwritten = 0;
1937     int ret = 0;
1938     int nr_pages, done = 0;
1939
1940     pagevec_init(&pvec);
1941
1942 next_step:
1943     index = 0;
1944
1945     while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1946             NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1947         int i;
1948
1949         for (i = 0; i < nr_pages; i++) {
1950             struct page *page = pvec.pages[i];
1951             bool submitted = false;
1952
1953             /* give a priority to WB_SYNC threads */
1954             if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1955                     wbc->sync_mode == WB_SYNC_NONE) {
1956                 done = 1;
1957                 break;
1958             }
1959
1960             /*
1961              * flushing sequence with step:
1962              * 0. indirect nodes
1963              * 1. dentry dnodes
1964              * 2. file dnodes
1965              */
1966             if (step == 0 && IS_DNODE(page))
1967                 continue;
1968             if (step == 1 && (!IS_DNODE(page) ||
1969                         is_cold_node(page)))
1970                 continue;
1971             if (step == 2 && (!IS_DNODE(page) ||
1972                         !is_cold_node(page)))
1973                 continue;
1974 lock_node:
1975             if (wbc->sync_mode == WB_SYNC_ALL)
1976                 lock_page(page);
1977             else if (!trylock_page(page))
1978                 continue;
1979
1980             if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1981 continue_unlock:
1982                 unlock_page(page);
1983                 continue;
1984             }
1985
1986             if (!PageDirty(page)) {
1987                 /* someone wrote it for us */
1988                 goto continue_unlock;
1989             }
1990
1991             /* flush inline_data/inode, if it's async context. */
1992             if (!do_balance)
1993                 goto write_node;
1994
1995             /* flush inline_data */
1996             if (page_private_inline(page)) {
1997                 clear_page_private_inline(page);
1998                 unlock_page(page);
1999                 flush_inline_data(sbi, ino_of_node(page));
2000                 goto lock_node;
2001             }
2002
2003             /* flush dirty inode */
2004             if (IS_INODE(page) && flush_dirty_inode(page))
2005                 goto lock_node;
2006 write_node:
2007             f2fs_wait_on_page_writeback(page, NODE, true, true);
2008
2009             if (!clear_page_dirty_for_io(page))
2010                 goto continue_unlock;
2011
2012             set_fsync_mark(page, 0);
2013             set_dentry_mark(page, 0);
2014
2015             ret = __write_node_page(page, false, &submitted,
2016                         wbc, do_balance, io_type, NULL);
2017             if (ret)
2018                 unlock_page(page);
2019             else if (submitted)
2020                 nwritten++;
2021
2022             if (--wbc->nr_to_write == 0)
2023                 break;
2024         }
2025         pagevec_release(&pvec);
2026         cond_resched();
2027
2028         if (wbc->nr_to_write == 0) {
2029             step = 2;
2030             break;
2031         }
2032     }
2033
2034     if (step < 2) {
2035         if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2036                 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2037             goto out;
2038         step++;
2039         goto next_step;
2040     }
2041 out:
2042     if (nwritten)
2043         f2fs_submit_merged_write(sbi, NODE);
2044
2045     if (unlikely(f2fs_cp_error(sbi)))
2046         return -EIO;
2047     return ret;
2048 }
2049
2050 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2051                         unsigned int seq_id)
2052 {
2053     struct fsync_node_entry *fn;
2054     struct page *page;
2055     struct list_head *head = &sbi->fsync_node_list;
2056     unsigned long flags;
2057     unsigned int cur_seq_id = 0;
2058     int ret2, ret = 0;
2059
2060     while (seq_id && cur_seq_id < seq_id) {
2061         spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2062         if (list_empty(head)) {
2063             spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2064             break;
2065         }
2066         fn = list_first_entry(head, struct fsync_node_entry, list);
2067         if (fn->seq_id > seq_id) {
2068             spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2069             break;
2070         }
2071         cur_seq_id = fn->seq_id;
2072         page = fn->page;
2073         get_page(page);
2074         spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2075
2076         f2fs_wait_on_page_writeback(page, NODE, true, false);
2077         if (TestClearPageError(page))
2078             ret = -EIO;
2079
2080         put_page(page);
2081
2082         if (ret)
2083             break;
2084     }
2085
2086     ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2087     if (!ret)
2088         ret = ret2;
2089
2090     return ret;
2091 }
2092
2093 static int f2fs_write_node_pages(struct address_space *mapping,
2094                 struct writeback_control *wbc)
2095 {
2096     struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2097     struct blk_plug plug;
2098     long diff;
2099
2100     if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2101         goto skip_write;
2102
2103     /* balancing f2fs's metadata in background */
2104     f2fs_balance_fs_bg(sbi, true);
2105
2106     /* collect a number of dirty node pages and write together */
2107     if (wbc->sync_mode != WB_SYNC_ALL &&
2108             get_pages(sbi, F2FS_DIRTY_NODES) <
2109                     nr_pages_to_skip(sbi, NODE))
2110         goto skip_write;
2111
2112     if (wbc->sync_mode == WB_SYNC_ALL)
2113         atomic_inc(&sbi->wb_sync_req[NODE]);
2114     else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2115         /* to avoid potential deadlock */
2116         if (current->plug)
2117             blk_finish_plug(current->plug);
2118         goto skip_write;
2119     }
2120
2121     trace_f2fs_writepages(mapping->host, wbc, NODE);
2122
2123     diff = nr_pages_to_write(sbi, NODE, wbc);
2124     blk_start_plug(&plug);
2125     f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2126     blk_finish_plug(&plug);
2127     wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2128
2129     if (wbc->sync_mode == WB_SYNC_ALL)
2130         atomic_dec(&sbi->wb_sync_req[NODE]);
2131     return 0;
2132
2133 skip_write:
2134     wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2135     trace_f2fs_writepages(mapping->host, wbc, NODE);
2136     return 0;
2137 }
2138
2139 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2140         struct folio *folio)
2141 {
2142     trace_f2fs_set_page_dirty(&folio->page, NODE);
2143
2144     if (!folio_test_uptodate(folio))
2145         folio_mark_uptodate(folio);
2146 #ifdef CONFIG_F2FS_CHECK_FS
2147     if (IS_INODE(&folio->page))
2148         f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2149 #endif
2150     if (!folio_test_dirty(folio)) {
2151         filemap_dirty_folio(mapping, folio);
2152         inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2153         set_page_private_reference(&folio->page);
2154         return true;
2155     }
2156     return false;
2157 }
2158
2159 /*
2160  * Structure of the f2fs node operations
2161  */
2162 const struct address_space_operations f2fs_node_aops = {
2163     .writepage  = f2fs_write_node_page,
2164     .writepages = f2fs_write_node_pages,
2165     .dirty_folio    = f2fs_dirty_node_folio,
2166     .invalidate_folio = f2fs_invalidate_folio,
2167     .release_folio  = f2fs_release_folio,
2168     .migrate_folio  = filemap_migrate_folio,
2169 };
2170
2171 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2172                         nid_t n)
2173 {
2174     return radix_tree_lookup(&nm_i->free_nid_root, n);
2175 }
2176
2177 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2178                 struct free_nid *i)
2179 {
2180     struct f2fs_nm_info *nm_i = NM_I(sbi);
2181     int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2182
2183     if (err)
2184         return err;
2185
2186     nm_i->nid_cnt[FREE_NID]++;
2187     list_add_tail(&i->list, &nm_i->free_nid_list);
2188     return 0;
2189 }
2190
2191 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2192             struct free_nid *i, enum nid_state state)
2193 {
2194     struct f2fs_nm_info *nm_i = NM_I(sbi);
2195
2196     f2fs_bug_on(sbi, state != i->state);
2197     nm_i->nid_cnt[state]--;
2198     if (state == FREE_NID)
2199         list_del(&i->list);
2200     radix_tree_delete(&nm_i->free_nid_root, i->nid);
2201 }
2202
2203 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2204             enum nid_state org_state, enum nid_state dst_state)
2205 {
2206     struct f2fs_nm_info *nm_i = NM_I(sbi);
2207
2208     f2fs_bug_on(sbi, org_state != i->state);
2209     i->state = dst_state;
2210     nm_i->nid_cnt[org_state]--;
2211     nm_i->nid_cnt[dst_state]++;
2212
2213     switch (dst_state) {
2214     case PREALLOC_NID:
2215         list_del(&i->list);
2216         break;
2217     case FREE_NID:
2218         list_add_tail(&i->list, &nm_i->free_nid_list);
2219         break;
2220     default:
2221         BUG_ON(1);
2222     }
2223 }
2224
2225 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2226 {
2227     struct f2fs_nm_info *nm_i = NM_I(sbi);
2228     unsigned int i;
2229     bool ret = true;
2230
2231     f2fs_down_read(&nm_i->nat_tree_lock);
2232     for (i = 0; i < nm_i->nat_blocks; i++) {
2233         if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2234             ret = false;
2235             break;
2236         }
2237     }
2238     f2fs_up_read(&nm_i->nat_tree_lock);
2239
2240     return ret;
2241 }
2242
2243 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2244                             bool set, bool build)
2245 {
2246     struct f2fs_nm_info *nm_i = NM_I(sbi);
2247     unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2248     unsigned int nid_ofs = nid - START_NID(nid);
2249
2250     if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2251         return;
2252
2253     if (set) {
2254         if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2255             return;
2256         __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2257         nm_i->free_nid_count[nat_ofs]++;
2258     } else {
2259         if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2260             return;
2261         __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2262         if (!build)
2263             nm_i->free_nid_count[nat_ofs]--;
2264     }
2265 }
2266
2267 /* return if the nid is recognized as free */
2268 static bool add_free_nid(struct f2fs_sb_info *sbi,
2269                 nid_t nid, bool build, bool update)
2270 {
2271     struct f2fs_nm_info *nm_i = NM_I(sbi);
2272     struct free_nid *i, *e;
2273     struct nat_entry *ne;
2274     int err = -EINVAL;
2275     bool ret = false;
2276
2277     /* 0 nid should not be used */
2278     if (unlikely(nid == 0))
2279         return false;
2280
2281     if (unlikely(f2fs_check_nid_range(sbi, nid)))
2282         return false;
2283
2284     i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2285     i->nid = nid;
2286     i->state = FREE_NID;
2287
2288     radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2289
2290     spin_lock(&nm_i->nid_list_lock);
2291
2292     if (build) {
2293         /*
2294          *   Thread A             Thread B
2295          *  - f2fs_create
2296          *   - f2fs_new_inode
2297          *    - f2fs_alloc_nid
2298          *     - __insert_nid_to_list(PREALLOC_NID)
2299          *                     - f2fs_balance_fs_bg
2300          *                      - f2fs_build_free_nids
2301          *                       - __f2fs_build_free_nids
2302          *                        - scan_nat_page
2303          *                         - add_free_nid
2304          *                          - __lookup_nat_cache
2305          *  - f2fs_add_link
2306          *   - f2fs_init_inode_metadata
2307          *    - f2fs_new_inode_page
2308          *     - f2fs_new_node_page
2309          *      - set_node_addr
2310          *  - f2fs_alloc_nid_done
2311          *   - __remove_nid_from_list(PREALLOC_NID)
2312          *                         - __insert_nid_to_list(FREE_NID)
2313          */
2314         ne = __lookup_nat_cache(nm_i, nid);
2315         if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2316                 nat_get_blkaddr(ne) != NULL_ADDR))
2317             goto err_out;
2318
2319         e = __lookup_free_nid_list(nm_i, nid);
2320         if (e) {
2321             if (e->state == FREE_NID)
2322                 ret = true;
2323             goto err_out;
2324         }
2325     }
2326     ret = true;
2327     err = __insert_free_nid(sbi, i);
2328 err_out:
2329     if (update) {
2330         update_free_nid_bitmap(sbi, nid, ret, build);
2331         if (!build)
2332             nm_i->available_nids++;
2333     }
2334     spin_unlock(&nm_i->nid_list_lock);
2335     radix_tree_preload_end();
2336
2337     if (err)
2338         kmem_cache_free(free_nid_slab, i);
2339     return ret;
2340 }
2341
2342 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2343 {
2344     struct f2fs_nm_info *nm_i = NM_I(sbi);
2345     struct free_nid *i;
2346     bool need_free = false;
2347
2348     spin_lock(&nm_i->nid_list_lock);
2349     i = __lookup_free_nid_list(nm_i, nid);
2350     if (i && i->state == FREE_NID) {
2351         __remove_free_nid(sbi, i, FREE_NID);
2352         need_free = true;
2353     }
2354     spin_unlock(&nm_i->nid_list_lock);
2355
2356     if (need_free)
2357         kmem_cache_free(free_nid_slab, i);
2358 }
2359
2360 static int scan_nat_page(struct f2fs_sb_info *sbi,
2361             struct page *nat_page, nid_t start_nid)
2362 {
2363     struct f2fs_nm_info *nm_i = NM_I(sbi);
2364     struct f2fs_nat_block *nat_blk = page_address(nat_page);
2365     block_t blk_addr;
2366     unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2367     int i;
2368
2369     __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2370
2371     i = start_nid % NAT_ENTRY_PER_BLOCK;
2372
2373     for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2374         if (unlikely(start_nid >= nm_i->max_nid))
2375             break;
2376
2377         blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2378
2379         if (blk_addr == NEW_ADDR)
2380             return -EINVAL;
2381
2382         if (blk_addr == NULL_ADDR) {
2383             add_free_nid(sbi, start_nid, true, true);
2384         } else {
2385             spin_lock(&NM_I(sbi)->nid_list_lock);
2386             update_free_nid_bitmap(sbi, start_nid, false, true);
2387             spin_unlock(&NM_I(sbi)->nid_list_lock);
2388         }
2389     }
2390
2391     return 0;
2392 }
2393
2394 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2395 {
2396     struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2397     struct f2fs_journal *journal = curseg->journal;
2398     int i;
2399
2400     down_read(&curseg->journal_rwsem);
2401     for (i = 0; i < nats_in_cursum(journal); i++) {
2402         block_t addr;
2403         nid_t nid;
2404
2405         addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2406         nid = le32_to_cpu(nid_in_journal(journal, i));
2407         if (addr == NULL_ADDR)
2408             add_free_nid(sbi, nid, true, false);
2409         else
2410             remove_free_nid(sbi, nid);
2411     }
2412     up_read(&curseg->journal_rwsem);
2413 }
2414
2415 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2416 {
2417     struct f2fs_nm_info *nm_i = NM_I(sbi);
2418     unsigned int i, idx;
2419     nid_t nid;
2420
2421     f2fs_down_read(&nm_i->nat_tree_lock);
2422
2423     for (i = 0; i < nm_i->nat_blocks; i++) {
2424         if (!test_bit_le(i, nm_i->nat_block_bitmap))
2425             continue;
2426         if (!nm_i->free_nid_count[i])
2427             continue;
2428         for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2429             idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2430                         NAT_ENTRY_PER_BLOCK, idx);
2431             if (idx >= NAT_ENTRY_PER_BLOCK)
2432                 break;
2433
2434             nid = i * NAT_ENTRY_PER_BLOCK + idx;
2435             add_free_nid(sbi, nid, true, false);
2436
2437             if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2438                 goto out;
2439         }
2440     }
2441 out:
2442     scan_curseg_cache(sbi);
2443
2444     f2fs_up_read(&nm_i->nat_tree_lock);
2445 }
2446
2447 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2448                         bool sync, bool mount)
2449 {
2450     struct f2fs_nm_info *nm_i = NM_I(sbi);
2451     int i = 0, ret;
2452     nid_t nid = nm_i->next_scan_nid;
2453
2454     if (unlikely(nid >= nm_i->max_nid))
2455         nid = 0;
2456
2457     if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2458         nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2459
2460     /* Enough entries */
2461     if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2462         return 0;
2463
2464     if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2465         return 0;
2466
2467     if (!mount) {
2468         /* try to find free nids in free_nid_bitmap */
2469         scan_free_nid_bits(sbi);
2470
2471         if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2472             return 0;
2473     }
2474
2475     /* readahead nat pages to be scanned */
2476     f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2477                             META_NAT, true);
2478
2479     f2fs_down_read(&nm_i->nat_tree_lock);
2480
2481     while (1) {
2482         if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2483                         nm_i->nat_block_bitmap)) {
2484             struct page *page = get_current_nat_page(sbi, nid);
2485
2486             if (IS_ERR(page)) {
2487                 ret = PTR_ERR(page);
2488             } else {
2489                 ret = scan_nat_page(sbi, page, nid);
2490                 f2fs_put_page(page, 1);
2491             }
2492
2493             if (ret) {
2494                 f2fs_up_read(&nm_i->nat_tree_lock);
2495                 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2496                 return ret;
2497             }
2498         }
2499
2500         nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2501         if (unlikely(nid >= nm_i->max_nid))
2502             nid = 0;
2503
2504         if (++i >= FREE_NID_PAGES)
2505             break;
2506     }
2507
2508     /* go to the next free nat pages to find free nids abundantly */
2509     nm_i->next_scan_nid = nid;
2510
2511     /* find free nids from current sum_pages */
2512     scan_curseg_cache(sbi);
2513
2514     f2fs_up_read(&nm_i->nat_tree_lock);
2515
2516     f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2517                     nm_i->ra_nid_pages, META_NAT, false);
2518
2519     return 0;
2520 }
2521
2522 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2523 {
2524     int ret;
2525
2526     mutex_lock(&NM_I(sbi)->build_lock);
2527     ret = __f2fs_build_free_nids(sbi, sync, mount);
2528     mutex_unlock(&NM_I(sbi)->build_lock);
2529
2530     return ret;
2531 }
2532
2533 /*
2534  * If this function returns success, caller can obtain a new nid
2535  * from second parameter of this function.
2536  * The returned nid could be used ino as well as nid when inode is created.
2537  */
2538 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2539 {
2540     struct f2fs_nm_info *nm_i = NM_I(sbi);
2541     struct free_nid *i = NULL;
2542 retry:
2543     if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2544         f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2545         return false;
2546     }
2547
2548     spin_lock(&nm_i->nid_list_lock);
2549
2550     if (unlikely(nm_i->available_nids == 0)) {
2551         spin_unlock(&nm_i->nid_list_lock);
2552         return false;
2553     }
2554
2555     /* We should not use stale free nids created by f2fs_build_free_nids */
2556     if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2557         f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2558         i = list_first_entry(&nm_i->free_nid_list,
2559                     struct free_nid, list);
2560         *nid = i->nid;
2561
2562         __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2563         nm_i->available_nids--;
2564
2565         update_free_nid_bitmap(sbi, *nid, false, false);
2566
2567         spin_unlock(&nm_i->nid_list_lock);
2568         return true;
2569     }
2570     spin_unlock(&nm_i->nid_list_lock);
2571
2572     /* Let's scan nat pages and its caches to get free nids */
2573     if (!f2fs_build_free_nids(sbi, true, false))
2574         goto retry;
2575     return false;
2576 }
2577
2578 /*
2579  * f2fs_alloc_nid() should be called prior to this function.
2580  */
2581 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2582 {
2583     struct f2fs_nm_info *nm_i = NM_I(sbi);
2584     struct free_nid *i;
2585
2586     spin_lock(&nm_i->nid_list_lock);
2587     i = __lookup_free_nid_list(nm_i, nid);
2588     f2fs_bug_on(sbi, !i);
2589     __remove_free_nid(sbi, i, PREALLOC_NID);
2590     spin_unlock(&nm_i->nid_list_lock);
2591
2592     kmem_cache_free(free_nid_slab, i);
2593 }
2594
2595 /*
2596  * f2fs_alloc_nid() should be called prior to this function.
2597  */
2598 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2599 {
2600     struct f2fs_nm_info *nm_i = NM_I(sbi);
2601     struct free_nid *i;
2602     bool need_free = false;
2603
2604     if (!nid)
2605         return;
2606
2607     spin_lock(&nm_i->nid_list_lock);
2608     i = __lookup_free_nid_list(nm_i, nid);
2609     f2fs_bug_on(sbi, !i);
2610
2611     if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2612         __remove_free_nid(sbi, i, PREALLOC_NID);
2613         need_free = true;
2614     } else {
2615         __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2616     }
2617
2618     nm_i->available_nids++;
2619
2620     update_free_nid_bitmap(sbi, nid, true, false);
2621
2622     spin_unlock(&nm_i->nid_list_lock);
2623
2624     if (need_free)
2625         kmem_cache_free(free_nid_slab, i);
2626 }
2627
2628 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2629 {
2630     struct f2fs_nm_info *nm_i = NM_I(sbi);
2631     int nr = nr_shrink;
2632
2633     if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2634         return 0;
2635
2636     if (!mutex_trylock(&nm_i->build_lock))
2637         return 0;
2638
2639     while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2640         struct free_nid *i, *next;
2641         unsigned int batch = SHRINK_NID_BATCH_SIZE;
2642
2643         spin_lock(&nm_i->nid_list_lock);
2644         list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2645             if (!nr_shrink || !batch ||
2646                 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2647                 break;
2648             __remove_free_nid(sbi, i, FREE_NID);
2649             kmem_cache_free(free_nid_slab, i);
2650             nr_shrink--;
2651             batch--;
2652         }
2653         spin_unlock(&nm_i->nid_list_lock);
2654     }
2655
2656     mutex_unlock(&nm_i->build_lock);
2657
2658     return nr - nr_shrink;
2659 }
2660
2661 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2662 {
2663     void *src_addr, *dst_addr;
2664     size_t inline_size;
2665     struct page *ipage;
2666     struct f2fs_inode *ri;
2667
2668     ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2669     if (IS_ERR(ipage))
2670         return PTR_ERR(ipage);
2671
2672     ri = F2FS_INODE(page);
2673     if (ri->i_inline & F2FS_INLINE_XATTR) {
2674         if (!f2fs_has_inline_xattr(inode)) {
2675             set_inode_flag(inode, FI_INLINE_XATTR);
2676             stat_inc_inline_xattr(inode);
2677         }
2678     } else {
2679         if (f2fs_has_inline_xattr(inode)) {
2680             stat_dec_inline_xattr(inode);
2681             clear_inode_flag(inode, FI_INLINE_XATTR);
2682         }
2683         goto update_inode;
2684     }
2685
2686     dst_addr = inline_xattr_addr(inode, ipage);
2687     src_addr = inline_xattr_addr(inode, page);
2688     inline_size = inline_xattr_size(inode);
2689
2690     f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2691     memcpy(dst_addr, src_addr, inline_size);
2692 update_inode:
2693     f2fs_update_inode(inode, ipage);
2694     f2fs_put_page(ipage, 1);
2695     return 0;
2696 }
2697
2698 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2699 {
2700     struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2701     nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2702     nid_t new_xnid;
2703     struct dnode_of_data dn;
2704     struct node_info ni;
2705     struct page *xpage;
2706     int err;
2707
2708     if (!prev_xnid)
2709         goto recover_xnid;
2710
2711     /* 1: invalidate the previous xattr nid */
2712     err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2713     if (err)
2714         return err;
2715
2716     f2fs_invalidate_blocks(sbi, ni.blk_addr);
2717     dec_valid_node_count(sbi, inode, false);
2718     set_node_addr(sbi, &ni, NULL_ADDR, false);
2719
2720 recover_xnid:
2721     /* 2: update xattr nid in inode */
2722     if (!f2fs_alloc_nid(sbi, &new_xnid))
2723         return -ENOSPC;
2724
2725     set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2726     xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2727     if (IS_ERR(xpage)) {
2728         f2fs_alloc_nid_failed(sbi, new_xnid);
2729         return PTR_ERR(xpage);
2730     }
2731
2732     f2fs_alloc_nid_done(sbi, new_xnid);
2733     f2fs_update_inode_page(inode);
2734
2735     /* 3: update and set xattr node page dirty */
2736     memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2737
2738     set_page_dirty(xpage);
2739     f2fs_put_page(xpage, 1);
2740
2741     return 0;
2742 }
2743
2744 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2745 {
2746     struct f2fs_inode *src, *dst;
2747     nid_t ino = ino_of_node(page);
2748     struct node_info old_ni, new_ni;
2749     struct page *ipage;
2750     int err;
2751
2752     err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2753     if (err)
2754         return err;
2755
2756     if (unlikely(old_ni.blk_addr != NULL_ADDR))
2757         return -EINVAL;
2758 retry:
2759     ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2760     if (!ipage) {
2761         memalloc_retry_wait(GFP_NOFS);
2762         goto retry;
2763     }
2764
2765     /* Should not use this inode from free nid list */
2766     remove_free_nid(sbi, ino);
2767
2768     if (!PageUptodate(ipage))
2769         SetPageUptodate(ipage);
2770     fill_node_footer(ipage, ino, ino, 0, true);
2771     set_cold_node(ipage, false);
2772
2773     src = F2FS_INODE(page);
2774     dst = F2FS_INODE(ipage);
2775
2776     memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2777     dst->i_size = 0;
2778     dst->i_blocks = cpu_to_le64(1);
2779     dst->i_links = cpu_to_le32(1);
2780     dst->i_xattr_nid = 0;
2781     dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2782     if (dst->i_inline & F2FS_EXTRA_ATTR) {
2783         dst->i_extra_isize = src->i_extra_isize;
2784
2785         if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2786             F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2787                             i_inline_xattr_size))
2788             dst->i_inline_xattr_size = src->i_inline_xattr_size;
2789
2790         if (f2fs_sb_has_project_quota(sbi) &&
2791             F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2792                                 i_projid))
2793             dst->i_projid = src->i_projid;
2794
2795         if (f2fs_sb_has_inode_crtime(sbi) &&
2796             F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2797                             i_crtime_nsec)) {
2798             dst->i_crtime = src->i_crtime;
2799             dst->i_crtime_nsec = src->i_crtime_nsec;
2800         }
2801     }
2802
2803     new_ni = old_ni;
2804     new_ni.ino = ino;
2805
2806     if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2807         WARN_ON(1);
2808     set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2809     inc_valid_inode_count(sbi);
2810     set_page_dirty(ipage);
2811     f2fs_put_page(ipage, 1);
2812     return 0;
2813 }
2814
2815 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2816             unsigned int segno, struct f2fs_summary_block *sum)
2817 {
2818     struct f2fs_node *rn;
2819     struct f2fs_summary *sum_entry;
2820     block_t addr;
2821     int i, idx, last_offset, nrpages;
2822
2823     /* scan the node segment */
2824     last_offset = sbi->blocks_per_seg;
2825     addr = START_BLOCK(sbi, segno);
2826     sum_entry = &sum->entries[0];
2827
2828     for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2829         nrpages = bio_max_segs(last_offset - i);
2830
2831         /* readahead node pages */
2832         f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2833
2834         for (idx = addr; idx < addr + nrpages; idx++) {
2835             struct page *page = f2fs_get_tmp_page(sbi, idx);
2836
2837             if (IS_ERR(page))
2838                 return PTR_ERR(page);
2839
2840             rn = F2FS_NODE(page);
2841             sum_entry->nid = rn->footer.nid;
2842             sum_entry->version = 0;
2843             sum_entry->ofs_in_node = 0;
2844             sum_entry++;
2845             f2fs_put_page(page, 1);
2846         }
2847
2848         invalidate_mapping_pages(META_MAPPING(sbi), addr,
2849                             addr + nrpages);
2850     }
2851     return 0;
2852 }
2853
2854 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2855 {
2856     struct f2fs_nm_info *nm_i = NM_I(sbi);
2857     struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2858     struct f2fs_journal *journal = curseg->journal;
2859     int i;
2860
2861     down_write(&curseg->journal_rwsem);
2862     for (i = 0; i < nats_in_cursum(journal); i++) {
2863         struct nat_entry *ne;
2864         struct f2fs_nat_entry raw_ne;
2865         nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2866
2867         if (f2fs_check_nid_range(sbi, nid))
2868             continue;
2869
2870         raw_ne = nat_in_journal(journal, i);
2871
2872         ne = __lookup_nat_cache(nm_i, nid);
2873         if (!ne) {
2874             ne = __alloc_nat_entry(sbi, nid, true);
2875             __init_nat_entry(nm_i, ne, &raw_ne, true);
2876         }
2877
2878         /*
2879          * if a free nat in journal has not been used after last
2880          * checkpoint, we should remove it from available nids,
2881          * since later we will add it again.
2882          */
2883         if (!get_nat_flag(ne, IS_DIRTY) &&
2884                 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2885             spin_lock(&nm_i->nid_list_lock);
2886             nm_i->available_nids--;
2887             spin_unlock(&nm_i->nid_list_lock);
2888         }
2889
2890         __set_nat_cache_dirty(nm_i, ne);
2891     }
2892     update_nats_in_cursum(journal, -i);
2893     up_write(&curseg->journal_rwsem);
2894 }
2895
2896 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2897                         struct list_head *head, int max)
2898 {
2899     struct nat_entry_set *cur;
2900
2901     if (nes->entry_cnt >= max)
2902         goto add_out;
2903
2904     list_for_each_entry(cur, head, set_list) {
2905         if (cur->entry_cnt >= nes->entry_cnt) {
2906             list_add(&nes->set_list, cur->set_list.prev);
2907             return;
2908         }
2909     }
2910 add_out:
2911     list_add_tail(&nes->set_list, head);
2912 }
2913
2914 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2915                             unsigned int valid)
2916 {
2917     if (valid == 0) {
2918         __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2919         __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2920         return;
2921     }
2922
2923     __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2924     if (valid == NAT_ENTRY_PER_BLOCK)
2925         __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2926     else
2927         __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2928 }
2929
2930 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2931                         struct page *page)
2932 {
2933     struct f2fs_nm_info *nm_i = NM_I(sbi);
2934     unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2935     struct f2fs_nat_block *nat_blk = page_address(page);
2936     int valid = 0;
2937     int i = 0;
2938
2939     if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2940         return;
2941
2942     if (nat_index == 0) {
2943         valid = 1;
2944         i = 1;
2945     }
2946     for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2947         if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2948             valid++;
2949     }
2950
2951     __update_nat_bits(nm_i, nat_index, valid);
2952 }
2953
2954 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2955 {
2956     struct f2fs_nm_info *nm_i = NM_I(sbi);
2957     unsigned int nat_ofs;
2958
2959     f2fs_down_read(&nm_i->nat_tree_lock);
2960
2961     for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2962         unsigned int valid = 0, nid_ofs = 0;
2963
2964         /* handle nid zero due to it should never be used */
2965         if (unlikely(nat_ofs == 0)) {
2966             valid = 1;
2967             nid_ofs = 1;
2968         }
2969
2970         for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2971             if (!test_bit_le(nid_ofs,
2972                     nm_i->free_nid_bitmap[nat_ofs]))
2973                 valid++;
2974         }
2975
2976         __update_nat_bits(nm_i, nat_ofs, valid);
2977     }
2978
2979     f2fs_up_read(&nm_i->nat_tree_lock);
2980 }
2981
2982 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2983         struct nat_entry_set *set, struct cp_control *cpc)
2984 {
2985     struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2986     struct f2fs_journal *journal = curseg->journal;
2987     nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2988     bool to_journal = true;
2989     struct f2fs_nat_block *nat_blk;
2990     struct nat_entry *ne, *cur;
2991     struct page *page = NULL;
2992
2993     /*
2994      * there are two steps to flush nat entries:
2995      * #1, flush nat entries to journal in current hot data summary block.
2996      * #2, flush nat entries to nat page.
2997      */
2998     if ((cpc->reason & CP_UMOUNT) ||
2999         !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3000         to_journal = false;
3001
3002     if (to_journal) {
3003         down_write(&curseg->journal_rwsem);
3004     } else {
3005         page = get_next_nat_page(sbi, start_nid);
3006         if (IS_ERR(page))
3007             return PTR_ERR(page);
3008
3009         nat_blk = page_address(page);
3010         f2fs_bug_on(sbi, !nat_blk);
3011     }
3012
3013     /* flush dirty nats in nat entry set */
3014     list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3015         struct f2fs_nat_entry *raw_ne;
3016         nid_t nid = nat_get_nid(ne);
3017         int offset;
3018
3019         f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3020
3021         if (to_journal) {
3022             offset = f2fs_lookup_journal_in_cursum(journal,
3023                             NAT_JOURNAL, nid, 1);
3024             f2fs_bug_on(sbi, offset < 0);
3025             raw_ne = &nat_in_journal(journal, offset);
3026             nid_in_journal(journal, offset) = cpu_to_le32(nid);
3027         } else {
3028             raw_ne = &nat_blk->entries[nid - start_nid];
3029         }
3030         raw_nat_from_node_info(raw_ne, &ne->ni);
3031         nat_reset_flag(ne);
3032         __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3033         if (nat_get_blkaddr(ne) == NULL_ADDR) {
3034             add_free_nid(sbi, nid, false, true);
3035         } else {
3036             spin_lock(&NM_I(sbi)->nid_list_lock);
3037             update_free_nid_bitmap(sbi, nid, false, false);
3038             spin_unlock(&NM_I(sbi)->nid_list_lock);
3039         }
3040     }
3041
3042     if (to_journal) {
3043         up_write(&curseg->journal_rwsem);
3044     } else {
3045         update_nat_bits(sbi, start_nid, page);
3046         f2fs_put_page(page, 1);
3047     }
3048
3049     /* Allow dirty nats by node block allocation in write_begin */
3050     if (!set->entry_cnt) {
3051         radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3052         kmem_cache_free(nat_entry_set_slab, set);
3053     }
3054     return 0;
3055 }
3056
3057 /*
3058  * This function is called during the checkpointing process.
3059  */
3060 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3061 {
3062     struct f2fs_nm_info *nm_i = NM_I(sbi);
3063     struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3064     struct f2fs_journal *journal = curseg->journal;
3065     struct nat_entry_set *setvec[SETVEC_SIZE];
3066     struct nat_entry_set *set, *tmp;
3067     unsigned int found;
3068     nid_t set_idx = 0;
3069     LIST_HEAD(sets);
3070     int err = 0;
3071
3072     /*
3073      * during unmount, let's flush nat_bits before checking
3074      * nat_cnt[DIRTY_NAT].
3075      */
3076     if (cpc->reason & CP_UMOUNT) {
3077         f2fs_down_write(&nm_i->nat_tree_lock);
3078         remove_nats_in_journal(sbi);
3079         f2fs_up_write(&nm_i->nat_tree_lock);
3080     }
3081
3082     if (!nm_i->nat_cnt[DIRTY_NAT])
3083         return 0;
3084
3085     f2fs_down_write(&nm_i->nat_tree_lock);
3086
3087     /*
3088      * if there are no enough space in journal to store dirty nat
3089      * entries, remove all entries from journal and merge them
3090      * into nat entry set.
3091      */
3092     if (cpc->reason & CP_UMOUNT ||
3093         !__has_cursum_space(journal,
3094             nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3095         remove_nats_in_journal(sbi);
3096
3097     while ((found = __gang_lookup_nat_set(nm_i,
3098                     set_idx, SETVEC_SIZE, setvec))) {
3099         unsigned idx;
3100
3101         set_idx = setvec[found - 1]->set + 1;
3102         for (idx = 0; idx < found; idx++)
3103             __adjust_nat_entry_set(setvec[idx], &sets,
3104                         MAX_NAT_JENTRIES(journal));
3105     }
3106
3107     /* flush dirty nats in nat entry set */
3108     list_for_each_entry_safe(set, tmp, &sets, set_list) {
3109         err = __flush_nat_entry_set(sbi, set, cpc);
3110         if (err)
3111             break;
3112     }
3113
3114     f2fs_up_write(&nm_i->nat_tree_lock);
3115     /* Allow dirty nats by node block allocation in write_begin */
3116
3117     return err;
3118 }
3119
3120 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3121 {
3122     struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3123     struct f2fs_nm_info *nm_i = NM_I(sbi);
3124     unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3125     unsigned int i;
3126     __u64 cp_ver = cur_cp_version(ckpt);
3127     block_t nat_bits_addr;
3128
3129     nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3130     nm_i->nat_bits = f2fs_kvzalloc(sbi,
3131             nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3132     if (!nm_i->nat_bits)
3133         return -ENOMEM;
3134
3135     nm_i->full_nat_bits = nm_i->nat_bits + 8;
3136     nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3137
3138     if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3139         return 0;
3140
3141     nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3142                         nm_i->nat_bits_blocks;
3143     for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3144         struct page *page;
3145
3146         page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3147         if (IS_ERR(page))
3148             return PTR_ERR(page);
3149
3150         memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3151                     page_address(page), F2FS_BLKSIZE);
3152         f2fs_put_page(page, 1);
3153     }
3154
3155     cp_ver |= (cur_cp_crc(ckpt) << 32);
3156     if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3157         clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3158         f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3159             cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3160         return 0;
3161     }
3162
3163     f2fs_notice(sbi, "Found nat_bits in checkpoint");
3164     return 0;
3165 }
3166
3167 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3168 {
3169     struct f2fs_nm_info *nm_i = NM_I(sbi);
3170     unsigned int i = 0;
3171     nid_t nid, last_nid;
3172
3173     if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3174         return;
3175
3176     for (i = 0; i < nm_i->nat_blocks; i++) {
3177         i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3178         if (i >= nm_i->nat_blocks)
3179             break;
3180
3181         __set_bit_le(i, nm_i->nat_block_bitmap);
3182
3183         nid = i * NAT_ENTRY_PER_BLOCK;
3184         last_nid = nid + NAT_ENTRY_PER_BLOCK;
3185
3186         spin_lock(&NM_I(sbi)->nid_list_lock);
3187         for (; nid < last_nid; nid++)
3188             update_free_nid_bitmap(sbi, nid, true, true);
3189         spin_unlock(&NM_I(sbi)->nid_list_lock);
3190     }
3191
3192     for (i = 0; i < nm_i->nat_blocks; i++) {
3193         i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3194         if (i >= nm_i->nat_blocks)
3195             break;
3196
3197         __set_bit_le(i, nm_i->nat_block_bitmap);
3198     }
3199 }
3200
3201 static int init_node_manager(struct f2fs_sb_info *sbi)
3202 {
3203     struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3204     struct f2fs_nm_info *nm_i = NM_I(sbi);
3205     unsigned char *version_bitmap;
3206     unsigned int nat_segs;
3207     int err;
3208
3209     nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3210
3211     /* segment_count_nat includes pair segment so divide to 2. */
3212     nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3213     nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3214     nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3215
3216     /* not used nids: 0, node, meta, (and root counted as valid node) */
3217     nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3218                         F2FS_RESERVED_NODE_NUM;
3219     nm_i->nid_cnt[FREE_NID] = 0;
3220     nm_i->nid_cnt[PREALLOC_NID] = 0;
3221     nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3222     nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3223     nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3224     nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3225
3226     INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3227     INIT_LIST_HEAD(&nm_i->free_nid_list);
3228     INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3229     INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3230     INIT_LIST_HEAD(&nm_i->nat_entries);
3231     spin_lock_init(&nm_i->nat_list_lock);
3232
3233     mutex_init(&nm_i->build_lock);
3234     spin_lock_init(&nm_i->nid_list_lock);
3235     init_f2fs_rwsem(&nm_i->nat_tree_lock);
3236
3237     nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3238     nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3239     version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3240     nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3241                     GFP_KERNEL);
3242     if (!nm_i->nat_bitmap)
3243         return -ENOMEM;
3244
3245     err = __get_nat_bitmaps(sbi);
3246     if (err)
3247         return err;
3248
3249 #ifdef CONFIG_F2FS_CHECK_FS
3250     nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3251                     GFP_KERNEL);
3252     if (!nm_i->nat_bitmap_mir)
3253         return -ENOMEM;
3254 #endif
3255
3256     return 0;
3257 }
3258
3259 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3260 {
3261     struct f2fs_nm_info *nm_i = NM_I(sbi);
3262     int i;
3263
3264     nm_i->free_nid_bitmap =
3265         f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3266                           nm_i->nat_blocks),
3267                   GFP_KERNEL);
3268     if (!nm_i->free_nid_bitmap)
3269         return -ENOMEM;
3270
3271     for (i = 0; i < nm_i->nat_blocks; i++) {
3272         nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3273             f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3274         if (!nm_i->free_nid_bitmap[i])
3275             return -ENOMEM;
3276     }
3277
3278     nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3279                                 GFP_KERNEL);
3280     if (!nm_i->nat_block_bitmap)
3281         return -ENOMEM;
3282
3283     nm_i->free_nid_count =
3284         f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3285                           nm_i->nat_blocks),
3286                   GFP_KERNEL);
3287     if (!nm_i->free_nid_count)
3288         return -ENOMEM;
3289     return 0;
3290 }
3291
3292 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3293 {
3294     int err;
3295
3296     sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3297                             GFP_KERNEL);
3298     if (!sbi->nm_info)
3299         return -ENOMEM;
3300
3301     err = init_node_manager(sbi);
3302     if (err)
3303         return err;
3304
3305     err = init_free_nid_cache(sbi);
3306     if (err)
3307         return err;
3308
3309     /* load free nid status from nat_bits table */
3310     load_free_nid_bitmap(sbi);
3311
3312     return f2fs_build_free_nids(sbi, true, true);
3313 }
3314
3315 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3316 {
3317     struct f2fs_nm_info *nm_i = NM_I(sbi);
3318     struct free_nid *i, *next_i;
3319     struct nat_entry *natvec[NATVEC_SIZE];
3320     struct nat_entry_set *setvec[SETVEC_SIZE];
3321     nid_t nid = 0;
3322     unsigned int found;
3323
3324     if (!nm_i)
3325         return;
3326
3327     /* destroy free nid list */
3328     spin_lock(&nm_i->nid_list_lock);
3329     list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3330         __remove_free_nid(sbi, i, FREE_NID);
3331         spin_unlock(&nm_i->nid_list_lock);
3332         kmem_cache_free(free_nid_slab, i);
3333         spin_lock(&nm_i->nid_list_lock);
3334     }
3335     f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3336     f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3337     f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3338     spin_unlock(&nm_i->nid_list_lock);
3339
3340     /* destroy nat cache */
3341     f2fs_down_write(&nm_i->nat_tree_lock);
3342     while ((found = __gang_lookup_nat_cache(nm_i,
3343                     nid, NATVEC_SIZE, natvec))) {
3344         unsigned idx;
3345
3346         nid = nat_get_nid(natvec[found - 1]) + 1;
3347         for (idx = 0; idx < found; idx++) {
3348             spin_lock(&nm_i->nat_list_lock);
3349             list_del(&natvec[idx]->list);
3350             spin_unlock(&nm_i->nat_list_lock);
3351
3352             __del_from_nat_cache(nm_i, natvec[idx]);
3353         }
3354     }
3355     f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3356
3357     /* destroy nat set cache */
3358     nid = 0;
3359     while ((found = __gang_lookup_nat_set(nm_i,
3360                     nid, SETVEC_SIZE, setvec))) {
3361         unsigned idx;
3362
3363         nid = setvec[found - 1]->set + 1;
3364         for (idx = 0; idx < found; idx++) {
3365             /* entry_cnt is not zero, when cp_error was occurred */
3366             f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3367             radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3368             kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3369         }
3370     }
3371     f2fs_up_write(&nm_i->nat_tree_lock);
3372
3373     kvfree(nm_i->nat_block_bitmap);
3374     if (nm_i->free_nid_bitmap) {
3375         int i;
3376
3377         for (i = 0; i < nm_i->nat_blocks; i++)
3378             kvfree(nm_i->free_nid_bitmap[i]);
3379         kvfree(nm_i->free_nid_bitmap);
3380     }
3381     kvfree(nm_i->free_nid_count);
3382
3383     kvfree(nm_i->nat_bitmap);
3384     kvfree(nm_i->nat_bits);
3385 #ifdef CONFIG_F2FS_CHECK_FS
3386     kvfree(nm_i->nat_bitmap_mir);
3387 #endif
3388     sbi->nm_info = NULL;
3389     kfree(nm_i);
3390 }
3391
3392 int __init f2fs_create_node_manager_caches(void)
3393 {
3394     nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3395             sizeof(struct nat_entry));
3396     if (!nat_entry_slab)
3397         goto fail;
3398
3399     free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3400             sizeof(struct free_nid));
3401     if (!free_nid_slab)
3402         goto destroy_nat_entry;
3403
3404     nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3405             sizeof(struct nat_entry_set));
3406     if (!nat_entry_set_slab)
3407         goto destroy_free_nid;
3408
3409     fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3410             sizeof(struct fsync_node_entry));
3411     if (!fsync_node_entry_slab)
3412         goto destroy_nat_entry_set;
3413     return 0;
3414
3415 destroy_nat_entry_set:
3416     kmem_cache_destroy(nat_entry_set_slab);
3417 destroy_free_nid:
3418     kmem_cache_destroy(free_nid_slab);
3419 destroy_nat_entry:
3420     kmem_cache_destroy(nat_entry_slab);
3421 fail:
3422     return -ENOMEM;
3423 }
3424
3425 void f2fs_destroy_node_manager_caches(void)
3426 {
3427     kmem_cache_destroy(fsync_node_entry_slab);
3428     kmem_cache_destroy(nat_entry_set_slab);
3429     kmem_cache_destroy(free_nid_slab);
3430     kmem_cache_destroy(nat_entry_slab);
3431 }