aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2010-10-23 04:26:47 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2010-10-23 04:26:47 -0400
commitab34c02afeafd047a831e6fe2a4dd875ce86bee0 (patch)
tree833e385e8387e3d4d4531ade23c1d92fba8ed075 /fs
parent7f3883962870dd28b5f2322ac44a9d03640ef448 (diff)
parent6b81e14e645016597c81e71cd27ee5c57c3a3c36 (diff)
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ryusuke/nilfs2
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ryusuke/nilfs2: (36 commits) nilfs2: eliminate sparse warning - "context imbalance" nilfs2: eliminate sparse warnings - "symbol not declared" nilfs2: get rid of bdi from nilfs object nilfs2: change license of exported header file nilfs2: add bdev freeze/thaw support nilfs2: accept 64-bit checkpoint numbers in cp mount option nilfs2: remove own inode allocator and destructor for metadata files nilfs2: get rid of back pointer to writable sb instance nilfs2: get rid of mi_nilfs back pointer to nilfs object nilfs2: see state of root dentry for mount check of snapshots nilfs2: use iget for all metadata files nilfs2: get rid of GCDAT inode nilfs2: add routines to redirect access to buffers of DAT file nilfs2: add routines to roll back state of DAT file nilfs2: add routines to save and restore bmap state nilfs2: do not allocate nilfs_mdt_info structure to gc-inodes nilfs2: allow nilfs_clear_inode to clear metadata file inodes nilfs2: get rid of snapshot mount flag nilfs2: simplify life cycle management of nilfs object nilfs2: do not allocate multiple super block instances for a device ...
Diffstat (limited to 'fs')
-rw-r--r--fs/nilfs2/Makefile2
-rw-r--r--fs/nilfs2/bmap.c22
-rw-r--r--fs/nilfs2/bmap.h10
-rw-r--r--fs/nilfs2/btnode.c17
-rw-r--r--fs/nilfs2/cpfile.c72
-rw-r--r--fs/nilfs2/cpfile.h4
-rw-r--r--fs/nilfs2/dat.c92
-rw-r--r--fs/nilfs2/dat.h4
-rw-r--r--fs/nilfs2/export.h17
-rw-r--r--fs/nilfs2/gcdat.c87
-rw-r--r--fs/nilfs2/gcinode.c134
-rw-r--r--fs/nilfs2/ifile.c51
-rw-r--r--fs/nilfs2/ifile.h4
-rw-r--r--fs/nilfs2/inode.c167
-rw-r--r--fs/nilfs2/ioctl.c23
-rw-r--r--fs/nilfs2/mdt.c313
-rw-r--r--fs/nilfs2/mdt.h32
-rw-r--r--fs/nilfs2/namei.c139
-rw-r--r--fs/nilfs2/nilfs.h38
-rw-r--r--fs/nilfs2/page.c55
-rw-r--r--fs/nilfs2/page.h6
-rw-r--r--fs/nilfs2/recovery.c19
-rw-r--r--fs/nilfs2/sb.h10
-rw-r--r--fs/nilfs2/segbuf.c3
-rw-r--r--fs/nilfs2/segment.c102
-rw-r--r--fs/nilfs2/segment.h10
-rw-r--r--fs/nilfs2/sufile.c77
-rw-r--r--fs/nilfs2/sufile.h6
-rw-r--r--fs/nilfs2/super.c609
-rw-r--r--fs/nilfs2/the_nilfs.c339
-rw-r--r--fs/nilfs2/the_nilfs.h101
31 files changed, 1313 insertions, 1252 deletions
diff --git a/fs/nilfs2/Makefile b/fs/nilfs2/Makefile
index df3e62c1ddc5..85c98737a146 100644
--- a/fs/nilfs2/Makefile
+++ b/fs/nilfs2/Makefile
@@ -2,4 +2,4 @@ obj-$(CONFIG_NILFS2_FS) += nilfs2.o
2nilfs2-y := inode.o file.o dir.o super.o namei.o page.o mdt.o \ 2nilfs2-y := inode.o file.o dir.o super.o namei.o page.o mdt.o \
3 btnode.o bmap.o btree.o direct.o dat.o recovery.o \ 3 btnode.o bmap.o btree.o direct.o dat.o recovery.o \
4 the_nilfs.o segbuf.o segment.o cpfile.o sufile.o \ 4 the_nilfs.o segbuf.o segment.o cpfile.o sufile.o \
5 ifile.o alloc.o gcinode.o ioctl.o gcdat.o 5 ifile.o alloc.o gcinode.o ioctl.o
diff --git a/fs/nilfs2/bmap.c b/fs/nilfs2/bmap.c
index 3dbdc1d356bf..8b782b062baa 100644
--- a/fs/nilfs2/bmap.c
+++ b/fs/nilfs2/bmap.c
@@ -533,18 +533,20 @@ void nilfs_bmap_init_gc(struct nilfs_bmap *bmap)
533 nilfs_btree_init_gc(bmap); 533 nilfs_btree_init_gc(bmap);
534} 534}
535 535
536void nilfs_bmap_init_gcdat(struct nilfs_bmap *gcbmap, struct nilfs_bmap *bmap) 536void nilfs_bmap_save(const struct nilfs_bmap *bmap,
537 struct nilfs_bmap_store *store)
537{ 538{
538 memcpy(gcbmap, bmap, sizeof(*bmap)); 539 memcpy(store->data, bmap->b_u.u_data, sizeof(store->data));
539 init_rwsem(&gcbmap->b_sem); 540 store->last_allocated_key = bmap->b_last_allocated_key;
540 lockdep_set_class(&bmap->b_sem, &nilfs_bmap_dat_lock_key); 541 store->last_allocated_ptr = bmap->b_last_allocated_ptr;
541 gcbmap->b_inode = &NILFS_BMAP_I(gcbmap)->vfs_inode; 542 store->state = bmap->b_state;
542} 543}
543 544
544void nilfs_bmap_commit_gcdat(struct nilfs_bmap *gcbmap, struct nilfs_bmap *bmap) 545void nilfs_bmap_restore(struct nilfs_bmap *bmap,
546 const struct nilfs_bmap_store *store)
545{ 547{
546 memcpy(bmap, gcbmap, sizeof(*bmap)); 548 memcpy(bmap->b_u.u_data, store->data, sizeof(store->data));
547 init_rwsem(&bmap->b_sem); 549 bmap->b_last_allocated_key = store->last_allocated_key;
548 lockdep_set_class(&bmap->b_sem, &nilfs_bmap_dat_lock_key); 550 bmap->b_last_allocated_ptr = store->last_allocated_ptr;
549 bmap->b_inode = &NILFS_BMAP_I(bmap)->vfs_inode; 551 bmap->b_state = store->state;
550} 552}
diff --git a/fs/nilfs2/bmap.h b/fs/nilfs2/bmap.h
index a20569b19929..bde1c0aa2e15 100644
--- a/fs/nilfs2/bmap.h
+++ b/fs/nilfs2/bmap.h
@@ -135,6 +135,12 @@ struct nilfs_bmap {
135/* state */ 135/* state */
136#define NILFS_BMAP_DIRTY 0x00000001 136#define NILFS_BMAP_DIRTY 0x00000001
137 137
138struct nilfs_bmap_store {
139 __le64 data[NILFS_BMAP_SIZE / sizeof(__le64)];
140 __u64 last_allocated_key;
141 __u64 last_allocated_ptr;
142 int state;
143};
138 144
139int nilfs_bmap_test_and_clear_dirty(struct nilfs_bmap *); 145int nilfs_bmap_test_and_clear_dirty(struct nilfs_bmap *);
140int nilfs_bmap_read(struct nilfs_bmap *, struct nilfs_inode *); 146int nilfs_bmap_read(struct nilfs_bmap *, struct nilfs_inode *);
@@ -153,9 +159,9 @@ int nilfs_bmap_lookup_at_level(struct nilfs_bmap *, __u64, int, __u64 *);
153int nilfs_bmap_mark(struct nilfs_bmap *, __u64, int); 159int nilfs_bmap_mark(struct nilfs_bmap *, __u64, int);
154 160
155void nilfs_bmap_init_gc(struct nilfs_bmap *); 161void nilfs_bmap_init_gc(struct nilfs_bmap *);
156void nilfs_bmap_init_gcdat(struct nilfs_bmap *, struct nilfs_bmap *);
157void nilfs_bmap_commit_gcdat(struct nilfs_bmap *, struct nilfs_bmap *);
158 162
163void nilfs_bmap_save(const struct nilfs_bmap *, struct nilfs_bmap_store *);
164void nilfs_bmap_restore(struct nilfs_bmap *, const struct nilfs_bmap_store *);
159 165
160static inline int nilfs_bmap_lookup(struct nilfs_bmap *bmap, __u64 key, 166static inline int nilfs_bmap_lookup(struct nilfs_bmap *bmap, __u64 key,
161 __u64 *ptr) 167 __u64 *ptr)
diff --git a/fs/nilfs2/btnode.c b/fs/nilfs2/btnode.c
index f78ab1044d1d..5115814cb745 100644
--- a/fs/nilfs2/btnode.c
+++ b/fs/nilfs2/btnode.c
@@ -37,15 +37,7 @@
37 37
38void nilfs_btnode_cache_init_once(struct address_space *btnc) 38void nilfs_btnode_cache_init_once(struct address_space *btnc)
39{ 39{
40 memset(btnc, 0, sizeof(*btnc)); 40 nilfs_mapping_init_once(btnc);
41 INIT_RADIX_TREE(&btnc->page_tree, GFP_ATOMIC);
42 spin_lock_init(&btnc->tree_lock);
43 INIT_LIST_HEAD(&btnc->private_list);
44 spin_lock_init(&btnc->private_lock);
45
46 spin_lock_init(&btnc->i_mmap_lock);
47 INIT_RAW_PRIO_TREE_ROOT(&btnc->i_mmap);
48 INIT_LIST_HEAD(&btnc->i_mmap_nonlinear);
49} 41}
50 42
51static const struct address_space_operations def_btnode_aops = { 43static const struct address_space_operations def_btnode_aops = {
@@ -55,12 +47,7 @@ static const struct address_space_operations def_btnode_aops = {
55void nilfs_btnode_cache_init(struct address_space *btnc, 47void nilfs_btnode_cache_init(struct address_space *btnc,
56 struct backing_dev_info *bdi) 48 struct backing_dev_info *bdi)
57{ 49{
58 btnc->host = NULL; /* can safely set to host inode ? */ 50 nilfs_mapping_init(btnc, bdi, &def_btnode_aops);
59 btnc->flags = 0;
60 mapping_set_gfp_mask(btnc, GFP_NOFS);
61 btnc->assoc_mapping = NULL;
62 btnc->backing_dev_info = bdi;
63 btnc->a_ops = &def_btnode_aops;
64} 51}
65 52
66void nilfs_btnode_cache_clear(struct address_space *btnc) 53void nilfs_btnode_cache_clear(struct address_space *btnc)
diff --git a/fs/nilfs2/cpfile.c b/fs/nilfs2/cpfile.c
index 18737818db63..5ff15a8a1024 100644
--- a/fs/nilfs2/cpfile.c
+++ b/fs/nilfs2/cpfile.c
@@ -863,26 +863,19 @@ int nilfs_cpfile_is_snapshot(struct inode *cpfile, __u64 cno)
863 */ 863 */
864int nilfs_cpfile_change_cpmode(struct inode *cpfile, __u64 cno, int mode) 864int nilfs_cpfile_change_cpmode(struct inode *cpfile, __u64 cno, int mode)
865{ 865{
866 struct the_nilfs *nilfs;
867 int ret; 866 int ret;
868 867
869 nilfs = NILFS_MDT(cpfile)->mi_nilfs;
870
871 switch (mode) { 868 switch (mode) {
872 case NILFS_CHECKPOINT: 869 case NILFS_CHECKPOINT:
873 /* 870 if (nilfs_checkpoint_is_mounted(cpfile->i_sb, cno))
874 * Check for protecting existing snapshot mounts: 871 /*
875 * ns_mount_mutex is used to make this operation atomic and 872 * Current implementation does not have to protect
876 * exclusive with a new mount job. Though it doesn't cover 873 * plain read-only mounts since they are exclusive
877 * umount, it's enough for the purpose. 874 * with a read/write mount and are protected from the
878 */ 875 * cleaner.
879 if (nilfs_checkpoint_is_mounted(nilfs, cno, 1)) { 876 */
880 /* Current implementation does not have to protect
881 plain read-only mounts since they are exclusive
882 with a read/write mount and are protected from the
883 cleaner. */
884 ret = -EBUSY; 877 ret = -EBUSY;
885 } else 878 else
886 ret = nilfs_cpfile_clear_snapshot(cpfile, cno); 879 ret = nilfs_cpfile_clear_snapshot(cpfile, cno);
887 return ret; 880 return ret;
888 case NILFS_SNAPSHOT: 881 case NILFS_SNAPSHOT:
@@ -933,27 +926,40 @@ int nilfs_cpfile_get_stat(struct inode *cpfile, struct nilfs_cpstat *cpstat)
933} 926}
934 927
935/** 928/**
936 * nilfs_cpfile_read - read cpfile inode 929 * nilfs_cpfile_read - read or get cpfile inode
937 * @cpfile: cpfile inode 930 * @sb: super block instance
938 * @raw_inode: on-disk cpfile inode
939 */
940int nilfs_cpfile_read(struct inode *cpfile, struct nilfs_inode *raw_inode)
941{
942 return nilfs_read_inode_common(cpfile, raw_inode);
943}
944
945/**
946 * nilfs_cpfile_new - create cpfile
947 * @nilfs: nilfs object
948 * @cpsize: size of a checkpoint entry 931 * @cpsize: size of a checkpoint entry
932 * @raw_inode: on-disk cpfile inode
933 * @inodep: buffer to store the inode
949 */ 934 */
950struct inode *nilfs_cpfile_new(struct the_nilfs *nilfs, size_t cpsize) 935int nilfs_cpfile_read(struct super_block *sb, size_t cpsize,
936 struct nilfs_inode *raw_inode, struct inode **inodep)
951{ 937{
952 struct inode *cpfile; 938 struct inode *cpfile;
939 int err;
940
941 cpfile = nilfs_iget_locked(sb, NULL, NILFS_CPFILE_INO);
942 if (unlikely(!cpfile))
943 return -ENOMEM;
944 if (!(cpfile->i_state & I_NEW))
945 goto out;
946
947 err = nilfs_mdt_init(cpfile, NILFS_MDT_GFP, 0);
948 if (err)
949 goto failed;
953 950
954 cpfile = nilfs_mdt_new(nilfs, NULL, NILFS_CPFILE_INO, 0); 951 nilfs_mdt_set_entry_size(cpfile, cpsize,
955 if (cpfile) 952 sizeof(struct nilfs_cpfile_header));
956 nilfs_mdt_set_entry_size(cpfile, cpsize, 953
957 sizeof(struct nilfs_cpfile_header)); 954 err = nilfs_read_inode_common(cpfile, raw_inode);
958 return cpfile; 955 if (err)
956 goto failed;
957
958 unlock_new_inode(cpfile);
959 out:
960 *inodep = cpfile;
961 return 0;
962 failed:
963 iget_failed(cpfile);
964 return err;
959} 965}
diff --git a/fs/nilfs2/cpfile.h b/fs/nilfs2/cpfile.h
index bc0809e0ab43..a242b9a314f9 100644
--- a/fs/nilfs2/cpfile.h
+++ b/fs/nilfs2/cpfile.h
@@ -40,7 +40,7 @@ int nilfs_cpfile_get_stat(struct inode *, struct nilfs_cpstat *);
40ssize_t nilfs_cpfile_get_cpinfo(struct inode *, __u64 *, int, void *, unsigned, 40ssize_t nilfs_cpfile_get_cpinfo(struct inode *, __u64 *, int, void *, unsigned,
41 size_t); 41 size_t);
42 42
43int nilfs_cpfile_read(struct inode *cpfile, struct nilfs_inode *raw_inode); 43int nilfs_cpfile_read(struct super_block *sb, size_t cpsize,
44struct inode *nilfs_cpfile_new(struct the_nilfs *nilfs, size_t cpsize); 44 struct nilfs_inode *raw_inode, struct inode **inodep);
45 45
46#endif /* _NILFS_CPFILE_H */ 46#endif /* _NILFS_CPFILE_H */
diff --git a/fs/nilfs2/dat.c b/fs/nilfs2/dat.c
index 013146755683..49c844dab33a 100644
--- a/fs/nilfs2/dat.c
+++ b/fs/nilfs2/dat.c
@@ -36,6 +36,7 @@
36struct nilfs_dat_info { 36struct nilfs_dat_info {
37 struct nilfs_mdt_info mi; 37 struct nilfs_mdt_info mi;
38 struct nilfs_palloc_cache palloc_cache; 38 struct nilfs_palloc_cache palloc_cache;
39 struct nilfs_shadow_map shadow;
39}; 40};
40 41
41static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat) 42static inline struct nilfs_dat_info *NILFS_DAT_I(struct inode *dat)
@@ -102,7 +103,8 @@ void nilfs_dat_abort_alloc(struct inode *dat, struct nilfs_palloc_req *req)
102 nilfs_palloc_abort_alloc_entry(dat, req); 103 nilfs_palloc_abort_alloc_entry(dat, req);
103} 104}
104 105
105void nilfs_dat_commit_free(struct inode *dat, struct nilfs_palloc_req *req) 106static void nilfs_dat_commit_free(struct inode *dat,
107 struct nilfs_palloc_req *req)
106{ 108{
107 struct nilfs_dat_entry *entry; 109 struct nilfs_dat_entry *entry;
108 void *kaddr; 110 void *kaddr;
@@ -327,6 +329,23 @@ int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
327 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh); 329 ret = nilfs_palloc_get_entry_block(dat, vblocknr, 0, &entry_bh);
328 if (ret < 0) 330 if (ret < 0)
329 return ret; 331 return ret;
332
333 /*
334 * The given disk block number (blocknr) is not yet written to
335 * the device at this point.
336 *
337 * To prevent nilfs_dat_translate() from returning the
338 * uncommited block number, this makes a copy of the entry
339 * buffer and redirects nilfs_dat_translate() to the copy.
340 */
341 if (!buffer_nilfs_redirected(entry_bh)) {
342 ret = nilfs_mdt_freeze_buffer(dat, entry_bh);
343 if (ret) {
344 brelse(entry_bh);
345 return ret;
346 }
347 }
348
330 kaddr = kmap_atomic(entry_bh->b_page, KM_USER0); 349 kaddr = kmap_atomic(entry_bh->b_page, KM_USER0);
331 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr); 350 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
332 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) { 351 if (unlikely(entry->de_blocknr == cpu_to_le64(0))) {
@@ -371,7 +390,7 @@ int nilfs_dat_move(struct inode *dat, __u64 vblocknr, sector_t blocknr)
371 */ 390 */
372int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp) 391int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
373{ 392{
374 struct buffer_head *entry_bh; 393 struct buffer_head *entry_bh, *bh;
375 struct nilfs_dat_entry *entry; 394 struct nilfs_dat_entry *entry;
376 sector_t blocknr; 395 sector_t blocknr;
377 void *kaddr; 396 void *kaddr;
@@ -381,6 +400,15 @@ int nilfs_dat_translate(struct inode *dat, __u64 vblocknr, sector_t *blocknrp)
381 if (ret < 0) 400 if (ret < 0)
382 return ret; 401 return ret;
383 402
403 if (!nilfs_doing_gc() && buffer_nilfs_redirected(entry_bh)) {
404 bh = nilfs_mdt_get_frozen_buffer(dat, entry_bh);
405 if (bh) {
406 WARN_ON(!buffer_uptodate(bh));
407 brelse(entry_bh);
408 entry_bh = bh;
409 }
410 }
411
384 kaddr = kmap_atomic(entry_bh->b_page, KM_USER0); 412 kaddr = kmap_atomic(entry_bh->b_page, KM_USER0);
385 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr); 413 entry = nilfs_palloc_block_get_entry(dat, vblocknr, entry_bh, kaddr);
386 blocknr = le64_to_cpu(entry->de_blocknr); 414 blocknr = le64_to_cpu(entry->de_blocknr);
@@ -436,38 +464,48 @@ ssize_t nilfs_dat_get_vinfo(struct inode *dat, void *buf, unsigned visz,
436} 464}
437 465
438/** 466/**
439 * nilfs_dat_read - read dat inode 467 * nilfs_dat_read - read or get dat inode
440 * @dat: dat inode 468 * @sb: super block instance
441 * @raw_inode: on-disk dat inode
442 */
443int nilfs_dat_read(struct inode *dat, struct nilfs_inode *raw_inode)
444{
445 return nilfs_read_inode_common(dat, raw_inode);
446}
447
448/**
449 * nilfs_dat_new - create dat file
450 * @nilfs: nilfs object
451 * @entry_size: size of a dat entry 469 * @entry_size: size of a dat entry
470 * @raw_inode: on-disk dat inode
471 * @inodep: buffer to store the inode
452 */ 472 */
453struct inode *nilfs_dat_new(struct the_nilfs *nilfs, size_t entry_size) 473int nilfs_dat_read(struct super_block *sb, size_t entry_size,
474 struct nilfs_inode *raw_inode, struct inode **inodep)
454{ 475{
455 static struct lock_class_key dat_lock_key; 476 static struct lock_class_key dat_lock_key;
456 struct inode *dat; 477 struct inode *dat;
457 struct nilfs_dat_info *di; 478 struct nilfs_dat_info *di;
458 int err; 479 int err;
459 480
460 dat = nilfs_mdt_new(nilfs, NULL, NILFS_DAT_INO, sizeof(*di)); 481 dat = nilfs_iget_locked(sb, NULL, NILFS_DAT_INO);
461 if (dat) { 482 if (unlikely(!dat))
462 err = nilfs_palloc_init_blockgroup(dat, entry_size); 483 return -ENOMEM;
463 if (unlikely(err)) { 484 if (!(dat->i_state & I_NEW))
464 nilfs_mdt_destroy(dat); 485 goto out;
465 return NULL;
466 }
467 486
468 di = NILFS_DAT_I(dat); 487 err = nilfs_mdt_init(dat, NILFS_MDT_GFP, sizeof(*di));
469 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key); 488 if (err)
470 nilfs_palloc_setup_cache(dat, &di->palloc_cache); 489 goto failed;
471 } 490
472 return dat; 491 err = nilfs_palloc_init_blockgroup(dat, entry_size);
492 if (err)
493 goto failed;
494
495 di = NILFS_DAT_I(dat);
496 lockdep_set_class(&di->mi.mi_sem, &dat_lock_key);
497 nilfs_palloc_setup_cache(dat, &di->palloc_cache);
498 nilfs_mdt_setup_shadow_map(dat, &di->shadow);
499
500 err = nilfs_read_inode_common(dat, raw_inode);
501 if (err)
502 goto failed;
503
504 unlock_new_inode(dat);
505 out:
506 *inodep = dat;
507 return 0;
508 failed:
509 iget_failed(dat);
510 return err;
473} 511}
diff --git a/fs/nilfs2/dat.h b/fs/nilfs2/dat.h
index d31c3aab0efe..cbd8e9732503 100644
--- a/fs/nilfs2/dat.h
+++ b/fs/nilfs2/dat.h
@@ -53,7 +53,7 @@ int nilfs_dat_freev(struct inode *, __u64 *, size_t);
53int nilfs_dat_move(struct inode *, __u64, sector_t); 53int nilfs_dat_move(struct inode *, __u64, sector_t);
54ssize_t nilfs_dat_get_vinfo(struct inode *, void *, unsigned, size_t); 54ssize_t nilfs_dat_get_vinfo(struct inode *, void *, unsigned, size_t);
55 55
56int nilfs_dat_read(struct inode *dat, struct nilfs_inode *raw_inode); 56int nilfs_dat_read(struct super_block *sb, size_t entry_size,
57struct inode *nilfs_dat_new(struct the_nilfs *nilfs, size_t entry_size); 57 struct nilfs_inode *raw_inode, struct inode **inodep);
58 58
59#endif /* _NILFS_DAT_H */ 59#endif /* _NILFS_DAT_H */
diff --git a/fs/nilfs2/export.h b/fs/nilfs2/export.h
new file mode 100644
index 000000000000..a71cc412b651
--- /dev/null
+++ b/fs/nilfs2/export.h
@@ -0,0 +1,17 @@
1#ifndef NILFS_EXPORT_H
2#define NILFS_EXPORT_H
3
4#include <linux/exportfs.h>
5
6extern const struct export_operations nilfs_export_ops;
7
8struct nilfs_fid {
9 u64 cno;
10 u64 ino;
11 u32 gen;
12
13 u32 parent_gen;
14 u64 parent_ino;
15} __attribute__ ((packed));
16
17#endif
diff --git a/fs/nilfs2/gcdat.c b/fs/nilfs2/gcdat.c
deleted file mode 100644
index 84a45d1d5464..000000000000
--- a/fs/nilfs2/gcdat.c
+++ /dev/null
@@ -1,87 +0,0 @@
1/*
2 * gcdat.c - NILFS shadow DAT inode for GC
3 *
4 * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
19 *
20 * Written by Seiji Kihara <kihara@osrg.net>, Amagai Yoshiji <amagai@osrg.net>,
21 * and Ryusuke Konishi <ryusuke@osrg.net>.
22 *
23 */
24
25#include <linux/buffer_head.h>
26#include "nilfs.h"
27#include "page.h"
28#include "mdt.h"
29
30int nilfs_init_gcdat_inode(struct the_nilfs *nilfs)
31{
32 struct inode *dat = nilfs->ns_dat, *gcdat = nilfs->ns_gc_dat;
33 struct nilfs_inode_info *dii = NILFS_I(dat), *gii = NILFS_I(gcdat);
34 int err;
35
36 gcdat->i_state = 0;
37 gcdat->i_blocks = dat->i_blocks;
38 gii->i_flags = dii->i_flags;
39 gii->i_state = dii->i_state | (1 << NILFS_I_GCDAT);
40 gii->i_cno = 0;
41 nilfs_bmap_init_gcdat(gii->i_bmap, dii->i_bmap);
42 err = nilfs_copy_dirty_pages(gcdat->i_mapping, dat->i_mapping);
43 if (unlikely(err))
44 return err;
45
46 return nilfs_copy_dirty_pages(&gii->i_btnode_cache,
47 &dii->i_btnode_cache);
48}
49
50void nilfs_commit_gcdat_inode(struct the_nilfs *nilfs)
51{
52 struct inode *dat = nilfs->ns_dat, *gcdat = nilfs->ns_gc_dat;
53 struct nilfs_inode_info *dii = NILFS_I(dat), *gii = NILFS_I(gcdat);
54 struct address_space *mapping = dat->i_mapping;
55 struct address_space *gmapping = gcdat->i_mapping;
56
57 down_write(&NILFS_MDT(dat)->mi_sem);
58 dat->i_blocks = gcdat->i_blocks;
59 dii->i_flags = gii->i_flags;
60 dii->i_state = gii->i_state & ~(1 << NILFS_I_GCDAT);
61
62 nilfs_bmap_commit_gcdat(gii->i_bmap, dii->i_bmap);
63
64 nilfs_palloc_clear_cache(dat);
65 nilfs_palloc_clear_cache(gcdat);
66 nilfs_clear_dirty_pages(mapping);
67 nilfs_copy_back_pages(mapping, gmapping);
68 /* note: mdt dirty flags should be cleared by segctor. */
69
70 nilfs_clear_dirty_pages(&dii->i_btnode_cache);
71 nilfs_copy_back_pages(&dii->i_btnode_cache, &gii->i_btnode_cache);
72
73 up_write(&NILFS_MDT(dat)->mi_sem);
74}
75
76void nilfs_clear_gcdat_inode(struct the_nilfs *nilfs)
77{
78 struct inode *gcdat = nilfs->ns_gc_dat;
79 struct nilfs_inode_info *gii = NILFS_I(gcdat);
80
81 gcdat->i_state = I_FREEING | I_CLEAR;
82 gii->i_flags = 0;
83
84 nilfs_palloc_clear_cache(gcdat);
85 truncate_inode_pages(gcdat->i_mapping, 0);
86 truncate_inode_pages(&gii->i_btnode_cache, 0);
87}
diff --git a/fs/nilfs2/gcinode.c b/fs/nilfs2/gcinode.c
index bed3a783129b..33ad25ddd5c4 100644
--- a/fs/nilfs2/gcinode.c
+++ b/fs/nilfs2/gcinode.c
@@ -28,13 +28,6 @@
28 * gcinodes), and this file provides lookup function of the dummy 28 * gcinodes), and this file provides lookup function of the dummy
29 * inodes and their buffer read function. 29 * inodes and their buffer read function.
30 * 30 *
31 * Since NILFS2 keeps up multiple checkpoints/snapshots across GC, it
32 * has to treat blocks that belong to a same file but have different
33 * checkpoint numbers. To avoid interference among generations, dummy
34 * inodes are managed separately from actual inodes, and their lookup
35 * function (nilfs_gc_iget) is designed to be specified with a
36 * checkpoint number argument as well as an inode number.
37 *
38 * Buffers and pages held by the dummy inodes will be released each 31 * Buffers and pages held by the dummy inodes will be released each
39 * time after they are copied to a new log. Dirty blocks made on the 32 * time after they are copied to a new log. Dirty blocks made on the
40 * current generation and the blocks to be moved by GC never overlap 33 * current generation and the blocks to be moved by GC never overlap
@@ -175,125 +168,46 @@ int nilfs_gccache_wait_and_mark_dirty(struct buffer_head *bh)
175 } 168 }
176 nilfs_btnode_mark_dirty(bh); 169 nilfs_btnode_mark_dirty(bh);
177 } else { 170 } else {
178 nilfs_mdt_mark_buffer_dirty(bh); 171 nilfs_mark_buffer_dirty(bh);
179 } 172 }
180 return 0; 173 return 0;
181} 174}
182 175
183/* 176int nilfs_init_gcinode(struct inode *inode)
184 * nilfs_init_gccache() - allocate and initialize gc_inode hash table
185 * @nilfs - the_nilfs
186 *
187 * Return Value: On success, 0.
188 * On error, a negative error code is returned.
189 */
190int nilfs_init_gccache(struct the_nilfs *nilfs)
191{ 177{
192 int loop; 178 struct nilfs_inode_info *ii = NILFS_I(inode);
193 179 struct the_nilfs *nilfs = NILFS_SB(inode->i_sb)->s_nilfs;
194 BUG_ON(nilfs->ns_gc_inodes_h);
195
196 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
197
198 nilfs->ns_gc_inodes_h =
199 kmalloc(sizeof(struct hlist_head) * NILFS_GCINODE_HASH_SIZE,
200 GFP_NOFS);
201 if (nilfs->ns_gc_inodes_h == NULL)
202 return -ENOMEM;
203
204 for (loop = 0; loop < NILFS_GCINODE_HASH_SIZE; loop++)
205 INIT_HLIST_HEAD(&nilfs->ns_gc_inodes_h[loop]);
206 return 0;
207}
208
209/*
210 * nilfs_destroy_gccache() - free gc_inode hash table
211 * @nilfs - the nilfs
212 */
213void nilfs_destroy_gccache(struct the_nilfs *nilfs)
214{
215 if (nilfs->ns_gc_inodes_h) {
216 nilfs_remove_all_gcinode(nilfs);
217 kfree(nilfs->ns_gc_inodes_h);
218 nilfs->ns_gc_inodes_h = NULL;
219 }
220}
221
222static struct inode *alloc_gcinode(struct the_nilfs *nilfs, ino_t ino,
223 __u64 cno)
224{
225 struct inode *inode;
226 struct nilfs_inode_info *ii;
227
228 inode = nilfs_mdt_new_common(nilfs, NULL, ino, GFP_NOFS, 0);
229 if (!inode)
230 return NULL;
231 180
232 inode->i_op = NULL; 181 inode->i_mode = S_IFREG;
233 inode->i_fop = NULL; 182 mapping_set_gfp_mask(inode->i_mapping, GFP_NOFS);
234 inode->i_mapping->a_ops = &def_gcinode_aops; 183 inode->i_mapping->a_ops = &def_gcinode_aops;
184 inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
235 185
236 ii = NILFS_I(inode);
237 ii->i_cno = cno;
238 ii->i_flags = 0; 186 ii->i_flags = 0;
239 ii->i_state = 1 << NILFS_I_GCINODE;
240 ii->i_bh = NULL;
241 nilfs_bmap_init_gc(ii->i_bmap); 187 nilfs_bmap_init_gc(ii->i_bmap);
242 188
243 return inode; 189 /*
244} 190 * Add the inode to GC inode list. Garbage Collection
245 191 * is serialized and no two processes manipulate the
246static unsigned long ihash(ino_t ino, __u64 cno) 192 * list simultaneously.
247{ 193 */
248 return hash_long((unsigned long)((ino << 2) + cno), 194 igrab(inode);
249 NILFS_GCINODE_HASH_BITS); 195 list_add(&NILFS_I(inode)->i_dirty, &nilfs->ns_gc_inodes);
250}
251
252/*
253 * nilfs_gc_iget() - find or create gc inode with specified (ino,cno)
254 */
255struct inode *nilfs_gc_iget(struct the_nilfs *nilfs, ino_t ino, __u64 cno)
256{
257 struct hlist_head *head = nilfs->ns_gc_inodes_h + ihash(ino, cno);
258 struct hlist_node *node;
259 struct inode *inode;
260
261 hlist_for_each_entry(inode, node, head, i_hash) {
262 if (inode->i_ino == ino && NILFS_I(inode)->i_cno == cno)
263 return inode;
264 }
265 196
266 inode = alloc_gcinode(nilfs, ino, cno); 197 return 0;
267 if (likely(inode)) {
268 hlist_add_head(&inode->i_hash, head);
269 list_add(&NILFS_I(inode)->i_dirty, &nilfs->ns_gc_inodes);
270 }
271 return inode;
272}
273
274/*
275 * nilfs_clear_gcinode() - clear and free a gc inode
276 */
277void nilfs_clear_gcinode(struct inode *inode)
278{
279 nilfs_mdt_destroy(inode);
280} 198}
281 199
282/* 200/**
283 * nilfs_remove_all_gcinode() - remove all inodes from the_nilfs 201 * nilfs_remove_all_gcinodes() - remove all unprocessed gc inodes
284 */ 202 */
285void nilfs_remove_all_gcinode(struct the_nilfs *nilfs) 203void nilfs_remove_all_gcinodes(struct the_nilfs *nilfs)
286{ 204{
287 struct hlist_head *head = nilfs->ns_gc_inodes_h; 205 struct list_head *head = &nilfs->ns_gc_inodes;
288 struct hlist_node *node, *n; 206 struct nilfs_inode_info *ii;
289 struct inode *inode;
290 int loop;
291 207
292 for (loop = 0; loop < NILFS_GCINODE_HASH_SIZE; loop++, head++) { 208 while (!list_empty(head)) {
293 hlist_for_each_entry_safe(inode, node, n, head, i_hash) { 209 ii = list_first_entry(head, struct nilfs_inode_info, i_dirty);
294 hlist_del_init(&inode->i_hash); 210 list_del_init(&ii->i_dirty);
295 list_del_init(&NILFS_I(inode)->i_dirty); 211 iput(&ii->vfs_inode);
296 nilfs_clear_gcinode(inode); /* might sleep */
297 }
298 } 212 }
299} 213}
diff --git a/fs/nilfs2/ifile.c b/fs/nilfs2/ifile.c
index 922d9dd42c8f..9f8a2da67f90 100644
--- a/fs/nilfs2/ifile.c
+++ b/fs/nilfs2/ifile.c
@@ -161,25 +161,46 @@ int nilfs_ifile_get_inode_block(struct inode *ifile, ino_t ino,
161} 161}
162 162
163/** 163/**
164 * nilfs_ifile_new - create inode file 164 * nilfs_ifile_read - read or get ifile inode
165 * @sbi: nilfs_sb_info struct 165 * @sb: super block instance
166 * @root: root object
166 * @inode_size: size of an inode 167 * @inode_size: size of an inode
168 * @raw_inode: on-disk ifile inode
169 * @inodep: buffer to store the inode
167 */ 170 */
168struct inode *nilfs_ifile_new(struct nilfs_sb_info *sbi, size_t inode_size) 171int nilfs_ifile_read(struct super_block *sb, struct nilfs_root *root,
172 size_t inode_size, struct nilfs_inode *raw_inode,
173 struct inode **inodep)
169{ 174{
170 struct inode *ifile; 175 struct inode *ifile;
171 int err; 176 int err;
172 177
173 ifile = nilfs_mdt_new(sbi->s_nilfs, sbi->s_super, NILFS_IFILE_INO, 178 ifile = nilfs_iget_locked(sb, root, NILFS_IFILE_INO);
174 sizeof(struct nilfs_ifile_info)); 179 if (unlikely(!ifile))
175 if (ifile) { 180 return -ENOMEM;
176 err = nilfs_palloc_init_blockgroup(ifile, inode_size); 181 if (!(ifile->i_state & I_NEW))
177 if (unlikely(err)) { 182 goto out;
178 nilfs_mdt_destroy(ifile); 183
179 return NULL; 184 err = nilfs_mdt_init(ifile, NILFS_MDT_GFP,
180 } 185 sizeof(struct nilfs_ifile_info));
181 nilfs_palloc_setup_cache(ifile, 186 if (err)
182 &NILFS_IFILE_I(ifile)->palloc_cache); 187 goto failed;
183 } 188
184 return ifile; 189 err = nilfs_palloc_init_blockgroup(ifile, inode_size);
190 if (err)
191 goto failed;
192
193 nilfs_palloc_setup_cache(ifile, &NILFS_IFILE_I(ifile)->palloc_cache);
194
195 err = nilfs_read_inode_common(ifile, raw_inode);
196 if (err)
197 goto failed;
198
199 unlock_new_inode(ifile);
200 out:
201 *inodep = ifile;
202 return 0;
203 failed:
204 iget_failed(ifile);
205 return err;
185} 206}
diff --git a/fs/nilfs2/ifile.h b/fs/nilfs2/ifile.h
index cbca32e498f2..59b6f2b51df6 100644
--- a/fs/nilfs2/ifile.h
+++ b/fs/nilfs2/ifile.h
@@ -49,6 +49,8 @@ int nilfs_ifile_create_inode(struct inode *, ino_t *, struct buffer_head **);
49int nilfs_ifile_delete_inode(struct inode *, ino_t); 49int nilfs_ifile_delete_inode(struct inode *, ino_t);
50int nilfs_ifile_get_inode_block(struct inode *, ino_t, struct buffer_head **); 50int nilfs_ifile_get_inode_block(struct inode *, ino_t, struct buffer_head **);
51 51
52struct inode *nilfs_ifile_new(struct nilfs_sb_info *sbi, size_t inode_size); 52int nilfs_ifile_read(struct super_block *sb, struct nilfs_root *root,
53 size_t inode_size, struct nilfs_inode *raw_inode,
54 struct inode **inodep);
53 55
54#endif /* _NILFS_IFILE_H */ 56#endif /* _NILFS_IFILE_H */
diff --git a/fs/nilfs2/inode.c b/fs/nilfs2/inode.c
index eccb2f2e2315..71d4bc8464e0 100644
--- a/fs/nilfs2/inode.c
+++ b/fs/nilfs2/inode.c
@@ -34,6 +34,12 @@
34#include "cpfile.h" 34#include "cpfile.h"
35#include "ifile.h" 35#include "ifile.h"
36 36
37struct nilfs_iget_args {
38 u64 ino;
39 __u64 cno;
40 struct nilfs_root *root;
41 int for_gc;
42};
37 43
38/** 44/**
39 * nilfs_get_block() - get a file block on the filesystem (callback function) 45 * nilfs_get_block() - get a file block on the filesystem (callback function)
@@ -279,6 +285,7 @@ struct inode *nilfs_new_inode(struct inode *dir, int mode)
279 struct nilfs_sb_info *sbi = NILFS_SB(sb); 285 struct nilfs_sb_info *sbi = NILFS_SB(sb);
280 struct inode *inode; 286 struct inode *inode;
281 struct nilfs_inode_info *ii; 287 struct nilfs_inode_info *ii;
288 struct nilfs_root *root;
282 int err = -ENOMEM; 289 int err = -ENOMEM;
283 ino_t ino; 290 ino_t ino;
284 291
@@ -289,15 +296,17 @@ struct inode *nilfs_new_inode(struct inode *dir, int mode)
289 mapping_set_gfp_mask(inode->i_mapping, 296 mapping_set_gfp_mask(inode->i_mapping,
290 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS); 297 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
291 298
299 root = NILFS_I(dir)->i_root;
292 ii = NILFS_I(inode); 300 ii = NILFS_I(inode);
293 ii->i_state = 1 << NILFS_I_NEW; 301 ii->i_state = 1 << NILFS_I_NEW;
302 ii->i_root = root;
294 303
295 err = nilfs_ifile_create_inode(sbi->s_ifile, &ino, &ii->i_bh); 304 err = nilfs_ifile_create_inode(root->ifile, &ino, &ii->i_bh);
296 if (unlikely(err)) 305 if (unlikely(err))
297 goto failed_ifile_create_inode; 306 goto failed_ifile_create_inode;
298 /* reference count of i_bh inherits from nilfs_mdt_read_block() */ 307 /* reference count of i_bh inherits from nilfs_mdt_read_block() */
299 308
300 atomic_inc(&sbi->s_inodes_count); 309 atomic_inc(&root->inodes_count);
301 inode_init_owner(inode, dir, mode); 310 inode_init_owner(inode, dir, mode);
302 inode->i_ino = ino; 311 inode->i_ino = ino;
303 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME; 312 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
@@ -320,7 +329,6 @@ struct inode *nilfs_new_inode(struct inode *dir, int mode)
320 /* ii->i_file_acl = 0; */ 329 /* ii->i_file_acl = 0; */
321 /* ii->i_dir_acl = 0; */ 330 /* ii->i_dir_acl = 0; */
322 ii->i_dir_start_lookup = 0; 331 ii->i_dir_start_lookup = 0;
323 ii->i_cno = 0;
324 nilfs_set_inode_flags(inode); 332 nilfs_set_inode_flags(inode);
325 spin_lock(&sbi->s_next_gen_lock); 333 spin_lock(&sbi->s_next_gen_lock);
326 inode->i_generation = sbi->s_next_generation++; 334 inode->i_generation = sbi->s_next_generation++;
@@ -350,16 +358,6 @@ struct inode *nilfs_new_inode(struct inode *dir, int mode)
350 return ERR_PTR(err); 358 return ERR_PTR(err);
351} 359}
352 360
353void nilfs_free_inode(struct inode *inode)
354{
355 struct super_block *sb = inode->i_sb;
356 struct nilfs_sb_info *sbi = NILFS_SB(sb);
357
358 /* XXX: check error code? Is there any thing I can do? */
359 (void) nilfs_ifile_delete_inode(sbi->s_ifile, inode->i_ino);
360 atomic_dec(&sbi->s_inodes_count);
361}
362
363void nilfs_set_inode_flags(struct inode *inode) 361void nilfs_set_inode_flags(struct inode *inode)
364{ 362{
365 unsigned int flags = NILFS_I(inode)->i_flags; 363 unsigned int flags = NILFS_I(inode)->i_flags;
@@ -410,7 +408,6 @@ int nilfs_read_inode_common(struct inode *inode,
410 0 : le32_to_cpu(raw_inode->i_dir_acl); 408 0 : le32_to_cpu(raw_inode->i_dir_acl);
411#endif 409#endif
412 ii->i_dir_start_lookup = 0; 410 ii->i_dir_start_lookup = 0;
413 ii->i_cno = 0;
414 inode->i_generation = le32_to_cpu(raw_inode->i_generation); 411 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
415 412
416 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) || 413 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
@@ -424,7 +421,8 @@ int nilfs_read_inode_common(struct inode *inode,
424 return 0; 421 return 0;
425} 422}
426 423
427static int __nilfs_read_inode(struct super_block *sb, unsigned long ino, 424static int __nilfs_read_inode(struct super_block *sb,
425 struct nilfs_root *root, unsigned long ino,
428 struct inode *inode) 426 struct inode *inode)
429{ 427{
430 struct nilfs_sb_info *sbi = NILFS_SB(sb); 428 struct nilfs_sb_info *sbi = NILFS_SB(sb);
@@ -434,11 +432,11 @@ static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
434 int err; 432 int err;
435 433
436 down_read(&NILFS_MDT(dat)->mi_sem); /* XXX */ 434 down_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
437 err = nilfs_ifile_get_inode_block(sbi->s_ifile, ino, &bh); 435 err = nilfs_ifile_get_inode_block(root->ifile, ino, &bh);
438 if (unlikely(err)) 436 if (unlikely(err))
439 goto bad_inode; 437 goto bad_inode;
440 438
441 raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, bh); 439 raw_inode = nilfs_ifile_map_inode(root->ifile, ino, bh);
442 440
443 err = nilfs_read_inode_common(inode, raw_inode); 441 err = nilfs_read_inode_common(inode, raw_inode);
444 if (err) 442 if (err)
@@ -461,14 +459,14 @@ static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
461 inode, inode->i_mode, 459 inode, inode->i_mode,
462 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code))); 460 huge_decode_dev(le64_to_cpu(raw_inode->i_device_code)));
463 } 461 }
464 nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh); 462 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
465 brelse(bh); 463 brelse(bh);
466 up_read(&NILFS_MDT(dat)->mi_sem); /* XXX */ 464 up_read(&NILFS_MDT(dat)->mi_sem); /* XXX */
467 nilfs_set_inode_flags(inode); 465 nilfs_set_inode_flags(inode);
468 return 0; 466 return 0;
469 467
470 failed_unmap: 468 failed_unmap:
471 nilfs_ifile_unmap_inode(sbi->s_ifile, ino, bh); 469 nilfs_ifile_unmap_inode(root->ifile, ino, bh);
472 brelse(bh); 470 brelse(bh);
473 471
474 bad_inode: 472 bad_inode:
@@ -476,18 +474,95 @@ static int __nilfs_read_inode(struct super_block *sb, unsigned long ino,
476 return err; 474 return err;
477} 475}
478 476
479struct inode *nilfs_iget(struct super_block *sb, unsigned long ino) 477static int nilfs_iget_test(struct inode *inode, void *opaque)
478{
479 struct nilfs_iget_args *args = opaque;
480 struct nilfs_inode_info *ii;
481
482 if (args->ino != inode->i_ino || args->root != NILFS_I(inode)->i_root)
483 return 0;
484
485 ii = NILFS_I(inode);
486 if (!test_bit(NILFS_I_GCINODE, &ii->i_state))
487 return !args->for_gc;
488
489 return args->for_gc && args->cno == ii->i_cno;
490}
491
492static int nilfs_iget_set(struct inode *inode, void *opaque)
493{
494 struct nilfs_iget_args *args = opaque;
495
496 inode->i_ino = args->ino;
497 if (args->for_gc) {
498 NILFS_I(inode)->i_state = 1 << NILFS_I_GCINODE;
499 NILFS_I(inode)->i_cno = args->cno;
500 NILFS_I(inode)->i_root = NULL;
501 } else {
502 if (args->root && args->ino == NILFS_ROOT_INO)
503 nilfs_get_root(args->root);
504 NILFS_I(inode)->i_root = args->root;
505 }
506 return 0;
507}
508
509struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
510 unsigned long ino)
511{
512 struct nilfs_iget_args args = {
513 .ino = ino, .root = root, .cno = 0, .for_gc = 0
514 };
515
516 return ilookup5(sb, ino, nilfs_iget_test, &args);
517}
518
519struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
520 unsigned long ino)
521{
522 struct nilfs_iget_args args = {
523 .ino = ino, .root = root, .cno = 0, .for_gc = 0
524 };
525
526 return iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
527}
528
529struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
530 unsigned long ino)
480{ 531{
481 struct inode *inode; 532 struct inode *inode;
482 int err; 533 int err;
483 534
484 inode = iget_locked(sb, ino); 535 inode = nilfs_iget_locked(sb, root, ino);
485 if (unlikely(!inode)) 536 if (unlikely(!inode))
486 return ERR_PTR(-ENOMEM); 537 return ERR_PTR(-ENOMEM);
487 if (!(inode->i_state & I_NEW)) 538 if (!(inode->i_state & I_NEW))
488 return inode; 539 return inode;
489 540
490 err = __nilfs_read_inode(sb, ino, inode); 541 err = __nilfs_read_inode(sb, root, ino, inode);
542 if (unlikely(err)) {
543 iget_failed(inode);
544 return ERR_PTR(err);
545 }
546 unlock_new_inode(inode);
547 return inode;
548}
549
550struct inode *nilfs_iget_for_gc(struct super_block *sb, unsigned long ino,
551 __u64 cno)
552{
553 struct nilfs_iget_args args = {
554 .ino = ino, .root = NULL, .cno = cno, .for_gc = 1
555 };
556 struct inode *inode;
557 int err;
558
559 inode = iget5_locked(sb, ino, nilfs_iget_test, nilfs_iget_set, &args);
560 if (unlikely(!inode))
561 return ERR_PTR(-ENOMEM);
562 if (!(inode->i_state & I_NEW))
563 return inode;
564
565 err = nilfs_init_gcinode(inode);
491 if (unlikely(err)) { 566 if (unlikely(err)) {
492 iget_failed(inode); 567 iget_failed(inode);
493 return ERR_PTR(err); 568 return ERR_PTR(err);
@@ -528,21 +603,20 @@ void nilfs_update_inode(struct inode *inode, struct buffer_head *ibh)
528{ 603{
529 ino_t ino = inode->i_ino; 604 ino_t ino = inode->i_ino;
530 struct nilfs_inode_info *ii = NILFS_I(inode); 605 struct nilfs_inode_info *ii = NILFS_I(inode);
531 struct super_block *sb = inode->i_sb; 606 struct inode *ifile = ii->i_root->ifile;
532 struct nilfs_sb_info *sbi = NILFS_SB(sb);
533 struct nilfs_inode *raw_inode; 607 struct nilfs_inode *raw_inode;
534 608
535 raw_inode = nilfs_ifile_map_inode(sbi->s_ifile, ino, ibh); 609 raw_inode = nilfs_ifile_map_inode(ifile, ino, ibh);
536 610
537 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state)) 611 if (test_and_clear_bit(NILFS_I_NEW, &ii->i_state))
538 memset(raw_inode, 0, NILFS_MDT(sbi->s_ifile)->mi_entry_size); 612 memset(raw_inode, 0, NILFS_MDT(ifile)->mi_entry_size);
539 set_bit(NILFS_I_INODE_DIRTY, &ii->i_state); 613 set_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
540 614
541 nilfs_write_inode_common(inode, raw_inode, 0); 615 nilfs_write_inode_common(inode, raw_inode, 0);
542 /* XXX: call with has_bmap = 0 is a workaround to avoid 616 /* XXX: call with has_bmap = 0 is a workaround to avoid
543 deadlock of bmap. This delays update of i_bmap to just 617 deadlock of bmap. This delays update of i_bmap to just
544 before writing */ 618 before writing */
545 nilfs_ifile_unmap_inode(sbi->s_ifile, ino, ibh); 619 nilfs_ifile_unmap_inode(ifile, ino, ibh);
546} 620}
547 621
548#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */ 622#define NILFS_MAX_TRUNCATE_BLOCKS 16384 /* 64MB for 4KB block */
@@ -617,6 +691,7 @@ void nilfs_truncate(struct inode *inode)
617static void nilfs_clear_inode(struct inode *inode) 691static void nilfs_clear_inode(struct inode *inode)
618{ 692{
619 struct nilfs_inode_info *ii = NILFS_I(inode); 693 struct nilfs_inode_info *ii = NILFS_I(inode);
694 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
620 695
621 /* 696 /*
622 * Free resources allocated in nilfs_read_inode(), here. 697 * Free resources allocated in nilfs_read_inode(), here.
@@ -625,10 +700,16 @@ static void nilfs_clear_inode(struct inode *inode)
625 brelse(ii->i_bh); 700 brelse(ii->i_bh);
626 ii->i_bh = NULL; 701 ii->i_bh = NULL;
627 702
703 if (mdi && mdi->mi_palloc_cache)
704 nilfs_palloc_destroy_cache(inode);
705
628 if (test_bit(NILFS_I_BMAP, &ii->i_state)) 706 if (test_bit(NILFS_I_BMAP, &ii->i_state))
629 nilfs_bmap_clear(ii->i_bmap); 707 nilfs_bmap_clear(ii->i_bmap);
630 708
631 nilfs_btnode_cache_clear(&ii->i_btnode_cache); 709 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
710
711 if (ii->i_root && inode->i_ino == NILFS_ROOT_INO)
712 nilfs_put_root(ii->i_root);
632} 713}
633 714
634void nilfs_evict_inode(struct inode *inode) 715void nilfs_evict_inode(struct inode *inode)
@@ -637,7 +718,7 @@ void nilfs_evict_inode(struct inode *inode)
637 struct super_block *sb = inode->i_sb; 718 struct super_block *sb = inode->i_sb;
638 struct nilfs_inode_info *ii = NILFS_I(inode); 719 struct nilfs_inode_info *ii = NILFS_I(inode);
639 720
640 if (inode->i_nlink || unlikely(is_bad_inode(inode))) { 721 if (inode->i_nlink || !ii->i_root || unlikely(is_bad_inode(inode))) {
641 if (inode->i_data.nrpages) 722 if (inode->i_data.nrpages)
642 truncate_inode_pages(&inode->i_data, 0); 723 truncate_inode_pages(&inode->i_data, 0);
643 end_writeback(inode); 724 end_writeback(inode);
@@ -649,12 +730,16 @@ void nilfs_evict_inode(struct inode *inode)
649 if (inode->i_data.nrpages) 730 if (inode->i_data.nrpages)
650 truncate_inode_pages(&inode->i_data, 0); 731 truncate_inode_pages(&inode->i_data, 0);
651 732
733 /* TODO: some of the following operations may fail. */
652 nilfs_truncate_bmap(ii, 0); 734 nilfs_truncate_bmap(ii, 0);
653 nilfs_mark_inode_dirty(inode); 735 nilfs_mark_inode_dirty(inode);
654 end_writeback(inode); 736 end_writeback(inode);
737
738 nilfs_ifile_delete_inode(ii->i_root->ifile, inode->i_ino);
739 atomic_dec(&ii->i_root->inodes_count);
740
655 nilfs_clear_inode(inode); 741 nilfs_clear_inode(inode);
656 nilfs_free_inode(inode); 742
657 /* nilfs_free_inode() marks inode buffer dirty */
658 if (IS_SYNC(inode)) 743 if (IS_SYNC(inode))
659 nilfs_set_transaction_flag(NILFS_TI_SYNC); 744 nilfs_set_transaction_flag(NILFS_TI_SYNC);
660 nilfs_transaction_commit(sb); 745 nilfs_transaction_commit(sb);
@@ -700,6 +785,17 @@ out_err:
700 return err; 785 return err;
701} 786}
702 787
788int nilfs_permission(struct inode *inode, int mask)
789{
790 struct nilfs_root *root = NILFS_I(inode)->i_root;
791
792 if ((mask & MAY_WRITE) && root &&
793 root->cno != NILFS_CPTREE_CURRENT_CNO)
794 return -EROFS; /* snapshot is not writable */
795
796 return generic_permission(inode, mask, NULL);
797}
798
703int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode, 799int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
704 struct buffer_head **pbh) 800 struct buffer_head **pbh)
705{ 801{
@@ -709,8 +805,8 @@ int nilfs_load_inode_block(struct nilfs_sb_info *sbi, struct inode *inode,
709 spin_lock(&sbi->s_inode_lock); 805 spin_lock(&sbi->s_inode_lock);
710 if (ii->i_bh == NULL) { 806 if (ii->i_bh == NULL) {
711 spin_unlock(&sbi->s_inode_lock); 807 spin_unlock(&sbi->s_inode_lock);
712 err = nilfs_ifile_get_inode_block(sbi->s_ifile, inode->i_ino, 808 err = nilfs_ifile_get_inode_block(ii->i_root->ifile,
713 pbh); 809 inode->i_ino, pbh);
714 if (unlikely(err)) 810 if (unlikely(err))
715 return err; 811 return err;
716 spin_lock(&sbi->s_inode_lock); 812 spin_lock(&sbi->s_inode_lock);
@@ -790,7 +886,7 @@ int nilfs_mark_inode_dirty(struct inode *inode)
790 } 886 }
791 nilfs_update_inode(inode, ibh); 887 nilfs_update_inode(inode, ibh);
792 nilfs_mdt_mark_buffer_dirty(ibh); 888 nilfs_mdt_mark_buffer_dirty(ibh);
793 nilfs_mdt_mark_dirty(sbi->s_ifile); 889 nilfs_mdt_mark_dirty(NILFS_I(inode)->i_root->ifile);
794 brelse(ibh); 890 brelse(ibh);
795 return 0; 891 return 0;
796} 892}
@@ -808,6 +904,7 @@ int nilfs_mark_inode_dirty(struct inode *inode)
808void nilfs_dirty_inode(struct inode *inode) 904void nilfs_dirty_inode(struct inode *inode)
809{ 905{
810 struct nilfs_transaction_info ti; 906 struct nilfs_transaction_info ti;
907 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
811 908
812 if (is_bad_inode(inode)) { 909 if (is_bad_inode(inode)) {
813 nilfs_warning(inode->i_sb, __func__, 910 nilfs_warning(inode->i_sb, __func__,
@@ -815,6 +912,10 @@ void nilfs_dirty_inode(struct inode *inode)
815 dump_stack(); 912 dump_stack();
816 return; 913 return;
817 } 914 }
915 if (mdi) {
916 nilfs_mdt_mark_dirty(inode);
917 return;
918 }
818 nilfs_transaction_begin(inode->i_sb, &ti, 0); 919 nilfs_transaction_begin(inode->i_sb, &ti, 0);
819 nilfs_mark_inode_dirty(inode); 920 nilfs_mark_inode_dirty(inode);
820 nilfs_transaction_commit(inode->i_sb); /* never fails */ 921 nilfs_transaction_commit(inode->i_sb); /* never fails */
diff --git a/fs/nilfs2/ioctl.c b/fs/nilfs2/ioctl.c
index 0442ee3b394f..3e90f86d5bfe 100644
--- a/fs/nilfs2/ioctl.c
+++ b/fs/nilfs2/ioctl.c
@@ -117,7 +117,7 @@ static int nilfs_ioctl_change_cpmode(struct inode *inode, struct file *filp,
117 if (copy_from_user(&cpmode, argp, sizeof(cpmode))) 117 if (copy_from_user(&cpmode, argp, sizeof(cpmode)))
118 goto out; 118 goto out;
119 119
120 mutex_lock(&nilfs->ns_mount_mutex); 120 down_read(&inode->i_sb->s_umount);
121 121
122 nilfs_transaction_begin(inode->i_sb, &ti, 0); 122 nilfs_transaction_begin(inode->i_sb, &ti, 0);
123 ret = nilfs_cpfile_change_cpmode( 123 ret = nilfs_cpfile_change_cpmode(
@@ -127,7 +127,7 @@ static int nilfs_ioctl_change_cpmode(struct inode *inode, struct file *filp,
127 else 127 else
128 nilfs_transaction_commit(inode->i_sb); /* never fails */ 128 nilfs_transaction_commit(inode->i_sb); /* never fails */
129 129
130 mutex_unlock(&nilfs->ns_mount_mutex); 130 up_read(&inode->i_sb->s_umount);
131out: 131out:
132 mnt_drop_write(filp->f_path.mnt); 132 mnt_drop_write(filp->f_path.mnt);
133 return ret; 133 return ret;
@@ -333,7 +333,7 @@ static int nilfs_ioctl_move_inode_block(struct inode *inode,
333 return 0; 333 return 0;
334} 334}
335 335
336static int nilfs_ioctl_move_blocks(struct the_nilfs *nilfs, 336static int nilfs_ioctl_move_blocks(struct super_block *sb,
337 struct nilfs_argv *argv, void *buf) 337 struct nilfs_argv *argv, void *buf)
338{ 338{
339 size_t nmembs = argv->v_nmembs; 339 size_t nmembs = argv->v_nmembs;
@@ -348,7 +348,7 @@ static int nilfs_ioctl_move_blocks(struct the_nilfs *nilfs,
348 for (i = 0, vdesc = buf; i < nmembs; ) { 348 for (i = 0, vdesc = buf; i < nmembs; ) {
349 ino = vdesc->vd_ino; 349 ino = vdesc->vd_ino;
350 cno = vdesc->vd_cno; 350 cno = vdesc->vd_cno;
351 inode = nilfs_gc_iget(nilfs, ino, cno); 351 inode = nilfs_iget_for_gc(sb, ino, cno);
352 if (unlikely(inode == NULL)) { 352 if (unlikely(inode == NULL)) {
353 ret = -ENOMEM; 353 ret = -ENOMEM;
354 goto failed; 354 goto failed;
@@ -356,11 +356,15 @@ static int nilfs_ioctl_move_blocks(struct the_nilfs *nilfs,
356 do { 356 do {
357 ret = nilfs_ioctl_move_inode_block(inode, vdesc, 357 ret = nilfs_ioctl_move_inode_block(inode, vdesc,
358 &buffers); 358 &buffers);
359 if (unlikely(ret < 0)) 359 if (unlikely(ret < 0)) {
360 iput(inode);
360 goto failed; 361 goto failed;
362 }
361 vdesc++; 363 vdesc++;
362 } while (++i < nmembs && 364 } while (++i < nmembs &&
363 vdesc->vd_ino == ino && vdesc->vd_cno == cno); 365 vdesc->vd_ino == ino && vdesc->vd_cno == cno);
366
367 iput(inode); /* The inode still remains in GC inode list */
364 } 368 }
365 369
366 list_for_each_entry_safe(bh, n, &buffers, b_assoc_buffers) { 370 list_for_each_entry_safe(bh, n, &buffers, b_assoc_buffers) {
@@ -566,7 +570,7 @@ static int nilfs_ioctl_clean_segments(struct inode *inode, struct file *filp,
566 } 570 }
567 571
568 /* 572 /*
569 * nilfs_ioctl_move_blocks() will call nilfs_gc_iget(), 573 * nilfs_ioctl_move_blocks() will call nilfs_iget_for_gc(),
570 * which will operates an inode list without blocking. 574 * which will operates an inode list without blocking.
571 * To protect the list from concurrent operations, 575 * To protect the list from concurrent operations,
572 * nilfs_ioctl_move_blocks should be atomic operation. 576 * nilfs_ioctl_move_blocks should be atomic operation.
@@ -576,15 +580,16 @@ static int nilfs_ioctl_clean_segments(struct inode *inode, struct file *filp,
576 goto out_free; 580 goto out_free;
577 } 581 }
578 582
579 ret = nilfs_ioctl_move_blocks(nilfs, &argv[0], kbufs[0]); 583 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
584
585 ret = nilfs_ioctl_move_blocks(inode->i_sb, &argv[0], kbufs[0]);
580 if (ret < 0) 586 if (ret < 0)
581 printk(KERN_ERR "NILFS: GC failed during preparation: " 587 printk(KERN_ERR "NILFS: GC failed during preparation: "
582 "cannot read source blocks: err=%d\n", ret); 588 "cannot read source blocks: err=%d\n", ret);
583 else 589 else
584 ret = nilfs_clean_segments(inode->i_sb, argv, kbufs); 590 ret = nilfs_clean_segments(inode->i_sb, argv, kbufs);
585 591
586 if (ret < 0) 592 nilfs_remove_all_gcinodes(nilfs);
587 nilfs_remove_all_gcinode(nilfs);
588 clear_nilfs_gc_running(nilfs); 593 clear_nilfs_gc_running(nilfs);
589 594
590out_free: 595out_free:
diff --git a/fs/nilfs2/mdt.c b/fs/nilfs2/mdt.c
index d01aff4957d9..39a5b84e2c9f 100644
--- a/fs/nilfs2/mdt.c
+++ b/fs/nilfs2/mdt.c
@@ -36,7 +36,6 @@
36 36
37#define NILFS_MDT_MAX_RA_BLOCKS (16 - 1) 37#define NILFS_MDT_MAX_RA_BLOCKS (16 - 1)
38 38
39#define INIT_UNUSED_INODE_FIELDS
40 39
41static int 40static int
42nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block, 41nilfs_mdt_insert_new_block(struct inode *inode, unsigned long block,
@@ -78,25 +77,11 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
78 struct buffer_head *, 77 struct buffer_head *,
79 void *)) 78 void *))
80{ 79{
81 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
82 struct super_block *sb = inode->i_sb; 80 struct super_block *sb = inode->i_sb;
83 struct nilfs_transaction_info ti; 81 struct nilfs_transaction_info ti;
84 struct buffer_head *bh; 82 struct buffer_head *bh;
85 int err; 83 int err;
86 84
87 if (!sb) {
88 /*
89 * Make sure this function is not called from any
90 * read-only context.
91 */
92 if (!nilfs->ns_writer) {
93 WARN_ON(1);
94 err = -EROFS;
95 goto out;
96 }
97 sb = nilfs->ns_writer->s_super;
98 }
99
100 nilfs_transaction_begin(sb, &ti, 0); 85 nilfs_transaction_begin(sb, &ti, 0);
101 86
102 err = -ENOMEM; 87 err = -ENOMEM;
@@ -112,7 +97,7 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
112 if (buffer_uptodate(bh)) 97 if (buffer_uptodate(bh))
113 goto failed_bh; 98 goto failed_bh;
114 99
115 bh->b_bdev = nilfs->ns_bdev; 100 bh->b_bdev = sb->s_bdev;
116 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block); 101 err = nilfs_mdt_insert_new_block(inode, block, bh, init_block);
117 if (likely(!err)) { 102 if (likely(!err)) {
118 get_bh(bh); 103 get_bh(bh);
@@ -129,7 +114,7 @@ static int nilfs_mdt_create_block(struct inode *inode, unsigned long block,
129 err = nilfs_transaction_commit(sb); 114 err = nilfs_transaction_commit(sb);
130 else 115 else
131 nilfs_transaction_abort(sb); 116 nilfs_transaction_abort(sb);
132 out: 117
133 return err; 118 return err;
134} 119}
135 120
@@ -167,9 +152,7 @@ nilfs_mdt_submit_block(struct inode *inode, unsigned long blkoff,
167 unlock_buffer(bh); 152 unlock_buffer(bh);
168 goto failed_bh; 153 goto failed_bh;
169 } 154 }
170 bh->b_bdev = NILFS_MDT(inode)->mi_nilfs->ns_bdev; 155 map_bh(bh, inode->i_sb, (sector_t)blknum);
171 bh->b_blocknr = (sector_t)blknum;
172 set_buffer_mapped(bh);
173 156
174 bh->b_end_io = end_buffer_read_sync; 157 bh->b_end_io = end_buffer_read_sync;
175 get_bh(bh); 158 get_bh(bh);
@@ -398,35 +381,24 @@ int nilfs_mdt_fetch_dirty(struct inode *inode)
398static int 381static int
399nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc) 382nilfs_mdt_write_page(struct page *page, struct writeback_control *wbc)
400{ 383{
401 struct inode *inode = container_of(page->mapping, 384 struct inode *inode;
402 struct inode, i_data); 385 struct super_block *sb;
403 struct super_block *sb = inode->i_sb;
404 struct the_nilfs *nilfs = NILFS_MDT(inode)->mi_nilfs;
405 struct nilfs_sb_info *writer = NULL;
406 int err = 0; 386 int err = 0;
407 387
408 redirty_page_for_writepage(wbc, page); 388 redirty_page_for_writepage(wbc, page);
409 unlock_page(page); 389 unlock_page(page);
410 390
411 if (page->mapping->assoc_mapping) 391 inode = page->mapping->host;
412 return 0; /* Do not request flush for shadow page cache */ 392 if (!inode)
413 if (!sb) { 393 return 0;
414 down_read(&nilfs->ns_writer_sem); 394
415 writer = nilfs->ns_writer; 395 sb = inode->i_sb;
416 if (!writer) {
417 up_read(&nilfs->ns_writer_sem);
418 return -EROFS;
419 }
420 sb = writer->s_super;
421 }
422 396
423 if (wbc->sync_mode == WB_SYNC_ALL) 397 if (wbc->sync_mode == WB_SYNC_ALL)
424 err = nilfs_construct_segment(sb); 398 err = nilfs_construct_segment(sb);
425 else if (wbc->for_reclaim) 399 else if (wbc->for_reclaim)
426 nilfs_flush_segment(sb, inode->i_ino); 400 nilfs_flush_segment(sb, inode->i_ino);
427 401
428 if (writer)
429 up_read(&nilfs->ns_writer_sem);
430 return err; 402 return err;
431} 403}
432 404
@@ -439,105 +411,27 @@ static const struct address_space_operations def_mdt_aops = {
439static const struct inode_operations def_mdt_iops; 411static const struct inode_operations def_mdt_iops;
440static const struct file_operations def_mdt_fops; 412static const struct file_operations def_mdt_fops;
441 413
442/* 414
443 * NILFS2 uses pseudo inodes for meta data files such as DAT, cpfile, sufile, 415int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz)
444 * ifile, or gcinodes. This allows the B-tree code and segment constructor
445 * to treat them like regular files, and this helps to simplify the
446 * implementation.
447 * On the other hand, some of the pseudo inodes have an irregular point:
448 * They don't have valid inode->i_sb pointer because their lifetimes are
449 * longer than those of the super block structs; they may continue for
450 * several consecutive mounts/umounts. This would need discussions.
451 */
452/**
453 * nilfs_mdt_new_common - allocate a pseudo inode for metadata file
454 * @nilfs: nilfs object
455 * @sb: super block instance the metadata file belongs to
456 * @ino: inode number
457 * @gfp_mask: gfp mask for data pages
458 * @objsz: size of the private object attached to inode->i_private
459 */
460struct inode *
461nilfs_mdt_new_common(struct the_nilfs *nilfs, struct super_block *sb,
462 ino_t ino, gfp_t gfp_mask, size_t objsz)
463{ 416{
464 struct inode *inode = nilfs_alloc_inode_common(nilfs); 417 struct nilfs_mdt_info *mi;
465 418
466 if (!inode) 419 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
467 return NULL; 420 if (!mi)
468 else { 421 return -ENOMEM;
469 struct address_space * const mapping = &inode->i_data;
470 struct nilfs_mdt_info *mi;
471
472 mi = kzalloc(max(sizeof(*mi), objsz), GFP_NOFS);
473 if (!mi) {
474 nilfs_destroy_inode(inode);
475 return NULL;
476 }
477 mi->mi_nilfs = nilfs;
478 init_rwsem(&mi->mi_sem);
479
480 inode->i_sb = sb; /* sb may be NULL for some meta data files */
481 inode->i_blkbits = nilfs->ns_blocksize_bits;
482 inode->i_flags = 0;
483 atomic_set(&inode->i_count, 1);
484 inode->i_nlink = 1;
485 inode->i_ino = ino;
486 inode->i_mode = S_IFREG;
487 inode->i_private = mi;
488
489#ifdef INIT_UNUSED_INODE_FIELDS
490 atomic_set(&inode->i_writecount, 0);
491 inode->i_size = 0;
492 inode->i_blocks = 0;
493 inode->i_bytes = 0;
494 inode->i_generation = 0;
495#ifdef CONFIG_QUOTA
496 memset(&inode->i_dquot, 0, sizeof(inode->i_dquot));
497#endif
498 inode->i_pipe = NULL;
499 inode->i_bdev = NULL;
500 inode->i_cdev = NULL;
501 inode->i_rdev = 0;
502#ifdef CONFIG_SECURITY
503 inode->i_security = NULL;
504#endif
505 inode->dirtied_when = 0;
506
507 INIT_LIST_HEAD(&inode->i_list);
508 INIT_LIST_HEAD(&inode->i_sb_list);
509 inode->i_state = 0;
510#endif
511
512 spin_lock_init(&inode->i_lock);
513 mutex_init(&inode->i_mutex);
514 init_rwsem(&inode->i_alloc_sem);
515
516 mapping->host = NULL; /* instead of inode */
517 mapping->flags = 0;
518 mapping_set_gfp_mask(mapping, gfp_mask);
519 mapping->assoc_mapping = NULL;
520 mapping->backing_dev_info = nilfs->ns_bdi;
521
522 inode->i_mapping = mapping;
523 }
524 422
525 return inode; 423 init_rwsem(&mi->mi_sem);
526} 424 inode->i_private = mi;
527 425
528struct inode *nilfs_mdt_new(struct the_nilfs *nilfs, struct super_block *sb, 426 inode->i_mode = S_IFREG;
529 ino_t ino, size_t objsz) 427 mapping_set_gfp_mask(inode->i_mapping, gfp_mask);
530{ 428 inode->i_mapping->backing_dev_info = inode->i_sb->s_bdi;
531 struct inode *inode;
532
533 inode = nilfs_mdt_new_common(nilfs, sb, ino, NILFS_MDT_GFP, objsz);
534 if (!inode)
535 return NULL;
536 429
537 inode->i_op = &def_mdt_iops; 430 inode->i_op = &def_mdt_iops;
538 inode->i_fop = &def_mdt_fops; 431 inode->i_fop = &def_mdt_fops;
539 inode->i_mapping->a_ops = &def_mdt_aops; 432 inode->i_mapping->a_ops = &def_mdt_aops;
540 return inode; 433
434 return 0;
541} 435}
542 436
543void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size, 437void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
@@ -550,34 +444,159 @@ void nilfs_mdt_set_entry_size(struct inode *inode, unsigned entry_size,
550 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size); 444 mi->mi_first_entry_offset = DIV_ROUND_UP(header_size, entry_size);
551} 445}
552 446
553void nilfs_mdt_set_shadow(struct inode *orig, struct inode *shadow) 447static const struct address_space_operations shadow_map_aops = {
448 .sync_page = block_sync_page,
449};
450
451/**
452 * nilfs_mdt_setup_shadow_map - setup shadow map and bind it to metadata file
453 * @inode: inode of the metadata file
454 * @shadow: shadow mapping
455 */
456int nilfs_mdt_setup_shadow_map(struct inode *inode,
457 struct nilfs_shadow_map *shadow)
554{ 458{
555 shadow->i_mapping->assoc_mapping = orig->i_mapping; 459 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
556 NILFS_I(shadow)->i_btnode_cache.assoc_mapping = 460 struct backing_dev_info *bdi = inode->i_sb->s_bdi;
557 &NILFS_I(orig)->i_btnode_cache; 461
462 INIT_LIST_HEAD(&shadow->frozen_buffers);
463 nilfs_mapping_init_once(&shadow->frozen_data);
464 nilfs_mapping_init(&shadow->frozen_data, bdi, &shadow_map_aops);
465 nilfs_mapping_init_once(&shadow->frozen_btnodes);
466 nilfs_mapping_init(&shadow->frozen_btnodes, bdi, &shadow_map_aops);
467 mi->mi_shadow = shadow;
468 return 0;
558} 469}
559 470
560static void nilfs_mdt_clear(struct inode *inode) 471/**
472 * nilfs_mdt_save_to_shadow_map - copy bmap and dirty pages to shadow map
473 * @inode: inode of the metadata file
474 */
475int nilfs_mdt_save_to_shadow_map(struct inode *inode)
561{ 476{
477 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
562 struct nilfs_inode_info *ii = NILFS_I(inode); 478 struct nilfs_inode_info *ii = NILFS_I(inode);
479 struct nilfs_shadow_map *shadow = mi->mi_shadow;
480 int ret;
563 481
564 invalidate_mapping_pages(inode->i_mapping, 0, -1); 482 ret = nilfs_copy_dirty_pages(&shadow->frozen_data, inode->i_mapping);
565 truncate_inode_pages(inode->i_mapping, 0); 483 if (ret)
484 goto out;
485
486 ret = nilfs_copy_dirty_pages(&shadow->frozen_btnodes,
487 &ii->i_btnode_cache);
488 if (ret)
489 goto out;
566 490
567 if (test_bit(NILFS_I_BMAP, &ii->i_state)) 491 nilfs_bmap_save(ii->i_bmap, &shadow->bmap_store);
568 nilfs_bmap_clear(ii->i_bmap); 492 out:
569 nilfs_btnode_cache_clear(&ii->i_btnode_cache); 493 return ret;
570} 494}
571 495
572void nilfs_mdt_destroy(struct inode *inode) 496int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh)
573{ 497{
574 struct nilfs_mdt_info *mdi = NILFS_MDT(inode); 498 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
499 struct buffer_head *bh_frozen;
500 struct page *page;
501 int blkbits = inode->i_blkbits;
502 int ret = -ENOMEM;
503
504 page = grab_cache_page(&shadow->frozen_data, bh->b_page->index);
505 if (!page)
506 return ret;
507
508 if (!page_has_buffers(page))
509 create_empty_buffers(page, 1 << blkbits, 0);
510
511 bh_frozen = nilfs_page_get_nth_block(page, bh_offset(bh) >> blkbits);
512 if (bh_frozen) {
513 if (!buffer_uptodate(bh_frozen))
514 nilfs_copy_buffer(bh_frozen, bh);
515 if (list_empty(&bh_frozen->b_assoc_buffers)) {
516 list_add_tail(&bh_frozen->b_assoc_buffers,
517 &shadow->frozen_buffers);
518 set_buffer_nilfs_redirected(bh);
519 } else {
520 brelse(bh_frozen); /* already frozen */
521 }
522 ret = 0;
523 }
524 unlock_page(page);
525 page_cache_release(page);
526 return ret;
527}
528
529struct buffer_head *
530nilfs_mdt_get_frozen_buffer(struct inode *inode, struct buffer_head *bh)
531{
532 struct nilfs_shadow_map *shadow = NILFS_MDT(inode)->mi_shadow;
533 struct buffer_head *bh_frozen = NULL;
534 struct page *page;
535 int n;
536
537 page = find_lock_page(&shadow->frozen_data, bh->b_page->index);
538 if (page) {
539 if (page_has_buffers(page)) {
540 n = bh_offset(bh) >> inode->i_blkbits;
541 bh_frozen = nilfs_page_get_nth_block(page, n);
542 }
543 unlock_page(page);
544 page_cache_release(page);
545 }
546 return bh_frozen;
547}
548
549static void nilfs_release_frozen_buffers(struct nilfs_shadow_map *shadow)
550{
551 struct list_head *head = &shadow->frozen_buffers;
552 struct buffer_head *bh;
553
554 while (!list_empty(head)) {
555 bh = list_first_entry(head, struct buffer_head,
556 b_assoc_buffers);
557 list_del_init(&bh->b_assoc_buffers);
558 brelse(bh); /* drop ref-count to make it releasable */
559 }
560}
561
562/**
563 * nilfs_mdt_restore_from_shadow_map - restore dirty pages and bmap state
564 * @inode: inode of the metadata file
565 */
566void nilfs_mdt_restore_from_shadow_map(struct inode *inode)
567{
568 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
569 struct nilfs_inode_info *ii = NILFS_I(inode);
570 struct nilfs_shadow_map *shadow = mi->mi_shadow;
571
572 down_write(&mi->mi_sem);
575 573
576 if (mdi->mi_palloc_cache) 574 if (mi->mi_palloc_cache)
577 nilfs_palloc_destroy_cache(inode); 575 nilfs_palloc_clear_cache(inode);
578 nilfs_mdt_clear(inode); 576
577 nilfs_clear_dirty_pages(inode->i_mapping);
578 nilfs_copy_back_pages(inode->i_mapping, &shadow->frozen_data);
579
580 nilfs_clear_dirty_pages(&ii->i_btnode_cache);
581 nilfs_copy_back_pages(&ii->i_btnode_cache, &shadow->frozen_btnodes);
582
583 nilfs_bmap_restore(ii->i_bmap, &shadow->bmap_store);
584
585 up_write(&mi->mi_sem);
586}
587
588/**
589 * nilfs_mdt_clear_shadow_map - truncate pages in shadow map caches
590 * @inode: inode of the metadata file
591 */
592void nilfs_mdt_clear_shadow_map(struct inode *inode)
593{
594 struct nilfs_mdt_info *mi = NILFS_MDT(inode);
595 struct nilfs_shadow_map *shadow = mi->mi_shadow;
579 596
580 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */ 597 down_write(&mi->mi_sem);
581 kfree(mdi); 598 nilfs_release_frozen_buffers(shadow);
582 nilfs_destroy_inode(inode); 599 truncate_inode_pages(&shadow->frozen_data, 0);
600 truncate_inode_pages(&shadow->frozen_btnodes, 0);
601 up_write(&mi->mi_sem);
583} 602}
diff --git a/fs/nilfs2/mdt.h b/fs/nilfs2/mdt.h
index 6c4bbb0470fc..b13734bf3521 100644
--- a/fs/nilfs2/mdt.h
+++ b/fs/nilfs2/mdt.h
@@ -28,26 +28,33 @@
28#include "nilfs.h" 28#include "nilfs.h"
29#include "page.h" 29#include "page.h"
30 30
31struct nilfs_shadow_map {
32 struct nilfs_bmap_store bmap_store;
33 struct address_space frozen_data;
34 struct address_space frozen_btnodes;
35 struct list_head frozen_buffers;
36};
37
31/** 38/**
32 * struct nilfs_mdt_info - on-memory private data of meta data files 39 * struct nilfs_mdt_info - on-memory private data of meta data files
33 * @mi_nilfs: back pointer to the_nilfs struct
34 * @mi_sem: reader/writer semaphore for meta data operations 40 * @mi_sem: reader/writer semaphore for meta data operations
35 * @mi_bgl: per-blockgroup locking 41 * @mi_bgl: per-blockgroup locking
36 * @mi_entry_size: size of an entry 42 * @mi_entry_size: size of an entry
37 * @mi_first_entry_offset: offset to the first entry 43 * @mi_first_entry_offset: offset to the first entry
38 * @mi_entries_per_block: number of entries in a block 44 * @mi_entries_per_block: number of entries in a block
39 * @mi_palloc_cache: persistent object allocator cache 45 * @mi_palloc_cache: persistent object allocator cache
46 * @mi_shadow: shadow of bmap and page caches
40 * @mi_blocks_per_group: number of blocks in a group 47 * @mi_blocks_per_group: number of blocks in a group
41 * @mi_blocks_per_desc_block: number of blocks per descriptor block 48 * @mi_blocks_per_desc_block: number of blocks per descriptor block
42 */ 49 */
43struct nilfs_mdt_info { 50struct nilfs_mdt_info {
44 struct the_nilfs *mi_nilfs;
45 struct rw_semaphore mi_sem; 51 struct rw_semaphore mi_sem;
46 struct blockgroup_lock *mi_bgl; 52 struct blockgroup_lock *mi_bgl;
47 unsigned mi_entry_size; 53 unsigned mi_entry_size;
48 unsigned mi_first_entry_offset; 54 unsigned mi_first_entry_offset;
49 unsigned long mi_entries_per_block; 55 unsigned long mi_entries_per_block;
50 struct nilfs_palloc_cache *mi_palloc_cache; 56 struct nilfs_palloc_cache *mi_palloc_cache;
57 struct nilfs_shadow_map *mi_shadow;
51 unsigned long mi_blocks_per_group; 58 unsigned long mi_blocks_per_group;
52 unsigned long mi_blocks_per_desc_block; 59 unsigned long mi_blocks_per_desc_block;
53}; 60};
@@ -59,9 +66,7 @@ static inline struct nilfs_mdt_info *NILFS_MDT(const struct inode *inode)
59 66
60static inline struct the_nilfs *NILFS_I_NILFS(struct inode *inode) 67static inline struct the_nilfs *NILFS_I_NILFS(struct inode *inode)
61{ 68{
62 struct super_block *sb = inode->i_sb; 69 return NILFS_SB(inode->i_sb)->s_nilfs;
63
64 return sb ? NILFS_SB(sb)->s_nilfs : NILFS_MDT(inode)->mi_nilfs;
65} 70}
66 71
67/* Default GFP flags using highmem */ 72/* Default GFP flags using highmem */
@@ -76,14 +81,17 @@ int nilfs_mdt_forget_block(struct inode *, unsigned long);
76int nilfs_mdt_mark_block_dirty(struct inode *, unsigned long); 81int nilfs_mdt_mark_block_dirty(struct inode *, unsigned long);
77int nilfs_mdt_fetch_dirty(struct inode *); 82int nilfs_mdt_fetch_dirty(struct inode *);
78 83
79struct inode *nilfs_mdt_new(struct the_nilfs *, struct super_block *, ino_t, 84int nilfs_mdt_init(struct inode *inode, gfp_t gfp_mask, size_t objsz);
80 size_t);
81struct inode *nilfs_mdt_new_common(struct the_nilfs *, struct super_block *,
82 ino_t, gfp_t, size_t);
83void nilfs_mdt_destroy(struct inode *);
84void nilfs_mdt_set_entry_size(struct inode *, unsigned, unsigned); 85void nilfs_mdt_set_entry_size(struct inode *, unsigned, unsigned);
85void nilfs_mdt_set_shadow(struct inode *, struct inode *);
86 86
87int nilfs_mdt_setup_shadow_map(struct inode *inode,
88 struct nilfs_shadow_map *shadow);
89int nilfs_mdt_save_to_shadow_map(struct inode *inode);
90void nilfs_mdt_restore_from_shadow_map(struct inode *inode);
91void nilfs_mdt_clear_shadow_map(struct inode *inode);
92int nilfs_mdt_freeze_buffer(struct inode *inode, struct buffer_head *bh);
93struct buffer_head *nilfs_mdt_get_frozen_buffer(struct inode *inode,
94 struct buffer_head *bh);
87 95
88#define nilfs_mdt_mark_buffer_dirty(bh) nilfs_mark_buffer_dirty(bh) 96#define nilfs_mdt_mark_buffer_dirty(bh) nilfs_mark_buffer_dirty(bh)
89 97
@@ -100,7 +108,7 @@ static inline void nilfs_mdt_clear_dirty(struct inode *inode)
100 108
101static inline __u64 nilfs_mdt_cno(struct inode *inode) 109static inline __u64 nilfs_mdt_cno(struct inode *inode)
102{ 110{
103 return NILFS_MDT(inode)->mi_nilfs->ns_cno; 111 return NILFS_I_NILFS(inode)->ns_cno;
104} 112}
105 113
106#define nilfs_mdt_bgl_lock(inode, bg) \ 114#define nilfs_mdt_bgl_lock(inode, bg) \
diff --git a/fs/nilfs2/namei.c b/fs/nilfs2/namei.c
index ad6ed2cf19b4..185d1607cb00 100644
--- a/fs/nilfs2/namei.c
+++ b/fs/nilfs2/namei.c
@@ -40,7 +40,11 @@
40 40
41#include <linux/pagemap.h> 41#include <linux/pagemap.h>
42#include "nilfs.h" 42#include "nilfs.h"
43#include "export.h"
43 44
45#define NILFS_FID_SIZE_NON_CONNECTABLE \
46 (offsetof(struct nilfs_fid, parent_gen) / 4)
47#define NILFS_FID_SIZE_CONNECTABLE (sizeof(struct nilfs_fid) / 4)
44 48
45static inline int nilfs_add_nondir(struct dentry *dentry, struct inode *inode) 49static inline int nilfs_add_nondir(struct dentry *dentry, struct inode *inode)
46{ 50{
@@ -70,29 +74,13 @@ nilfs_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
70 ino = nilfs_inode_by_name(dir, &dentry->d_name); 74 ino = nilfs_inode_by_name(dir, &dentry->d_name);
71 inode = NULL; 75 inode = NULL;
72 if (ino) { 76 if (ino) {
73 inode = nilfs_iget(dir->i_sb, ino); 77 inode = nilfs_iget(dir->i_sb, NILFS_I(dir)->i_root, ino);
74 if (IS_ERR(inode)) 78 if (IS_ERR(inode))
75 return ERR_CAST(inode); 79 return ERR_CAST(inode);
76 } 80 }
77 return d_splice_alias(inode, dentry); 81 return d_splice_alias(inode, dentry);
78} 82}
79 83
80struct dentry *nilfs_get_parent(struct dentry *child)
81{
82 unsigned long ino;
83 struct inode *inode;
84 struct qstr dotdot = {.name = "..", .len = 2};
85
86 ino = nilfs_inode_by_name(child->d_inode, &dotdot);
87 if (!ino)
88 return ERR_PTR(-ENOENT);
89
90 inode = nilfs_iget(child->d_inode->i_sb, ino);
91 if (IS_ERR(inode))
92 return ERR_CAST(inode);
93 return d_obtain_alias(inode);
94}
95
96/* 84/*
97 * By the time this is called, we already have created 85 * By the time this is called, we already have created
98 * the directory cache entry for the new file, but it 86 * the directory cache entry for the new file, but it
@@ -468,6 +456,115 @@ out:
468 return err; 456 return err;
469} 457}
470 458
459/*
460 * Export operations
461 */
462static struct dentry *nilfs_get_parent(struct dentry *child)
463{
464 unsigned long ino;
465 struct inode *inode;
466 struct qstr dotdot = {.name = "..", .len = 2};
467 struct nilfs_root *root;
468
469 ino = nilfs_inode_by_name(child->d_inode, &dotdot);
470 if (!ino)
471 return ERR_PTR(-ENOENT);
472
473 root = NILFS_I(child->d_inode)->i_root;
474
475 inode = nilfs_iget(child->d_inode->i_sb, root, ino);
476 if (IS_ERR(inode))
477 return ERR_CAST(inode);
478
479 return d_obtain_alias(inode);
480}
481
482static struct dentry *nilfs_get_dentry(struct super_block *sb, u64 cno,
483 u64 ino, u32 gen)
484{
485 struct nilfs_root *root;
486 struct inode *inode;
487
488 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO)
489 return ERR_PTR(-ESTALE);
490
491 root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
492 if (!root)
493 return ERR_PTR(-ESTALE);
494
495 inode = nilfs_iget(sb, root, ino);
496 nilfs_put_root(root);
497
498 if (IS_ERR(inode))
499 return ERR_CAST(inode);
500 if (gen && inode->i_generation != gen) {
501 iput(inode);
502 return ERR_PTR(-ESTALE);
503 }
504 return d_obtain_alias(inode);
505}
506
507static struct dentry *nilfs_fh_to_dentry(struct super_block *sb, struct fid *fh,
508 int fh_len, int fh_type)
509{
510 struct nilfs_fid *fid = (struct nilfs_fid *)fh;
511
512 if ((fh_len != NILFS_FID_SIZE_NON_CONNECTABLE &&
513 fh_len != NILFS_FID_SIZE_CONNECTABLE) ||
514 (fh_type != FILEID_NILFS_WITH_PARENT &&
515 fh_type != FILEID_NILFS_WITHOUT_PARENT))
516 return NULL;
517
518 return nilfs_get_dentry(sb, fid->cno, fid->ino, fid->gen);
519}
520
521static struct dentry *nilfs_fh_to_parent(struct super_block *sb, struct fid *fh,
522 int fh_len, int fh_type)
523{
524 struct nilfs_fid *fid = (struct nilfs_fid *)fh;
525
526 if (fh_len != NILFS_FID_SIZE_CONNECTABLE ||
527 fh_type != FILEID_NILFS_WITH_PARENT)
528 return NULL;
529
530 return nilfs_get_dentry(sb, fid->cno, fid->parent_ino, fid->parent_gen);
531}
532
533static int nilfs_encode_fh(struct dentry *dentry, __u32 *fh, int *lenp,
534 int connectable)
535{
536 struct nilfs_fid *fid = (struct nilfs_fid *)fh;
537 struct inode *inode = dentry->d_inode;
538 struct nilfs_root *root = NILFS_I(inode)->i_root;
539 int type;
540
541 if (*lenp < NILFS_FID_SIZE_NON_CONNECTABLE ||
542 (connectable && *lenp < NILFS_FID_SIZE_CONNECTABLE))
543 return 255;
544
545 fid->cno = root->cno;
546 fid->ino = inode->i_ino;
547 fid->gen = inode->i_generation;
548
549 if (connectable && !S_ISDIR(inode->i_mode)) {
550 struct inode *parent;
551
552 spin_lock(&dentry->d_lock);
553 parent = dentry->d_parent->d_inode;
554 fid->parent_ino = parent->i_ino;
555 fid->parent_gen = parent->i_generation;
556 spin_unlock(&dentry->d_lock);
557
558 type = FILEID_NILFS_WITH_PARENT;
559 *lenp = NILFS_FID_SIZE_CONNECTABLE;
560 } else {
561 type = FILEID_NILFS_WITHOUT_PARENT;
562 *lenp = NILFS_FID_SIZE_NON_CONNECTABLE;
563 }
564
565 return type;
566}
567
471const struct inode_operations nilfs_dir_inode_operations = { 568const struct inode_operations nilfs_dir_inode_operations = {
472 .create = nilfs_create, 569 .create = nilfs_create,
473 .lookup = nilfs_lookup, 570 .lookup = nilfs_lookup,
@@ -491,4 +588,12 @@ const struct inode_operations nilfs_symlink_inode_operations = {
491 .readlink = generic_readlink, 588 .readlink = generic_readlink,
492 .follow_link = page_follow_link_light, 589 .follow_link = page_follow_link_light,
493 .put_link = page_put_link, 590 .put_link = page_put_link,
591 .permission = nilfs_permission,
592};
593
594const struct export_operations nilfs_export_ops = {
595 .encode_fh = nilfs_encode_fh,
596 .fh_to_dentry = nilfs_fh_to_dentry,
597 .fh_to_parent = nilfs_fh_to_parent,
598 .get_parent = nilfs_get_parent,
494}; 599};
diff --git a/fs/nilfs2/nilfs.h b/fs/nilfs2/nilfs.h
index d3d54046e5f8..f7560da5a567 100644
--- a/fs/nilfs2/nilfs.h
+++ b/fs/nilfs2/nilfs.h
@@ -59,6 +59,7 @@ struct nilfs_inode_info {
59#endif 59#endif
60 struct buffer_head *i_bh; /* i_bh contains a new or dirty 60 struct buffer_head *i_bh; /* i_bh contains a new or dirty
61 disk inode */ 61 disk inode */
62 struct nilfs_root *i_root;
62 struct inode vfs_inode; 63 struct inode vfs_inode;
63}; 64};
64 65
@@ -100,7 +101,6 @@ enum {
100 NILFS_I_INODE_DIRTY, /* write_inode is requested */ 101 NILFS_I_INODE_DIRTY, /* write_inode is requested */
101 NILFS_I_BMAP, /* has bmap and btnode_cache */ 102 NILFS_I_BMAP, /* has bmap and btnode_cache */
102 NILFS_I_GCINODE, /* inode for GC, on memory only */ 103 NILFS_I_GCINODE, /* inode for GC, on memory only */
103 NILFS_I_GCDAT, /* shadow DAT, on memory only */
104}; 104};
105 105
106/* 106/*
@@ -192,7 +192,7 @@ static inline int nilfs_doing_construction(void)
192 192
193static inline struct inode *nilfs_dat_inode(const struct the_nilfs *nilfs) 193static inline struct inode *nilfs_dat_inode(const struct the_nilfs *nilfs)
194{ 194{
195 return nilfs_doing_gc() ? nilfs->ns_gc_dat : nilfs->ns_dat; 195 return nilfs->ns_dat;
196} 196}
197 197
198/* 198/*
@@ -200,12 +200,9 @@ static inline struct inode *nilfs_dat_inode(const struct the_nilfs *nilfs)
200 */ 200 */
201#ifdef CONFIG_NILFS_POSIX_ACL 201#ifdef CONFIG_NILFS_POSIX_ACL
202#error "NILFS: not yet supported POSIX ACL" 202#error "NILFS: not yet supported POSIX ACL"
203extern int nilfs_permission(struct inode *, int, struct nameidata *);
204extern int nilfs_acl_chmod(struct inode *); 203extern int nilfs_acl_chmod(struct inode *);
205extern int nilfs_init_acl(struct inode *, struct inode *); 204extern int nilfs_init_acl(struct inode *, struct inode *);
206#else 205#else
207#define nilfs_permission NULL
208
209static inline int nilfs_acl_chmod(struct inode *inode) 206static inline int nilfs_acl_chmod(struct inode *inode)
210{ 207{
211 return 0; 208 return 0;
@@ -247,11 +244,19 @@ extern int nilfs_get_block(struct inode *, sector_t, struct buffer_head *, int);
247extern void nilfs_set_inode_flags(struct inode *); 244extern void nilfs_set_inode_flags(struct inode *);
248extern int nilfs_read_inode_common(struct inode *, struct nilfs_inode *); 245extern int nilfs_read_inode_common(struct inode *, struct nilfs_inode *);
249extern void nilfs_write_inode_common(struct inode *, struct nilfs_inode *, int); 246extern void nilfs_write_inode_common(struct inode *, struct nilfs_inode *, int);
250extern struct inode *nilfs_iget(struct super_block *, unsigned long); 247struct inode *nilfs_ilookup(struct super_block *sb, struct nilfs_root *root,
248 unsigned long ino);
249struct inode *nilfs_iget_locked(struct super_block *sb, struct nilfs_root *root,
250 unsigned long ino);
251struct inode *nilfs_iget(struct super_block *sb, struct nilfs_root *root,
252 unsigned long ino);
253extern struct inode *nilfs_iget_for_gc(struct super_block *sb,
254 unsigned long ino, __u64 cno);
251extern void nilfs_update_inode(struct inode *, struct buffer_head *); 255extern void nilfs_update_inode(struct inode *, struct buffer_head *);
252extern void nilfs_truncate(struct inode *); 256extern void nilfs_truncate(struct inode *);
253extern void nilfs_evict_inode(struct inode *); 257extern void nilfs_evict_inode(struct inode *);
254extern int nilfs_setattr(struct dentry *, struct iattr *); 258extern int nilfs_setattr(struct dentry *, struct iattr *);
259int nilfs_permission(struct inode *inode, int mask);
255extern int nilfs_load_inode_block(struct nilfs_sb_info *, struct inode *, 260extern int nilfs_load_inode_block(struct nilfs_sb_info *, struct inode *,
256 struct buffer_head **); 261 struct buffer_head **);
257extern int nilfs_inode_dirty(struct inode *); 262extern int nilfs_inode_dirty(struct inode *);
@@ -260,11 +265,7 @@ extern int nilfs_set_file_dirty(struct nilfs_sb_info *, struct inode *,
260extern int nilfs_mark_inode_dirty(struct inode *); 265extern int nilfs_mark_inode_dirty(struct inode *);
261extern void nilfs_dirty_inode(struct inode *); 266extern void nilfs_dirty_inode(struct inode *);
262 267
263/* namei.c */
264extern struct dentry *nilfs_get_parent(struct dentry *);
265
266/* super.c */ 268/* super.c */
267extern struct inode *nilfs_alloc_inode_common(struct the_nilfs *);
268extern struct inode *nilfs_alloc_inode(struct super_block *); 269extern struct inode *nilfs_alloc_inode(struct super_block *);
269extern void nilfs_destroy_inode(struct inode *); 270extern void nilfs_destroy_inode(struct inode *);
270extern void nilfs_error(struct super_block *, const char *, const char *, ...) 271extern void nilfs_error(struct super_block *, const char *, const char *, ...)
@@ -283,8 +284,9 @@ extern struct nilfs_super_block **nilfs_prepare_super(struct nilfs_sb_info *,
283 int flip); 284 int flip);
284extern int nilfs_commit_super(struct nilfs_sb_info *, int); 285extern int nilfs_commit_super(struct nilfs_sb_info *, int);
285extern int nilfs_cleanup_super(struct nilfs_sb_info *); 286extern int nilfs_cleanup_super(struct nilfs_sb_info *);
286extern int nilfs_attach_checkpoint(struct nilfs_sb_info *, __u64); 287int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
287extern void nilfs_detach_checkpoint(struct nilfs_sb_info *); 288 struct nilfs_root **root);
289int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno);
288 290
289/* gcinode.c */ 291/* gcinode.c */
290int nilfs_gccache_submit_read_data(struct inode *, sector_t, sector_t, __u64, 292int nilfs_gccache_submit_read_data(struct inode *, sector_t, sector_t, __u64,
@@ -292,16 +294,8 @@ int nilfs_gccache_submit_read_data(struct inode *, sector_t, sector_t, __u64,
292int nilfs_gccache_submit_read_node(struct inode *, sector_t, __u64, 294int nilfs_gccache_submit_read_node(struct inode *, sector_t, __u64,
293 struct buffer_head **); 295 struct buffer_head **);
294int nilfs_gccache_wait_and_mark_dirty(struct buffer_head *); 296int nilfs_gccache_wait_and_mark_dirty(struct buffer_head *);
295int nilfs_init_gccache(struct the_nilfs *); 297int nilfs_init_gcinode(struct inode *inode);
296void nilfs_destroy_gccache(struct the_nilfs *); 298void nilfs_remove_all_gcinodes(struct the_nilfs *nilfs);
297void nilfs_clear_gcinode(struct inode *);
298struct inode *nilfs_gc_iget(struct the_nilfs *, ino_t, __u64);
299void nilfs_remove_all_gcinode(struct the_nilfs *);
300
301/* gcdat.c */
302int nilfs_init_gcdat_inode(struct the_nilfs *);
303void nilfs_commit_gcdat_inode(struct the_nilfs *);
304void nilfs_clear_gcdat_inode(struct the_nilfs *);
305 299
306/* 300/*
307 * Inodes and files operations 301 * Inodes and files operations
diff --git a/fs/nilfs2/page.c b/fs/nilfs2/page.c
index aab11db2cb08..a6c3c2e817f8 100644
--- a/fs/nilfs2/page.c
+++ b/fs/nilfs2/page.c
@@ -79,8 +79,8 @@ struct buffer_head *nilfs_grab_buffer(struct inode *inode,
79{ 79{
80 int blkbits = inode->i_blkbits; 80 int blkbits = inode->i_blkbits;
81 pgoff_t index = blkoff >> (PAGE_CACHE_SHIFT - blkbits); 81 pgoff_t index = blkoff >> (PAGE_CACHE_SHIFT - blkbits);
82 struct page *page, *opage; 82 struct page *page;
83 struct buffer_head *bh, *obh; 83 struct buffer_head *bh;
84 84
85 page = grab_cache_page(mapping, index); 85 page = grab_cache_page(mapping, index);
86 if (unlikely(!page)) 86 if (unlikely(!page))
@@ -92,30 +92,6 @@ struct buffer_head *nilfs_grab_buffer(struct inode *inode,
92 page_cache_release(page); 92 page_cache_release(page);
93 return NULL; 93 return NULL;
94 } 94 }
95 if (!buffer_uptodate(bh) && mapping->assoc_mapping != NULL) {
96 /*
97 * Shadow page cache uses assoc_mapping to point its original
98 * page cache. The following code tries the original cache
99 * if the given cache is a shadow and it didn't hit.
100 */
101 opage = find_lock_page(mapping->assoc_mapping, index);
102 if (!opage)
103 return bh;
104
105 obh = __nilfs_get_page_block(opage, blkoff, index, blkbits,
106 b_state);
107 if (buffer_uptodate(obh)) {
108 nilfs_copy_buffer(bh, obh);
109 if (buffer_dirty(obh)) {
110 nilfs_mark_buffer_dirty(bh);
111 if (!buffer_nilfs_node(bh) && NILFS_MDT(inode))
112 nilfs_mdt_mark_dirty(inode);
113 }
114 }
115 brelse(obh);
116 unlock_page(opage);
117 page_cache_release(opage);
118 }
119 return bh; 95 return bh;
120} 96}
121 97
@@ -131,6 +107,7 @@ void nilfs_forget_buffer(struct buffer_head *bh)
131 lock_buffer(bh); 107 lock_buffer(bh);
132 clear_buffer_nilfs_volatile(bh); 108 clear_buffer_nilfs_volatile(bh);
133 clear_buffer_nilfs_checked(bh); 109 clear_buffer_nilfs_checked(bh);
110 clear_buffer_nilfs_redirected(bh);
134 clear_buffer_dirty(bh); 111 clear_buffer_dirty(bh);
135 if (nilfs_page_buffers_clean(page)) 112 if (nilfs_page_buffers_clean(page))
136 __nilfs_clear_page_dirty(page); 113 __nilfs_clear_page_dirty(page);
@@ -483,6 +460,7 @@ void nilfs_clear_dirty_pages(struct address_space *mapping)
483 clear_buffer_dirty(bh); 460 clear_buffer_dirty(bh);
484 clear_buffer_nilfs_volatile(bh); 461 clear_buffer_nilfs_volatile(bh);
485 clear_buffer_nilfs_checked(bh); 462 clear_buffer_nilfs_checked(bh);
463 clear_buffer_nilfs_redirected(bh);
486 clear_buffer_uptodate(bh); 464 clear_buffer_uptodate(bh);
487 clear_buffer_mapped(bh); 465 clear_buffer_mapped(bh);
488 unlock_buffer(bh); 466 unlock_buffer(bh);
@@ -513,6 +491,31 @@ unsigned nilfs_page_count_clean_buffers(struct page *page,
513 } 491 }
514 return nc; 492 return nc;
515} 493}
494
495void nilfs_mapping_init_once(struct address_space *mapping)
496{
497 memset(mapping, 0, sizeof(*mapping));
498 INIT_RADIX_TREE(&mapping->page_tree, GFP_ATOMIC);
499 spin_lock_init(&mapping->tree_lock);
500 INIT_LIST_HEAD(&mapping->private_list);
501 spin_lock_init(&mapping->private_lock);
502
503 spin_lock_init(&mapping->i_mmap_lock);
504 INIT_RAW_PRIO_TREE_ROOT(&mapping->i_mmap);
505 INIT_LIST_HEAD(&mapping->i_mmap_nonlinear);
506}
507
508void nilfs_mapping_init(struct address_space *mapping,
509 struct backing_dev_info *bdi,
510 const struct address_space_operations *aops)
511{
512 mapping->host = NULL;
513 mapping->flags = 0;
514 mapping_set_gfp_mask(mapping, GFP_NOFS);
515 mapping->assoc_mapping = NULL;
516 mapping->backing_dev_info = bdi;
517 mapping->a_ops = aops;
518}
516 519
517/* 520/*
518 * NILFS2 needs clear_page_dirty() in the following two cases: 521 * NILFS2 needs clear_page_dirty() in the following two cases:
diff --git a/fs/nilfs2/page.h b/fs/nilfs2/page.h
index f53d8da41ed7..fb9e8a8a2038 100644
--- a/fs/nilfs2/page.h
+++ b/fs/nilfs2/page.h
@@ -35,12 +35,14 @@ enum {
35 BH_NILFS_Node, 35 BH_NILFS_Node,
36 BH_NILFS_Volatile, 36 BH_NILFS_Volatile,
37 BH_NILFS_Checked, 37 BH_NILFS_Checked,
38 BH_NILFS_Redirected,
38}; 39};
39 40
40BUFFER_FNS(NILFS_Allocated, nilfs_allocated) /* nilfs private buffers */ 41BUFFER_FNS(NILFS_Allocated, nilfs_allocated) /* nilfs private buffers */
41BUFFER_FNS(NILFS_Node, nilfs_node) /* nilfs node buffers */ 42BUFFER_FNS(NILFS_Node, nilfs_node) /* nilfs node buffers */
42BUFFER_FNS(NILFS_Volatile, nilfs_volatile) 43BUFFER_FNS(NILFS_Volatile, nilfs_volatile)
43BUFFER_FNS(NILFS_Checked, nilfs_checked) /* buffer is verified */ 44BUFFER_FNS(NILFS_Checked, nilfs_checked) /* buffer is verified */
45BUFFER_FNS(NILFS_Redirected, nilfs_redirected) /* redirected to a copy */
44 46
45 47
46void nilfs_mark_buffer_dirty(struct buffer_head *bh); 48void nilfs_mark_buffer_dirty(struct buffer_head *bh);
@@ -59,6 +61,10 @@ void nilfs_free_private_page(struct page *);
59int nilfs_copy_dirty_pages(struct address_space *, struct address_space *); 61int nilfs_copy_dirty_pages(struct address_space *, struct address_space *);
60void nilfs_copy_back_pages(struct address_space *, struct address_space *); 62void nilfs_copy_back_pages(struct address_space *, struct address_space *);
61void nilfs_clear_dirty_pages(struct address_space *); 63void nilfs_clear_dirty_pages(struct address_space *);
64void nilfs_mapping_init_once(struct address_space *mapping);
65void nilfs_mapping_init(struct address_space *mapping,
66 struct backing_dev_info *bdi,
67 const struct address_space_operations *aops);
62unsigned nilfs_page_count_clean_buffers(struct page *, unsigned, unsigned); 68unsigned nilfs_page_count_clean_buffers(struct page *, unsigned, unsigned);
63 69
64#define NILFS_PAGE_BUG(page, m, a...) \ 70#define NILFS_PAGE_BUG(page, m, a...) \
diff --git a/fs/nilfs2/recovery.c b/fs/nilfs2/recovery.c
index d0c35ef39f6a..5d2711c28da7 100644
--- a/fs/nilfs2/recovery.c
+++ b/fs/nilfs2/recovery.c
@@ -440,7 +440,6 @@ static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
440 segnum[2] = ri->ri_segnum; 440 segnum[2] = ri->ri_segnum;
441 segnum[3] = ri->ri_nextnum; 441 segnum[3] = ri->ri_nextnum;
442 442
443 nilfs_attach_writer(nilfs, sbi);
444 /* 443 /*
445 * Releasing the next segment of the latest super root. 444 * Releasing the next segment of the latest super root.
446 * The next segment is invalidated by this recovery. 445 * The next segment is invalidated by this recovery.
@@ -480,7 +479,6 @@ static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs,
480 479
481 failed: 480 failed:
482 /* No need to recover sufile because it will be destroyed on error */ 481 /* No need to recover sufile because it will be destroyed on error */
483 nilfs_detach_writer(nilfs, sbi);
484 return err; 482 return err;
485} 483}
486 484
@@ -504,6 +502,7 @@ static int nilfs_recovery_copy_block(struct the_nilfs *nilfs,
504 502
505static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs, 503static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
506 struct nilfs_sb_info *sbi, 504 struct nilfs_sb_info *sbi,
505 struct nilfs_root *root,
507 struct list_head *head, 506 struct list_head *head,
508 unsigned long *nr_salvaged_blocks) 507 unsigned long *nr_salvaged_blocks)
509{ 508{
@@ -515,7 +514,7 @@ static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
515 int err = 0, err2 = 0; 514 int err = 0, err2 = 0;
516 515
517 list_for_each_entry_safe(rb, n, head, list) { 516 list_for_each_entry_safe(rb, n, head, list) {
518 inode = nilfs_iget(sbi->s_super, rb->ino); 517 inode = nilfs_iget(sbi->s_super, root, rb->ino);
519 if (IS_ERR(inode)) { 518 if (IS_ERR(inode)) {
520 err = PTR_ERR(inode); 519 err = PTR_ERR(inode);
521 inode = NULL; 520 inode = NULL;
@@ -578,6 +577,7 @@ static int nilfs_recover_dsync_blocks(struct the_nilfs *nilfs,
578 */ 577 */
579static int nilfs_do_roll_forward(struct the_nilfs *nilfs, 578static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
580 struct nilfs_sb_info *sbi, 579 struct nilfs_sb_info *sbi,
580 struct nilfs_root *root,
581 struct nilfs_recovery_info *ri) 581 struct nilfs_recovery_info *ri)
582{ 582{
583 struct buffer_head *bh_sum = NULL; 583 struct buffer_head *bh_sum = NULL;
@@ -597,7 +597,6 @@ static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
597 }; 597 };
598 int state = RF_INIT_ST; 598 int state = RF_INIT_ST;
599 599
600 nilfs_attach_writer(nilfs, sbi);
601 pseg_start = ri->ri_lsegs_start; 600 pseg_start = ri->ri_lsegs_start;
602 seg_seq = ri->ri_lsegs_start_seq; 601 seg_seq = ri->ri_lsegs_start_seq;
603 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start); 602 segnum = nilfs_get_segnum_of_block(nilfs, pseg_start);
@@ -649,7 +648,7 @@ static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
649 goto failed; 648 goto failed;
650 if (flags & NILFS_SS_LOGEND) { 649 if (flags & NILFS_SS_LOGEND) {
651 err = nilfs_recover_dsync_blocks( 650 err = nilfs_recover_dsync_blocks(
652 nilfs, sbi, &dsync_blocks, 651 nilfs, sbi, root, &dsync_blocks,
653 &nsalvaged_blocks); 652 &nsalvaged_blocks);
654 if (unlikely(err)) 653 if (unlikely(err))
655 goto failed; 654 goto failed;
@@ -688,7 +687,6 @@ static int nilfs_do_roll_forward(struct the_nilfs *nilfs,
688 out: 687 out:
689 brelse(bh_sum); 688 brelse(bh_sum);
690 dispose_recovery_list(&dsync_blocks); 689 dispose_recovery_list(&dsync_blocks);
691 nilfs_detach_writer(nilfs, sbi);
692 return err; 690 return err;
693 691
694 confused: 692 confused:
@@ -746,19 +744,20 @@ int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
746 struct nilfs_sb_info *sbi, 744 struct nilfs_sb_info *sbi,
747 struct nilfs_recovery_info *ri) 745 struct nilfs_recovery_info *ri)
748{ 746{
747 struct nilfs_root *root;
749 int err; 748 int err;
750 749
751 if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0) 750 if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0)
752 return 0; 751 return 0;
753 752
754 err = nilfs_attach_checkpoint(sbi, ri->ri_cno); 753 err = nilfs_attach_checkpoint(sbi, ri->ri_cno, true, &root);
755 if (unlikely(err)) { 754 if (unlikely(err)) {
756 printk(KERN_ERR 755 printk(KERN_ERR
757 "NILFS: error loading the latest checkpoint.\n"); 756 "NILFS: error loading the latest checkpoint.\n");
758 return err; 757 return err;
759 } 758 }
760 759
761 err = nilfs_do_roll_forward(nilfs, sbi, ri); 760 err = nilfs_do_roll_forward(nilfs, sbi, root, ri);
762 if (unlikely(err)) 761 if (unlikely(err))
763 goto failed; 762 goto failed;
764 763
@@ -770,7 +769,7 @@ int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
770 goto failed; 769 goto failed;
771 } 770 }
772 771
773 err = nilfs_attach_segment_constructor(sbi); 772 err = nilfs_attach_segment_constructor(sbi, root);
774 if (unlikely(err)) 773 if (unlikely(err))
775 goto failed; 774 goto failed;
776 775
@@ -788,7 +787,7 @@ int nilfs_salvage_orphan_logs(struct the_nilfs *nilfs,
788 } 787 }
789 788
790 failed: 789 failed:
791 nilfs_detach_checkpoint(sbi); 790 nilfs_put_root(root);
792 return err; 791 return err;
793} 792}
794 793
diff --git a/fs/nilfs2/sb.h b/fs/nilfs2/sb.h
index 0776ccc2504a..35a07157b980 100644
--- a/fs/nilfs2/sb.h
+++ b/fs/nilfs2/sb.h
@@ -42,11 +42,6 @@ struct nilfs_sc_info;
42 * NILFS super-block data in memory 42 * NILFS super-block data in memory
43 */ 43 */
44struct nilfs_sb_info { 44struct nilfs_sb_info {
45 /* Snapshot status */
46 __u64 s_snapshot_cno; /* Checkpoint number */
47 atomic_t s_inodes_count;
48 atomic_t s_blocks_count; /* Reserved (might be deleted) */
49
50 /* Mount options */ 45 /* Mount options */
51 unsigned long s_mount_opt; 46 unsigned long s_mount_opt;
52 uid_t s_resuid; 47 uid_t s_resuid;
@@ -59,8 +54,6 @@ struct nilfs_sb_info {
59 /* Fundamental members */ 54 /* Fundamental members */
60 struct super_block *s_super; /* reverse pointer to super_block */ 55 struct super_block *s_super; /* reverse pointer to super_block */
61 struct the_nilfs *s_nilfs; 56 struct the_nilfs *s_nilfs;
62 struct list_head s_list; /* list head for nilfs->ns_supers */
63 atomic_t s_count; /* reference count */
64 57
65 /* Segment constructor */ 58 /* Segment constructor */
66 struct list_head s_dirty_files; /* dirty files list */ 59 struct list_head s_dirty_files; /* dirty files list */
@@ -68,9 +61,6 @@ struct nilfs_sb_info {
68 spinlock_t s_inode_lock; /* Lock for the nilfs inode. 61 spinlock_t s_inode_lock; /* Lock for the nilfs inode.
69 It covers s_dirty_files list */ 62 It covers s_dirty_files list */
70 63
71 /* Metadata files */
72 struct inode *s_ifile; /* index file inode */
73
74 /* Inode allocator */ 64 /* Inode allocator */
75 spinlock_t s_next_gen_lock; 65 spinlock_t s_next_gen_lock;
76 u32 s_next_generation; 66 u32 s_next_generation;
diff --git a/fs/nilfs2/segbuf.c b/fs/nilfs2/segbuf.c
index 4588fb9e93df..0f83e93935b2 100644
--- a/fs/nilfs2/segbuf.c
+++ b/fs/nilfs2/segbuf.c
@@ -371,7 +371,8 @@ static int nilfs_segbuf_submit_bio(struct nilfs_segment_buffer *segbuf,
371 struct bio *bio = wi->bio; 371 struct bio *bio = wi->bio;
372 int err; 372 int err;
373 373
374 if (segbuf->sb_nbio > 0 && bdi_write_congested(wi->nilfs->ns_bdi)) { 374 if (segbuf->sb_nbio > 0 &&
375 bdi_write_congested(segbuf->sb_super->s_bdi)) {
375 wait_for_completion(&segbuf->sb_bio_event); 376 wait_for_completion(&segbuf->sb_bio_event);
376 segbuf->sb_nbio--; 377 segbuf->sb_nbio--;
377 if (unlikely(atomic_read(&segbuf->sb_err))) { 378 if (unlikely(atomic_read(&segbuf->sb_err))) {
diff --git a/fs/nilfs2/segment.c b/fs/nilfs2/segment.c
index 9fd051a33c4f..d926af626177 100644
--- a/fs/nilfs2/segment.c
+++ b/fs/nilfs2/segment.c
@@ -191,6 +191,8 @@ int nilfs_transaction_begin(struct super_block *sb,
191 if (ret > 0) 191 if (ret > 0)
192 return 0; 192 return 0;
193 193
194 vfs_check_frozen(sb, SB_FREEZE_WRITE);
195
194 sbi = NILFS_SB(sb); 196 sbi = NILFS_SB(sb);
195 nilfs = sbi->s_nilfs; 197 nilfs = sbi->s_nilfs;
196 down_read(&nilfs->ns_segctor_sem); 198 down_read(&nilfs->ns_segctor_sem);
@@ -366,8 +368,7 @@ static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
366 368
367 if (nilfs_doing_gc()) 369 if (nilfs_doing_gc())
368 flags = NILFS_SS_GC; 370 flags = NILFS_SS_GC;
369 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, 371 err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
370 sci->sc_sbi->s_nilfs->ns_cno);
371 if (unlikely(err)) 372 if (unlikely(err))
372 return err; 373 return err;
373 374
@@ -440,17 +441,26 @@ static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
440 struct nilfs_finfo *finfo; 441 struct nilfs_finfo *finfo;
441 struct nilfs_inode_info *ii; 442 struct nilfs_inode_info *ii;
442 struct nilfs_segment_buffer *segbuf; 443 struct nilfs_segment_buffer *segbuf;
444 __u64 cno;
443 445
444 if (sci->sc_blk_cnt == 0) 446 if (sci->sc_blk_cnt == 0)
445 return; 447 return;
446 448
447 ii = NILFS_I(inode); 449 ii = NILFS_I(inode);
450
451 if (test_bit(NILFS_I_GCINODE, &ii->i_state))
452 cno = ii->i_cno;
453 else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
454 cno = 0;
455 else
456 cno = sci->sc_cno;
457
448 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr, 458 finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
449 sizeof(*finfo)); 459 sizeof(*finfo));
450 finfo->fi_ino = cpu_to_le64(inode->i_ino); 460 finfo->fi_ino = cpu_to_le64(inode->i_ino);
451 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt); 461 finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
452 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt); 462 finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
453 finfo->fi_cno = cpu_to_le64(ii->i_cno); 463 finfo->fi_cno = cpu_to_le64(cno);
454 464
455 segbuf = sci->sc_curseg; 465 segbuf = sci->sc_curseg;
456 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset + 466 segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
@@ -755,12 +765,12 @@ static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
755 } 765 }
756} 766}
757 767
758static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi) 768static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
769 struct nilfs_root *root)
759{ 770{
760 struct the_nilfs *nilfs = sbi->s_nilfs;
761 int ret = 0; 771 int ret = 0;
762 772
763 if (nilfs_mdt_fetch_dirty(sbi->s_ifile)) 773 if (nilfs_mdt_fetch_dirty(root->ifile))
764 ret++; 774 ret++;
765 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile)) 775 if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
766 ret++; 776 ret++;
@@ -785,7 +795,7 @@ static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
785 struct nilfs_sb_info *sbi = sci->sc_sbi; 795 struct nilfs_sb_info *sbi = sci->sc_sbi;
786 int ret = 0; 796 int ret = 0;
787 797
788 if (nilfs_test_metadata_dirty(sbi)) 798 if (nilfs_test_metadata_dirty(sbi->s_nilfs, sci->sc_root))
789 set_bit(NILFS_SC_DIRTY, &sci->sc_flags); 799 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
790 800
791 spin_lock(&sbi->s_inode_lock); 801 spin_lock(&sbi->s_inode_lock);
@@ -801,7 +811,7 @@ static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
801 struct nilfs_sb_info *sbi = sci->sc_sbi; 811 struct nilfs_sb_info *sbi = sci->sc_sbi;
802 struct the_nilfs *nilfs = sbi->s_nilfs; 812 struct the_nilfs *nilfs = sbi->s_nilfs;
803 813
804 nilfs_mdt_clear_dirty(sbi->s_ifile); 814 nilfs_mdt_clear_dirty(sci->sc_root->ifile);
805 nilfs_mdt_clear_dirty(nilfs->ns_cpfile); 815 nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
806 nilfs_mdt_clear_dirty(nilfs->ns_sufile); 816 nilfs_mdt_clear_dirty(nilfs->ns_sufile);
807 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs)); 817 nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
@@ -848,9 +858,9 @@ static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
848 raw_cp->cp_snapshot_list.ssl_next = 0; 858 raw_cp->cp_snapshot_list.ssl_next = 0;
849 raw_cp->cp_snapshot_list.ssl_prev = 0; 859 raw_cp->cp_snapshot_list.ssl_prev = 0;
850 raw_cp->cp_inodes_count = 860 raw_cp->cp_inodes_count =
851 cpu_to_le64(atomic_read(&sbi->s_inodes_count)); 861 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
852 raw_cp->cp_blocks_count = 862 raw_cp->cp_blocks_count =
853 cpu_to_le64(atomic_read(&sbi->s_blocks_count)); 863 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
854 raw_cp->cp_nblk_inc = 864 raw_cp->cp_nblk_inc =
855 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc); 865 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
856 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime); 866 raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
@@ -861,7 +871,8 @@ static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
861 else 871 else
862 nilfs_checkpoint_set_minor(raw_cp); 872 nilfs_checkpoint_set_minor(raw_cp);
863 873
864 nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1); 874 nilfs_write_inode_common(sci->sc_root->ifile,
875 &raw_cp->cp_ifile_inode, 1);
865 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp); 876 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
866 return 0; 877 return 0;
867 878
@@ -886,13 +897,12 @@ static void nilfs_fill_in_file_bmap(struct inode *ifile,
886 } 897 }
887} 898}
888 899
889static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci, 900static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
890 struct inode *ifile)
891{ 901{
892 struct nilfs_inode_info *ii; 902 struct nilfs_inode_info *ii;
893 903
894 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) { 904 list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
895 nilfs_fill_in_file_bmap(ifile, ii); 905 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
896 set_bit(NILFS_I_COLLECTED, &ii->i_state); 906 set_bit(NILFS_I_COLLECTED, &ii->i_state);
897 } 907 }
898} 908}
@@ -1135,7 +1145,7 @@ static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1135 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED; 1145 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1136 /* Fall through */ 1146 /* Fall through */
1137 case NILFS_ST_IFILE: 1147 case NILFS_ST_IFILE:
1138 err = nilfs_segctor_scan_file(sci, sbi->s_ifile, 1148 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1139 &nilfs_sc_file_ops); 1149 &nilfs_sc_file_ops);
1140 if (unlikely(err)) 1150 if (unlikely(err))
1141 break; 1151 break;
@@ -1900,6 +1910,7 @@ static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1900 set_buffer_uptodate(bh); 1910 set_buffer_uptodate(bh);
1901 clear_buffer_dirty(bh); 1911 clear_buffer_dirty(bh);
1902 clear_buffer_nilfs_volatile(bh); 1912 clear_buffer_nilfs_volatile(bh);
1913 clear_buffer_nilfs_redirected(bh);
1903 if (bh == segbuf->sb_super_root) { 1914 if (bh == segbuf->sb_super_root) {
1904 if (bh->b_page != bd_page) { 1915 if (bh->b_page != bd_page) {
1905 end_page_writeback(bd_page); 1916 end_page_writeback(bd_page);
@@ -1936,11 +1947,9 @@ static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1936 1947
1937 nilfs_drop_collected_inodes(&sci->sc_dirty_files); 1948 nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1938 1949
1939 if (nilfs_doing_gc()) { 1950 if (nilfs_doing_gc())
1940 nilfs_drop_collected_inodes(&sci->sc_gc_inodes); 1951 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1941 if (update_sr) 1952 else
1942 nilfs_commit_gcdat_inode(nilfs);
1943 } else
1944 nilfs->ns_nongc_ctime = sci->sc_seg_ctime; 1953 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1945 1954
1946 sci->sc_nblk_inc += sci->sc_nblk_this_inc; 1955 sci->sc_nblk_inc += sci->sc_nblk_this_inc;
@@ -1976,7 +1985,7 @@ static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1976 struct nilfs_sb_info *sbi) 1985 struct nilfs_sb_info *sbi)
1977{ 1986{
1978 struct nilfs_inode_info *ii, *n; 1987 struct nilfs_inode_info *ii, *n;
1979 __u64 cno = sbi->s_nilfs->ns_cno; 1988 struct inode *ifile = sci->sc_root->ifile;
1980 1989
1981 spin_lock(&sbi->s_inode_lock); 1990 spin_lock(&sbi->s_inode_lock);
1982 retry: 1991 retry:
@@ -1987,14 +1996,14 @@ static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
1987 1996
1988 spin_unlock(&sbi->s_inode_lock); 1997 spin_unlock(&sbi->s_inode_lock);
1989 err = nilfs_ifile_get_inode_block( 1998 err = nilfs_ifile_get_inode_block(
1990 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh); 1999 ifile, ii->vfs_inode.i_ino, &ibh);
1991 if (unlikely(err)) { 2000 if (unlikely(err)) {
1992 nilfs_warning(sbi->s_super, __func__, 2001 nilfs_warning(sbi->s_super, __func__,
1993 "failed to get inode block.\n"); 2002 "failed to get inode block.\n");
1994 return err; 2003 return err;
1995 } 2004 }
1996 nilfs_mdt_mark_buffer_dirty(ibh); 2005 nilfs_mdt_mark_buffer_dirty(ibh);
1997 nilfs_mdt_mark_dirty(sbi->s_ifile); 2006 nilfs_mdt_mark_dirty(ifile);
1998 spin_lock(&sbi->s_inode_lock); 2007 spin_lock(&sbi->s_inode_lock);
1999 if (likely(!ii->i_bh)) 2008 if (likely(!ii->i_bh))
2000 ii->i_bh = ibh; 2009 ii->i_bh = ibh;
@@ -2002,7 +2011,6 @@ static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2002 brelse(ibh); 2011 brelse(ibh);
2003 goto retry; 2012 goto retry;
2004 } 2013 }
2005 ii->i_cno = cno;
2006 2014
2007 clear_bit(NILFS_I_QUEUED, &ii->i_state); 2015 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2008 set_bit(NILFS_I_BUSY, &ii->i_state); 2016 set_bit(NILFS_I_BUSY, &ii->i_state);
@@ -2011,8 +2019,6 @@ static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2011 } 2019 }
2012 spin_unlock(&sbi->s_inode_lock); 2020 spin_unlock(&sbi->s_inode_lock);
2013 2021
2014 NILFS_I(sbi->s_ifile)->i_cno = cno;
2015
2016 return 0; 2022 return 0;
2017} 2023}
2018 2024
@@ -2021,19 +2027,13 @@ static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2021{ 2027{
2022 struct nilfs_transaction_info *ti = current->journal_info; 2028 struct nilfs_transaction_info *ti = current->journal_info;
2023 struct nilfs_inode_info *ii, *n; 2029 struct nilfs_inode_info *ii, *n;
2024 __u64 cno = sbi->s_nilfs->ns_cno;
2025 2030
2026 spin_lock(&sbi->s_inode_lock); 2031 spin_lock(&sbi->s_inode_lock);
2027 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) { 2032 list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2028 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) || 2033 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2029 test_bit(NILFS_I_DIRTY, &ii->i_state)) { 2034 test_bit(NILFS_I_DIRTY, &ii->i_state))
2030 /* The current checkpoint number (=nilfs->ns_cno) is
2031 changed between check-in and check-out only if the
2032 super root is written out. So, we can update i_cno
2033 for the inodes that remain in the dirty list. */
2034 ii->i_cno = cno;
2035 continue; 2035 continue;
2036 } 2036
2037 clear_bit(NILFS_I_BUSY, &ii->i_state); 2037 clear_bit(NILFS_I_BUSY, &ii->i_state);
2038 brelse(ii->i_bh); 2038 brelse(ii->i_bh);
2039 ii->i_bh = NULL; 2039 ii->i_bh = NULL;
@@ -2054,12 +2054,13 @@ static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2054 int err; 2054 int err;
2055 2055
2056 sci->sc_stage.scnt = NILFS_ST_INIT; 2056 sci->sc_stage.scnt = NILFS_ST_INIT;
2057 sci->sc_cno = nilfs->ns_cno;
2057 2058
2058 err = nilfs_segctor_check_in_files(sci, sbi); 2059 err = nilfs_segctor_check_in_files(sci, sbi);
2059 if (unlikely(err)) 2060 if (unlikely(err))
2060 goto out; 2061 goto out;
2061 2062
2062 if (nilfs_test_metadata_dirty(sbi)) 2063 if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2063 set_bit(NILFS_SC_DIRTY, &sci->sc_flags); 2064 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2064 2065
2065 if (nilfs_segctor_clean(sci)) 2066 if (nilfs_segctor_clean(sci))
@@ -2091,7 +2092,7 @@ static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2091 goto failed; 2092 goto failed;
2092 2093
2093 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED) 2094 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2094 nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile); 2095 nilfs_segctor_fill_in_file_bmap(sci);
2095 2096
2096 if (mode == SC_LSEG_SR && 2097 if (mode == SC_LSEG_SR &&
2097 sci->sc_stage.scnt >= NILFS_ST_CPFILE) { 2098 sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
@@ -2452,9 +2453,8 @@ nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2452 list_for_each_entry_safe(ii, n, head, i_dirty) { 2453 list_for_each_entry_safe(ii, n, head, i_dirty) {
2453 if (!test_bit(NILFS_I_UPDATED, &ii->i_state)) 2454 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2454 continue; 2455 continue;
2455 hlist_del_init(&ii->vfs_inode.i_hash);
2456 list_del_init(&ii->i_dirty); 2456 list_del_init(&ii->i_dirty);
2457 nilfs_clear_gcinode(&ii->vfs_inode); 2457 iput(&ii->vfs_inode);
2458 } 2458 }
2459} 2459}
2460 2460
@@ -2472,13 +2472,15 @@ int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2472 2472
2473 nilfs_transaction_lock(sbi, &ti, 1); 2473 nilfs_transaction_lock(sbi, &ti, 1);
2474 2474
2475 err = nilfs_init_gcdat_inode(nilfs); 2475 err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2476 if (unlikely(err)) 2476 if (unlikely(err))
2477 goto out_unlock; 2477 goto out_unlock;
2478 2478
2479 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs); 2479 err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2480 if (unlikely(err)) 2480 if (unlikely(err)) {
2481 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2481 goto out_unlock; 2482 goto out_unlock;
2483 }
2482 2484
2483 sci->sc_freesegs = kbufs[4]; 2485 sci->sc_freesegs = kbufs[4];
2484 sci->sc_nfreesegs = argv[4].v_nmembs; 2486 sci->sc_nfreesegs = argv[4].v_nmembs;
@@ -2510,7 +2512,7 @@ int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2510 out_unlock: 2512 out_unlock:
2511 sci->sc_freesegs = NULL; 2513 sci->sc_freesegs = NULL;
2512 sci->sc_nfreesegs = 0; 2514 sci->sc_nfreesegs = 0;
2513 nilfs_clear_gcdat_inode(nilfs); 2515 nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2514 nilfs_transaction_unlock(sbi); 2516 nilfs_transaction_unlock(sbi);
2515 return err; 2517 return err;
2516} 2518}
@@ -2672,6 +2674,8 @@ static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2672} 2674}
2673 2675
2674static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci) 2676static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2677 __acquires(&sci->sc_state_lock)
2678 __releases(&sci->sc_state_lock)
2675{ 2679{
2676 sci->sc_state |= NILFS_SEGCTOR_QUIT; 2680 sci->sc_state |= NILFS_SEGCTOR_QUIT;
2677 2681
@@ -2686,7 +2690,8 @@ static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2686/* 2690/*
2687 * Setup & clean-up functions 2691 * Setup & clean-up functions
2688 */ 2692 */
2689static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi) 2693static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi,
2694 struct nilfs_root *root)
2690{ 2695{
2691 struct nilfs_sc_info *sci; 2696 struct nilfs_sc_info *sci;
2692 2697
@@ -2697,6 +2702,9 @@ static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2697 sci->sc_sbi = sbi; 2702 sci->sc_sbi = sbi;
2698 sci->sc_super = sbi->s_super; 2703 sci->sc_super = sbi->s_super;
2699 2704
2705 nilfs_get_root(root);
2706 sci->sc_root = root;
2707
2700 init_waitqueue_head(&sci->sc_wait_request); 2708 init_waitqueue_head(&sci->sc_wait_request);
2701 init_waitqueue_head(&sci->sc_wait_daemon); 2709 init_waitqueue_head(&sci->sc_wait_daemon);
2702 init_waitqueue_head(&sci->sc_wait_task); 2710 init_waitqueue_head(&sci->sc_wait_task);
@@ -2771,6 +2779,8 @@ static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2771 WARN_ON(!list_empty(&sci->sc_segbufs)); 2779 WARN_ON(!list_empty(&sci->sc_segbufs));
2772 WARN_ON(!list_empty(&sci->sc_write_logs)); 2780 WARN_ON(!list_empty(&sci->sc_write_logs));
2773 2781
2782 nilfs_put_root(sci->sc_root);
2783
2774 down_write(&sbi->s_nilfs->ns_segctor_sem); 2784 down_write(&sbi->s_nilfs->ns_segctor_sem);
2775 2785
2776 del_timer_sync(&sci->sc_timer); 2786 del_timer_sync(&sci->sc_timer);
@@ -2780,6 +2790,7 @@ static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2780/** 2790/**
2781 * nilfs_attach_segment_constructor - attach a segment constructor 2791 * nilfs_attach_segment_constructor - attach a segment constructor
2782 * @sbi: nilfs_sb_info 2792 * @sbi: nilfs_sb_info
2793 * @root: root object of the current filesystem tree
2783 * 2794 *
2784 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info, 2795 * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2785 * initializes it, and starts the segment constructor. 2796 * initializes it, and starts the segment constructor.
@@ -2789,9 +2800,9 @@ static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2789 * 2800 *
2790 * %-ENOMEM - Insufficient memory available. 2801 * %-ENOMEM - Insufficient memory available.
2791 */ 2802 */
2792int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi) 2803int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
2804 struct nilfs_root *root)
2793{ 2805{
2794 struct the_nilfs *nilfs = sbi->s_nilfs;
2795 int err; 2806 int err;
2796 2807
2797 if (NILFS_SC(sbi)) { 2808 if (NILFS_SC(sbi)) {
@@ -2803,14 +2814,12 @@ int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
2803 nilfs_detach_segment_constructor(sbi); 2814 nilfs_detach_segment_constructor(sbi);
2804 } 2815 }
2805 2816
2806 sbi->s_sc_info = nilfs_segctor_new(sbi); 2817 sbi->s_sc_info = nilfs_segctor_new(sbi, root);
2807 if (!sbi->s_sc_info) 2818 if (!sbi->s_sc_info)
2808 return -ENOMEM; 2819 return -ENOMEM;
2809 2820
2810 nilfs_attach_writer(nilfs, sbi);
2811 err = nilfs_segctor_start_thread(NILFS_SC(sbi)); 2821 err = nilfs_segctor_start_thread(NILFS_SC(sbi));
2812 if (err) { 2822 if (err) {
2813 nilfs_detach_writer(nilfs, sbi);
2814 kfree(sbi->s_sc_info); 2823 kfree(sbi->s_sc_info);
2815 sbi->s_sc_info = NULL; 2824 sbi->s_sc_info = NULL;
2816 } 2825 }
@@ -2847,5 +2856,4 @@ void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2847 up_write(&nilfs->ns_segctor_sem); 2856 up_write(&nilfs->ns_segctor_sem);
2848 2857
2849 nilfs_dispose_list(sbi, &garbage_list, 1); 2858 nilfs_dispose_list(sbi, &garbage_list, 1);
2850 nilfs_detach_writer(nilfs, sbi);
2851} 2859}
diff --git a/fs/nilfs2/segment.h b/fs/nilfs2/segment.h
index 17c487bd8152..cd8056e7cbed 100644
--- a/fs/nilfs2/segment.h
+++ b/fs/nilfs2/segment.h
@@ -29,6 +29,8 @@
29#include <linux/nilfs2_fs.h> 29#include <linux/nilfs2_fs.h>
30#include "sb.h" 30#include "sb.h"
31 31
32struct nilfs_root;
33
32/** 34/**
33 * struct nilfs_recovery_info - Recovery information 35 * struct nilfs_recovery_info - Recovery information
34 * @ri_need_recovery: Recovery status 36 * @ri_need_recovery: Recovery status
@@ -87,6 +89,7 @@ struct nilfs_segsum_pointer {
87 * struct nilfs_sc_info - Segment constructor information 89 * struct nilfs_sc_info - Segment constructor information
88 * @sc_super: Back pointer to super_block struct 90 * @sc_super: Back pointer to super_block struct
89 * @sc_sbi: Back pointer to nilfs_sb_info struct 91 * @sc_sbi: Back pointer to nilfs_sb_info struct
92 * @sc_root: root object of the current filesystem tree
90 * @sc_nblk_inc: Block count of current generation 93 * @sc_nblk_inc: Block count of current generation
91 * @sc_dirty_files: List of files to be written 94 * @sc_dirty_files: List of files to be written
92 * @sc_gc_inodes: List of GC inodes having blocks to be written 95 * @sc_gc_inodes: List of GC inodes having blocks to be written
@@ -107,6 +110,7 @@ struct nilfs_segsum_pointer {
107 * @sc_datablk_cnt: Data block count of a file 110 * @sc_datablk_cnt: Data block count of a file
108 * @sc_nblk_this_inc: Number of blocks included in the current logical segment 111 * @sc_nblk_this_inc: Number of blocks included in the current logical segment
109 * @sc_seg_ctime: Creation time 112 * @sc_seg_ctime: Creation time
113 * @sc_cno: checkpoint number of current log
110 * @sc_flags: Internal flags 114 * @sc_flags: Internal flags
111 * @sc_state_lock: spinlock for sc_state and so on 115 * @sc_state_lock: spinlock for sc_state and so on
112 * @sc_state: Segctord state flags 116 * @sc_state: Segctord state flags
@@ -128,6 +132,7 @@ struct nilfs_segsum_pointer {
128struct nilfs_sc_info { 132struct nilfs_sc_info {
129 struct super_block *sc_super; 133 struct super_block *sc_super;
130 struct nilfs_sb_info *sc_sbi; 134 struct nilfs_sb_info *sc_sbi;
135 struct nilfs_root *sc_root;
131 136
132 unsigned long sc_nblk_inc; 137 unsigned long sc_nblk_inc;
133 138
@@ -156,7 +161,7 @@ struct nilfs_sc_info {
156 unsigned long sc_datablk_cnt; 161 unsigned long sc_datablk_cnt;
157 unsigned long sc_nblk_this_inc; 162 unsigned long sc_nblk_this_inc;
158 time_t sc_seg_ctime; 163 time_t sc_seg_ctime;
159 164 __u64 sc_cno;
160 unsigned long sc_flags; 165 unsigned long sc_flags;
161 166
162 spinlock_t sc_state_lock; 167 spinlock_t sc_state_lock;
@@ -230,7 +235,8 @@ extern void nilfs_flush_segment(struct super_block *, ino_t);
230extern int nilfs_clean_segments(struct super_block *, struct nilfs_argv *, 235extern int nilfs_clean_segments(struct super_block *, struct nilfs_argv *,
231 void **); 236 void **);
232 237
233extern int nilfs_attach_segment_constructor(struct nilfs_sb_info *); 238int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
239 struct nilfs_root *root);
234extern void nilfs_detach_segment_constructor(struct nilfs_sb_info *); 240extern void nilfs_detach_segment_constructor(struct nilfs_sb_info *);
235 241
236/* recovery.c */ 242/* recovery.c */
diff --git a/fs/nilfs2/sufile.c b/fs/nilfs2/sufile.c
index 3c6cc6005c2e..1d6f488ccae8 100644
--- a/fs/nilfs2/sufile.c
+++ b/fs/nilfs2/sufile.c
@@ -505,7 +505,7 @@ int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
505{ 505{
506 struct buffer_head *header_bh; 506 struct buffer_head *header_bh;
507 struct nilfs_sufile_header *header; 507 struct nilfs_sufile_header *header;
508 struct the_nilfs *nilfs = NILFS_MDT(sufile)->mi_nilfs; 508 struct the_nilfs *nilfs = NILFS_I_NILFS(sufile);
509 void *kaddr; 509 void *kaddr;
510 int ret; 510 int ret;
511 511
@@ -583,7 +583,7 @@ ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
583 struct nilfs_segment_usage *su; 583 struct nilfs_segment_usage *su;
584 struct nilfs_suinfo *si = buf; 584 struct nilfs_suinfo *si = buf;
585 size_t susz = NILFS_MDT(sufile)->mi_entry_size; 585 size_t susz = NILFS_MDT(sufile)->mi_entry_size;
586 struct the_nilfs *nilfs = NILFS_MDT(sufile)->mi_nilfs; 586 struct the_nilfs *nilfs = NILFS_I_NILFS(sufile);
587 void *kaddr; 587 void *kaddr;
588 unsigned long nsegs, segusages_per_block; 588 unsigned long nsegs, segusages_per_block;
589 ssize_t n; 589 ssize_t n;
@@ -635,46 +635,55 @@ ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
635} 635}
636 636
637/** 637/**
638 * nilfs_sufile_read - read sufile inode 638 * nilfs_sufile_read - read or get sufile inode
639 * @sufile: sufile inode 639 * @sb: super block instance
640 * @susize: size of a segment usage entry
640 * @raw_inode: on-disk sufile inode 641 * @raw_inode: on-disk sufile inode
642 * @inodep: buffer to store the inode
641 */ 643 */
642int nilfs_sufile_read(struct inode *sufile, struct nilfs_inode *raw_inode) 644int nilfs_sufile_read(struct super_block *sb, size_t susize,
645 struct nilfs_inode *raw_inode, struct inode **inodep)
643{ 646{
644 struct nilfs_sufile_info *sui = NILFS_SUI(sufile); 647 struct inode *sufile;
648 struct nilfs_sufile_info *sui;
645 struct buffer_head *header_bh; 649 struct buffer_head *header_bh;
646 struct nilfs_sufile_header *header; 650 struct nilfs_sufile_header *header;
647 void *kaddr; 651 void *kaddr;
648 int ret; 652 int err;
649 653
650 ret = nilfs_read_inode_common(sufile, raw_inode); 654 sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
651 if (ret < 0) 655 if (unlikely(!sufile))
652 return ret; 656 return -ENOMEM;
657 if (!(sufile->i_state & I_NEW))
658 goto out;
653 659
654 ret = nilfs_sufile_get_header_block(sufile, &header_bh); 660 err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
655 if (!ret) { 661 if (err)
656 kaddr = kmap_atomic(header_bh->b_page, KM_USER0); 662 goto failed;
657 header = kaddr + bh_offset(header_bh);
658 sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
659 kunmap_atomic(kaddr, KM_USER0);
660 brelse(header_bh);
661 }
662 return ret;
663}
664 663
665/** 664 nilfs_mdt_set_entry_size(sufile, susize,
666 * nilfs_sufile_new - create sufile 665 sizeof(struct nilfs_sufile_header));
667 * @nilfs: nilfs object 666
668 * @susize: size of a segment usage entry 667 err = nilfs_read_inode_common(sufile, raw_inode);
669 */ 668 if (err)
670struct inode *nilfs_sufile_new(struct the_nilfs *nilfs, size_t susize) 669 goto failed;
671{ 670
672 struct inode *sufile; 671 err = nilfs_sufile_get_header_block(sufile, &header_bh);
672 if (err)
673 goto failed;
673 674
674 sufile = nilfs_mdt_new(nilfs, NULL, NILFS_SUFILE_INO, 675 sui = NILFS_SUI(sufile);
675 sizeof(struct nilfs_sufile_info)); 676 kaddr = kmap_atomic(header_bh->b_page, KM_USER0);
676 if (sufile) 677 header = kaddr + bh_offset(header_bh);
677 nilfs_mdt_set_entry_size(sufile, susize, 678 sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
678 sizeof(struct nilfs_sufile_header)); 679 kunmap_atomic(kaddr, KM_USER0);
679 return sufile; 680 brelse(header_bh);
681
682 unlock_new_inode(sufile);
683 out:
684 *inodep = sufile;
685 return 0;
686 failed:
687 iget_failed(sufile);
688 return err;
680} 689}
diff --git a/fs/nilfs2/sufile.h b/fs/nilfs2/sufile.h
index 15163b8aff7d..a943fbacb45b 100644
--- a/fs/nilfs2/sufile.h
+++ b/fs/nilfs2/sufile.h
@@ -31,7 +31,7 @@
31 31
32static inline unsigned long nilfs_sufile_get_nsegments(struct inode *sufile) 32static inline unsigned long nilfs_sufile_get_nsegments(struct inode *sufile)
33{ 33{
34 return NILFS_MDT(sufile)->mi_nilfs->ns_nsegments; 34 return NILFS_I_NILFS(sufile)->ns_nsegments;
35} 35}
36 36
37unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile); 37unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile);
@@ -61,8 +61,8 @@ void nilfs_sufile_do_cancel_free(struct inode *, __u64, struct buffer_head *,
61void nilfs_sufile_do_set_error(struct inode *, __u64, struct buffer_head *, 61void nilfs_sufile_do_set_error(struct inode *, __u64, struct buffer_head *,
62 struct buffer_head *); 62 struct buffer_head *);
63 63
64int nilfs_sufile_read(struct inode *sufile, struct nilfs_inode *raw_inode); 64int nilfs_sufile_read(struct super_block *sb, size_t susize,
65struct inode *nilfs_sufile_new(struct the_nilfs *nilfs, size_t susize); 65 struct nilfs_inode *raw_inode, struct inode **inodep);
66 66
67/** 67/**
68 * nilfs_sufile_scrap - make a segment garbage 68 * nilfs_sufile_scrap - make a segment garbage
diff --git a/fs/nilfs2/super.c b/fs/nilfs2/super.c
index f3b75206e956..35ae03c0db86 100644
--- a/fs/nilfs2/super.c
+++ b/fs/nilfs2/super.c
@@ -48,10 +48,10 @@
48#include <linux/vfs.h> 48#include <linux/vfs.h>
49#include <linux/writeback.h> 49#include <linux/writeback.h>
50#include <linux/kobject.h> 50#include <linux/kobject.h>
51#include <linux/exportfs.h>
52#include <linux/seq_file.h> 51#include <linux/seq_file.h>
53#include <linux/mount.h> 52#include <linux/mount.h>
54#include "nilfs.h" 53#include "nilfs.h"
54#include "export.h"
55#include "mdt.h" 55#include "mdt.h"
56#include "alloc.h" 56#include "alloc.h"
57#include "btree.h" 57#include "btree.h"
@@ -68,11 +68,12 @@ MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem "
68 "(NILFS)"); 68 "(NILFS)");
69MODULE_LICENSE("GPL"); 69MODULE_LICENSE("GPL");
70 70
71struct kmem_cache *nilfs_inode_cachep; 71static struct kmem_cache *nilfs_inode_cachep;
72struct kmem_cache *nilfs_transaction_cachep; 72struct kmem_cache *nilfs_transaction_cachep;
73struct kmem_cache *nilfs_segbuf_cachep; 73struct kmem_cache *nilfs_segbuf_cachep;
74struct kmem_cache *nilfs_btree_path_cache; 74struct kmem_cache *nilfs_btree_path_cache;
75 75
76static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount);
76static int nilfs_remount(struct super_block *sb, int *flags, char *data); 77static int nilfs_remount(struct super_block *sb, int *flags, char *data);
77 78
78static void nilfs_set_error(struct nilfs_sb_info *sbi) 79static void nilfs_set_error(struct nilfs_sb_info *sbi)
@@ -146,7 +147,7 @@ void nilfs_warning(struct super_block *sb, const char *function,
146} 147}
147 148
148 149
149struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs) 150struct inode *nilfs_alloc_inode(struct super_block *sb)
150{ 151{
151 struct nilfs_inode_info *ii; 152 struct nilfs_inode_info *ii;
152 153
@@ -155,18 +156,20 @@ struct inode *nilfs_alloc_inode_common(struct the_nilfs *nilfs)
155 return NULL; 156 return NULL;
156 ii->i_bh = NULL; 157 ii->i_bh = NULL;
157 ii->i_state = 0; 158 ii->i_state = 0;
159 ii->i_cno = 0;
158 ii->vfs_inode.i_version = 1; 160 ii->vfs_inode.i_version = 1;
159 nilfs_btnode_cache_init(&ii->i_btnode_cache, nilfs->ns_bdi); 161 nilfs_btnode_cache_init(&ii->i_btnode_cache, sb->s_bdi);
160 return &ii->vfs_inode; 162 return &ii->vfs_inode;
161} 163}
162 164
163struct inode *nilfs_alloc_inode(struct super_block *sb)
164{
165 return nilfs_alloc_inode_common(NILFS_SB(sb)->s_nilfs);
166}
167
168void nilfs_destroy_inode(struct inode *inode) 165void nilfs_destroy_inode(struct inode *inode)
169{ 166{
167 struct nilfs_mdt_info *mdi = NILFS_MDT(inode);
168
169 if (mdi) {
170 kfree(mdi->mi_bgl); /* kfree(NULL) is safe */
171 kfree(mdi);
172 }
170 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode)); 173 kmem_cache_free(nilfs_inode_cachep, NILFS_I(inode));
171} 174}
172 175
@@ -340,16 +343,15 @@ static void nilfs_put_super(struct super_block *sb)
340 nilfs_cleanup_super(sbi); 343 nilfs_cleanup_super(sbi);
341 up_write(&nilfs->ns_sem); 344 up_write(&nilfs->ns_sem);
342 } 345 }
343 down_write(&nilfs->ns_super_sem);
344 if (nilfs->ns_current == sbi)
345 nilfs->ns_current = NULL;
346 up_write(&nilfs->ns_super_sem);
347 346
348 nilfs_detach_checkpoint(sbi); 347 iput(nilfs->ns_sufile);
349 put_nilfs(sbi->s_nilfs); 348 iput(nilfs->ns_cpfile);
349 iput(nilfs->ns_dat);
350
351 destroy_nilfs(nilfs);
350 sbi->s_super = NULL; 352 sbi->s_super = NULL;
351 sb->s_fs_info = NULL; 353 sb->s_fs_info = NULL;
352 nilfs_put_sbinfo(sbi); 354 kfree(sbi);
353} 355}
354 356
355static int nilfs_sync_fs(struct super_block *sb, int wait) 357static int nilfs_sync_fs(struct super_block *sb, int wait)
@@ -376,21 +378,22 @@ static int nilfs_sync_fs(struct super_block *sb, int wait)
376 return err; 378 return err;
377} 379}
378 380
379int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno) 381int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno, int curr_mnt,
382 struct nilfs_root **rootp)
380{ 383{
381 struct the_nilfs *nilfs = sbi->s_nilfs; 384 struct the_nilfs *nilfs = sbi->s_nilfs;
385 struct nilfs_root *root;
382 struct nilfs_checkpoint *raw_cp; 386 struct nilfs_checkpoint *raw_cp;
383 struct buffer_head *bh_cp; 387 struct buffer_head *bh_cp;
384 int err; 388 int err = -ENOMEM;
385 389
386 down_write(&nilfs->ns_super_sem); 390 root = nilfs_find_or_create_root(
387 list_add(&sbi->s_list, &nilfs->ns_supers); 391 nilfs, curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno);
388 up_write(&nilfs->ns_super_sem); 392 if (!root)
393 return err;
389 394
390 err = -ENOMEM; 395 if (root->ifile)
391 sbi->s_ifile = nilfs_ifile_new(sbi, nilfs->ns_inode_size); 396 goto reuse; /* already attached checkpoint */
392 if (!sbi->s_ifile)
393 goto delist;
394 397
395 down_read(&nilfs->ns_segctor_sem); 398 down_read(&nilfs->ns_segctor_sem);
396 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp, 399 err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, cno, 0, &raw_cp,
@@ -406,45 +409,64 @@ int nilfs_attach_checkpoint(struct nilfs_sb_info *sbi, __u64 cno)
406 } 409 }
407 goto failed; 410 goto failed;
408 } 411 }
409 err = nilfs_read_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode); 412
410 if (unlikely(err)) 413 err = nilfs_ifile_read(sbi->s_super, root, nilfs->ns_inode_size,
414 &raw_cp->cp_ifile_inode, &root->ifile);
415 if (err)
411 goto failed_bh; 416 goto failed_bh;
412 atomic_set(&sbi->s_inodes_count, le64_to_cpu(raw_cp->cp_inodes_count)); 417
413 atomic_set(&sbi->s_blocks_count, le64_to_cpu(raw_cp->cp_blocks_count)); 418 atomic_set(&root->inodes_count, le64_to_cpu(raw_cp->cp_inodes_count));
419 atomic_set(&root->blocks_count, le64_to_cpu(raw_cp->cp_blocks_count));
414 420
415 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); 421 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
422
423 reuse:
424 *rootp = root;
416 return 0; 425 return 0;
417 426
418 failed_bh: 427 failed_bh:
419 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp); 428 nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, cno, bh_cp);
420 failed: 429 failed:
421 nilfs_mdt_destroy(sbi->s_ifile); 430 nilfs_put_root(root);
422 sbi->s_ifile = NULL; 431
432 return err;
433}
423 434
424 delist: 435static int nilfs_freeze(struct super_block *sb)
425 down_write(&nilfs->ns_super_sem); 436{
426 list_del_init(&sbi->s_list); 437 struct nilfs_sb_info *sbi = NILFS_SB(sb);
427 up_write(&nilfs->ns_super_sem); 438 struct the_nilfs *nilfs = sbi->s_nilfs;
439 int err;
440
441 if (sb->s_flags & MS_RDONLY)
442 return 0;
428 443
444 /* Mark super block clean */
445 down_write(&nilfs->ns_sem);
446 err = nilfs_cleanup_super(sbi);
447 up_write(&nilfs->ns_sem);
429 return err; 448 return err;
430} 449}
431 450
432void nilfs_detach_checkpoint(struct nilfs_sb_info *sbi) 451static int nilfs_unfreeze(struct super_block *sb)
433{ 452{
453 struct nilfs_sb_info *sbi = NILFS_SB(sb);
434 struct the_nilfs *nilfs = sbi->s_nilfs; 454 struct the_nilfs *nilfs = sbi->s_nilfs;
435 455
436 nilfs_mdt_destroy(sbi->s_ifile); 456 if (sb->s_flags & MS_RDONLY)
437 sbi->s_ifile = NULL; 457 return 0;
438 down_write(&nilfs->ns_super_sem); 458
439 list_del_init(&sbi->s_list); 459 down_write(&nilfs->ns_sem);
440 up_write(&nilfs->ns_super_sem); 460 nilfs_setup_super(sbi, false);
461 up_write(&nilfs->ns_sem);
462 return 0;
441} 463}
442 464
443static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf) 465static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
444{ 466{
445 struct super_block *sb = dentry->d_sb; 467 struct super_block *sb = dentry->d_sb;
446 struct nilfs_sb_info *sbi = NILFS_SB(sb); 468 struct nilfs_root *root = NILFS_I(dentry->d_inode)->i_root;
447 struct the_nilfs *nilfs = sbi->s_nilfs; 469 struct the_nilfs *nilfs = root->nilfs;
448 u64 id = huge_encode_dev(sb->s_bdev->bd_dev); 470 u64 id = huge_encode_dev(sb->s_bdev->bd_dev);
449 unsigned long long blocks; 471 unsigned long long blocks;
450 unsigned long overhead; 472 unsigned long overhead;
@@ -480,7 +502,7 @@ static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf)
480 buf->f_bfree = nfreeblocks; 502 buf->f_bfree = nfreeblocks;
481 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ? 503 buf->f_bavail = (buf->f_bfree >= nrsvblocks) ?
482 (buf->f_bfree - nrsvblocks) : 0; 504 (buf->f_bfree - nrsvblocks) : 0;
483 buf->f_files = atomic_read(&sbi->s_inodes_count); 505 buf->f_files = atomic_read(&root->inodes_count);
484 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */ 506 buf->f_ffree = 0; /* nilfs_count_free_inodes(sb); */
485 buf->f_namelen = NILFS_NAME_LEN; 507 buf->f_namelen = NILFS_NAME_LEN;
486 buf->f_fsid.val[0] = (u32)id; 508 buf->f_fsid.val[0] = (u32)id;
@@ -493,12 +515,12 @@ static int nilfs_show_options(struct seq_file *seq, struct vfsmount *vfs)
493{ 515{
494 struct super_block *sb = vfs->mnt_sb; 516 struct super_block *sb = vfs->mnt_sb;
495 struct nilfs_sb_info *sbi = NILFS_SB(sb); 517 struct nilfs_sb_info *sbi = NILFS_SB(sb);
518 struct nilfs_root *root = NILFS_I(vfs->mnt_root->d_inode)->i_root;
496 519
497 if (!nilfs_test_opt(sbi, BARRIER)) 520 if (!nilfs_test_opt(sbi, BARRIER))
498 seq_puts(seq, ",nobarrier"); 521 seq_puts(seq, ",nobarrier");
499 if (nilfs_test_opt(sbi, SNAPSHOT)) 522 if (root->cno != NILFS_CPTREE_CURRENT_CNO)
500 seq_printf(seq, ",cp=%llu", 523 seq_printf(seq, ",cp=%llu", (unsigned long long)root->cno);
501 (unsigned long long int)sbi->s_snapshot_cno);
502 if (nilfs_test_opt(sbi, ERRORS_PANIC)) 524 if (nilfs_test_opt(sbi, ERRORS_PANIC))
503 seq_puts(seq, ",errors=panic"); 525 seq_puts(seq, ",errors=panic");
504 if (nilfs_test_opt(sbi, ERRORS_CONT)) 526 if (nilfs_test_opt(sbi, ERRORS_CONT))
@@ -524,6 +546,8 @@ static const struct super_operations nilfs_sops = {
524 .put_super = nilfs_put_super, 546 .put_super = nilfs_put_super,
525 /* .write_super = nilfs_write_super, */ 547 /* .write_super = nilfs_write_super, */
526 .sync_fs = nilfs_sync_fs, 548 .sync_fs = nilfs_sync_fs,
549 .freeze_fs = nilfs_freeze,
550 .unfreeze_fs = nilfs_unfreeze,
527 /* .write_super_lockfs */ 551 /* .write_super_lockfs */
528 /* .unlockfs */ 552 /* .unlockfs */
529 .statfs = nilfs_statfs, 553 .statfs = nilfs_statfs,
@@ -532,48 +556,6 @@ static const struct super_operations nilfs_sops = {
532 .show_options = nilfs_show_options 556 .show_options = nilfs_show_options
533}; 557};
534 558
535static struct inode *
536nilfs_nfs_get_inode(struct super_block *sb, u64 ino, u32 generation)
537{
538 struct inode *inode;
539
540 if (ino < NILFS_FIRST_INO(sb) && ino != NILFS_ROOT_INO &&
541 ino != NILFS_SKETCH_INO)
542 return ERR_PTR(-ESTALE);
543
544 inode = nilfs_iget(sb, ino);
545 if (IS_ERR(inode))
546 return ERR_CAST(inode);
547 if (generation && inode->i_generation != generation) {
548 iput(inode);
549 return ERR_PTR(-ESTALE);
550 }
551
552 return inode;
553}
554
555static struct dentry *
556nilfs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
557 int fh_type)
558{
559 return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
560 nilfs_nfs_get_inode);
561}
562
563static struct dentry *
564nilfs_fh_to_parent(struct super_block *sb, struct fid *fid, int fh_len,
565 int fh_type)
566{
567 return generic_fh_to_parent(sb, fid, fh_len, fh_type,
568 nilfs_nfs_get_inode);
569}
570
571static const struct export_operations nilfs_export_ops = {
572 .fh_to_dentry = nilfs_fh_to_dentry,
573 .fh_to_parent = nilfs_fh_to_parent,
574 .get_parent = nilfs_get_parent,
575};
576
577enum { 559enum {
578 Opt_err_cont, Opt_err_panic, Opt_err_ro, 560 Opt_err_cont, Opt_err_panic, Opt_err_ro,
579 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery, 561 Opt_barrier, Opt_nobarrier, Opt_snapshot, Opt_order, Opt_norecovery,
@@ -599,7 +581,6 @@ static int parse_options(char *options, struct super_block *sb, int is_remount)
599 struct nilfs_sb_info *sbi = NILFS_SB(sb); 581 struct nilfs_sb_info *sbi = NILFS_SB(sb);
600 char *p; 582 char *p;
601 substring_t args[MAX_OPT_ARGS]; 583 substring_t args[MAX_OPT_ARGS];
602 int option;
603 584
604 if (!options) 585 if (!options)
605 return 1; 586 return 1;
@@ -637,30 +618,12 @@ static int parse_options(char *options, struct super_block *sb, int is_remount)
637 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT); 618 nilfs_write_opt(sbi, ERROR_MODE, ERRORS_CONT);
638 break; 619 break;
639 case Opt_snapshot: 620 case Opt_snapshot:
640 if (match_int(&args[0], &option) || option <= 0)
641 return 0;
642 if (is_remount) { 621 if (is_remount) {
643 if (!nilfs_test_opt(sbi, SNAPSHOT)) { 622 printk(KERN_ERR
644 printk(KERN_ERR 623 "NILFS: \"%s\" option is invalid "
645 "NILFS: cannot change regular " 624 "for remount.\n", p);
646 "mount to snapshot.\n");
647 return 0;
648 } else if (option != sbi->s_snapshot_cno) {
649 printk(KERN_ERR
650 "NILFS: cannot remount to a "
651 "different snapshot.\n");
652 return 0;
653 }
654 break;
655 }
656 if (!(sb->s_flags & MS_RDONLY)) {
657 printk(KERN_ERR "NILFS: cannot mount snapshot "
658 "read/write. A read-only option is "
659 "required.\n");
660 return 0; 625 return 0;
661 } 626 }
662 sbi->s_snapshot_cno = option;
663 nilfs_set_opt(sbi, SNAPSHOT);
664 break; 627 break;
665 case Opt_norecovery: 628 case Opt_norecovery:
666 nilfs_set_opt(sbi, NORECOVERY); 629 nilfs_set_opt(sbi, NORECOVERY);
@@ -688,7 +651,7 @@ nilfs_set_default_options(struct nilfs_sb_info *sbi,
688 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER; 651 NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER;
689} 652}
690 653
691static int nilfs_setup_super(struct nilfs_sb_info *sbi) 654static int nilfs_setup_super(struct nilfs_sb_info *sbi, int is_mount)
692{ 655{
693 struct the_nilfs *nilfs = sbi->s_nilfs; 656 struct the_nilfs *nilfs = sbi->s_nilfs;
694 struct nilfs_super_block **sbp; 657 struct nilfs_super_block **sbp;
@@ -700,6 +663,9 @@ static int nilfs_setup_super(struct nilfs_sb_info *sbi)
700 if (!sbp) 663 if (!sbp)
701 return -EIO; 664 return -EIO;
702 665
666 if (!is_mount)
667 goto skip_mount_setup;
668
703 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count); 669 max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count);
704 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count); 670 mnt_count = le16_to_cpu(sbp[0]->s_mnt_count);
705 671
@@ -716,9 +682,11 @@ static int nilfs_setup_super(struct nilfs_sb_info *sbi)
716 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT); 682 sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT);
717 683
718 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1); 684 sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1);
685 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
686
687skip_mount_setup:
719 sbp[0]->s_state = 688 sbp[0]->s_state =
720 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS); 689 cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS);
721 sbp[0]->s_mtime = cpu_to_le64(get_seconds());
722 /* synchronize sbp[1] with sbp[0] */ 690 /* synchronize sbp[1] with sbp[0] */
723 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize); 691 memcpy(sbp[1], sbp[0], nilfs->ns_sbsize);
724 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL); 692 return nilfs_commit_super(sbi, NILFS_SB_COMMIT_ALL);
@@ -785,22 +753,156 @@ int nilfs_check_feature_compatibility(struct super_block *sb,
785 return 0; 753 return 0;
786} 754}
787 755
756static int nilfs_get_root_dentry(struct super_block *sb,
757 struct nilfs_root *root,
758 struct dentry **root_dentry)
759{
760 struct inode *inode;
761 struct dentry *dentry;
762 int ret = 0;
763
764 inode = nilfs_iget(sb, root, NILFS_ROOT_INO);
765 if (IS_ERR(inode)) {
766 printk(KERN_ERR "NILFS: get root inode failed\n");
767 ret = PTR_ERR(inode);
768 goto out;
769 }
770 if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) {
771 iput(inode);
772 printk(KERN_ERR "NILFS: corrupt root inode.\n");
773 ret = -EINVAL;
774 goto out;
775 }
776
777 if (root->cno == NILFS_CPTREE_CURRENT_CNO) {
778 dentry = d_find_alias(inode);
779 if (!dentry) {
780 dentry = d_alloc_root(inode);
781 if (!dentry) {
782 iput(inode);
783 ret = -ENOMEM;
784 goto failed_dentry;
785 }
786 } else {
787 iput(inode);
788 }
789 } else {
790 dentry = d_obtain_alias(inode);
791 if (IS_ERR(dentry)) {
792 ret = PTR_ERR(dentry);
793 goto failed_dentry;
794 }
795 }
796 *root_dentry = dentry;
797 out:
798 return ret;
799
800 failed_dentry:
801 printk(KERN_ERR "NILFS: get root dentry failed\n");
802 goto out;
803}
804
805static int nilfs_attach_snapshot(struct super_block *s, __u64 cno,
806 struct dentry **root_dentry)
807{
808 struct the_nilfs *nilfs = NILFS_SB(s)->s_nilfs;
809 struct nilfs_root *root;
810 int ret;
811
812 down_read(&nilfs->ns_segctor_sem);
813 ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno);
814 up_read(&nilfs->ns_segctor_sem);
815 if (ret < 0) {
816 ret = (ret == -ENOENT) ? -EINVAL : ret;
817 goto out;
818 } else if (!ret) {
819 printk(KERN_ERR "NILFS: The specified checkpoint is "
820 "not a snapshot (checkpoint number=%llu).\n",
821 (unsigned long long)cno);
822 ret = -EINVAL;
823 goto out;
824 }
825
826 ret = nilfs_attach_checkpoint(NILFS_SB(s), cno, false, &root);
827 if (ret) {
828 printk(KERN_ERR "NILFS: error loading snapshot "
829 "(checkpoint number=%llu).\n",
830 (unsigned long long)cno);
831 goto out;
832 }
833 ret = nilfs_get_root_dentry(s, root, root_dentry);
834 nilfs_put_root(root);
835 out:
836 return ret;
837}
838
839static int nilfs_tree_was_touched(struct dentry *root_dentry)
840{
841 return atomic_read(&root_dentry->d_count) > 1;
842}
843
844/**
845 * nilfs_try_to_shrink_tree() - try to shrink dentries of a checkpoint
846 * @root_dentry: root dentry of the tree to be shrunk
847 *
848 * This function returns true if the tree was in-use.
849 */
850static int nilfs_try_to_shrink_tree(struct dentry *root_dentry)
851{
852 if (have_submounts(root_dentry))
853 return true;
854 shrink_dcache_parent(root_dentry);
855 return nilfs_tree_was_touched(root_dentry);
856}
857
858int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno)
859{
860 struct the_nilfs *nilfs = NILFS_SB(sb)->s_nilfs;
861 struct nilfs_root *root;
862 struct inode *inode;
863 struct dentry *dentry;
864 int ret;
865
866 if (cno < 0 || cno > nilfs->ns_cno)
867 return false;
868
869 if (cno >= nilfs_last_cno(nilfs))
870 return true; /* protect recent checkpoints */
871
872 ret = false;
873 root = nilfs_lookup_root(NILFS_SB(sb)->s_nilfs, cno);
874 if (root) {
875 inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO);
876 if (inode) {
877 dentry = d_find_alias(inode);
878 if (dentry) {
879 if (nilfs_tree_was_touched(dentry))
880 ret = nilfs_try_to_shrink_tree(dentry);
881 dput(dentry);
882 }
883 iput(inode);
884 }
885 nilfs_put_root(root);
886 }
887 return ret;
888}
889
788/** 890/**
789 * nilfs_fill_super() - initialize a super block instance 891 * nilfs_fill_super() - initialize a super block instance
790 * @sb: super_block 892 * @sb: super_block
791 * @data: mount options 893 * @data: mount options
792 * @silent: silent mode flag 894 * @silent: silent mode flag
793 * @nilfs: the_nilfs struct
794 * 895 *
795 * This function is called exclusively by nilfs->ns_mount_mutex. 896 * This function is called exclusively by nilfs->ns_mount_mutex.
796 * So, the recovery process is protected from other simultaneous mounts. 897 * So, the recovery process is protected from other simultaneous mounts.
797 */ 898 */
798static int 899static int
799nilfs_fill_super(struct super_block *sb, void *data, int silent, 900nilfs_fill_super(struct super_block *sb, void *data, int silent)
800 struct the_nilfs *nilfs)
801{ 901{
902 struct the_nilfs *nilfs;
802 struct nilfs_sb_info *sbi; 903 struct nilfs_sb_info *sbi;
803 struct inode *root; 904 struct nilfs_root *fsroot;
905 struct backing_dev_info *bdi;
804 __u64 cno; 906 __u64 cno;
805 int err; 907 int err;
806 908
@@ -809,19 +911,21 @@ nilfs_fill_super(struct super_block *sb, void *data, int silent,
809 return -ENOMEM; 911 return -ENOMEM;
810 912
811 sb->s_fs_info = sbi; 913 sb->s_fs_info = sbi;
914 sbi->s_super = sb;
812 915
813 get_nilfs(nilfs); 916 nilfs = alloc_nilfs(sb->s_bdev);
917 if (!nilfs) {
918 err = -ENOMEM;
919 goto failed_sbi;
920 }
814 sbi->s_nilfs = nilfs; 921 sbi->s_nilfs = nilfs;
815 sbi->s_super = sb;
816 atomic_set(&sbi->s_count, 1);
817 922
818 err = init_nilfs(nilfs, sbi, (char *)data); 923 err = init_nilfs(nilfs, sbi, (char *)data);
819 if (err) 924 if (err)
820 goto failed_sbi; 925 goto failed_nilfs;
821 926
822 spin_lock_init(&sbi->s_inode_lock); 927 spin_lock_init(&sbi->s_inode_lock);
823 INIT_LIST_HEAD(&sbi->s_dirty_files); 928 INIT_LIST_HEAD(&sbi->s_dirty_files);
824 INIT_LIST_HEAD(&sbi->s_list);
825 929
826 /* 930 /*
827 * Following initialization is overlapped because 931 * Following initialization is overlapped because
@@ -837,94 +941,59 @@ nilfs_fill_super(struct super_block *sb, void *data, int silent,
837 sb->s_export_op = &nilfs_export_ops; 941 sb->s_export_op = &nilfs_export_ops;
838 sb->s_root = NULL; 942 sb->s_root = NULL;
839 sb->s_time_gran = 1; 943 sb->s_time_gran = 1;
840 sb->s_bdi = nilfs->ns_bdi; 944
945 bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
946 sb->s_bdi = bdi ? : &default_backing_dev_info;
841 947
842 err = load_nilfs(nilfs, sbi); 948 err = load_nilfs(nilfs, sbi);
843 if (err) 949 if (err)
844 goto failed_sbi; 950 goto failed_nilfs;
845 951
846 cno = nilfs_last_cno(nilfs); 952 cno = nilfs_last_cno(nilfs);
847 953 err = nilfs_attach_checkpoint(sbi, cno, true, &fsroot);
848 if (sb->s_flags & MS_RDONLY) {
849 if (nilfs_test_opt(sbi, SNAPSHOT)) {
850 down_read(&nilfs->ns_segctor_sem);
851 err = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile,
852 sbi->s_snapshot_cno);
853 up_read(&nilfs->ns_segctor_sem);
854 if (err < 0) {
855 if (err == -ENOENT)
856 err = -EINVAL;
857 goto failed_sbi;
858 }
859 if (!err) {
860 printk(KERN_ERR
861 "NILFS: The specified checkpoint is "
862 "not a snapshot "
863 "(checkpoint number=%llu).\n",
864 (unsigned long long)sbi->s_snapshot_cno);
865 err = -EINVAL;
866 goto failed_sbi;
867 }
868 cno = sbi->s_snapshot_cno;
869 }
870 }
871
872 err = nilfs_attach_checkpoint(sbi, cno);
873 if (err) { 954 if (err) {
874 printk(KERN_ERR "NILFS: error loading a checkpoint" 955 printk(KERN_ERR "NILFS: error loading last checkpoint "
875 " (checkpoint number=%llu).\n", (unsigned long long)cno); 956 "(checkpoint number=%llu).\n", (unsigned long long)cno);
876 goto failed_sbi; 957 goto failed_unload;
877 } 958 }
878 959
879 if (!(sb->s_flags & MS_RDONLY)) { 960 if (!(sb->s_flags & MS_RDONLY)) {
880 err = nilfs_attach_segment_constructor(sbi); 961 err = nilfs_attach_segment_constructor(sbi, fsroot);
881 if (err) 962 if (err)
882 goto failed_checkpoint; 963 goto failed_checkpoint;
883 } 964 }
884 965
885 root = nilfs_iget(sb, NILFS_ROOT_INO); 966 err = nilfs_get_root_dentry(sb, fsroot, &sb->s_root);
886 if (IS_ERR(root)) { 967 if (err)
887 printk(KERN_ERR "NILFS: get root inode failed\n");
888 err = PTR_ERR(root);
889 goto failed_segctor;
890 }
891 if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
892 iput(root);
893 printk(KERN_ERR "NILFS: corrupt root inode.\n");
894 err = -EINVAL;
895 goto failed_segctor;
896 }
897 sb->s_root = d_alloc_root(root);
898 if (!sb->s_root) {
899 iput(root);
900 printk(KERN_ERR "NILFS: get root dentry failed\n");
901 err = -ENOMEM;
902 goto failed_segctor; 968 goto failed_segctor;
903 } 969
970 nilfs_put_root(fsroot);
904 971
905 if (!(sb->s_flags & MS_RDONLY)) { 972 if (!(sb->s_flags & MS_RDONLY)) {
906 down_write(&nilfs->ns_sem); 973 down_write(&nilfs->ns_sem);
907 nilfs_setup_super(sbi); 974 nilfs_setup_super(sbi, true);
908 up_write(&nilfs->ns_sem); 975 up_write(&nilfs->ns_sem);
909 } 976 }
910 977
911 down_write(&nilfs->ns_super_sem);
912 if (!nilfs_test_opt(sbi, SNAPSHOT))
913 nilfs->ns_current = sbi;
914 up_write(&nilfs->ns_super_sem);
915
916 return 0; 978 return 0;
917 979
918 failed_segctor: 980 failed_segctor:
919 nilfs_detach_segment_constructor(sbi); 981 nilfs_detach_segment_constructor(sbi);
920 982
921 failed_checkpoint: 983 failed_checkpoint:
922 nilfs_detach_checkpoint(sbi); 984 nilfs_put_root(fsroot);
985
986 failed_unload:
987 iput(nilfs->ns_sufile);
988 iput(nilfs->ns_cpfile);
989 iput(nilfs->ns_dat);
990
991 failed_nilfs:
992 destroy_nilfs(nilfs);
923 993
924 failed_sbi: 994 failed_sbi:
925 put_nilfs(nilfs);
926 sb->s_fs_info = NULL; 995 sb->s_fs_info = NULL;
927 nilfs_put_sbinfo(sbi); 996 kfree(sbi);
928 return err; 997 return err;
929} 998}
930 999
@@ -934,13 +1003,10 @@ static int nilfs_remount(struct super_block *sb, int *flags, char *data)
934 struct the_nilfs *nilfs = sbi->s_nilfs; 1003 struct the_nilfs *nilfs = sbi->s_nilfs;
935 unsigned long old_sb_flags; 1004 unsigned long old_sb_flags;
936 struct nilfs_mount_options old_opts; 1005 struct nilfs_mount_options old_opts;
937 int was_snapshot, err; 1006 int err;
938 1007
939 down_write(&nilfs->ns_super_sem);
940 old_sb_flags = sb->s_flags; 1008 old_sb_flags = sb->s_flags;
941 old_opts.mount_opt = sbi->s_mount_opt; 1009 old_opts.mount_opt = sbi->s_mount_opt;
942 old_opts.snapshot_cno = sbi->s_snapshot_cno;
943 was_snapshot = nilfs_test_opt(sbi, SNAPSHOT);
944 1010
945 if (!parse_options(data, sb, 1)) { 1011 if (!parse_options(data, sb, 1)) {
946 err = -EINVAL; 1012 err = -EINVAL;
@@ -949,11 +1015,6 @@ static int nilfs_remount(struct super_block *sb, int *flags, char *data)
949 sb->s_flags = (sb->s_flags & ~MS_POSIXACL); 1015 sb->s_flags = (sb->s_flags & ~MS_POSIXACL);
950 1016
951 err = -EINVAL; 1017 err = -EINVAL;
952 if (was_snapshot && !(*flags & MS_RDONLY)) {
953 printk(KERN_ERR "NILFS (device %s): cannot remount snapshot "
954 "read/write.\n", sb->s_id);
955 goto restore_opts;
956 }
957 1018
958 if (!nilfs_valid_fs(nilfs)) { 1019 if (!nilfs_valid_fs(nilfs)) {
959 printk(KERN_WARNING "NILFS (device %s): couldn't " 1020 printk(KERN_WARNING "NILFS (device %s): couldn't "
@@ -978,6 +1039,7 @@ static int nilfs_remount(struct super_block *sb, int *flags, char *data)
978 up_write(&nilfs->ns_sem); 1039 up_write(&nilfs->ns_sem);
979 } else { 1040 } else {
980 __u64 features; 1041 __u64 features;
1042 struct nilfs_root *root;
981 1043
982 /* 1044 /*
983 * Mounting a RDONLY partition read-write, so reread and 1045 * Mounting a RDONLY partition read-write, so reread and
@@ -999,23 +1061,21 @@ static int nilfs_remount(struct super_block *sb, int *flags, char *data)
999 1061
1000 sb->s_flags &= ~MS_RDONLY; 1062 sb->s_flags &= ~MS_RDONLY;
1001 1063
1002 err = nilfs_attach_segment_constructor(sbi); 1064 root = NILFS_I(sb->s_root->d_inode)->i_root;
1065 err = nilfs_attach_segment_constructor(sbi, root);
1003 if (err) 1066 if (err)
1004 goto restore_opts; 1067 goto restore_opts;
1005 1068
1006 down_write(&nilfs->ns_sem); 1069 down_write(&nilfs->ns_sem);
1007 nilfs_setup_super(sbi); 1070 nilfs_setup_super(sbi, true);
1008 up_write(&nilfs->ns_sem); 1071 up_write(&nilfs->ns_sem);
1009 } 1072 }
1010 out: 1073 out:
1011 up_write(&nilfs->ns_super_sem);
1012 return 0; 1074 return 0;
1013 1075
1014 restore_opts: 1076 restore_opts:
1015 sb->s_flags = old_sb_flags; 1077 sb->s_flags = old_sb_flags;
1016 sbi->s_mount_opt = old_opts.mount_opt; 1078 sbi->s_mount_opt = old_opts.mount_opt;
1017 sbi->s_snapshot_cno = old_opts.snapshot_cno;
1018 up_write(&nilfs->ns_super_sem);
1019 return err; 1079 return err;
1020} 1080}
1021 1081
@@ -1035,7 +1095,7 @@ static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1035{ 1095{
1036 char *p, *options = data; 1096 char *p, *options = data;
1037 substring_t args[MAX_OPT_ARGS]; 1097 substring_t args[MAX_OPT_ARGS];
1038 int option, token; 1098 int token;
1039 int ret = 0; 1099 int ret = 0;
1040 1100
1041 do { 1101 do {
@@ -1043,16 +1103,18 @@ static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1043 if (p != NULL && *p) { 1103 if (p != NULL && *p) {
1044 token = match_token(p, tokens, args); 1104 token = match_token(p, tokens, args);
1045 if (token == Opt_snapshot) { 1105 if (token == Opt_snapshot) {
1046 if (!(sd->flags & MS_RDONLY)) 1106 if (!(sd->flags & MS_RDONLY)) {
1047 ret++; 1107 ret++;
1048 else { 1108 } else {
1049 ret = match_int(&args[0], &option); 1109 sd->cno = simple_strtoull(args[0].from,
1050 if (!ret) { 1110 NULL, 0);
1051 if (option > 0) 1111 /*
1052 sd->cno = option; 1112 * No need to see the end pointer;
1053 else 1113 * match_token() has done syntax
1054 ret++; 1114 * checking.
1055 } 1115 */
1116 if (sd->cno == 0)
1117 ret++;
1056 } 1118 }
1057 } 1119 }
1058 if (ret) 1120 if (ret)
@@ -1069,18 +1131,14 @@ static int nilfs_identify(char *data, struct nilfs_super_data *sd)
1069 1131
1070static int nilfs_set_bdev_super(struct super_block *s, void *data) 1132static int nilfs_set_bdev_super(struct super_block *s, void *data)
1071{ 1133{
1072 struct nilfs_super_data *sd = data; 1134 s->s_bdev = data;
1073
1074 s->s_bdev = sd->bdev;
1075 s->s_dev = s->s_bdev->bd_dev; 1135 s->s_dev = s->s_bdev->bd_dev;
1076 return 0; 1136 return 0;
1077} 1137}
1078 1138
1079static int nilfs_test_bdev_super(struct super_block *s, void *data) 1139static int nilfs_test_bdev_super(struct super_block *s, void *data)
1080{ 1140{
1081 struct nilfs_super_data *sd = data; 1141 return (void *)s->s_bdev == data;
1082
1083 return sd->sbi && s->s_fs_info == (void *)sd->sbi;
1084} 1142}
1085 1143
1086static int 1144static int
@@ -1090,8 +1148,8 @@ nilfs_get_sb(struct file_system_type *fs_type, int flags,
1090 struct nilfs_super_data sd; 1148 struct nilfs_super_data sd;
1091 struct super_block *s; 1149 struct super_block *s;
1092 fmode_t mode = FMODE_READ; 1150 fmode_t mode = FMODE_READ;
1093 struct the_nilfs *nilfs; 1151 struct dentry *root_dentry;
1094 int err, need_to_close = 1; 1152 int err, s_new = false;
1095 1153
1096 if (!(flags & MS_RDONLY)) 1154 if (!(flags & MS_RDONLY))
1097 mode |= FMODE_WRITE; 1155 mode |= FMODE_WRITE;
@@ -1100,12 +1158,6 @@ nilfs_get_sb(struct file_system_type *fs_type, int flags,
1100 if (IS_ERR(sd.bdev)) 1158 if (IS_ERR(sd.bdev))
1101 return PTR_ERR(sd.bdev); 1159 return PTR_ERR(sd.bdev);
1102 1160
1103 /*
1104 * To get mount instance using sget() vfs-routine, NILFS needs
1105 * much more information than normal filesystems to identify mount
1106 * instance. For snapshot mounts, not only a mount type (ro-mount
1107 * or rw-mount) but also a checkpoint number is required.
1108 */
1109 sd.cno = 0; 1161 sd.cno = 0;
1110 sd.flags = flags; 1162 sd.flags = flags;
1111 if (nilfs_identify((char *)data, &sd)) { 1163 if (nilfs_identify((char *)data, &sd)) {
@@ -1113,93 +1165,86 @@ nilfs_get_sb(struct file_system_type *fs_type, int flags,
1113 goto failed; 1165 goto failed;
1114 } 1166 }
1115 1167
1116 nilfs = find_or_create_nilfs(sd.bdev);
1117 if (!nilfs) {
1118 err = -ENOMEM;
1119 goto failed;
1120 }
1121
1122 mutex_lock(&nilfs->ns_mount_mutex);
1123
1124 if (!sd.cno) {
1125 /*
1126 * Check if an exclusive mount exists or not.
1127 * Snapshot mounts coexist with a current mount
1128 * (i.e. rw-mount or ro-mount), whereas rw-mount and
1129 * ro-mount are mutually exclusive.
1130 */
1131 down_read(&nilfs->ns_super_sem);
1132 if (nilfs->ns_current &&
1133 ((nilfs->ns_current->s_super->s_flags ^ flags)
1134 & MS_RDONLY)) {
1135 up_read(&nilfs->ns_super_sem);
1136 err = -EBUSY;
1137 goto failed_unlock;
1138 }
1139 up_read(&nilfs->ns_super_sem);
1140 }
1141
1142 /*
1143 * Find existing nilfs_sb_info struct
1144 */
1145 sd.sbi = nilfs_find_sbinfo(nilfs, !(flags & MS_RDONLY), sd.cno);
1146
1147 /* 1168 /*
1148 * Get super block instance holding the nilfs_sb_info struct. 1169 * once the super is inserted into the list by sget, s_umount
1149 * A new instance is allocated if no existing mount is present or 1170 * will protect the lockfs code from trying to start a snapshot
1150 * existing instance has been unmounted. 1171 * while we are mounting
1151 */ 1172 */
1152 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, &sd); 1173 mutex_lock(&sd.bdev->bd_fsfreeze_mutex);
1153 if (sd.sbi) 1174 if (sd.bdev->bd_fsfreeze_count > 0) {
1154 nilfs_put_sbinfo(sd.sbi); 1175 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1155 1176 err = -EBUSY;
1177 goto failed;
1178 }
1179 s = sget(fs_type, nilfs_test_bdev_super, nilfs_set_bdev_super, sd.bdev);
1180 mutex_unlock(&sd.bdev->bd_fsfreeze_mutex);
1156 if (IS_ERR(s)) { 1181 if (IS_ERR(s)) {
1157 err = PTR_ERR(s); 1182 err = PTR_ERR(s);
1158 goto failed_unlock; 1183 goto failed;
1159 } 1184 }
1160 1185
1161 if (!s->s_root) { 1186 if (!s->s_root) {
1162 char b[BDEVNAME_SIZE]; 1187 char b[BDEVNAME_SIZE];
1163 1188
1189 s_new = true;
1190
1164 /* New superblock instance created */ 1191 /* New superblock instance created */
1165 s->s_flags = flags; 1192 s->s_flags = flags;
1166 s->s_mode = mode; 1193 s->s_mode = mode;
1167 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id)); 1194 strlcpy(s->s_id, bdevname(sd.bdev, b), sizeof(s->s_id));
1168 sb_set_blocksize(s, block_size(sd.bdev)); 1195 sb_set_blocksize(s, block_size(sd.bdev));
1169 1196
1170 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0, 1197 err = nilfs_fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1171 nilfs);
1172 if (err) 1198 if (err)
1173 goto cancel_new; 1199 goto failed_super;
1174 1200
1175 s->s_flags |= MS_ACTIVE; 1201 s->s_flags |= MS_ACTIVE;
1176 need_to_close = 0; 1202 } else if (!sd.cno) {
1203 int busy = false;
1204
1205 if (nilfs_tree_was_touched(s->s_root)) {
1206 busy = nilfs_try_to_shrink_tree(s->s_root);
1207 if (busy && (flags ^ s->s_flags) & MS_RDONLY) {
1208 printk(KERN_ERR "NILFS: the device already "
1209 "has a %s mount.\n",
1210 (s->s_flags & MS_RDONLY) ?
1211 "read-only" : "read/write");
1212 err = -EBUSY;
1213 goto failed_super;
1214 }
1215 }
1216 if (!busy) {
1217 /*
1218 * Try remount to setup mount states if the current
1219 * tree is not mounted and only snapshots use this sb.
1220 */
1221 err = nilfs_remount(s, &flags, data);
1222 if (err)
1223 goto failed_super;
1224 }
1177 } 1225 }
1178 1226
1179 mutex_unlock(&nilfs->ns_mount_mutex); 1227 if (sd.cno) {
1180 put_nilfs(nilfs); 1228 err = nilfs_attach_snapshot(s, sd.cno, &root_dentry);
1181 if (need_to_close) 1229 if (err)
1230 goto failed_super;
1231 } else {
1232 root_dentry = dget(s->s_root);
1233 }
1234
1235 if (!s_new)
1182 close_bdev_exclusive(sd.bdev, mode); 1236 close_bdev_exclusive(sd.bdev, mode);
1183 simple_set_mnt(mnt, s);
1184 return 0;
1185 1237
1186 failed_unlock: 1238 mnt->mnt_sb = s;
1187 mutex_unlock(&nilfs->ns_mount_mutex); 1239 mnt->mnt_root = root_dentry;
1188 put_nilfs(nilfs); 1240 return 0;
1189 failed:
1190 close_bdev_exclusive(sd.bdev, mode);
1191 return err;
1192 1241
1193 cancel_new: 1242 failed_super:
1194 /* Abandoning the newly allocated superblock */
1195 mutex_unlock(&nilfs->ns_mount_mutex);
1196 put_nilfs(nilfs);
1197 deactivate_locked_super(s); 1243 deactivate_locked_super(s);
1198 /* 1244
1199 * deactivate_locked_super() invokes close_bdev_exclusive(). 1245 failed:
1200 * We must finish all post-cleaning before this call; 1246 if (!s_new)
1201 * put_nilfs() needs the block device. 1247 close_bdev_exclusive(sd.bdev, mode);
1202 */
1203 return err; 1248 return err;
1204} 1249}
1205 1250
diff --git a/fs/nilfs2/the_nilfs.c b/fs/nilfs2/the_nilfs.c
index d27715103376..0254be2d73c6 100644
--- a/fs/nilfs2/the_nilfs.c
+++ b/fs/nilfs2/the_nilfs.c
@@ -35,9 +35,6 @@
35#include "segbuf.h" 35#include "segbuf.h"
36 36
37 37
38static LIST_HEAD(nilfs_objects);
39static DEFINE_SPINLOCK(nilfs_lock);
40
41static int nilfs_valid_sb(struct nilfs_super_block *sbp); 38static int nilfs_valid_sb(struct nilfs_super_block *sbp);
42 39
43void nilfs_set_last_segment(struct the_nilfs *nilfs, 40void nilfs_set_last_segment(struct the_nilfs *nilfs,
@@ -61,16 +58,13 @@ void nilfs_set_last_segment(struct the_nilfs *nilfs,
61} 58}
62 59
63/** 60/**
64 * alloc_nilfs - allocate the_nilfs structure 61 * alloc_nilfs - allocate a nilfs object
65 * @bdev: block device to which the_nilfs is related 62 * @bdev: block device to which the_nilfs is related
66 * 63 *
67 * alloc_nilfs() allocates memory for the_nilfs and
68 * initializes its reference count and locks.
69 *
70 * Return Value: On success, pointer to the_nilfs is returned. 64 * Return Value: On success, pointer to the_nilfs is returned.
71 * On error, NULL is returned. 65 * On error, NULL is returned.
72 */ 66 */
73static struct the_nilfs *alloc_nilfs(struct block_device *bdev) 67struct the_nilfs *alloc_nilfs(struct block_device *bdev)
74{ 68{
75 struct the_nilfs *nilfs; 69 struct the_nilfs *nilfs;
76 70
@@ -79,103 +73,38 @@ static struct the_nilfs *alloc_nilfs(struct block_device *bdev)
79 return NULL; 73 return NULL;
80 74
81 nilfs->ns_bdev = bdev; 75 nilfs->ns_bdev = bdev;
82 atomic_set(&nilfs->ns_count, 1);
83 atomic_set(&nilfs->ns_ndirtyblks, 0); 76 atomic_set(&nilfs->ns_ndirtyblks, 0);
84 init_rwsem(&nilfs->ns_sem); 77 init_rwsem(&nilfs->ns_sem);
85 init_rwsem(&nilfs->ns_super_sem); 78 INIT_LIST_HEAD(&nilfs->ns_gc_inodes);
86 mutex_init(&nilfs->ns_mount_mutex);
87 init_rwsem(&nilfs->ns_writer_sem);
88 INIT_LIST_HEAD(&nilfs->ns_list);
89 INIT_LIST_HEAD(&nilfs->ns_supers);
90 spin_lock_init(&nilfs->ns_last_segment_lock); 79 spin_lock_init(&nilfs->ns_last_segment_lock);
91 nilfs->ns_gc_inodes_h = NULL; 80 nilfs->ns_cptree = RB_ROOT;
81 spin_lock_init(&nilfs->ns_cptree_lock);
92 init_rwsem(&nilfs->ns_segctor_sem); 82 init_rwsem(&nilfs->ns_segctor_sem);
93 83
94 return nilfs; 84 return nilfs;
95} 85}
96 86
97/** 87/**
98 * find_or_create_nilfs - find or create nilfs object 88 * destroy_nilfs - destroy nilfs object
99 * @bdev: block device to which the_nilfs is related 89 * @nilfs: nilfs object to be released
100 *
101 * find_nilfs() looks up an existent nilfs object created on the
102 * device and gets the reference count of the object. If no nilfs object
103 * is found on the device, a new nilfs object is allocated.
104 *
105 * Return Value: On success, pointer to the nilfs object is returned.
106 * On error, NULL is returned.
107 */
108struct the_nilfs *find_or_create_nilfs(struct block_device *bdev)
109{
110 struct the_nilfs *nilfs, *new = NULL;
111
112 retry:
113 spin_lock(&nilfs_lock);
114 list_for_each_entry(nilfs, &nilfs_objects, ns_list) {
115 if (nilfs->ns_bdev == bdev) {
116 get_nilfs(nilfs);
117 spin_unlock(&nilfs_lock);
118 if (new)
119 put_nilfs(new);
120 return nilfs; /* existing object */
121 }
122 }
123 if (new) {
124 list_add_tail(&new->ns_list, &nilfs_objects);
125 spin_unlock(&nilfs_lock);
126 return new; /* new object */
127 }
128 spin_unlock(&nilfs_lock);
129
130 new = alloc_nilfs(bdev);
131 if (new)
132 goto retry;
133 return NULL; /* insufficient memory */
134}
135
136/**
137 * put_nilfs - release a reference to the_nilfs
138 * @nilfs: the_nilfs structure to be released
139 *
140 * put_nilfs() decrements a reference counter of the_nilfs.
141 * If the reference count reaches zero, the_nilfs is freed.
142 */ 90 */
143void put_nilfs(struct the_nilfs *nilfs) 91void destroy_nilfs(struct the_nilfs *nilfs)
144{ 92{
145 spin_lock(&nilfs_lock);
146 if (!atomic_dec_and_test(&nilfs->ns_count)) {
147 spin_unlock(&nilfs_lock);
148 return;
149 }
150 list_del_init(&nilfs->ns_list);
151 spin_unlock(&nilfs_lock);
152
153 /*
154 * Increment of ns_count never occurs below because the caller
155 * of get_nilfs() holds at least one reference to the_nilfs.
156 * Thus its exclusion control is not required here.
157 */
158
159 might_sleep(); 93 might_sleep();
160 if (nilfs_loaded(nilfs)) {
161 nilfs_mdt_destroy(nilfs->ns_sufile);
162 nilfs_mdt_destroy(nilfs->ns_cpfile);
163 nilfs_mdt_destroy(nilfs->ns_dat);
164 nilfs_mdt_destroy(nilfs->ns_gc_dat);
165 }
166 if (nilfs_init(nilfs)) { 94 if (nilfs_init(nilfs)) {
167 nilfs_destroy_gccache(nilfs);
168 brelse(nilfs->ns_sbh[0]); 95 brelse(nilfs->ns_sbh[0]);
169 brelse(nilfs->ns_sbh[1]); 96 brelse(nilfs->ns_sbh[1]);
170 } 97 }
171 kfree(nilfs); 98 kfree(nilfs);
172} 99}
173 100
174static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block) 101static int nilfs_load_super_root(struct the_nilfs *nilfs,
102 struct super_block *sb, sector_t sr_block)
175{ 103{
176 struct buffer_head *bh_sr; 104 struct buffer_head *bh_sr;
177 struct nilfs_super_root *raw_sr; 105 struct nilfs_super_root *raw_sr;
178 struct nilfs_super_block **sbp = nilfs->ns_sbp; 106 struct nilfs_super_block **sbp = nilfs->ns_sbp;
107 struct nilfs_inode *rawi;
179 unsigned dat_entry_size, segment_usage_size, checkpoint_size; 108 unsigned dat_entry_size, segment_usage_size, checkpoint_size;
180 unsigned inode_size; 109 unsigned inode_size;
181 int err; 110 int err;
@@ -192,40 +121,22 @@ static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
192 121
193 inode_size = nilfs->ns_inode_size; 122 inode_size = nilfs->ns_inode_size;
194 123
195 err = -ENOMEM; 124 rawi = (void *)bh_sr->b_data + NILFS_SR_DAT_OFFSET(inode_size);
196 nilfs->ns_dat = nilfs_dat_new(nilfs, dat_entry_size); 125 err = nilfs_dat_read(sb, dat_entry_size, rawi, &nilfs->ns_dat);
197 if (unlikely(!nilfs->ns_dat)) 126 if (err)
198 goto failed; 127 goto failed;
199 128
200 nilfs->ns_gc_dat = nilfs_dat_new(nilfs, dat_entry_size); 129 rawi = (void *)bh_sr->b_data + NILFS_SR_CPFILE_OFFSET(inode_size);
201 if (unlikely(!nilfs->ns_gc_dat)) 130 err = nilfs_cpfile_read(sb, checkpoint_size, rawi, &nilfs->ns_cpfile);
131 if (err)
202 goto failed_dat; 132 goto failed_dat;
203 133
204 nilfs->ns_cpfile = nilfs_cpfile_new(nilfs, checkpoint_size); 134 rawi = (void *)bh_sr->b_data + NILFS_SR_SUFILE_OFFSET(inode_size);
205 if (unlikely(!nilfs->ns_cpfile)) 135 err = nilfs_sufile_read(sb, segment_usage_size, rawi,
206 goto failed_gc_dat; 136 &nilfs->ns_sufile);
207 137 if (err)
208 nilfs->ns_sufile = nilfs_sufile_new(nilfs, segment_usage_size);
209 if (unlikely(!nilfs->ns_sufile))
210 goto failed_cpfile; 138 goto failed_cpfile;
211 139
212 nilfs_mdt_set_shadow(nilfs->ns_dat, nilfs->ns_gc_dat);
213
214 err = nilfs_dat_read(nilfs->ns_dat, (void *)bh_sr->b_data +
215 NILFS_SR_DAT_OFFSET(inode_size));
216 if (unlikely(err))
217 goto failed_sufile;
218
219 err = nilfs_cpfile_read(nilfs->ns_cpfile, (void *)bh_sr->b_data +
220 NILFS_SR_CPFILE_OFFSET(inode_size));
221 if (unlikely(err))
222 goto failed_sufile;
223
224 err = nilfs_sufile_read(nilfs->ns_sufile, (void *)bh_sr->b_data +
225 NILFS_SR_SUFILE_OFFSET(inode_size));
226 if (unlikely(err))
227 goto failed_sufile;
228
229 raw_sr = (struct nilfs_super_root *)bh_sr->b_data; 140 raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
230 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime); 141 nilfs->ns_nongc_ctime = le64_to_cpu(raw_sr->sr_nongc_ctime);
231 142
@@ -233,17 +144,11 @@ static int nilfs_load_super_root(struct the_nilfs *nilfs, sector_t sr_block)
233 brelse(bh_sr); 144 brelse(bh_sr);
234 return err; 145 return err;
235 146
236 failed_sufile:
237 nilfs_mdt_destroy(nilfs->ns_sufile);
238
239 failed_cpfile: 147 failed_cpfile:
240 nilfs_mdt_destroy(nilfs->ns_cpfile); 148 iput(nilfs->ns_cpfile);
241
242 failed_gc_dat:
243 nilfs_mdt_destroy(nilfs->ns_gc_dat);
244 149
245 failed_dat: 150 failed_dat:
246 nilfs_mdt_destroy(nilfs->ns_dat); 151 iput(nilfs->ns_dat);
247 goto failed; 152 goto failed;
248} 153}
249 154
@@ -306,15 +211,6 @@ int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
306 int valid_fs = nilfs_valid_fs(nilfs); 211 int valid_fs = nilfs_valid_fs(nilfs);
307 int err; 212 int err;
308 213
309 if (nilfs_loaded(nilfs)) {
310 if (valid_fs ||
311 ((s_flags & MS_RDONLY) && nilfs_test_opt(sbi, NORECOVERY)))
312 return 0;
313 printk(KERN_ERR "NILFS: the filesystem is in an incomplete "
314 "recovery state.\n");
315 return -EINVAL;
316 }
317
318 if (!valid_fs) { 214 if (!valid_fs) {
319 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n"); 215 printk(KERN_WARNING "NILFS warning: mounting unchecked fs\n");
320 if (s_flags & MS_RDONLY) { 216 if (s_flags & MS_RDONLY) {
@@ -375,7 +271,7 @@ int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
375 goto scan_error; 271 goto scan_error;
376 } 272 }
377 273
378 err = nilfs_load_super_root(nilfs, ri.ri_super_root); 274 err = nilfs_load_super_root(nilfs, sbi->s_super, ri.ri_super_root);
379 if (unlikely(err)) { 275 if (unlikely(err)) {
380 printk(KERN_ERR "NILFS: error loading super root.\n"); 276 printk(KERN_ERR "NILFS: error loading super root.\n");
381 goto failed; 277 goto failed;
@@ -443,10 +339,9 @@ int load_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
443 goto failed; 339 goto failed;
444 340
445 failed_unload: 341 failed_unload:
446 nilfs_mdt_destroy(nilfs->ns_cpfile); 342 iput(nilfs->ns_cpfile);
447 nilfs_mdt_destroy(nilfs->ns_sufile); 343 iput(nilfs->ns_sufile);
448 nilfs_mdt_destroy(nilfs->ns_dat); 344 iput(nilfs->ns_dat);
449 nilfs_mdt_destroy(nilfs->ns_gc_dat);
450 345
451 failed: 346 failed:
452 nilfs_clear_recovery_info(&ri); 347 nilfs_clear_recovery_info(&ri);
@@ -468,8 +363,8 @@ static unsigned long long nilfs_max_size(unsigned int blkbits)
468static int nilfs_store_disk_layout(struct the_nilfs *nilfs, 363static int nilfs_store_disk_layout(struct the_nilfs *nilfs,
469 struct nilfs_super_block *sbp) 364 struct nilfs_super_block *sbp)
470{ 365{
471 if (le32_to_cpu(sbp->s_rev_level) != NILFS_CURRENT_REV) { 366 if (le32_to_cpu(sbp->s_rev_level) < NILFS_MIN_SUPP_REV) {
472 printk(KERN_ERR "NILFS: revision mismatch " 367 printk(KERN_ERR "NILFS: unsupported revision "
473 "(superblock rev.=%d.%d, current rev.=%d.%d). " 368 "(superblock rev.=%d.%d, current rev.=%d.%d). "
474 "Please check the version of mkfs.nilfs.\n", 369 "Please check the version of mkfs.nilfs.\n",
475 le32_to_cpu(sbp->s_rev_level), 370 le32_to_cpu(sbp->s_rev_level),
@@ -631,12 +526,7 @@ static int nilfs_load_super_block(struct the_nilfs *nilfs,
631 * 526 *
632 * init_nilfs() performs common initialization per block device (e.g. 527 * init_nilfs() performs common initialization per block device (e.g.
633 * reading the super block, getting disk layout information, initializing 528 * reading the super block, getting disk layout information, initializing
634 * shared fields in the_nilfs). It takes on some portion of the jobs 529 * shared fields in the_nilfs).
635 * typically done by a fill_super() routine. This division arises from
636 * the nature that multiple NILFS instances may be simultaneously
637 * mounted on a device.
638 * For multiple mounts on the same device, only the first mount
639 * invokes these tasks.
640 * 530 *
641 * Return Value: On success, 0 is returned. On error, a negative error 531 * Return Value: On success, 0 is returned. On error, a negative error
642 * code is returned. 532 * code is returned.
@@ -645,32 +535,10 @@ int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
645{ 535{
646 struct super_block *sb = sbi->s_super; 536 struct super_block *sb = sbi->s_super;
647 struct nilfs_super_block *sbp; 537 struct nilfs_super_block *sbp;
648 struct backing_dev_info *bdi;
649 int blocksize; 538 int blocksize;
650 int err; 539 int err;
651 540
652 down_write(&nilfs->ns_sem); 541 down_write(&nilfs->ns_sem);
653 if (nilfs_init(nilfs)) {
654 /* Load values from existing the_nilfs */
655 sbp = nilfs->ns_sbp[0];
656 err = nilfs_store_magic_and_option(sb, sbp, data);
657 if (err)
658 goto out;
659
660 err = nilfs_check_feature_compatibility(sb, sbp);
661 if (err)
662 goto out;
663
664 blocksize = BLOCK_SIZE << le32_to_cpu(sbp->s_log_block_size);
665 if (sb->s_blocksize != blocksize &&
666 !sb_set_blocksize(sb, blocksize)) {
667 printk(KERN_ERR "NILFS: blocksize %d unfit to device\n",
668 blocksize);
669 err = -EINVAL;
670 }
671 sb->s_maxbytes = nilfs_max_size(sb->s_blocksize_bits);
672 goto out;
673 }
674 542
675 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE); 543 blocksize = sb_min_blocksize(sb, NILFS_MIN_BLOCK_SIZE);
676 if (!blocksize) { 544 if (!blocksize) {
@@ -729,18 +597,10 @@ int init_nilfs(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, char *data)
729 597
730 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state); 598 nilfs->ns_mount_state = le16_to_cpu(sbp->s_state);
731 599
732 bdi = nilfs->ns_bdev->bd_inode->i_mapping->backing_dev_info;
733 nilfs->ns_bdi = bdi ? : &default_backing_dev_info;
734
735 err = nilfs_store_log_cursor(nilfs, sbp); 600 err = nilfs_store_log_cursor(nilfs, sbp);
736 if (err) 601 if (err)
737 goto failed_sbh; 602 goto failed_sbh;
738 603
739 /* Initialize gcinode cache */
740 err = nilfs_init_gccache(nilfs);
741 if (err)
742 goto failed_sbh;
743
744 set_nilfs_init(nilfs); 604 set_nilfs_init(nilfs);
745 err = 0; 605 err = 0;
746 out: 606 out:
@@ -812,79 +672,92 @@ int nilfs_near_disk_full(struct the_nilfs *nilfs)
812 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs; 672 return ncleansegs <= nilfs->ns_nrsvsegs + nincsegs;
813} 673}
814 674
815/** 675struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno)
816 * nilfs_find_sbinfo - find existing nilfs_sb_info structure
817 * @nilfs: nilfs object
818 * @rw_mount: mount type (non-zero value for read/write mount)
819 * @cno: checkpoint number (zero for read-only mount)
820 *
821 * nilfs_find_sbinfo() returns the nilfs_sb_info structure which
822 * @rw_mount and @cno (in case of snapshots) matched. If no instance
823 * was found, NULL is returned. Although the super block instance can
824 * be unmounted after this function returns, the nilfs_sb_info struct
825 * is kept on memory until nilfs_put_sbinfo() is called.
826 */
827struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *nilfs,
828 int rw_mount, __u64 cno)
829{ 676{
830 struct nilfs_sb_info *sbi; 677 struct rb_node *n;
831 678 struct nilfs_root *root;
832 down_read(&nilfs->ns_super_sem); 679
833 /* 680 spin_lock(&nilfs->ns_cptree_lock);
834 * The SNAPSHOT flag and sb->s_flags are supposed to be 681 n = nilfs->ns_cptree.rb_node;
835 * protected with nilfs->ns_super_sem. 682 while (n) {
836 */ 683 root = rb_entry(n, struct nilfs_root, rb_node);
837 sbi = nilfs->ns_current; 684
838 if (rw_mount) { 685 if (cno < root->cno) {
839 if (sbi && !(sbi->s_super->s_flags & MS_RDONLY)) 686 n = n->rb_left;
840 goto found; /* read/write mount */ 687 } else if (cno > root->cno) {
841 else 688 n = n->rb_right;
842 goto out; 689 } else {
843 } else if (cno == 0) { 690 atomic_inc(&root->count);
844 if (sbi && (sbi->s_super->s_flags & MS_RDONLY)) 691 spin_unlock(&nilfs->ns_cptree_lock);
845 goto found; /* read-only mount */ 692 return root;
846 else 693 }
847 goto out;
848 } 694 }
695 spin_unlock(&nilfs->ns_cptree_lock);
849 696
850 list_for_each_entry(sbi, &nilfs->ns_supers, s_list) {
851 if (nilfs_test_opt(sbi, SNAPSHOT) &&
852 sbi->s_snapshot_cno == cno)
853 goto found; /* snapshot mount */
854 }
855 out:
856 up_read(&nilfs->ns_super_sem);
857 return NULL; 697 return NULL;
858
859 found:
860 atomic_inc(&sbi->s_count);
861 up_read(&nilfs->ns_super_sem);
862 return sbi;
863} 698}
864 699
865int nilfs_checkpoint_is_mounted(struct the_nilfs *nilfs, __u64 cno, 700struct nilfs_root *
866 int snapshot_mount) 701nilfs_find_or_create_root(struct the_nilfs *nilfs, __u64 cno)
867{ 702{
868 struct nilfs_sb_info *sbi; 703 struct rb_node **p, *parent;
869 int ret = 0; 704 struct nilfs_root *root, *new;
870 705
871 down_read(&nilfs->ns_super_sem); 706 root = nilfs_lookup_root(nilfs, cno);
872 if (cno == 0 || cno > nilfs->ns_cno) 707 if (root)
873 goto out_unlock; 708 return root;
874 709
875 list_for_each_entry(sbi, &nilfs->ns_supers, s_list) { 710 new = kmalloc(sizeof(*root), GFP_KERNEL);
876 if (sbi->s_snapshot_cno == cno && 711 if (!new)
877 (!snapshot_mount || nilfs_test_opt(sbi, SNAPSHOT))) { 712 return NULL;
878 /* exclude read-only mounts */ 713
879 ret++; 714 spin_lock(&nilfs->ns_cptree_lock);
880 break; 715
716 p = &nilfs->ns_cptree.rb_node;
717 parent = NULL;
718
719 while (*p) {
720 parent = *p;
721 root = rb_entry(parent, struct nilfs_root, rb_node);
722
723 if (cno < root->cno) {
724 p = &(*p)->rb_left;
725 } else if (cno > root->cno) {
726 p = &(*p)->rb_right;
727 } else {
728 atomic_inc(&root->count);
729 spin_unlock(&nilfs->ns_cptree_lock);
730 kfree(new);
731 return root;
881 } 732 }
882 } 733 }
883 /* for protecting recent checkpoints */
884 if (cno >= nilfs_last_cno(nilfs))
885 ret++;
886 734
887 out_unlock: 735 new->cno = cno;
888 up_read(&nilfs->ns_super_sem); 736 new->ifile = NULL;
889 return ret; 737 new->nilfs = nilfs;
738 atomic_set(&new->count, 1);
739 atomic_set(&new->inodes_count, 0);
740 atomic_set(&new->blocks_count, 0);
741
742 rb_link_node(&new->rb_node, parent, p);
743 rb_insert_color(&new->rb_node, &nilfs->ns_cptree);
744
745 spin_unlock(&nilfs->ns_cptree_lock);
746
747 return new;
748}
749
750void nilfs_put_root(struct nilfs_root *root)
751{
752 if (atomic_dec_and_test(&root->count)) {
753 struct the_nilfs *nilfs = root->nilfs;
754
755 spin_lock(&nilfs->ns_cptree_lock);
756 rb_erase(&root->rb_node, &nilfs->ns_cptree);
757 spin_unlock(&nilfs->ns_cptree_lock);
758 if (root->ifile)
759 iput(root->ifile);
760
761 kfree(root);
762 }
890} 763}
diff --git a/fs/nilfs2/the_nilfs.h b/fs/nilfs2/the_nilfs.h
index f785a7b0ab99..69226e14b745 100644
--- a/fs/nilfs2/the_nilfs.h
+++ b/fs/nilfs2/the_nilfs.h
@@ -26,6 +26,7 @@
26 26
27#include <linux/types.h> 27#include <linux/types.h>
28#include <linux/buffer_head.h> 28#include <linux/buffer_head.h>
29#include <linux/rbtree.h>
29#include <linux/fs.h> 30#include <linux/fs.h>
30#include <linux/blkdev.h> 31#include <linux/blkdev.h>
31#include <linux/backing-dev.h> 32#include <linux/backing-dev.h>
@@ -45,22 +46,13 @@ enum {
45/** 46/**
46 * struct the_nilfs - struct to supervise multiple nilfs mount points 47 * struct the_nilfs - struct to supervise multiple nilfs mount points
47 * @ns_flags: flags 48 * @ns_flags: flags
48 * @ns_count: reference count
49 * @ns_list: list head for nilfs_list
50 * @ns_bdev: block device 49 * @ns_bdev: block device
51 * @ns_bdi: backing dev info
52 * @ns_writer: back pointer to writable nilfs_sb_info
53 * @ns_sem: semaphore for shared states 50 * @ns_sem: semaphore for shared states
54 * @ns_super_sem: semaphore for global operations across super block instances
55 * @ns_mount_mutex: mutex protecting mount process of nilfs
56 * @ns_writer_sem: semaphore protecting ns_writer attach/detach
57 * @ns_current: back pointer to current mount
58 * @ns_sbh: buffer heads of on-disk super blocks 51 * @ns_sbh: buffer heads of on-disk super blocks
59 * @ns_sbp: pointers to super block data 52 * @ns_sbp: pointers to super block data
60 * @ns_sbwtime: previous write time of super block 53 * @ns_sbwtime: previous write time of super block
61 * @ns_sbwcount: write count of super block 54 * @ns_sbwcount: write count of super block
62 * @ns_sbsize: size of valid data in super block 55 * @ns_sbsize: size of valid data in super block
63 * @ns_supers: list of nilfs super block structs
64 * @ns_seg_seq: segment sequence counter 56 * @ns_seg_seq: segment sequence counter
65 * @ns_segnum: index number of the latest full segment. 57 * @ns_segnum: index number of the latest full segment.
66 * @ns_nextnum: index number of the full segment index to be used next 58 * @ns_nextnum: index number of the full segment index to be used next
@@ -79,9 +71,9 @@ enum {
79 * @ns_dat: DAT file inode 71 * @ns_dat: DAT file inode
80 * @ns_cpfile: checkpoint file inode 72 * @ns_cpfile: checkpoint file inode
81 * @ns_sufile: segusage file inode 73 * @ns_sufile: segusage file inode
82 * @ns_gc_dat: shadow inode of the DAT file inode for GC 74 * @ns_cptree: rb-tree of all mounted checkpoints (nilfs_root)
75 * @ns_cptree_lock: lock protecting @ns_cptree
83 * @ns_gc_inodes: dummy inodes to keep live blocks 76 * @ns_gc_inodes: dummy inodes to keep live blocks
84 * @ns_gc_inodes_h: hash list to keep dummy inode holding live blocks
85 * @ns_blocksize_bits: bit length of block size 77 * @ns_blocksize_bits: bit length of block size
86 * @ns_blocksize: block size 78 * @ns_blocksize: block size
87 * @ns_nsegments: number of segments in filesystem 79 * @ns_nsegments: number of segments in filesystem
@@ -95,22 +87,9 @@ enum {
95 */ 87 */
96struct the_nilfs { 88struct the_nilfs {
97 unsigned long ns_flags; 89 unsigned long ns_flags;
98 atomic_t ns_count;
99 struct list_head ns_list;
100 90
101 struct block_device *ns_bdev; 91 struct block_device *ns_bdev;
102 struct backing_dev_info *ns_bdi;
103 struct nilfs_sb_info *ns_writer;
104 struct rw_semaphore ns_sem; 92 struct rw_semaphore ns_sem;
105 struct rw_semaphore ns_super_sem;
106 struct mutex ns_mount_mutex;
107 struct rw_semaphore ns_writer_sem;
108
109 /*
110 * components protected by ns_super_sem
111 */
112 struct nilfs_sb_info *ns_current;
113 struct list_head ns_supers;
114 93
115 /* 94 /*
116 * used for 95 * used for
@@ -163,11 +142,13 @@ struct the_nilfs {
163 struct inode *ns_dat; 142 struct inode *ns_dat;
164 struct inode *ns_cpfile; 143 struct inode *ns_cpfile;
165 struct inode *ns_sufile; 144 struct inode *ns_sufile;
166 struct inode *ns_gc_dat;
167 145
168 /* GC inode list and hash table head */ 146 /* Checkpoint tree */
147 struct rb_root ns_cptree;
148 spinlock_t ns_cptree_lock;
149
150 /* GC inode list */
169 struct list_head ns_gc_inodes; 151 struct list_head ns_gc_inodes;
170 struct hlist_head *ns_gc_inodes_h;
171 152
172 /* Disk layout information (static) */ 153 /* Disk layout information (static) */
173 unsigned int ns_blocksize_bits; 154 unsigned int ns_blocksize_bits;
@@ -182,9 +163,6 @@ struct the_nilfs {
182 u32 ns_crc_seed; 163 u32 ns_crc_seed;
183}; 164};
184 165
185#define NILFS_GCINODE_HASH_BITS 8
186#define NILFS_GCINODE_HASH_SIZE (1<<NILFS_GCINODE_HASH_BITS)
187
188#define THE_NILFS_FNS(bit, name) \ 166#define THE_NILFS_FNS(bit, name) \
189static inline void set_nilfs_##name(struct the_nilfs *nilfs) \ 167static inline void set_nilfs_##name(struct the_nilfs *nilfs) \
190{ \ 168{ \
@@ -205,6 +183,32 @@ THE_NILFS_FNS(DISCONTINUED, discontinued)
205THE_NILFS_FNS(GC_RUNNING, gc_running) 183THE_NILFS_FNS(GC_RUNNING, gc_running)
206THE_NILFS_FNS(SB_DIRTY, sb_dirty) 184THE_NILFS_FNS(SB_DIRTY, sb_dirty)
207 185
186/**
187 * struct nilfs_root - nilfs root object
188 * @cno: checkpoint number
189 * @rb_node: red-black tree node
190 * @count: refcount of this structure
191 * @nilfs: nilfs object
192 * @ifile: inode file
193 * @root: root inode
194 * @inodes_count: number of inodes
195 * @blocks_count: number of blocks (Reserved)
196 */
197struct nilfs_root {
198 __u64 cno;
199 struct rb_node rb_node;
200
201 atomic_t count;
202 struct the_nilfs *nilfs;
203 struct inode *ifile;
204
205 atomic_t inodes_count;
206 atomic_t blocks_count;
207};
208
209/* Special checkpoint number */
210#define NILFS_CPTREE_CURRENT_CNO 0
211
208/* Minimum interval of periodical update of superblocks (in seconds) */ 212/* Minimum interval of periodical update of superblocks (in seconds) */
209#define NILFS_SB_FREQ 10 213#define NILFS_SB_FREQ 10
210 214
@@ -221,46 +225,25 @@ static inline int nilfs_sb_will_flip(struct the_nilfs *nilfs)
221} 225}
222 226
223void nilfs_set_last_segment(struct the_nilfs *, sector_t, u64, __u64); 227void nilfs_set_last_segment(struct the_nilfs *, sector_t, u64, __u64);
224struct the_nilfs *find_or_create_nilfs(struct block_device *); 228struct the_nilfs *alloc_nilfs(struct block_device *bdev);
225void put_nilfs(struct the_nilfs *); 229void destroy_nilfs(struct the_nilfs *nilfs);
226int init_nilfs(struct the_nilfs *, struct nilfs_sb_info *, char *); 230int init_nilfs(struct the_nilfs *, struct nilfs_sb_info *, char *);
227int load_nilfs(struct the_nilfs *, struct nilfs_sb_info *); 231int load_nilfs(struct the_nilfs *, struct nilfs_sb_info *);
228int nilfs_discard_segments(struct the_nilfs *, __u64 *, size_t); 232int nilfs_discard_segments(struct the_nilfs *, __u64 *, size_t);
229int nilfs_count_free_blocks(struct the_nilfs *, sector_t *); 233int nilfs_count_free_blocks(struct the_nilfs *, sector_t *);
234struct nilfs_root *nilfs_lookup_root(struct the_nilfs *nilfs, __u64 cno);
235struct nilfs_root *nilfs_find_or_create_root(struct the_nilfs *nilfs,
236 __u64 cno);
237void nilfs_put_root(struct nilfs_root *root);
230struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *, int, __u64); 238struct nilfs_sb_info *nilfs_find_sbinfo(struct the_nilfs *, int, __u64);
231int nilfs_checkpoint_is_mounted(struct the_nilfs *, __u64, int);
232int nilfs_near_disk_full(struct the_nilfs *); 239int nilfs_near_disk_full(struct the_nilfs *);
233void nilfs_fall_back_super_block(struct the_nilfs *); 240void nilfs_fall_back_super_block(struct the_nilfs *);
234void nilfs_swap_super_block(struct the_nilfs *); 241void nilfs_swap_super_block(struct the_nilfs *);
235 242
236 243
237static inline void get_nilfs(struct the_nilfs *nilfs) 244static inline void nilfs_get_root(struct nilfs_root *root)
238{
239 /* Caller must have at least one reference of the_nilfs. */
240 atomic_inc(&nilfs->ns_count);
241}
242
243static inline void
244nilfs_attach_writer(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
245{
246 down_write(&nilfs->ns_writer_sem);
247 nilfs->ns_writer = sbi;
248 up_write(&nilfs->ns_writer_sem);
249}
250
251static inline void
252nilfs_detach_writer(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi)
253{
254 down_write(&nilfs->ns_writer_sem);
255 if (sbi == nilfs->ns_writer)
256 nilfs->ns_writer = NULL;
257 up_write(&nilfs->ns_writer_sem);
258}
259
260static inline void nilfs_put_sbinfo(struct nilfs_sb_info *sbi)
261{ 245{
262 if (atomic_dec_and_test(&sbi->s_count)) 246 atomic_inc(&root->count);
263 kfree(sbi);
264} 247}
265 248
266static inline int nilfs_valid_fs(struct the_nilfs *nilfs) 249static inline int nilfs_valid_fs(struct the_nilfs *nilfs)
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-02 11:05:28 -0500