git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/fs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #ifndef NO_BCACHEFS_FS
   3
   4 #include "bcachefs.h"
   5 #include "acl.h"
   6 #include "btree_update.h"
   7 #include "buckets.h"
   8 #include "chardev.h"
   9 #include "dirent.h"
  10 #include "extents.h"
  11 #include "fs.h"
  12 #include "fs-common.h"
  13 #include "fs-io.h"
  14 #include "fs-ioctl.h"
  15 #include "fsck.h"
  16 #include "inode.h"
  17 #include "io.h"
  18 #include "journal.h"
  19 #include "keylist.h"
  20 #include "quota.h"
  21 #include "super.h"
  22 #include "xattr.h"
  23
  24 #include <linux/aio.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/exportfs.h>
  27 #include <linux/module.h>
  28 #include <linux/posix_acl.h>
  29 #include <linux/random.h>
  30 #include <linux/statfs.h>
  31 #include <linux/xattr.h>
  32
  33 static struct kmem_cache *bch2_inode_cache;
  34
  35 static void bch2_vfs_inode_init(struct bch_fs *,
  36                                 struct bch_inode_info *,
  37                                 struct bch_inode_unpacked *);
  38
  39 static void journal_seq_copy(struct bch_inode_info *dst,
  40                              u64 journal_seq)
  41 {
  42         u64 old, v = READ_ONCE(dst->ei_journal_seq);
  43
  44         do {
  45                 old = v;
  46
  47                 if (old >= journal_seq)
  48                         break;
  49         } while ((v = cmpxchg(&dst->ei_journal_seq, old, journal_seq)) != old);
  50 }
  51
  52 void bch2_inode_update_after_write(struct bch_fs *c,
  53                                    struct bch_inode_info *inode,
  54                                    struct bch_inode_unpacked *bi,
  55                                    unsigned fields)
  56 {
  57         set_nlink(&inode->v, bch2_inode_nlink_get(bi));
  58         i_uid_write(&inode->v, bi->bi_uid);
  59         i_gid_write(&inode->v, bi->bi_gid);
  60         inode->v.i_mode = bi->bi_mode;
  61
  62         if (fields & ATTR_ATIME)
  63                 inode->v.i_atime = bch2_time_to_timespec(c, bi->bi_atime);
  64         if (fields & ATTR_MTIME)
  65                 inode->v.i_mtime = bch2_time_to_timespec(c, bi->bi_mtime);
  66         if (fields & ATTR_CTIME)
  67                 inode->v.i_ctime = bch2_time_to_timespec(c, bi->bi_ctime);
  68
  69         inode->ei_inode         = *bi;
  70
  71         bch2_inode_flags_to_vfs(inode);
  72 }
  73
  74 int __must_check bch2_write_inode(struct bch_fs *c,
  75                                   struct bch_inode_info *inode,
  76                                   inode_set_fn set,
  77                                   void *p, unsigned fields)
  78 {
  79         struct btree_trans trans;
  80         struct btree_iter *iter;
  81         struct bch_inode_unpacked inode_u;
  82         int ret;
  83
  84         bch2_trans_init(&trans, c, 0, 0);
  85 retry:
  86         bch2_trans_begin(&trans);
  87
  88         iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
  89                                BTREE_ITER_INTENT);
  90         ret   = PTR_ERR_OR_ZERO(iter) ?:
  91                 (set ? set(inode, &inode_u, p) : 0) ?:
  92                 bch2_inode_write(&trans, iter, &inode_u) ?:
  93                 bch2_trans_commit(&trans, NULL,
  94                                   &inode->ei_journal_seq,
  95                                   BTREE_INSERT_ATOMIC|
  96                                   BTREE_INSERT_NOUNLOCK|
  97                                   BTREE_INSERT_NOFAIL);
  98         if (ret == -EINTR)
  99                 goto retry;
 100
 101         /*
 102          * the btree node lock protects inode->ei_inode, not ei_update_lock;
 103          * this is important for inode updates via bchfs_write_index_update
 104          */
 105         if (!ret)
 106                 bch2_inode_update_after_write(c, inode, &inode_u, fields);
 107
 108         bch2_trans_exit(&trans);
 109         return ret < 0 ? ret : 0;
 110 }
 111
 112 int bch2_fs_quota_transfer(struct bch_fs *c,
 113                            struct bch_inode_info *inode,
 114                            struct bch_qid new_qid,
 115                            unsigned qtypes,
 116                            enum quota_acct_mode mode)
 117 {
 118         unsigned i;
 119         int ret;
 120
 121         qtypes &= enabled_qtypes(c);
 122
 123         for (i = 0; i < QTYP_NR; i++)
 124                 if (new_qid.q[i] == inode->ei_qid.q[i])
 125                         qtypes &= ~(1U << i);
 126
 127         if (!qtypes)
 128                 return 0;
 129
 130         mutex_lock(&inode->ei_quota_lock);
 131
 132         ret = bch2_quota_transfer(c, qtypes, new_qid,
 133                                   inode->ei_qid,
 134                                   inode->v.i_blocks +
 135                                   inode->ei_quota_reserved,
 136                                   mode);
 137         if (!ret)
 138                 for (i = 0; i < QTYP_NR; i++)
 139                         if (qtypes & (1 << i))
 140                                 inode->ei_qid.q[i] = new_qid.q[i];
 141
 142         mutex_unlock(&inode->ei_quota_lock);
 143
 144         return ret;
 145 }
 146
 147 struct inode *bch2_vfs_inode_get(struct bch_fs *c, u64 inum)
 148 {
 149         struct bch_inode_unpacked inode_u;
 150         struct bch_inode_info *inode;
 151         int ret;
 152
 153         inode = to_bch_ei(iget_locked(c->vfs_sb, inum));
 154         if (unlikely(!inode))
 155                 return ERR_PTR(-ENOMEM);
 156         if (!(inode->v.i_state & I_NEW))
 157                 return &inode->v;
 158
 159         ret = bch2_inode_find_by_inum(c, inum, &inode_u);
 160         if (ret) {
 161                 iget_failed(&inode->v);
 162                 return ERR_PTR(ret);
 163         }
 164
 165         bch2_vfs_inode_init(c, inode, &inode_u);
 166
 167         inode->ei_journal_seq = bch2_inode_journal_seq(&c->journal, inum);
 168
 169         unlock_new_inode(&inode->v);
 170
 171         return &inode->v;
 172 }
 173
 174 static struct bch_inode_info *
 175 __bch2_create(struct bch_inode_info *dir, struct dentry *dentry,
 176               umode_t mode, dev_t rdev, bool tmpfile)
 177 {
 178         struct bch_fs *c = dir->v.i_sb->s_fs_info;
 179         struct user_namespace *ns = dir->v.i_sb->s_user_ns;
 180         struct btree_trans trans;
 181         struct bch_inode_unpacked dir_u;
 182         struct bch_inode_info *inode, *old;
 183         struct bch_inode_unpacked inode_u;
 184         struct posix_acl *default_acl = NULL, *acl = NULL;
 185         u64 journal_seq = 0;
 186         int ret;
 187
 188         /*
 189          * preallocate acls + vfs inode before btree transaction, so that
 190          * nothing can fail after the transaction succeeds:
 191          */
 192 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 193         ret = posix_acl_create(&dir->v, &mode, &default_acl, &acl);
 194         if (ret)
 195                 return ERR_PTR(ret);
 196 #endif
 197         inode = to_bch_ei(new_inode(c->vfs_sb));
 198         if (unlikely(!inode)) {
 199                 inode = ERR_PTR(-ENOMEM);
 200                 goto err;
 201         }
 202
 203         bch2_inode_init_early(c, &inode_u);
 204
 205         if (!tmpfile)
 206                 mutex_lock(&dir->ei_update_lock);
 207
 208         bch2_trans_init(&trans, c, 8, 1024);
 209 retry:
 210         bch2_trans_begin(&trans);
 211
 212         ret   = bch2_create_trans(&trans, dir->v.i_ino, &dir_u, &inode_u,
 213                                   !tmpfile ? &dentry->d_name : NULL,
 214                                   from_kuid(ns, current_fsuid()),
 215                                   from_kgid(ns, current_fsgid()),
 216                                   mode, rdev,
 217                                   default_acl, acl) ?:
 218                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, 1,
 219                                 KEY_TYPE_QUOTA_PREALLOC);
 220         if (unlikely(ret))
 221                 goto err_before_quota;
 222
 223         ret   = bch2_trans_commit(&trans, NULL, &journal_seq,
 224                                   BTREE_INSERT_ATOMIC|
 225                                   BTREE_INSERT_NOUNLOCK);
 226         if (unlikely(ret)) {
 227                 bch2_quota_acct(c, bch_qid(&inode_u), Q_INO, -1,
 228                                 KEY_TYPE_QUOTA_WARN);
 229 err_before_quota:
 230                 if (ret == -EINTR)
 231                         goto retry;
 232                 goto err_trans;
 233         }
 234
 235         if (!tmpfile) {
 236                 bch2_inode_update_after_write(c, dir, &dir_u,
 237                                               ATTR_MTIME|ATTR_CTIME);
 238                 journal_seq_copy(dir, journal_seq);
 239                 mutex_unlock(&dir->ei_update_lock);
 240         }
 241
 242         bch2_vfs_inode_init(c, inode, &inode_u);
 243         journal_seq_copy(inode, journal_seq);
 244
 245         set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 246         set_cached_acl(&inode->v, ACL_TYPE_DEFAULT, default_acl);
 247
 248         /*
 249          * we must insert the new inode into the inode cache before calling
 250          * bch2_trans_exit() and dropping locks, else we could race with another
 251          * thread pulling the inode in and modifying it:
 252          */
 253
 254         old = to_bch_ei(insert_inode_locked2(&inode->v));
 255         if (unlikely(old)) {
 256                 /*
 257                  * We raced, another process pulled the new inode into cache
 258                  * before us:
 259                  */
 260                 journal_seq_copy(old, journal_seq);
 261                 make_bad_inode(&inode->v);
 262                 iput(&inode->v);
 263
 264                 inode = old;
 265         } else {
 266                 /*
 267                  * we really don't want insert_inode_locked2() to be setting
 268                  * I_NEW...
 269                  */
 270                 unlock_new_inode(&inode->v);
 271         }
 272
 273         bch2_trans_exit(&trans);
 274 err:
 275         posix_acl_release(default_acl);
 276         posix_acl_release(acl);
 277         return inode;
 278 err_trans:
 279         if (!tmpfile)
 280                 mutex_unlock(&dir->ei_update_lock);
 281
 282         bch2_trans_exit(&trans);
 283         make_bad_inode(&inode->v);
 284         iput(&inode->v);
 285         inode = ERR_PTR(ret);
 286         goto err;
 287 }
 288
 289 /* methods */
 290
 291 static struct dentry *bch2_lookup(struct inode *vdir, struct dentry *dentry,
 292                                   unsigned int flags)
 293 {
 294         struct bch_fs *c = vdir->i_sb->s_fs_info;
 295         struct bch_inode_info *dir = to_bch_ei(vdir);
 296         struct inode *vinode = NULL;
 297         u64 inum;
 298
 299         inum = bch2_dirent_lookup(c, dir->v.i_ino,
 300                                   &dir->ei_str_hash,
 301                                   &dentry->d_name);
 302
 303         if (inum)
 304                 vinode = bch2_vfs_inode_get(c, inum);
 305
 306         return d_splice_alias(vinode, dentry);
 307 }
 308
 309 static int bch2_create(struct inode *vdir, struct dentry *dentry,
 310                        umode_t mode, bool excl)
 311 {
 312         struct bch_inode_info *inode =
 313                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFREG, 0, false);
 314
 315         if (IS_ERR(inode))
 316                 return PTR_ERR(inode);
 317
 318         d_instantiate(dentry, &inode->v);
 319         return 0;
 320 }
 321
 322 static int __bch2_link(struct bch_fs *c,
 323                        struct bch_inode_info *inode,
 324                        struct bch_inode_info *dir,
 325                        struct dentry *dentry)
 326 {
 327         struct btree_trans trans;
 328         struct bch_inode_unpacked inode_u;
 329         int ret;
 330
 331         mutex_lock(&inode->ei_update_lock);
 332         bch2_trans_init(&trans, c, 4, 1024);
 333
 334         do {
 335                 bch2_trans_begin(&trans);
 336                 ret   = bch2_link_trans(&trans,
 337                                         dir->v.i_ino,
 338                                         inode->v.i_ino, &inode_u,
 339                                         &dentry->d_name) ?:
 340                         bch2_trans_commit(&trans, NULL,
 341                                         &inode->ei_journal_seq,
 342                                         BTREE_INSERT_ATOMIC|
 343                                         BTREE_INSERT_NOUNLOCK);
 344         } while (ret == -EINTR);
 345
 346         if (likely(!ret))
 347                 bch2_inode_update_after_write(c, inode, &inode_u, ATTR_CTIME);
 348
 349         bch2_trans_exit(&trans);
 350         mutex_unlock(&inode->ei_update_lock);
 351         return ret;
 352 }
 353
 354 static int bch2_link(struct dentry *old_dentry, struct inode *vdir,
 355                      struct dentry *dentry)
 356 {
 357         struct bch_fs *c = vdir->i_sb->s_fs_info;
 358         struct bch_inode_info *dir = to_bch_ei(vdir);
 359         struct bch_inode_info *inode = to_bch_ei(old_dentry->d_inode);
 360         int ret;
 361
 362         lockdep_assert_held(&inode->v.i_rwsem);
 363
 364         ret = __bch2_link(c, inode, dir, dentry);
 365         if (unlikely(ret))
 366                 return ret;
 367
 368         ihold(&inode->v);
 369         d_instantiate(dentry, &inode->v);
 370         return 0;
 371 }
 372
 373 static int bch2_unlink(struct inode *vdir, struct dentry *dentry)
 374 {
 375         struct bch_fs *c = vdir->i_sb->s_fs_info;
 376         struct bch_inode_info *dir = to_bch_ei(vdir);
 377         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 378         struct bch_inode_unpacked dir_u, inode_u;
 379         struct btree_trans trans;
 380         int ret;
 381
 382         bch2_lock_inodes(INODE_UPDATE_LOCK, dir, inode);
 383         bch2_trans_init(&trans, c, 4, 1024);
 384
 385         do {
 386                 bch2_trans_begin(&trans);
 387
 388                 ret   = bch2_unlink_trans(&trans,
 389                                           dir->v.i_ino, &dir_u,
 390                                           &inode_u, &dentry->d_name) ?:
 391                         bch2_trans_commit(&trans, NULL,
 392                                           &dir->ei_journal_seq,
 393                                           BTREE_INSERT_ATOMIC|
 394                                           BTREE_INSERT_NOUNLOCK|
 395                                           BTREE_INSERT_NOFAIL);
 396         } while (ret == -EINTR);
 397
 398         if (likely(!ret)) {
 399                 BUG_ON(inode_u.bi_inum != inode->v.i_ino);
 400
 401                 journal_seq_copy(inode, dir->ei_journal_seq);
 402                 bch2_inode_update_after_write(c, dir, &dir_u,
 403                                               ATTR_MTIME|ATTR_CTIME);
 404                 bch2_inode_update_after_write(c, inode, &inode_u,
 405                                               ATTR_MTIME);
 406         }
 407
 408         bch2_trans_exit(&trans);
 409         bch2_unlock_inodes(INODE_UPDATE_LOCK, dir, inode);
 410
 411         return ret;
 412 }
 413
 414 static int bch2_symlink(struct inode *vdir, struct dentry *dentry,
 415                         const char *symname)
 416 {
 417         struct bch_fs *c = vdir->i_sb->s_fs_info;
 418         struct bch_inode_info *dir = to_bch_ei(vdir), *inode;
 419         int ret;
 420
 421         inode = __bch2_create(dir, dentry, S_IFLNK|S_IRWXUGO, 0, true);
 422         if (unlikely(IS_ERR(inode)))
 423                 return PTR_ERR(inode);
 424
 425         inode_lock(&inode->v);
 426         ret = page_symlink(&inode->v, symname, strlen(symname) + 1);
 427         inode_unlock(&inode->v);
 428
 429         if (unlikely(ret))
 430                 goto err;
 431
 432         ret = filemap_write_and_wait_range(inode->v.i_mapping, 0, LLONG_MAX);
 433         if (unlikely(ret))
 434                 goto err;
 435
 436         journal_seq_copy(dir, inode->ei_journal_seq);
 437
 438         ret = __bch2_link(c, inode, dir, dentry);
 439         if (unlikely(ret))
 440                 goto err;
 441
 442         d_instantiate(dentry, &inode->v);
 443         return 0;
 444 err:
 445         iput(&inode->v);
 446         return ret;
 447 }
 448
 449 static int bch2_mkdir(struct inode *vdir, struct dentry *dentry, umode_t mode)
 450 {
 451         struct bch_inode_info *inode =
 452                 __bch2_create(to_bch_ei(vdir), dentry, mode|S_IFDIR, 0, false);
 453
 454         if (IS_ERR(inode))
 455                 return PTR_ERR(inode);
 456
 457         d_instantiate(dentry, &inode->v);
 458         return 0;
 459 }
 460
 461 static int bch2_rmdir(struct inode *vdir, struct dentry *dentry)
 462 {
 463         return bch2_unlink(vdir, dentry);
 464 }
 465
 466 static int bch2_mknod(struct inode *vdir, struct dentry *dentry,
 467                       umode_t mode, dev_t rdev)
 468 {
 469         struct bch_inode_info *inode =
 470                 __bch2_create(to_bch_ei(vdir), dentry, mode, rdev, false);
 471
 472         if (IS_ERR(inode))
 473                 return PTR_ERR(inode);
 474
 475         d_instantiate(dentry, &inode->v);
 476         return 0;
 477 }
 478
 479 static int bch2_rename2(struct inode *src_vdir, struct dentry *src_dentry,
 480                         struct inode *dst_vdir, struct dentry *dst_dentry,
 481                         unsigned flags)
 482 {
 483         struct bch_fs *c = src_vdir->i_sb->s_fs_info;
 484         struct bch_inode_info *src_dir = to_bch_ei(src_vdir);
 485         struct bch_inode_info *dst_dir = to_bch_ei(dst_vdir);
 486         struct bch_inode_info *src_inode = to_bch_ei(src_dentry->d_inode);
 487         struct bch_inode_info *dst_inode = to_bch_ei(dst_dentry->d_inode);
 488         struct bch_inode_unpacked dst_dir_u, src_dir_u;
 489         struct bch_inode_unpacked src_inode_u, dst_inode_u;
 490         struct btree_trans trans;
 491         enum bch_rename_mode mode = flags & RENAME_EXCHANGE
 492                 ? BCH_RENAME_EXCHANGE
 493                 : dst_dentry->d_inode
 494                 ? BCH_RENAME_OVERWRITE : BCH_RENAME;
 495         u64 journal_seq = 0;
 496         int ret;
 497
 498         if (flags & ~(RENAME_NOREPLACE|RENAME_EXCHANGE))
 499                 return -EINVAL;
 500
 501         if (mode == BCH_RENAME_OVERWRITE) {
 502                 ret = filemap_write_and_wait_range(src_inode->v.i_mapping,
 503                                                    0, LLONG_MAX);
 504                 if (ret)
 505                         return ret;
 506         }
 507
 508         bch2_trans_init(&trans, c, 8, 2048);
 509
 510         bch2_lock_inodes(INODE_UPDATE_LOCK,
 511                          src_dir,
 512                          dst_dir,
 513                          src_inode,
 514                          dst_inode);
 515
 516         if (inode_attr_changing(dst_dir, src_inode, Inode_opt_project)) {
 517                 ret = bch2_fs_quota_transfer(c, src_inode,
 518                                              dst_dir->ei_qid,
 519                                              1 << QTYP_PRJ,
 520                                              KEY_TYPE_QUOTA_PREALLOC);
 521                 if (ret)
 522                         goto err;
 523         }
 524
 525         if (mode == BCH_RENAME_EXCHANGE &&
 526             inode_attr_changing(src_dir, dst_inode, Inode_opt_project)) {
 527                 ret = bch2_fs_quota_transfer(c, dst_inode,
 528                                              src_dir->ei_qid,
 529                                              1 << QTYP_PRJ,
 530                                              KEY_TYPE_QUOTA_PREALLOC);
 531                 if (ret)
 532                         goto err;
 533         }
 534
 535 retry:
 536         bch2_trans_begin(&trans);
 537         ret   = bch2_rename_trans(&trans,
 538                                   src_dir->v.i_ino, &src_dir_u,
 539                                   dst_dir->v.i_ino, &dst_dir_u,
 540                                   &src_inode_u,
 541                                   &dst_inode_u,
 542                                   &src_dentry->d_name,
 543                                   &dst_dentry->d_name,
 544                                   mode) ?:
 545                 bch2_trans_commit(&trans, NULL,
 546                                   &journal_seq,
 547                                   BTREE_INSERT_ATOMIC|
 548                                   BTREE_INSERT_NOUNLOCK);
 549         if (ret == -EINTR)
 550                 goto retry;
 551         if (unlikely(ret))
 552                 goto err;
 553
 554         BUG_ON(src_inode->v.i_ino != src_inode_u.bi_inum);
 555         BUG_ON(dst_inode &&
 556                dst_inode->v.i_ino != dst_inode_u.bi_inum);
 557
 558         bch2_inode_update_after_write(c, src_dir, &src_dir_u,
 559                                       ATTR_MTIME|ATTR_CTIME);
 560         journal_seq_copy(src_dir, journal_seq);
 561
 562         if (src_dir != dst_dir) {
 563                 bch2_inode_update_after_write(c, dst_dir, &dst_dir_u,
 564                                               ATTR_MTIME|ATTR_CTIME);
 565                 journal_seq_copy(dst_dir, journal_seq);
 566         }
 567
 568         bch2_inode_update_after_write(c, src_inode, &src_inode_u,
 569                                       ATTR_CTIME);
 570         journal_seq_copy(src_inode, journal_seq);
 571
 572         if (dst_inode) {
 573                 bch2_inode_update_after_write(c, dst_inode, &dst_inode_u,
 574                                               ATTR_CTIME);
 575                 journal_seq_copy(dst_inode, journal_seq);
 576         }
 577 err:
 578         bch2_trans_exit(&trans);
 579
 580         bch2_fs_quota_transfer(c, src_inode,
 581                                bch_qid(&src_inode->ei_inode),
 582                                1 << QTYP_PRJ,
 583                                KEY_TYPE_QUOTA_NOCHECK);
 584         if (dst_inode)
 585                 bch2_fs_quota_transfer(c, dst_inode,
 586                                        bch_qid(&dst_inode->ei_inode),
 587                                        1 << QTYP_PRJ,
 588                                        KEY_TYPE_QUOTA_NOCHECK);
 589
 590         bch2_unlock_inodes(INODE_UPDATE_LOCK,
 591                            src_dir,
 592                            dst_dir,
 593                            src_inode,
 594                            dst_inode);
 595
 596         return ret;
 597 }
 598
 599 void bch2_setattr_copy(struct bch_inode_info *inode,
 600                        struct bch_inode_unpacked *bi,
 601                        struct iattr *attr)
 602 {
 603         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 604         unsigned int ia_valid = attr->ia_valid;
 605
 606         if (ia_valid & ATTR_UID)
 607                 bi->bi_uid = from_kuid(c->vfs_sb->s_user_ns, attr->ia_uid);
 608         if (ia_valid & ATTR_GID)
 609                 bi->bi_gid = from_kgid(c->vfs_sb->s_user_ns, attr->ia_gid);
 610
 611         if (ia_valid & ATTR_ATIME)
 612                 bi->bi_atime = timespec_to_bch2_time(c, attr->ia_atime);
 613         if (ia_valid & ATTR_MTIME)
 614                 bi->bi_mtime = timespec_to_bch2_time(c, attr->ia_mtime);
 615         if (ia_valid & ATTR_CTIME)
 616                 bi->bi_ctime = timespec_to_bch2_time(c, attr->ia_ctime);
 617
 618         if (ia_valid & ATTR_MODE) {
 619                 umode_t mode = attr->ia_mode;
 620                 kgid_t gid = ia_valid & ATTR_GID
 621                         ? attr->ia_gid
 622                         : inode->v.i_gid;
 623
 624                 if (!in_group_p(gid) &&
 625                     !capable_wrt_inode_uidgid(&inode->v, CAP_FSETID))
 626                         mode &= ~S_ISGID;
 627                 bi->bi_mode = mode;
 628         }
 629 }
 630
 631 static int bch2_setattr_nonsize(struct bch_inode_info *inode,
 632                                 struct iattr *attr)
 633 {
 634         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 635         struct bch_qid qid;
 636         struct btree_trans trans;
 637         struct btree_iter *inode_iter;
 638         struct bch_inode_unpacked inode_u;
 639         struct posix_acl *acl = NULL;
 640         int ret;
 641
 642         mutex_lock(&inode->ei_update_lock);
 643
 644         qid = inode->ei_qid;
 645
 646         if (attr->ia_valid & ATTR_UID)
 647                 qid.q[QTYP_USR] = from_kuid(&init_user_ns, attr->ia_uid);
 648
 649         if (attr->ia_valid & ATTR_GID)
 650                 qid.q[QTYP_GRP] = from_kgid(&init_user_ns, attr->ia_gid);
 651
 652         ret = bch2_fs_quota_transfer(c, inode, qid, ~0,
 653                                      KEY_TYPE_QUOTA_PREALLOC);
 654         if (ret)
 655                 goto err;
 656
 657         bch2_trans_init(&trans, c, 0, 0);
 658 retry:
 659         bch2_trans_begin(&trans);
 660         kfree(acl);
 661         acl = NULL;
 662
 663         inode_iter = bch2_inode_peek(&trans, &inode_u, inode->v.i_ino,
 664                                      BTREE_ITER_INTENT);
 665         ret = PTR_ERR_OR_ZERO(inode_iter);
 666         if (ret)
 667                 goto btree_err;
 668
 669         bch2_setattr_copy(inode, &inode_u, attr);
 670
 671         if (attr->ia_valid & ATTR_MODE) {
 672                 ret = bch2_acl_chmod(&trans, inode, inode_u.bi_mode, &acl);
 673                 if (ret)
 674                         goto btree_err;
 675         }
 676
 677         ret =   bch2_inode_write(&trans, inode_iter, &inode_u) ?:
 678                 bch2_trans_commit(&trans, NULL,
 679                                   &inode->ei_journal_seq,
 680                                   BTREE_INSERT_ATOMIC|
 681                                   BTREE_INSERT_NOUNLOCK|
 682                                   BTREE_INSERT_NOFAIL);
 683 btree_err:
 684         if (ret == -EINTR)
 685                 goto retry;
 686         if (unlikely(ret))
 687                 goto err_trans;
 688
 689         bch2_inode_update_after_write(c, inode, &inode_u, attr->ia_valid);
 690
 691         if (acl)
 692                 set_cached_acl(&inode->v, ACL_TYPE_ACCESS, acl);
 693 err_trans:
 694         bch2_trans_exit(&trans);
 695 err:
 696         mutex_unlock(&inode->ei_update_lock);
 697
 698         return ret;
 699 }
 700
 701 static int bch2_getattr(const struct path *path, struct kstat *stat,
 702                         u32 request_mask, unsigned query_flags)
 703 {
 704         struct bch_inode_info *inode = to_bch_ei(d_inode(path->dentry));
 705         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 706
 707         stat->dev       = inode->v.i_sb->s_dev;
 708         stat->ino       = inode->v.i_ino;
 709         stat->mode      = inode->v.i_mode;
 710         stat->nlink     = inode->v.i_nlink;
 711         stat->uid       = inode->v.i_uid;
 712         stat->gid       = inode->v.i_gid;
 713         stat->rdev      = inode->v.i_rdev;
 714         stat->size      = i_size_read(&inode->v);
 715         stat->atime     = inode->v.i_atime;
 716         stat->mtime     = inode->v.i_mtime;
 717         stat->ctime     = inode->v.i_ctime;
 718         stat->blksize   = block_bytes(c);
 719         stat->blocks    = inode->v.i_blocks;
 720
 721         if (request_mask & STATX_BTIME) {
 722                 stat->result_mask |= STATX_BTIME;
 723                 stat->btime = bch2_time_to_timespec(c, inode->ei_inode.bi_otime);
 724         }
 725
 726         if (inode->ei_inode.bi_flags & BCH_INODE_IMMUTABLE)
 727                 stat->attributes |= STATX_ATTR_IMMUTABLE;
 728         if (inode->ei_inode.bi_flags & BCH_INODE_APPEND)
 729                 stat->attributes |= STATX_ATTR_APPEND;
 730         if (inode->ei_inode.bi_flags & BCH_INODE_NODUMP)
 731                 stat->attributes |= STATX_ATTR_NODUMP;
 732
 733         return 0;
 734 }
 735
 736 static int bch2_setattr(struct dentry *dentry, struct iattr *iattr)
 737 {
 738         struct bch_inode_info *inode = to_bch_ei(dentry->d_inode);
 739         int ret;
 740
 741         lockdep_assert_held(&inode->v.i_rwsem);
 742
 743         ret = setattr_prepare(dentry, iattr);
 744         if (ret)
 745                 return ret;
 746
 747         return iattr->ia_valid & ATTR_SIZE
 748                 ? bch2_truncate(inode, iattr)
 749                 : bch2_setattr_nonsize(inode, iattr);
 750 }
 751
 752 static int bch2_tmpfile(struct inode *vdir, struct dentry *dentry, umode_t mode)
 753 {
 754         struct bch_inode_info *inode =
 755                 __bch2_create(to_bch_ei(vdir), dentry, mode, 0, true);
 756
 757         if (IS_ERR(inode))
 758                 return PTR_ERR(inode);
 759
 760         d_mark_tmpfile(dentry, &inode->v);
 761         d_instantiate(dentry, &inode->v);
 762         return 0;
 763 }
 764
 765 static int bch2_fill_extent(struct bch_fs *c,
 766                             struct fiemap_extent_info *info,
 767                             struct bkey_s_c k, unsigned flags)
 768 {
 769         if (bkey_extent_is_data(k.k)) {
 770                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 771                 const union bch_extent_entry *entry;
 772                 struct extent_ptr_decoded p;
 773                 int ret;
 774
 775                 if (k.k->type == KEY_TYPE_reflink_v)
 776                         flags |= FIEMAP_EXTENT_SHARED;
 777
 778                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
 779                         int flags2 = 0;
 780                         u64 offset = p.ptr.offset;
 781
 782                         if (p.crc.compression_type)
 783                                 flags2 |= FIEMAP_EXTENT_ENCODED;
 784                         else
 785                                 offset += p.crc.offset;
 786
 787                         if ((offset & (c->opts.block_size - 1)) ||
 788                             (k.k->size & (c->opts.block_size - 1)))
 789                                 flags2 |= FIEMAP_EXTENT_NOT_ALIGNED;
 790
 791                         ret = fiemap_fill_next_extent(info,
 792                                                 bkey_start_offset(k.k) << 9,
 793                                                 offset << 9,
 794                                                 k.k->size << 9, flags|flags2);
 795                         if (ret)
 796                                 return ret;
 797                 }
 798
 799                 return 0;
 800         } else if (k.k->type == KEY_TYPE_reservation) {
 801                 return fiemap_fill_next_extent(info,
 802                                                bkey_start_offset(k.k) << 9,
 803                                                0, k.k->size << 9,
 804                                                flags|
 805                                                FIEMAP_EXTENT_DELALLOC|
 806                                                FIEMAP_EXTENT_UNWRITTEN);
 807         } else {
 808                 BUG();
 809         }
 810 }
 811
 812 static int bch2_fiemap(struct inode *vinode, struct fiemap_extent_info *info,
 813                        u64 start, u64 len)
 814 {
 815         struct bch_fs *c = vinode->i_sb->s_fs_info;
 816         struct bch_inode_info *ei = to_bch_ei(vinode);
 817         struct btree_trans trans;
 818         struct btree_iter *iter;
 819         struct bkey_s_c k;
 820         BKEY_PADDED(k) cur, prev;
 821         struct bpos end = POS(ei->v.i_ino, (start + len) >> 9);
 822         unsigned offset_into_extent, sectors;
 823         bool have_extent = false;
 824         int ret = 0;
 825
 826         if (start + len < start)
 827                 return -EINVAL;
 828
 829         bch2_trans_init(&trans, c, 0, 0);
 830
 831         iter = bch2_trans_get_iter(&trans, BTREE_ID_EXTENTS,
 832                                    POS(ei->v.i_ino, start >> 9), 0);
 833 retry:
 834         while ((k = bch2_btree_iter_peek(iter)).k &&
 835                !(ret = bkey_err(k)) &&
 836                bkey_cmp(iter->pos, end) < 0) {
 837                 if (!bkey_extent_is_data(k.k) &&
 838                     k.k->type != KEY_TYPE_reservation) {
 839                         bch2_btree_iter_next(iter);
 840                         continue;
 841                 }
 842
 843                 bkey_reassemble(&cur.k, k);
 844                 k = bkey_i_to_s_c(&cur.k);
 845
 846                 offset_into_extent      = iter->pos.offset -
 847                         bkey_start_offset(k.k);
 848                 sectors                 = k.k->size - offset_into_extent;
 849
 850                 ret = bch2_read_indirect_extent(&trans,
 851                                         &offset_into_extent, &cur.k);
 852                 if (ret)
 853                         break;
 854
 855                 sectors = min(sectors, k.k->size - offset_into_extent);
 856
 857                 if (offset_into_extent)
 858                         bch2_cut_front(POS(k.k->p.inode,
 859                                            bkey_start_offset(k.k) +
 860                                            offset_into_extent),
 861                                        &cur.k);
 862                 bch2_key_resize(&cur.k.k, sectors);
 863                 cur.k.k.p = iter->pos;
 864                 cur.k.k.p.offset += cur.k.k.size;
 865
 866                 if (have_extent) {
 867                         ret = bch2_fill_extent(c, info,
 868                                         bkey_i_to_s_c(&prev.k), 0);
 869                         if (ret)
 870                                 break;
 871                 }
 872
 873                 bkey_copy(&prev.k, &cur.k);
 874                 have_extent = true;
 875
 876                 if (k.k->type == KEY_TYPE_reflink_v)
 877                         bch2_btree_iter_set_pos(iter, k.k->p);
 878                 else
 879                         bch2_btree_iter_next(iter);
 880         }
 881
 882         if (ret == -EINTR)
 883                 goto retry;
 884
 885         if (!ret && have_extent)
 886                 ret = bch2_fill_extent(c, info, bkey_i_to_s_c(&prev.k),
 887                                        FIEMAP_EXTENT_LAST);
 888
 889         ret = bch2_trans_exit(&trans) ?: ret;
 890         return ret < 0 ? ret : 0;
 891 }
 892
 893 static const struct vm_operations_struct bch_vm_ops = {
 894         .fault          = filemap_fault,
 895         .map_pages      = filemap_map_pages,
 896         .page_mkwrite   = bch2_page_mkwrite,
 897 };
 898
 899 static int bch2_mmap(struct file *file, struct vm_area_struct *vma)
 900 {
 901         file_accessed(file);
 902
 903         vma->vm_ops = &bch_vm_ops;
 904         return 0;
 905 }
 906
 907 /* Directories: */
 908
 909 static loff_t bch2_dir_llseek(struct file *file, loff_t offset, int whence)
 910 {
 911         return generic_file_llseek_size(file, offset, whence,
 912                                         S64_MAX, S64_MAX);
 913 }
 914
 915 static int bch2_vfs_readdir(struct file *file, struct dir_context *ctx)
 916 {
 917         struct bch_inode_info *inode = file_bch_inode(file);
 918         struct bch_fs *c = inode->v.i_sb->s_fs_info;
 919
 920         if (!dir_emit_dots(file, ctx))
 921                 return 0;
 922
 923         return bch2_readdir(c, inode->v.i_ino, ctx);
 924 }
 925
 926 static const struct file_operations bch_file_operations = {
 927         .llseek         = bch2_llseek,
 928         .read_iter      = generic_file_read_iter,
 929         .write_iter     = bch2_write_iter,
 930         .mmap           = bch2_mmap,
 931         .open           = generic_file_open,
 932         .fsync          = bch2_fsync,
 933         .splice_read    = generic_file_splice_read,
 934         .splice_write   = iter_file_splice_write,
 935         .fallocate      = bch2_fallocate_dispatch,
 936         .unlocked_ioctl = bch2_fs_file_ioctl,
 937 #ifdef CONFIG_COMPAT
 938         .compat_ioctl   = bch2_compat_fs_ioctl,
 939 #endif
 940         .remap_file_range = bch2_remap_file_range,
 941 };
 942
 943 static const struct inode_operations bch_file_inode_operations = {
 944         .getattr        = bch2_getattr,
 945         .setattr        = bch2_setattr,
 946         .fiemap         = bch2_fiemap,
 947         .listxattr      = bch2_xattr_list,
 948 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 949         .get_acl        = bch2_get_acl,
 950         .set_acl        = bch2_set_acl,
 951 #endif
 952 };
 953
 954 static const struct inode_operations bch_dir_inode_operations = {
 955         .lookup         = bch2_lookup,
 956         .create         = bch2_create,
 957         .link           = bch2_link,
 958         .unlink         = bch2_unlink,
 959         .symlink        = bch2_symlink,
 960         .mkdir          = bch2_mkdir,
 961         .rmdir          = bch2_rmdir,
 962         .mknod          = bch2_mknod,
 963         .rename         = bch2_rename2,
 964         .getattr        = bch2_getattr,
 965         .setattr        = bch2_setattr,
 966         .tmpfile        = bch2_tmpfile,
 967         .listxattr      = bch2_xattr_list,
 968 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 969         .get_acl        = bch2_get_acl,
 970         .set_acl        = bch2_set_acl,
 971 #endif
 972 };
 973
 974 static const struct file_operations bch_dir_file_operations = {
 975         .llseek         = bch2_dir_llseek,
 976         .read           = generic_read_dir,
 977         .iterate        = bch2_vfs_readdir,
 978         .fsync          = bch2_fsync,
 979         .unlocked_ioctl = bch2_fs_file_ioctl,
 980 #ifdef CONFIG_COMPAT
 981         .compat_ioctl   = bch2_compat_fs_ioctl,
 982 #endif
 983 };
 984
 985 static const struct inode_operations bch_symlink_inode_operations = {
 986         .get_link       = page_get_link,
 987         .getattr        = bch2_getattr,
 988         .setattr        = bch2_setattr,
 989         .listxattr      = bch2_xattr_list,
 990 #ifdef CONFIG_BCACHEFS_POSIX_ACL
 991         .get_acl        = bch2_get_acl,
 992         .set_acl        = bch2_set_acl,
 993 #endif
 994 };
 995
 996 static const struct inode_operations bch_special_inode_operations = {
 997         .getattr        = bch2_getattr,
 998         .setattr        = bch2_setattr,
 999         .listxattr      = bch2_xattr_list,
1000 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1001         .get_acl        = bch2_get_acl,
1002         .set_acl        = bch2_set_acl,
1003 #endif
1004 };
1005
1006 static const struct address_space_operations bch_address_space_operations = {
1007         .writepage      = bch2_writepage,
1008         .readpage       = bch2_readpage,
1009         .writepages     = bch2_writepages,
1010         .readpages      = bch2_readpages,
1011         .set_page_dirty = __set_page_dirty_nobuffers,
1012         .write_begin    = bch2_write_begin,
1013         .write_end      = bch2_write_end,
1014         .invalidatepage = bch2_invalidatepage,
1015         .releasepage    = bch2_releasepage,
1016         .direct_IO      = bch2_direct_IO,
1017 #ifdef CONFIG_MIGRATION
1018         .migratepage    = bch2_migrate_page,
1019 #endif
1020         .error_remove_page = generic_error_remove_page,
1021 };
1022
1023 static struct inode *bch2_nfs_get_inode(struct super_block *sb,
1024                 u64 ino, u32 generation)
1025 {
1026         struct bch_fs *c = sb->s_fs_info;
1027         struct inode *vinode;
1028
1029         if (ino < BCACHEFS_ROOT_INO)
1030                 return ERR_PTR(-ESTALE);
1031
1032         vinode = bch2_vfs_inode_get(c, ino);
1033         if (IS_ERR(vinode))
1034                 return ERR_CAST(vinode);
1035         if (generation && vinode->i_generation != generation) {
1036                 /* we didn't find the right inode.. */
1037                 iput(vinode);
1038                 return ERR_PTR(-ESTALE);
1039         }
1040         return vinode;
1041 }
1042
1043 static struct dentry *bch2_fh_to_dentry(struct super_block *sb, struct fid *fid,
1044                 int fh_len, int fh_type)
1045 {
1046         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1047                                     bch2_nfs_get_inode);
1048 }
1049
1050 static struct dentry *bch2_fh_to_parent(struct super_block *sb, struct fid *fid,
1051                 int fh_len, int fh_type)
1052 {
1053         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1054                                     bch2_nfs_get_inode);
1055 }
1056
1057 static const struct export_operations bch_export_ops = {
1058         .fh_to_dentry   = bch2_fh_to_dentry,
1059         .fh_to_parent   = bch2_fh_to_parent,
1060         //.get_parent   = bch2_get_parent,
1061 };
1062
1063 static void bch2_vfs_inode_init(struct bch_fs *c,
1064                                 struct bch_inode_info *inode,
1065                                 struct bch_inode_unpacked *bi)
1066 {
1067         bch2_inode_update_after_write(c, inode, bi, ~0);
1068
1069         inode->v.i_blocks       = bi->bi_sectors;
1070         inode->v.i_ino          = bi->bi_inum;
1071         inode->v.i_rdev         = bi->bi_dev;
1072         inode->v.i_generation   = bi->bi_generation;
1073         inode->v.i_size         = bi->bi_size;
1074
1075         inode->ei_journal_seq   = 0;
1076         inode->ei_quota_reserved = 0;
1077         inode->ei_str_hash      = bch2_hash_info_init(c, bi);
1078         inode->ei_qid           = bch_qid(bi);
1079
1080         inode->v.i_mapping->a_ops = &bch_address_space_operations;
1081
1082         switch (inode->v.i_mode & S_IFMT) {
1083         case S_IFREG:
1084                 inode->v.i_op   = &bch_file_inode_operations;
1085                 inode->v.i_fop  = &bch_file_operations;
1086                 break;
1087         case S_IFDIR:
1088                 inode->v.i_op   = &bch_dir_inode_operations;
1089                 inode->v.i_fop  = &bch_dir_file_operations;
1090                 break;
1091         case S_IFLNK:
1092                 inode_nohighmem(&inode->v);
1093                 inode->v.i_op   = &bch_symlink_inode_operations;
1094                 break;
1095         default:
1096                 init_special_inode(&inode->v, inode->v.i_mode, inode->v.i_rdev);
1097                 inode->v.i_op   = &bch_special_inode_operations;
1098                 break;
1099         }
1100 }
1101
1102 static struct inode *bch2_alloc_inode(struct super_block *sb)
1103 {
1104         struct bch_inode_info *inode;
1105
1106         inode = kmem_cache_alloc(bch2_inode_cache, GFP_NOFS);
1107         if (!inode)
1108                 return NULL;
1109
1110         inode_init_once(&inode->v);
1111         mutex_init(&inode->ei_update_lock);
1112         mutex_init(&inode->ei_quota_lock);
1113         inode->ei_journal_seq = 0;
1114
1115         return &inode->v;
1116 }
1117
1118 static void bch2_i_callback(struct rcu_head *head)
1119 {
1120         struct inode *vinode = container_of(head, struct inode, i_rcu);
1121         struct bch_inode_info *inode = to_bch_ei(vinode);
1122
1123         kmem_cache_free(bch2_inode_cache, inode);
1124 }
1125
1126 static void bch2_destroy_inode(struct inode *vinode)
1127 {
1128         call_rcu(&vinode->i_rcu, bch2_i_callback);
1129 }
1130
1131 static int inode_update_times_fn(struct bch_inode_info *inode,
1132                                  struct bch_inode_unpacked *bi,
1133                                  void *p)
1134 {
1135         struct bch_fs *c = inode->v.i_sb->s_fs_info;
1136
1137         bi->bi_atime    = timespec_to_bch2_time(c, inode->v.i_atime);
1138         bi->bi_mtime    = timespec_to_bch2_time(c, inode->v.i_mtime);
1139         bi->bi_ctime    = timespec_to_bch2_time(c, inode->v.i_ctime);
1140
1141         return 0;
1142 }
1143
1144 static int bch2_vfs_write_inode(struct inode *vinode,
1145                                 struct writeback_control *wbc)
1146 {
1147         struct bch_fs *c = vinode->i_sb->s_fs_info;
1148         struct bch_inode_info *inode = to_bch_ei(vinode);
1149         int ret;
1150
1151         mutex_lock(&inode->ei_update_lock);
1152         ret = bch2_write_inode(c, inode, inode_update_times_fn, NULL,
1153                                ATTR_ATIME|ATTR_MTIME|ATTR_CTIME);
1154         mutex_unlock(&inode->ei_update_lock);
1155
1156         return ret;
1157 }
1158
1159 static void bch2_evict_inode(struct inode *vinode)
1160 {
1161         struct bch_fs *c = vinode->i_sb->s_fs_info;
1162         struct bch_inode_info *inode = to_bch_ei(vinode);
1163
1164         truncate_inode_pages_final(&inode->v.i_data);
1165
1166         clear_inode(&inode->v);
1167
1168         BUG_ON(!is_bad_inode(&inode->v) && inode->ei_quota_reserved);
1169
1170         if (!inode->v.i_nlink && !is_bad_inode(&inode->v)) {
1171                 bch2_quota_acct(c, inode->ei_qid, Q_SPC, -((s64) inode->v.i_blocks),
1172                                 KEY_TYPE_QUOTA_WARN);
1173                 bch2_quota_acct(c, inode->ei_qid, Q_INO, -1,
1174                                 KEY_TYPE_QUOTA_WARN);
1175                 bch2_inode_rm(c, inode->v.i_ino);
1176         }
1177 }
1178
1179 static int bch2_statfs(struct dentry *dentry, struct kstatfs *buf)
1180 {
1181         struct super_block *sb = dentry->d_sb;
1182         struct bch_fs *c = sb->s_fs_info;
1183         struct bch_fs_usage_short usage = bch2_fs_usage_read_short(c);
1184         unsigned shift = sb->s_blocksize_bits - 9;
1185         u64 fsid;
1186
1187         buf->f_type     = BCACHEFS_STATFS_MAGIC;
1188         buf->f_bsize    = sb->s_blocksize;
1189         buf->f_blocks   = usage.capacity >> shift;
1190         buf->f_bfree    = (usage.capacity - usage.used) >> shift;
1191         buf->f_bavail   = buf->f_bfree;
1192         buf->f_files    = usage.nr_inodes;
1193         buf->f_ffree    = U64_MAX;
1194
1195         fsid = le64_to_cpup((void *) c->sb.user_uuid.b) ^
1196                le64_to_cpup((void *) c->sb.user_uuid.b + sizeof(u64));
1197         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
1198         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
1199         buf->f_namelen  = BCH_NAME_MAX;
1200
1201         return 0;
1202 }
1203
1204 static int bch2_sync_fs(struct super_block *sb, int wait)
1205 {
1206         struct bch_fs *c = sb->s_fs_info;
1207
1208         if (c->opts.journal_flush_disabled)
1209                 return 0;
1210
1211         if (!wait) {
1212                 bch2_journal_flush_async(&c->journal, NULL);
1213                 return 0;
1214         }
1215
1216         return bch2_journal_flush(&c->journal);
1217 }
1218
1219 static struct bch_fs *bch2_path_to_fs(const char *dev)
1220 {
1221         struct bch_fs *c;
1222         struct block_device *bdev = lookup_bdev(dev);
1223
1224         if (IS_ERR(bdev))
1225                 return ERR_CAST(bdev);
1226
1227         c = bch2_bdev_to_fs(bdev);
1228         bdput(bdev);
1229         return c ?: ERR_PTR(-ENOENT);
1230 }
1231
1232 static struct bch_fs *__bch2_open_as_blockdevs(const char *dev_name, char * const *devs,
1233                                                unsigned nr_devs, struct bch_opts opts)
1234 {
1235         struct bch_fs *c, *c1, *c2;
1236         size_t i;
1237
1238         if (!nr_devs)
1239                 return ERR_PTR(-EINVAL);
1240
1241         c = bch2_fs_open(devs, nr_devs, opts);
1242
1243         if (IS_ERR(c) && PTR_ERR(c) == -EBUSY) {
1244                 /*
1245                  * Already open?
1246                  * Look up each block device, make sure they all belong to a
1247                  * filesystem and they all belong to the _same_ filesystem
1248                  */
1249
1250                 c1 = bch2_path_to_fs(devs[0]);
1251                 if (IS_ERR(c1))
1252                         return c;
1253
1254                 for (i = 1; i < nr_devs; i++) {
1255                         c2 = bch2_path_to_fs(devs[i]);
1256                         if (!IS_ERR(c2))
1257                                 closure_put(&c2->cl);
1258
1259                         if (c1 != c2) {
1260                                 closure_put(&c1->cl);
1261                                 return c;
1262                         }
1263                 }
1264
1265                 c = c1;
1266         }
1267
1268         if (IS_ERR(c))
1269                 return c;
1270
1271         mutex_lock(&c->state_lock);
1272
1273         if (!test_bit(BCH_FS_STARTED, &c->flags)) {
1274                 mutex_unlock(&c->state_lock);
1275                 closure_put(&c->cl);
1276                 pr_err("err mounting %s: incomplete filesystem", dev_name);
1277                 return ERR_PTR(-EINVAL);
1278         }
1279
1280         mutex_unlock(&c->state_lock);
1281
1282         set_bit(BCH_FS_BDEV_MOUNTED, &c->flags);
1283         return c;
1284 }
1285
1286 static struct bch_fs *bch2_open_as_blockdevs(const char *_dev_name,
1287                                              struct bch_opts opts)
1288 {
1289         char *dev_name = NULL, **devs = NULL, *s;
1290         struct bch_fs *c = ERR_PTR(-ENOMEM);
1291         size_t i, nr_devs = 0;
1292
1293         dev_name = kstrdup(_dev_name, GFP_KERNEL);
1294         if (!dev_name)
1295                 goto err;
1296
1297         for (s = dev_name; s; s = strchr(s + 1, ':'))
1298                 nr_devs++;
1299
1300         devs = kcalloc(nr_devs, sizeof(const char *), GFP_KERNEL);
1301         if (!devs)
1302                 goto err;
1303
1304         for (i = 0, s = dev_name;
1305              s;
1306              (s = strchr(s, ':')) && (*s++ = '\0'))
1307                 devs[i++] = s;
1308
1309         c = __bch2_open_as_blockdevs(_dev_name, devs, nr_devs, opts);
1310 err:
1311         kfree(devs);
1312         kfree(dev_name);
1313         return c;
1314 }
1315
1316 static int bch2_remount(struct super_block *sb, int *flags, char *data)
1317 {
1318         struct bch_fs *c = sb->s_fs_info;
1319         struct bch_opts opts = bch2_opts_empty();
1320         int ret;
1321
1322         opt_set(opts, read_only, (*flags & SB_RDONLY) != 0);
1323
1324         ret = bch2_parse_mount_opts(&opts, data);
1325         if (ret)
1326                 return ret;
1327
1328         if (opts.read_only != c->opts.read_only) {
1329                 mutex_lock(&c->state_lock);
1330
1331                 if (opts.read_only) {
1332                         bch2_fs_read_only(c);
1333
1334                         sb->s_flags |= SB_RDONLY;
1335                 } else {
1336                         ret = bch2_fs_read_write(c);
1337                         if (ret) {
1338                                 bch_err(c, "error going rw: %i", ret);
1339                                 mutex_unlock(&c->state_lock);
1340                                 return -EINVAL;
1341                         }
1342
1343                         sb->s_flags &= ~SB_RDONLY;
1344                 }
1345
1346                 c->opts.read_only = opts.read_only;
1347
1348                 mutex_unlock(&c->state_lock);
1349         }
1350
1351         if (opts.errors >= 0)
1352                 c->opts.errors = opts.errors;
1353
1354         return ret;
1355 }
1356
1357 static int bch2_show_options(struct seq_file *seq, struct dentry *root)
1358 {
1359         struct bch_fs *c = root->d_sb->s_fs_info;
1360         enum bch_opt_id i;
1361         char buf[512];
1362
1363         for (i = 0; i < bch2_opts_nr; i++) {
1364                 const struct bch_option *opt = &bch2_opt_table[i];
1365                 u64 v = bch2_opt_get_by_id(&c->opts, i);
1366
1367                 if (!(opt->mode & OPT_MOUNT))
1368                         continue;
1369
1370                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
1371                         continue;
1372
1373                 bch2_opt_to_text(&PBUF(buf), c, opt, v,
1374                                  OPT_SHOW_MOUNT_STYLE);
1375                 seq_putc(seq, ',');
1376                 seq_puts(seq, buf);
1377         }
1378
1379         return 0;
1380
1381 }
1382
1383 static const struct super_operations bch_super_operations = {
1384         .alloc_inode    = bch2_alloc_inode,
1385         .destroy_inode  = bch2_destroy_inode,
1386         .write_inode    = bch2_vfs_write_inode,
1387         .evict_inode    = bch2_evict_inode,
1388         .sync_fs        = bch2_sync_fs,
1389         .statfs         = bch2_statfs,
1390         .show_options   = bch2_show_options,
1391         .remount_fs     = bch2_remount,
1392 #if 0
1393         .put_super      = bch2_put_super,
1394         .freeze_fs      = bch2_freeze,
1395         .unfreeze_fs    = bch2_unfreeze,
1396 #endif
1397 };
1398
1399 static int bch2_test_super(struct super_block *s, void *data)
1400 {
1401         return s->s_fs_info == data;
1402 }
1403
1404 static int bch2_set_super(struct super_block *s, void *data)
1405 {
1406         s->s_fs_info = data;
1407         return 0;
1408 }
1409
1410 static struct dentry *bch2_mount(struct file_system_type *fs_type,
1411                                  int flags, const char *dev_name, void *data)
1412 {
1413         struct bch_fs *c;
1414         struct bch_dev *ca;
1415         struct super_block *sb;
1416         struct inode *vinode;
1417         struct bch_opts opts = bch2_opts_empty();
1418         unsigned i;
1419         int ret;
1420
1421         opt_set(opts, read_only, (flags & SB_RDONLY) != 0);
1422
1423         ret = bch2_parse_mount_opts(&opts, data);
1424         if (ret)
1425                 return ERR_PTR(ret);
1426
1427         c = bch2_open_as_blockdevs(dev_name, opts);
1428         if (IS_ERR(c))
1429                 return ERR_CAST(c);
1430
1431         sb = sget(fs_type, bch2_test_super, bch2_set_super, flags|SB_NOSEC, c);
1432         if (IS_ERR(sb)) {
1433                 closure_put(&c->cl);
1434                 return ERR_CAST(sb);
1435         }
1436
1437         BUG_ON(sb->s_fs_info != c);
1438
1439         if (sb->s_root) {
1440                 closure_put(&c->cl);
1441
1442                 if ((flags ^ sb->s_flags) & SB_RDONLY) {
1443                         ret = -EBUSY;
1444                         goto err_put_super;
1445                 }
1446                 goto out;
1447         }
1448
1449         sb->s_blocksize         = block_bytes(c);
1450         sb->s_blocksize_bits    = ilog2(block_bytes(c));
1451         sb->s_maxbytes          = MAX_LFS_FILESIZE;
1452         sb->s_op                = &bch_super_operations;
1453         sb->s_export_op         = &bch_export_ops;
1454 #ifdef CONFIG_BCACHEFS_QUOTA
1455         sb->s_qcop              = &bch2_quotactl_operations;
1456         sb->s_quota_types       = QTYPE_MASK_USR|QTYPE_MASK_GRP|QTYPE_MASK_PRJ;
1457 #endif
1458         sb->s_xattr             = bch2_xattr_handlers;
1459         sb->s_magic             = BCACHEFS_STATFS_MAGIC;
1460         sb->s_time_gran         = c->sb.time_precision;
1461         c->vfs_sb               = sb;
1462         strlcpy(sb->s_id, c->name, sizeof(sb->s_id));
1463
1464         ret = super_setup_bdi(sb);
1465         if (ret)
1466                 goto err_put_super;
1467
1468         sb->s_bdi->congested_fn         = bch2_congested;
1469         sb->s_bdi->congested_data       = c;
1470         sb->s_bdi->ra_pages             = VM_READAHEAD_PAGES;
1471
1472         for_each_online_member(ca, c, i) {
1473                 struct block_device *bdev = ca->disk_sb.bdev;
1474
1475                 /* XXX: create an anonymous device for multi device filesystems */
1476                 sb->s_bdev      = bdev;
1477                 sb->s_dev       = bdev->bd_dev;
1478                 percpu_ref_put(&ca->io_ref);
1479                 break;
1480         }
1481
1482 #ifdef CONFIG_BCACHEFS_POSIX_ACL
1483         if (c->opts.acl)
1484                 sb->s_flags     |= SB_POSIXACL;
1485 #endif
1486
1487         vinode = bch2_vfs_inode_get(c, BCACHEFS_ROOT_INO);
1488         if (IS_ERR(vinode)) {
1489                 bch_err(c, "error mounting: error getting root inode %i",
1490                         (int) PTR_ERR(vinode));
1491                 ret = PTR_ERR(vinode);
1492                 goto err_put_super;
1493         }
1494
1495         sb->s_root = d_make_root(vinode);
1496         if (!sb->s_root) {
1497                 bch_err(c, "error mounting: error allocating root dentry");
1498                 ret = -ENOMEM;
1499                 goto err_put_super;
1500         }
1501
1502         sb->s_flags |= SB_ACTIVE;
1503 out:
1504         return dget(sb->s_root);
1505
1506 err_put_super:
1507         deactivate_locked_super(sb);
1508         return ERR_PTR(ret);
1509 }
1510
1511 static void bch2_kill_sb(struct super_block *sb)
1512 {
1513         struct bch_fs *c = sb->s_fs_info;
1514
1515         generic_shutdown_super(sb);
1516
1517         if (test_bit(BCH_FS_BDEV_MOUNTED, &c->flags))
1518                 bch2_fs_stop(c);
1519         else
1520                 closure_put(&c->cl);
1521 }
1522
1523 static struct file_system_type bcache_fs_type = {
1524         .owner          = THIS_MODULE,
1525         .name           = "bcachefs",
1526         .mount          = bch2_mount,
1527         .kill_sb        = bch2_kill_sb,
1528         .fs_flags       = FS_REQUIRES_DEV,
1529 };
1530
1531 MODULE_ALIAS_FS("bcachefs");
1532
1533 void bch2_vfs_exit(void)
1534 {
1535         unregister_filesystem(&bcache_fs_type);
1536         if (bch2_inode_cache)
1537                 kmem_cache_destroy(bch2_inode_cache);
1538 }
1539
1540 int __init bch2_vfs_init(void)
1541 {
1542         int ret = -ENOMEM;
1543
1544         bch2_inode_cache = KMEM_CACHE(bch_inode_info, 0);
1545         if (!bch2_inode_cache)
1546                 goto err;
1547
1548         ret = register_filesystem(&bcache_fs_type);
1549         if (ret)
1550                 goto err;
1551
1552         return 0;
1553 err:
1554         bch2_vfs_exit();
1555         return ret;
1556 }
1557
1558 #endif /* NO_BCACHEFS_FS */