| 1 | // SPDX-License-Identifier: GPL-2.0+ |
| 2 | /* |
| 3 | * NILFS module and super block management. |
| 4 | * |
| 5 | * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. |
| 6 | * |
| 7 | * Written by Ryusuke Konishi. |
| 8 | */ |
| 9 | /* |
| 10 | * linux/fs/ext2/super.c |
| 11 | * |
| 12 | * Copyright (C) 1992, 1993, 1994, 1995 |
| 13 | * Remy Card (card@masi.ibp.fr) |
| 14 | * Laboratoire MASI - Institut Blaise Pascal |
| 15 | * Universite Pierre et Marie Curie (Paris VI) |
| 16 | * |
| 17 | * from |
| 18 | * |
| 19 | * linux/fs/minix/inode.c |
| 20 | * |
| 21 | * Copyright (C) 1991, 1992 Linus Torvalds |
| 22 | * |
| 23 | * Big-endian to little-endian byte-swapping/bitmaps by |
| 24 | * David S. Miller (davem@caip.rutgers.edu), 1995 |
| 25 | */ |
| 26 | |
| 27 | #include <linux/module.h> |
| 28 | #include <linux/string.h> |
| 29 | #include <linux/slab.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/blkdev.h> |
| 32 | #include <linux/crc32.h> |
| 33 | #include <linux/vfs.h> |
| 34 | #include <linux/writeback.h> |
| 35 | #include <linux/seq_file.h> |
| 36 | #include <linux/mount.h> |
| 37 | #include <linux/fs_context.h> |
| 38 | #include <linux/fs_parser.h> |
| 39 | #include "nilfs.h" |
| 40 | #include "export.h" |
| 41 | #include "mdt.h" |
| 42 | #include "alloc.h" |
| 43 | #include "btree.h" |
| 44 | #include "btnode.h" |
| 45 | #include "page.h" |
| 46 | #include "cpfile.h" |
| 47 | #include "sufile.h" /* nilfs_sufile_resize(), nilfs_sufile_set_alloc_range() */ |
| 48 | #include "ifile.h" |
| 49 | #include "dat.h" |
| 50 | #include "segment.h" |
| 51 | #include "segbuf.h" |
| 52 | |
| 53 | MODULE_AUTHOR("NTT Corp." ); |
| 54 | MODULE_DESCRIPTION("A New Implementation of the Log-structured Filesystem " |
| 55 | "(NILFS)" ); |
| 56 | MODULE_LICENSE("GPL" ); |
| 57 | |
| 58 | static struct kmem_cache *nilfs_inode_cachep; |
| 59 | struct kmem_cache *nilfs_transaction_cachep; |
| 60 | struct kmem_cache *nilfs_segbuf_cachep; |
| 61 | struct kmem_cache *nilfs_btree_path_cache; |
| 62 | |
| 63 | static int nilfs_setup_super(struct super_block *sb, int is_mount); |
| 64 | |
| 65 | void __nilfs_msg(struct super_block *sb, const char *fmt, ...) |
| 66 | { |
| 67 | struct va_format vaf; |
| 68 | va_list args; |
| 69 | int level; |
| 70 | |
| 71 | va_start(args, fmt); |
| 72 | |
| 73 | level = printk_get_level(buffer: fmt); |
| 74 | vaf.fmt = printk_skip_level(buffer: fmt); |
| 75 | vaf.va = &args; |
| 76 | |
| 77 | if (sb) |
| 78 | printk("%c%cNILFS (%s): %pV\n" , |
| 79 | KERN_SOH_ASCII, level, sb->s_id, &vaf); |
| 80 | else |
| 81 | printk("%c%cNILFS: %pV\n" , |
| 82 | KERN_SOH_ASCII, level, &vaf); |
| 83 | |
| 84 | va_end(args); |
| 85 | } |
| 86 | |
| 87 | static void nilfs_set_error(struct super_block *sb) |
| 88 | { |
| 89 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 90 | struct nilfs_super_block **sbp; |
| 91 | |
| 92 | down_write(sem: &nilfs->ns_sem); |
| 93 | if (!(nilfs->ns_mount_state & NILFS_ERROR_FS)) { |
| 94 | nilfs->ns_mount_state |= NILFS_ERROR_FS; |
| 95 | sbp = nilfs_prepare_super(sb, flip: 0); |
| 96 | if (likely(sbp)) { |
| 97 | sbp[0]->s_state |= cpu_to_le16(NILFS_ERROR_FS); |
| 98 | if (sbp[1]) |
| 99 | sbp[1]->s_state |= cpu_to_le16(NILFS_ERROR_FS); |
| 100 | nilfs_commit_super(sb, flag: NILFS_SB_COMMIT_ALL); |
| 101 | } |
| 102 | } |
| 103 | up_write(sem: &nilfs->ns_sem); |
| 104 | } |
| 105 | |
| 106 | /** |
| 107 | * __nilfs_error() - report failure condition on a filesystem |
| 108 | * @sb: super block instance |
| 109 | * @function: name of calling function |
| 110 | * @fmt: format string for message to be output |
| 111 | * @...: optional arguments to @fmt |
| 112 | * |
| 113 | * __nilfs_error() sets an ERROR_FS flag on the superblock as well as |
| 114 | * reporting an error message. This function should be called when |
| 115 | * NILFS detects incoherences or defects of meta data on disk. |
| 116 | * |
| 117 | * This implements the body of nilfs_error() macro. Normally, |
| 118 | * nilfs_error() should be used. As for sustainable errors such as a |
| 119 | * single-shot I/O error, nilfs_err() should be used instead. |
| 120 | * |
| 121 | * Callers should not add a trailing newline since this will do it. |
| 122 | */ |
| 123 | void __nilfs_error(struct super_block *sb, const char *function, |
| 124 | const char *fmt, ...) |
| 125 | { |
| 126 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 127 | struct va_format vaf; |
| 128 | va_list args; |
| 129 | |
| 130 | va_start(args, fmt); |
| 131 | |
| 132 | vaf.fmt = fmt; |
| 133 | vaf.va = &args; |
| 134 | |
| 135 | printk(KERN_CRIT "NILFS error (device %s): %s: %pV\n" , |
| 136 | sb->s_id, function, &vaf); |
| 137 | |
| 138 | va_end(args); |
| 139 | |
| 140 | if (!sb_rdonly(sb)) { |
| 141 | nilfs_set_error(sb); |
| 142 | |
| 143 | if (nilfs_test_opt(nilfs, ERRORS_RO)) { |
| 144 | printk(KERN_CRIT "Remounting filesystem read-only\n" ); |
| 145 | sb->s_flags |= SB_RDONLY; |
| 146 | } |
| 147 | } |
| 148 | |
| 149 | if (nilfs_test_opt(nilfs, ERRORS_PANIC)) |
| 150 | panic(fmt: "NILFS (device %s): panic forced after error\n" , |
| 151 | sb->s_id); |
| 152 | } |
| 153 | |
| 154 | struct inode *nilfs_alloc_inode(struct super_block *sb) |
| 155 | { |
| 156 | struct nilfs_inode_info *ii; |
| 157 | |
| 158 | ii = alloc_inode_sb(sb, nilfs_inode_cachep, GFP_NOFS); |
| 159 | if (!ii) |
| 160 | return NULL; |
| 161 | ii->i_bh = NULL; |
| 162 | ii->i_state = 0; |
| 163 | ii->i_type = 0; |
| 164 | ii->i_cno = 0; |
| 165 | ii->i_assoc_inode = NULL; |
| 166 | ii->i_bmap = &ii->i_bmap_data; |
| 167 | return &ii->vfs_inode; |
| 168 | } |
| 169 | |
| 170 | static void nilfs_free_inode(struct inode *inode) |
| 171 | { |
| 172 | if (nilfs_is_metadata_file_inode(inode)) |
| 173 | nilfs_mdt_destroy(inode); |
| 174 | |
| 175 | kmem_cache_free(s: nilfs_inode_cachep, objp: NILFS_I(inode)); |
| 176 | } |
| 177 | |
| 178 | static int nilfs_sync_super(struct super_block *sb, int flag) |
| 179 | { |
| 180 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 181 | int err; |
| 182 | |
| 183 | retry: |
| 184 | set_buffer_dirty(nilfs->ns_sbh[0]); |
| 185 | if (nilfs_test_opt(nilfs, BARRIER)) { |
| 186 | err = __sync_dirty_buffer(bh: nilfs->ns_sbh[0], |
| 187 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA); |
| 188 | } else { |
| 189 | err = sync_dirty_buffer(bh: nilfs->ns_sbh[0]); |
| 190 | } |
| 191 | |
| 192 | if (unlikely(err)) { |
| 193 | nilfs_err(sb, "unable to write superblock: err=%d" , err); |
| 194 | if (err == -EIO && nilfs->ns_sbh[1]) { |
| 195 | /* |
| 196 | * sbp[0] points to newer log than sbp[1], |
| 197 | * so copy sbp[0] to sbp[1] to take over sbp[0]. |
| 198 | */ |
| 199 | memcpy(nilfs->ns_sbp[1], nilfs->ns_sbp[0], |
| 200 | nilfs->ns_sbsize); |
| 201 | nilfs_fall_back_super_block(nilfs); |
| 202 | goto retry; |
| 203 | } |
| 204 | } else { |
| 205 | struct nilfs_super_block *sbp = nilfs->ns_sbp[0]; |
| 206 | |
| 207 | nilfs->ns_sbwcount++; |
| 208 | |
| 209 | /* |
| 210 | * The latest segment becomes trailable from the position |
| 211 | * written in superblock. |
| 212 | */ |
| 213 | clear_nilfs_discontinued(nilfs); |
| 214 | |
| 215 | /* update GC protection for recent segments */ |
| 216 | if (nilfs->ns_sbh[1]) { |
| 217 | if (flag == NILFS_SB_COMMIT_ALL) { |
| 218 | set_buffer_dirty(nilfs->ns_sbh[1]); |
| 219 | if (sync_dirty_buffer(bh: nilfs->ns_sbh[1]) < 0) |
| 220 | goto out; |
| 221 | } |
| 222 | if (le64_to_cpu(nilfs->ns_sbp[1]->s_last_cno) < |
| 223 | le64_to_cpu(nilfs->ns_sbp[0]->s_last_cno)) |
| 224 | sbp = nilfs->ns_sbp[1]; |
| 225 | } |
| 226 | |
| 227 | spin_lock(lock: &nilfs->ns_last_segment_lock); |
| 228 | nilfs->ns_prot_seq = le64_to_cpu(sbp->s_last_seq); |
| 229 | spin_unlock(lock: &nilfs->ns_last_segment_lock); |
| 230 | } |
| 231 | out: |
| 232 | return err; |
| 233 | } |
| 234 | |
| 235 | void nilfs_set_log_cursor(struct nilfs_super_block *sbp, |
| 236 | struct the_nilfs *nilfs) |
| 237 | { |
| 238 | sector_t nfreeblocks; |
| 239 | |
| 240 | /* nilfs->ns_sem must be locked by the caller. */ |
| 241 | nilfs_count_free_blocks(nilfs, &nfreeblocks); |
| 242 | sbp->s_free_blocks_count = cpu_to_le64(nfreeblocks); |
| 243 | |
| 244 | spin_lock(lock: &nilfs->ns_last_segment_lock); |
| 245 | sbp->s_last_seq = cpu_to_le64(nilfs->ns_last_seq); |
| 246 | sbp->s_last_pseg = cpu_to_le64(nilfs->ns_last_pseg); |
| 247 | sbp->s_last_cno = cpu_to_le64(nilfs->ns_last_cno); |
| 248 | spin_unlock(lock: &nilfs->ns_last_segment_lock); |
| 249 | } |
| 250 | |
| 251 | struct nilfs_super_block **nilfs_prepare_super(struct super_block *sb, |
| 252 | int flip) |
| 253 | { |
| 254 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 255 | struct nilfs_super_block **sbp = nilfs->ns_sbp; |
| 256 | |
| 257 | /* nilfs->ns_sem must be locked by the caller. */ |
| 258 | if (sbp[0]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) { |
| 259 | if (sbp[1] && |
| 260 | sbp[1]->s_magic == cpu_to_le16(NILFS_SUPER_MAGIC)) { |
| 261 | memcpy(sbp[0], sbp[1], nilfs->ns_sbsize); |
| 262 | } else { |
| 263 | nilfs_crit(sb, "superblock broke" ); |
| 264 | return NULL; |
| 265 | } |
| 266 | } else if (sbp[1] && |
| 267 | sbp[1]->s_magic != cpu_to_le16(NILFS_SUPER_MAGIC)) { |
| 268 | memcpy(sbp[1], sbp[0], nilfs->ns_sbsize); |
| 269 | } |
| 270 | |
| 271 | if (flip && sbp[1]) |
| 272 | nilfs_swap_super_block(nilfs); |
| 273 | |
| 274 | return sbp; |
| 275 | } |
| 276 | |
| 277 | int nilfs_commit_super(struct super_block *sb, int flag) |
| 278 | { |
| 279 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 280 | struct nilfs_super_block **sbp = nilfs->ns_sbp; |
| 281 | time64_t t; |
| 282 | |
| 283 | /* nilfs->ns_sem must be locked by the caller. */ |
| 284 | t = ktime_get_real_seconds(); |
| 285 | nilfs->ns_sbwtime = t; |
| 286 | sbp[0]->s_wtime = cpu_to_le64(t); |
| 287 | sbp[0]->s_sum = 0; |
| 288 | sbp[0]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed, |
| 289 | (unsigned char *)sbp[0], |
| 290 | nilfs->ns_sbsize)); |
| 291 | if (flag == NILFS_SB_COMMIT_ALL && sbp[1]) { |
| 292 | sbp[1]->s_wtime = sbp[0]->s_wtime; |
| 293 | sbp[1]->s_sum = 0; |
| 294 | sbp[1]->s_sum = cpu_to_le32(crc32_le(nilfs->ns_crc_seed, |
| 295 | (unsigned char *)sbp[1], |
| 296 | nilfs->ns_sbsize)); |
| 297 | } |
| 298 | clear_nilfs_sb_dirty(nilfs); |
| 299 | nilfs->ns_flushed_device = 1; |
| 300 | /* make sure store to ns_flushed_device cannot be reordered */ |
| 301 | smp_wmb(); |
| 302 | return nilfs_sync_super(sb, flag); |
| 303 | } |
| 304 | |
| 305 | /** |
| 306 | * nilfs_cleanup_super() - write filesystem state for cleanup |
| 307 | * @sb: super block instance to be unmounted or degraded to read-only |
| 308 | * |
| 309 | * This function restores state flags in the on-disk super block. |
| 310 | * This will set "clean" flag (i.e. NILFS_VALID_FS) unless the |
| 311 | * filesystem was not clean previously. |
| 312 | * |
| 313 | * Return: 0 on success, %-EIO if I/O error or superblock is corrupted. |
| 314 | */ |
| 315 | int nilfs_cleanup_super(struct super_block *sb) |
| 316 | { |
| 317 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 318 | struct nilfs_super_block **sbp; |
| 319 | int flag = NILFS_SB_COMMIT; |
| 320 | int ret = -EIO; |
| 321 | |
| 322 | sbp = nilfs_prepare_super(sb, flip: 0); |
| 323 | if (sbp) { |
| 324 | sbp[0]->s_state = cpu_to_le16(nilfs->ns_mount_state); |
| 325 | nilfs_set_log_cursor(sbp: sbp[0], nilfs); |
| 326 | if (sbp[1] && sbp[0]->s_last_cno == sbp[1]->s_last_cno) { |
| 327 | /* |
| 328 | * make the "clean" flag also to the opposite |
| 329 | * super block if both super blocks point to |
| 330 | * the same checkpoint. |
| 331 | */ |
| 332 | sbp[1]->s_state = sbp[0]->s_state; |
| 333 | flag = NILFS_SB_COMMIT_ALL; |
| 334 | } |
| 335 | ret = nilfs_commit_super(sb, flag); |
| 336 | } |
| 337 | return ret; |
| 338 | } |
| 339 | |
| 340 | /** |
| 341 | * nilfs_move_2nd_super - relocate secondary super block |
| 342 | * @sb: super block instance |
| 343 | * @sb2off: new offset of the secondary super block (in bytes) |
| 344 | * |
| 345 | * Return: 0 on success, or a negative error code on failure. |
| 346 | */ |
| 347 | static int nilfs_move_2nd_super(struct super_block *sb, loff_t sb2off) |
| 348 | { |
| 349 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 350 | struct buffer_head *nsbh; |
| 351 | struct nilfs_super_block *nsbp; |
| 352 | sector_t blocknr, newblocknr; |
| 353 | unsigned long offset; |
| 354 | int sb2i; /* array index of the secondary superblock */ |
| 355 | int ret = 0; |
| 356 | |
| 357 | /* nilfs->ns_sem must be locked by the caller. */ |
| 358 | if (nilfs->ns_sbh[1] && |
| 359 | nilfs->ns_sbh[1]->b_blocknr > nilfs->ns_first_data_block) { |
| 360 | sb2i = 1; |
| 361 | blocknr = nilfs->ns_sbh[1]->b_blocknr; |
| 362 | } else if (nilfs->ns_sbh[0]->b_blocknr > nilfs->ns_first_data_block) { |
| 363 | sb2i = 0; |
| 364 | blocknr = nilfs->ns_sbh[0]->b_blocknr; |
| 365 | } else { |
| 366 | sb2i = -1; |
| 367 | blocknr = 0; |
| 368 | } |
| 369 | if (sb2i >= 0 && (u64)blocknr << nilfs->ns_blocksize_bits == sb2off) |
| 370 | goto out; /* super block location is unchanged */ |
| 371 | |
| 372 | /* Get new super block buffer */ |
| 373 | newblocknr = sb2off >> nilfs->ns_blocksize_bits; |
| 374 | offset = sb2off & (nilfs->ns_blocksize - 1); |
| 375 | nsbh = sb_getblk(sb, block: newblocknr); |
| 376 | if (!nsbh) { |
| 377 | nilfs_warn(sb, |
| 378 | "unable to move secondary superblock to block %llu" , |
| 379 | (unsigned long long)newblocknr); |
| 380 | ret = -EIO; |
| 381 | goto out; |
| 382 | } |
| 383 | nsbp = (void *)nsbh->b_data + offset; |
| 384 | |
| 385 | lock_buffer(bh: nsbh); |
| 386 | if (sb2i >= 0) { |
| 387 | /* |
| 388 | * The position of the second superblock only changes by 4KiB, |
| 389 | * which is larger than the maximum superblock data size |
| 390 | * (= 1KiB), so there is no need to use memmove() to allow |
| 391 | * overlap between source and destination. |
| 392 | */ |
| 393 | memcpy(nsbp, nilfs->ns_sbp[sb2i], nilfs->ns_sbsize); |
| 394 | |
| 395 | /* |
| 396 | * Zero fill after copy to avoid overwriting in case of move |
| 397 | * within the same block. |
| 398 | */ |
| 399 | memset(nsbh->b_data, 0, offset); |
| 400 | memset((void *)nsbp + nilfs->ns_sbsize, 0, |
| 401 | nsbh->b_size - offset - nilfs->ns_sbsize); |
| 402 | } else { |
| 403 | memset(nsbh->b_data, 0, nsbh->b_size); |
| 404 | } |
| 405 | set_buffer_uptodate(nsbh); |
| 406 | unlock_buffer(bh: nsbh); |
| 407 | |
| 408 | if (sb2i >= 0) { |
| 409 | brelse(bh: nilfs->ns_sbh[sb2i]); |
| 410 | nilfs->ns_sbh[sb2i] = nsbh; |
| 411 | nilfs->ns_sbp[sb2i] = nsbp; |
| 412 | } else if (nilfs->ns_sbh[0]->b_blocknr < nilfs->ns_first_data_block) { |
| 413 | /* secondary super block will be restored to index 1 */ |
| 414 | nilfs->ns_sbh[1] = nsbh; |
| 415 | nilfs->ns_sbp[1] = nsbp; |
| 416 | } else { |
| 417 | brelse(bh: nsbh); |
| 418 | } |
| 419 | out: |
| 420 | return ret; |
| 421 | } |
| 422 | |
| 423 | /** |
| 424 | * nilfs_resize_fs - resize the filesystem |
| 425 | * @sb: super block instance |
| 426 | * @newsize: new size of the filesystem (in bytes) |
| 427 | * |
| 428 | * Return: 0 on success, or a negative error code on failure. |
| 429 | */ |
| 430 | int nilfs_resize_fs(struct super_block *sb, __u64 newsize) |
| 431 | { |
| 432 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 433 | struct nilfs_super_block **sbp; |
| 434 | __u64 devsize, newnsegs; |
| 435 | loff_t sb2off; |
| 436 | int ret; |
| 437 | |
| 438 | ret = -ERANGE; |
| 439 | devsize = bdev_nr_bytes(bdev: sb->s_bdev); |
| 440 | if (newsize > devsize) |
| 441 | goto out; |
| 442 | |
| 443 | /* |
| 444 | * Prevent underflow in second superblock position calculation. |
| 445 | * The exact minimum size check is done in nilfs_sufile_resize(). |
| 446 | */ |
| 447 | if (newsize < 4096) { |
| 448 | ret = -ENOSPC; |
| 449 | goto out; |
| 450 | } |
| 451 | |
| 452 | /* |
| 453 | * Write lock is required to protect some functions depending |
| 454 | * on the number of segments, the number of reserved segments, |
| 455 | * and so forth. |
| 456 | */ |
| 457 | down_write(sem: &nilfs->ns_segctor_sem); |
| 458 | |
| 459 | sb2off = NILFS_SB2_OFFSET_BYTES(newsize); |
| 460 | newnsegs = sb2off >> nilfs->ns_blocksize_bits; |
| 461 | newnsegs = div64_ul(newnsegs, nilfs->ns_blocks_per_segment); |
| 462 | |
| 463 | ret = nilfs_sufile_resize(sufile: nilfs->ns_sufile, newnsegs); |
| 464 | up_write(sem: &nilfs->ns_segctor_sem); |
| 465 | if (ret < 0) |
| 466 | goto out; |
| 467 | |
| 468 | ret = nilfs_construct_segment(sb); |
| 469 | if (ret < 0) |
| 470 | goto out; |
| 471 | |
| 472 | down_write(sem: &nilfs->ns_sem); |
| 473 | nilfs_move_2nd_super(sb, sb2off); |
| 474 | ret = -EIO; |
| 475 | sbp = nilfs_prepare_super(sb, flip: 0); |
| 476 | if (likely(sbp)) { |
| 477 | nilfs_set_log_cursor(sbp: sbp[0], nilfs); |
| 478 | /* |
| 479 | * Drop NILFS_RESIZE_FS flag for compatibility with |
| 480 | * mount-time resize which may be implemented in a |
| 481 | * future release. |
| 482 | */ |
| 483 | sbp[0]->s_state = cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & |
| 484 | ~NILFS_RESIZE_FS); |
| 485 | sbp[0]->s_dev_size = cpu_to_le64(newsize); |
| 486 | sbp[0]->s_nsegments = cpu_to_le64(nilfs->ns_nsegments); |
| 487 | if (sbp[1]) |
| 488 | memcpy(sbp[1], sbp[0], nilfs->ns_sbsize); |
| 489 | ret = nilfs_commit_super(sb, flag: NILFS_SB_COMMIT_ALL); |
| 490 | } |
| 491 | up_write(sem: &nilfs->ns_sem); |
| 492 | |
| 493 | /* |
| 494 | * Reset the range of allocatable segments last. This order |
| 495 | * is important in the case of expansion because the secondary |
| 496 | * superblock must be protected from log write until migration |
| 497 | * completes. |
| 498 | */ |
| 499 | if (!ret) |
| 500 | nilfs_sufile_set_alloc_range(sufile: nilfs->ns_sufile, start: 0, end: newnsegs - 1); |
| 501 | out: |
| 502 | return ret; |
| 503 | } |
| 504 | |
| 505 | static void nilfs_put_super(struct super_block *sb) |
| 506 | { |
| 507 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 508 | |
| 509 | nilfs_detach_log_writer(sb); |
| 510 | |
| 511 | if (!sb_rdonly(sb)) { |
| 512 | down_write(sem: &nilfs->ns_sem); |
| 513 | nilfs_cleanup_super(sb); |
| 514 | up_write(sem: &nilfs->ns_sem); |
| 515 | } |
| 516 | |
| 517 | nilfs_sysfs_delete_device_group(nilfs); |
| 518 | iput(nilfs->ns_sufile); |
| 519 | iput(nilfs->ns_cpfile); |
| 520 | iput(nilfs->ns_dat); |
| 521 | |
| 522 | destroy_nilfs(nilfs); |
| 523 | sb->s_fs_info = NULL; |
| 524 | } |
| 525 | |
| 526 | static int nilfs_sync_fs(struct super_block *sb, int wait) |
| 527 | { |
| 528 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 529 | struct nilfs_super_block **sbp; |
| 530 | int err = 0; |
| 531 | |
| 532 | /* This function is called when super block should be written back */ |
| 533 | if (wait) |
| 534 | err = nilfs_construct_segment(sb); |
| 535 | |
| 536 | down_write(sem: &nilfs->ns_sem); |
| 537 | if (nilfs_sb_dirty(nilfs)) { |
| 538 | sbp = nilfs_prepare_super(sb, flip: nilfs_sb_will_flip(nilfs)); |
| 539 | if (likely(sbp)) { |
| 540 | nilfs_set_log_cursor(sbp: sbp[0], nilfs); |
| 541 | nilfs_commit_super(sb, flag: NILFS_SB_COMMIT); |
| 542 | } |
| 543 | } |
| 544 | up_write(sem: &nilfs->ns_sem); |
| 545 | |
| 546 | if (!err) |
| 547 | err = nilfs_flush_device(nilfs); |
| 548 | |
| 549 | return err; |
| 550 | } |
| 551 | |
| 552 | int nilfs_attach_checkpoint(struct super_block *sb, __u64 cno, int curr_mnt, |
| 553 | struct nilfs_root **rootp) |
| 554 | { |
| 555 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 556 | struct nilfs_root *root; |
| 557 | int err = -ENOMEM; |
| 558 | |
| 559 | root = nilfs_find_or_create_root( |
| 560 | nilfs, cno: curr_mnt ? NILFS_CPTREE_CURRENT_CNO : cno); |
| 561 | if (!root) |
| 562 | return err; |
| 563 | |
| 564 | if (root->ifile) |
| 565 | goto reuse; /* already attached checkpoint */ |
| 566 | |
| 567 | down_read(sem: &nilfs->ns_segctor_sem); |
| 568 | err = nilfs_ifile_read(sb, root, cno, inode_size: nilfs->ns_inode_size); |
| 569 | up_read(sem: &nilfs->ns_segctor_sem); |
| 570 | if (unlikely(err)) |
| 571 | goto failed; |
| 572 | |
| 573 | reuse: |
| 574 | *rootp = root; |
| 575 | return 0; |
| 576 | |
| 577 | failed: |
| 578 | if (err == -EINVAL) |
| 579 | nilfs_err(sb, "Invalid checkpoint (checkpoint number=%llu)" , |
| 580 | (unsigned long long)cno); |
| 581 | nilfs_put_root(root); |
| 582 | |
| 583 | return err; |
| 584 | } |
| 585 | |
| 586 | static int nilfs_freeze(struct super_block *sb) |
| 587 | { |
| 588 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 589 | int err; |
| 590 | |
| 591 | if (sb_rdonly(sb)) |
| 592 | return 0; |
| 593 | |
| 594 | /* Mark super block clean */ |
| 595 | down_write(sem: &nilfs->ns_sem); |
| 596 | err = nilfs_cleanup_super(sb); |
| 597 | up_write(sem: &nilfs->ns_sem); |
| 598 | return err; |
| 599 | } |
| 600 | |
| 601 | static int nilfs_unfreeze(struct super_block *sb) |
| 602 | { |
| 603 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 604 | |
| 605 | if (sb_rdonly(sb)) |
| 606 | return 0; |
| 607 | |
| 608 | down_write(sem: &nilfs->ns_sem); |
| 609 | nilfs_setup_super(sb, is_mount: false); |
| 610 | up_write(sem: &nilfs->ns_sem); |
| 611 | return 0; |
| 612 | } |
| 613 | |
| 614 | static int nilfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
| 615 | { |
| 616 | struct super_block *sb = dentry->d_sb; |
| 617 | struct nilfs_root *root = NILFS_I(inode: d_inode(dentry))->i_root; |
| 618 | struct the_nilfs *nilfs = root->nilfs; |
| 619 | u64 id = huge_encode_dev(dev: sb->s_bdev->bd_dev); |
| 620 | unsigned long long blocks; |
| 621 | unsigned long overhead; |
| 622 | unsigned long nrsvblocks; |
| 623 | sector_t nfreeblocks; |
| 624 | u64 nmaxinodes, nfreeinodes; |
| 625 | int err; |
| 626 | |
| 627 | /* |
| 628 | * Compute all of the segment blocks |
| 629 | * |
| 630 | * The blocks before first segment and after last segment |
| 631 | * are excluded. |
| 632 | */ |
| 633 | blocks = nilfs->ns_blocks_per_segment * nilfs->ns_nsegments |
| 634 | - nilfs->ns_first_data_block; |
| 635 | nrsvblocks = nilfs->ns_nrsvsegs * nilfs->ns_blocks_per_segment; |
| 636 | |
| 637 | /* |
| 638 | * Compute the overhead |
| 639 | * |
| 640 | * When distributing meta data blocks outside segment structure, |
| 641 | * We must count them as the overhead. |
| 642 | */ |
| 643 | overhead = 0; |
| 644 | |
| 645 | err = nilfs_count_free_blocks(nilfs, &nfreeblocks); |
| 646 | if (unlikely(err)) |
| 647 | return err; |
| 648 | |
| 649 | err = nilfs_ifile_count_free_inodes(root->ifile, |
| 650 | &nmaxinodes, &nfreeinodes); |
| 651 | if (unlikely(err)) { |
| 652 | nilfs_warn(sb, "failed to count free inodes: err=%d" , err); |
| 653 | if (err == -ERANGE) { |
| 654 | /* |
| 655 | * If nilfs_palloc_count_max_entries() returns |
| 656 | * -ERANGE error code then we simply treat |
| 657 | * curent inodes count as maximum possible and |
| 658 | * zero as free inodes value. |
| 659 | */ |
| 660 | nmaxinodes = atomic64_read(v: &root->inodes_count); |
| 661 | nfreeinodes = 0; |
| 662 | err = 0; |
| 663 | } else |
| 664 | return err; |
| 665 | } |
| 666 | |
| 667 | buf->f_type = NILFS_SUPER_MAGIC; |
| 668 | buf->f_bsize = sb->s_blocksize; |
| 669 | buf->f_blocks = blocks - overhead; |
| 670 | buf->f_bfree = nfreeblocks; |
| 671 | buf->f_bavail = (buf->f_bfree >= nrsvblocks) ? |
| 672 | (buf->f_bfree - nrsvblocks) : 0; |
| 673 | buf->f_files = nmaxinodes; |
| 674 | buf->f_ffree = nfreeinodes; |
| 675 | buf->f_namelen = NILFS_NAME_LEN; |
| 676 | buf->f_fsid = u64_to_fsid(v: id); |
| 677 | |
| 678 | return 0; |
| 679 | } |
| 680 | |
| 681 | static int nilfs_show_options(struct seq_file *seq, struct dentry *dentry) |
| 682 | { |
| 683 | struct super_block *sb = dentry->d_sb; |
| 684 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 685 | struct nilfs_root *root = NILFS_I(inode: d_inode(dentry))->i_root; |
| 686 | |
| 687 | if (!nilfs_test_opt(nilfs, BARRIER)) |
| 688 | seq_puts(m: seq, s: ",nobarrier" ); |
| 689 | if (root->cno != NILFS_CPTREE_CURRENT_CNO) |
| 690 | seq_printf(m: seq, fmt: ",cp=%llu" , (unsigned long long)root->cno); |
| 691 | if (nilfs_test_opt(nilfs, ERRORS_PANIC)) |
| 692 | seq_puts(m: seq, s: ",errors=panic" ); |
| 693 | if (nilfs_test_opt(nilfs, ERRORS_CONT)) |
| 694 | seq_puts(m: seq, s: ",errors=continue" ); |
| 695 | if (nilfs_test_opt(nilfs, STRICT_ORDER)) |
| 696 | seq_puts(m: seq, s: ",order=strict" ); |
| 697 | if (nilfs_test_opt(nilfs, NORECOVERY)) |
| 698 | seq_puts(m: seq, s: ",norecovery" ); |
| 699 | if (nilfs_test_opt(nilfs, DISCARD)) |
| 700 | seq_puts(m: seq, s: ",discard" ); |
| 701 | |
| 702 | return 0; |
| 703 | } |
| 704 | |
| 705 | static const struct super_operations nilfs_sops = { |
| 706 | .alloc_inode = nilfs_alloc_inode, |
| 707 | .free_inode = nilfs_free_inode, |
| 708 | .dirty_inode = nilfs_dirty_inode, |
| 709 | .evict_inode = nilfs_evict_inode, |
| 710 | .put_super = nilfs_put_super, |
| 711 | .sync_fs = nilfs_sync_fs, |
| 712 | .freeze_fs = nilfs_freeze, |
| 713 | .unfreeze_fs = nilfs_unfreeze, |
| 714 | .statfs = nilfs_statfs, |
| 715 | .show_options = nilfs_show_options |
| 716 | }; |
| 717 | |
| 718 | enum { |
| 719 | Opt_err, Opt_barrier, Opt_snapshot, Opt_order, Opt_norecovery, |
| 720 | Opt_discard, |
| 721 | }; |
| 722 | |
| 723 | static const struct constant_table nilfs_param_err[] = { |
| 724 | {"continue" , NILFS_MOUNT_ERRORS_CONT}, |
| 725 | {"panic" , NILFS_MOUNT_ERRORS_PANIC}, |
| 726 | {"remount-ro" , NILFS_MOUNT_ERRORS_RO}, |
| 727 | {} |
| 728 | }; |
| 729 | |
| 730 | static const struct fs_parameter_spec nilfs_param_spec[] = { |
| 731 | fsparam_enum ("errors" , Opt_err, nilfs_param_err), |
| 732 | fsparam_flag_no ("barrier" , Opt_barrier), |
| 733 | fsparam_u64 ("cp" , Opt_snapshot), |
| 734 | fsparam_string ("order" , Opt_order), |
| 735 | fsparam_flag ("norecovery" , Opt_norecovery), |
| 736 | fsparam_flag_no ("discard" , Opt_discard), |
| 737 | {} |
| 738 | }; |
| 739 | |
| 740 | struct nilfs_fs_context { |
| 741 | unsigned long ns_mount_opt; |
| 742 | __u64 cno; |
| 743 | }; |
| 744 | |
| 745 | static int nilfs_parse_param(struct fs_context *fc, struct fs_parameter *param) |
| 746 | { |
| 747 | struct nilfs_fs_context *nilfs = fc->fs_private; |
| 748 | int is_remount = fc->purpose == FS_CONTEXT_FOR_RECONFIGURE; |
| 749 | struct fs_parse_result result; |
| 750 | int opt; |
| 751 | |
| 752 | opt = fs_parse(fc, desc: nilfs_param_spec, param, result: &result); |
| 753 | if (opt < 0) |
| 754 | return opt; |
| 755 | |
| 756 | switch (opt) { |
| 757 | case Opt_barrier: |
| 758 | if (result.negated) |
| 759 | nilfs_clear_opt(nilfs, BARRIER); |
| 760 | else |
| 761 | nilfs_set_opt(nilfs, BARRIER); |
| 762 | break; |
| 763 | case Opt_order: |
| 764 | if (strcmp(param->string, "relaxed" ) == 0) |
| 765 | /* Ordered data semantics */ |
| 766 | nilfs_clear_opt(nilfs, STRICT_ORDER); |
| 767 | else if (strcmp(param->string, "strict" ) == 0) |
| 768 | /* Strict in-order semantics */ |
| 769 | nilfs_set_opt(nilfs, STRICT_ORDER); |
| 770 | else |
| 771 | return -EINVAL; |
| 772 | break; |
| 773 | case Opt_err: |
| 774 | nilfs->ns_mount_opt &= ~NILFS_MOUNT_ERROR_MODE; |
| 775 | nilfs->ns_mount_opt |= result.uint_32; |
| 776 | break; |
| 777 | case Opt_snapshot: |
| 778 | if (is_remount) { |
| 779 | struct super_block *sb = fc->root->d_sb; |
| 780 | |
| 781 | nilfs_err(sb, |
| 782 | "\"%s\" option is invalid for remount" , |
| 783 | param->key); |
| 784 | return -EINVAL; |
| 785 | } |
| 786 | if (result.uint_64 == 0) { |
| 787 | nilfs_err(NULL, |
| 788 | "invalid option \"cp=0\": invalid checkpoint number 0" ); |
| 789 | return -EINVAL; |
| 790 | } |
| 791 | nilfs->cno = result.uint_64; |
| 792 | break; |
| 793 | case Opt_norecovery: |
| 794 | nilfs_set_opt(nilfs, NORECOVERY); |
| 795 | break; |
| 796 | case Opt_discard: |
| 797 | if (result.negated) |
| 798 | nilfs_clear_opt(nilfs, DISCARD); |
| 799 | else |
| 800 | nilfs_set_opt(nilfs, DISCARD); |
| 801 | break; |
| 802 | default: |
| 803 | return -EINVAL; |
| 804 | } |
| 805 | |
| 806 | return 0; |
| 807 | } |
| 808 | |
| 809 | static int nilfs_setup_super(struct super_block *sb, int is_mount) |
| 810 | { |
| 811 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 812 | struct nilfs_super_block **sbp; |
| 813 | int max_mnt_count; |
| 814 | int mnt_count; |
| 815 | |
| 816 | /* nilfs->ns_sem must be locked by the caller. */ |
| 817 | sbp = nilfs_prepare_super(sb, flip: 0); |
| 818 | if (!sbp) |
| 819 | return -EIO; |
| 820 | |
| 821 | if (!is_mount) |
| 822 | goto skip_mount_setup; |
| 823 | |
| 824 | max_mnt_count = le16_to_cpu(sbp[0]->s_max_mnt_count); |
| 825 | mnt_count = le16_to_cpu(sbp[0]->s_mnt_count); |
| 826 | |
| 827 | if (nilfs->ns_mount_state & NILFS_ERROR_FS) { |
| 828 | nilfs_warn(sb, "mounting fs with errors" ); |
| 829 | #if 0 |
| 830 | } else if (max_mnt_count >= 0 && mnt_count >= max_mnt_count) { |
| 831 | nilfs_warn(sb, "maximal mount count reached" ); |
| 832 | #endif |
| 833 | } |
| 834 | if (!max_mnt_count) |
| 835 | sbp[0]->s_max_mnt_count = cpu_to_le16(NILFS_DFL_MAX_MNT_COUNT); |
| 836 | |
| 837 | sbp[0]->s_mnt_count = cpu_to_le16(mnt_count + 1); |
| 838 | sbp[0]->s_mtime = cpu_to_le64(ktime_get_real_seconds()); |
| 839 | |
| 840 | skip_mount_setup: |
| 841 | sbp[0]->s_state = |
| 842 | cpu_to_le16(le16_to_cpu(sbp[0]->s_state) & ~NILFS_VALID_FS); |
| 843 | /* synchronize sbp[1] with sbp[0] */ |
| 844 | if (sbp[1]) |
| 845 | memcpy(sbp[1], sbp[0], nilfs->ns_sbsize); |
| 846 | return nilfs_commit_super(sb, flag: NILFS_SB_COMMIT_ALL); |
| 847 | } |
| 848 | |
| 849 | struct nilfs_super_block *nilfs_read_super_block(struct super_block *sb, |
| 850 | u64 pos, int blocksize, |
| 851 | struct buffer_head **pbh) |
| 852 | { |
| 853 | unsigned long long sb_index = pos; |
| 854 | unsigned long offset; |
| 855 | |
| 856 | offset = do_div(sb_index, blocksize); |
| 857 | *pbh = sb_bread(sb, block: sb_index); |
| 858 | if (!*pbh) |
| 859 | return NULL; |
| 860 | return (struct nilfs_super_block *)((char *)(*pbh)->b_data + offset); |
| 861 | } |
| 862 | |
| 863 | int nilfs_store_magic(struct super_block *sb, |
| 864 | struct nilfs_super_block *sbp) |
| 865 | { |
| 866 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 867 | |
| 868 | sb->s_magic = le16_to_cpu(sbp->s_magic); |
| 869 | |
| 870 | /* FS independent flags */ |
| 871 | #ifdef NILFS_ATIME_DISABLE |
| 872 | sb->s_flags |= SB_NOATIME; |
| 873 | #endif |
| 874 | |
| 875 | nilfs->ns_resuid = le16_to_cpu(sbp->s_def_resuid); |
| 876 | nilfs->ns_resgid = le16_to_cpu(sbp->s_def_resgid); |
| 877 | nilfs->ns_interval = le32_to_cpu(sbp->s_c_interval); |
| 878 | nilfs->ns_watermark = le32_to_cpu(sbp->s_c_block_max); |
| 879 | |
| 880 | return 0; |
| 881 | } |
| 882 | |
| 883 | int nilfs_check_feature_compatibility(struct super_block *sb, |
| 884 | struct nilfs_super_block *sbp) |
| 885 | { |
| 886 | __u64 features; |
| 887 | |
| 888 | features = le64_to_cpu(sbp->s_feature_incompat) & |
| 889 | ~NILFS_FEATURE_INCOMPAT_SUPP; |
| 890 | if (features) { |
| 891 | nilfs_err(sb, |
| 892 | "couldn't mount because of unsupported optional features (%llx)" , |
| 893 | (unsigned long long)features); |
| 894 | return -EINVAL; |
| 895 | } |
| 896 | features = le64_to_cpu(sbp->s_feature_compat_ro) & |
| 897 | ~NILFS_FEATURE_COMPAT_RO_SUPP; |
| 898 | if (!sb_rdonly(sb) && features) { |
| 899 | nilfs_err(sb, |
| 900 | "couldn't mount RDWR because of unsupported optional features (%llx)" , |
| 901 | (unsigned long long)features); |
| 902 | return -EINVAL; |
| 903 | } |
| 904 | return 0; |
| 905 | } |
| 906 | |
| 907 | static int nilfs_get_root_dentry(struct super_block *sb, |
| 908 | struct nilfs_root *root, |
| 909 | struct dentry **root_dentry) |
| 910 | { |
| 911 | struct inode *inode; |
| 912 | struct dentry *dentry; |
| 913 | int ret = 0; |
| 914 | |
| 915 | inode = nilfs_iget(sb, root, NILFS_ROOT_INO); |
| 916 | if (IS_ERR(ptr: inode)) { |
| 917 | ret = PTR_ERR(ptr: inode); |
| 918 | nilfs_err(sb, "error %d getting root inode" , ret); |
| 919 | goto out; |
| 920 | } |
| 921 | if (!S_ISDIR(inode->i_mode) || !inode->i_blocks || !inode->i_size) { |
| 922 | iput(inode); |
| 923 | nilfs_err(sb, "corrupt root inode" ); |
| 924 | ret = -EINVAL; |
| 925 | goto out; |
| 926 | } |
| 927 | |
| 928 | if (root->cno == NILFS_CPTREE_CURRENT_CNO) { |
| 929 | dentry = d_find_alias(inode); |
| 930 | if (!dentry) { |
| 931 | dentry = d_make_root(inode); |
| 932 | if (!dentry) { |
| 933 | ret = -ENOMEM; |
| 934 | goto failed_dentry; |
| 935 | } |
| 936 | } else { |
| 937 | iput(inode); |
| 938 | } |
| 939 | } else { |
| 940 | dentry = d_obtain_root(inode); |
| 941 | if (IS_ERR(ptr: dentry)) { |
| 942 | ret = PTR_ERR(ptr: dentry); |
| 943 | goto failed_dentry; |
| 944 | } |
| 945 | } |
| 946 | *root_dentry = dentry; |
| 947 | out: |
| 948 | return ret; |
| 949 | |
| 950 | failed_dentry: |
| 951 | nilfs_err(sb, "error %d getting root dentry" , ret); |
| 952 | goto out; |
| 953 | } |
| 954 | |
| 955 | static int nilfs_attach_snapshot(struct super_block *s, __u64 cno, |
| 956 | struct dentry **root_dentry) |
| 957 | { |
| 958 | struct the_nilfs *nilfs = s->s_fs_info; |
| 959 | struct nilfs_root *root; |
| 960 | int ret; |
| 961 | |
| 962 | mutex_lock(&nilfs->ns_snapshot_mount_mutex); |
| 963 | |
| 964 | down_read(sem: &nilfs->ns_segctor_sem); |
| 965 | ret = nilfs_cpfile_is_snapshot(nilfs->ns_cpfile, cno); |
| 966 | up_read(sem: &nilfs->ns_segctor_sem); |
| 967 | if (ret < 0) { |
| 968 | ret = (ret == -ENOENT) ? -EINVAL : ret; |
| 969 | goto out; |
| 970 | } else if (!ret) { |
| 971 | nilfs_err(s, |
| 972 | "The specified checkpoint is not a snapshot (checkpoint number=%llu)" , |
| 973 | (unsigned long long)cno); |
| 974 | ret = -EINVAL; |
| 975 | goto out; |
| 976 | } |
| 977 | |
| 978 | ret = nilfs_attach_checkpoint(sb: s, cno, curr_mnt: false, rootp: &root); |
| 979 | if (ret) { |
| 980 | nilfs_err(s, |
| 981 | "error %d while loading snapshot (checkpoint number=%llu)" , |
| 982 | ret, (unsigned long long)cno); |
| 983 | goto out; |
| 984 | } |
| 985 | ret = nilfs_get_root_dentry(sb: s, root, root_dentry); |
| 986 | nilfs_put_root(root); |
| 987 | out: |
| 988 | mutex_unlock(lock: &nilfs->ns_snapshot_mount_mutex); |
| 989 | return ret; |
| 990 | } |
| 991 | |
| 992 | /** |
| 993 | * nilfs_tree_is_busy() - try to shrink dentries of a checkpoint |
| 994 | * @root_dentry: root dentry of the tree to be shrunk |
| 995 | * |
| 996 | * Return: true if the tree was in-use, false otherwise. |
| 997 | */ |
| 998 | static bool nilfs_tree_is_busy(struct dentry *root_dentry) |
| 999 | { |
| 1000 | shrink_dcache_parent(root_dentry); |
| 1001 | return d_count(dentry: root_dentry) > 1; |
| 1002 | } |
| 1003 | |
| 1004 | int nilfs_checkpoint_is_mounted(struct super_block *sb, __u64 cno) |
| 1005 | { |
| 1006 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 1007 | struct nilfs_root *root; |
| 1008 | struct inode *inode; |
| 1009 | struct dentry *dentry; |
| 1010 | int ret; |
| 1011 | |
| 1012 | if (cno > nilfs->ns_cno) |
| 1013 | return false; |
| 1014 | |
| 1015 | if (cno >= nilfs_last_cno(nilfs)) |
| 1016 | return true; /* protect recent checkpoints */ |
| 1017 | |
| 1018 | ret = false; |
| 1019 | root = nilfs_lookup_root(nilfs, cno); |
| 1020 | if (root) { |
| 1021 | inode = nilfs_ilookup(sb, root, NILFS_ROOT_INO); |
| 1022 | if (inode) { |
| 1023 | dentry = d_find_alias(inode); |
| 1024 | if (dentry) { |
| 1025 | ret = nilfs_tree_is_busy(root_dentry: dentry); |
| 1026 | dput(dentry); |
| 1027 | } |
| 1028 | iput(inode); |
| 1029 | } |
| 1030 | nilfs_put_root(root); |
| 1031 | } |
| 1032 | return ret; |
| 1033 | } |
| 1034 | |
| 1035 | /** |
| 1036 | * nilfs_fill_super() - initialize a super block instance |
| 1037 | * @sb: super_block |
| 1038 | * @fc: filesystem context |
| 1039 | * |
| 1040 | * This function is called exclusively by nilfs->ns_mount_mutex. |
| 1041 | * So, the recovery process is protected from other simultaneous mounts. |
| 1042 | * |
| 1043 | * Return: 0 on success, or a negative error code on failure. |
| 1044 | */ |
| 1045 | static int |
| 1046 | nilfs_fill_super(struct super_block *sb, struct fs_context *fc) |
| 1047 | { |
| 1048 | struct the_nilfs *nilfs; |
| 1049 | struct nilfs_root *fsroot; |
| 1050 | struct nilfs_fs_context *ctx = fc->fs_private; |
| 1051 | __u64 cno; |
| 1052 | int err; |
| 1053 | |
| 1054 | nilfs = alloc_nilfs(sb); |
| 1055 | if (!nilfs) |
| 1056 | return -ENOMEM; |
| 1057 | |
| 1058 | sb->s_fs_info = nilfs; |
| 1059 | |
| 1060 | err = init_nilfs(nilfs, sb); |
| 1061 | if (err) |
| 1062 | goto failed_nilfs; |
| 1063 | |
| 1064 | /* Copy in parsed mount options */ |
| 1065 | nilfs->ns_mount_opt = ctx->ns_mount_opt; |
| 1066 | |
| 1067 | sb->s_op = &nilfs_sops; |
| 1068 | sb->s_export_op = &nilfs_export_ops; |
| 1069 | sb->s_root = NULL; |
| 1070 | sb->s_time_gran = 1; |
| 1071 | sb->s_max_links = NILFS_LINK_MAX; |
| 1072 | |
| 1073 | sb->s_bdi = bdi_get(bdi: sb->s_bdev->bd_disk->bdi); |
| 1074 | |
| 1075 | err = load_nilfs(nilfs, sb); |
| 1076 | if (err) |
| 1077 | goto failed_nilfs; |
| 1078 | |
| 1079 | super_set_uuid(sb, uuid: nilfs->ns_sbp[0]->s_uuid, |
| 1080 | len: sizeof(nilfs->ns_sbp[0]->s_uuid)); |
| 1081 | super_set_sysfs_name_bdev(sb); |
| 1082 | |
| 1083 | cno = nilfs_last_cno(nilfs); |
| 1084 | err = nilfs_attach_checkpoint(sb, cno, curr_mnt: true, rootp: &fsroot); |
| 1085 | if (err) { |
| 1086 | nilfs_err(sb, |
| 1087 | "error %d while loading last checkpoint (checkpoint number=%llu)" , |
| 1088 | err, (unsigned long long)cno); |
| 1089 | goto failed_unload; |
| 1090 | } |
| 1091 | |
| 1092 | if (!sb_rdonly(sb)) { |
| 1093 | err = nilfs_attach_log_writer(sb, root: fsroot); |
| 1094 | if (err) |
| 1095 | goto failed_checkpoint; |
| 1096 | } |
| 1097 | |
| 1098 | err = nilfs_get_root_dentry(sb, root: fsroot, root_dentry: &sb->s_root); |
| 1099 | if (err) |
| 1100 | goto failed_segctor; |
| 1101 | |
| 1102 | nilfs_put_root(root: fsroot); |
| 1103 | |
| 1104 | if (!sb_rdonly(sb)) { |
| 1105 | down_write(sem: &nilfs->ns_sem); |
| 1106 | nilfs_setup_super(sb, is_mount: true); |
| 1107 | up_write(sem: &nilfs->ns_sem); |
| 1108 | } |
| 1109 | |
| 1110 | return 0; |
| 1111 | |
| 1112 | failed_segctor: |
| 1113 | nilfs_detach_log_writer(sb); |
| 1114 | |
| 1115 | failed_checkpoint: |
| 1116 | nilfs_put_root(root: fsroot); |
| 1117 | |
| 1118 | failed_unload: |
| 1119 | nilfs_sysfs_delete_device_group(nilfs); |
| 1120 | iput(nilfs->ns_sufile); |
| 1121 | iput(nilfs->ns_cpfile); |
| 1122 | iput(nilfs->ns_dat); |
| 1123 | |
| 1124 | failed_nilfs: |
| 1125 | destroy_nilfs(nilfs); |
| 1126 | return err; |
| 1127 | } |
| 1128 | |
| 1129 | static int nilfs_reconfigure(struct fs_context *fc) |
| 1130 | { |
| 1131 | struct nilfs_fs_context *ctx = fc->fs_private; |
| 1132 | struct super_block *sb = fc->root->d_sb; |
| 1133 | struct the_nilfs *nilfs = sb->s_fs_info; |
| 1134 | int err; |
| 1135 | |
| 1136 | sync_filesystem(sb); |
| 1137 | |
| 1138 | err = -EINVAL; |
| 1139 | |
| 1140 | if (!nilfs_valid_fs(nilfs)) { |
| 1141 | nilfs_warn(sb, |
| 1142 | "couldn't remount because the filesystem is in an incomplete recovery state" ); |
| 1143 | goto ignore_opts; |
| 1144 | } |
| 1145 | if ((bool)(fc->sb_flags & SB_RDONLY) == sb_rdonly(sb)) |
| 1146 | goto out; |
| 1147 | if (fc->sb_flags & SB_RDONLY) { |
| 1148 | sb->s_flags |= SB_RDONLY; |
| 1149 | |
| 1150 | /* |
| 1151 | * Remounting a valid RW partition RDONLY, so set |
| 1152 | * the RDONLY flag and then mark the partition as valid again. |
| 1153 | */ |
| 1154 | down_write(sem: &nilfs->ns_sem); |
| 1155 | nilfs_cleanup_super(sb); |
| 1156 | up_write(sem: &nilfs->ns_sem); |
| 1157 | } else { |
| 1158 | __u64 features; |
| 1159 | struct nilfs_root *root; |
| 1160 | |
| 1161 | /* |
| 1162 | * Mounting a RDONLY partition read-write, so reread and |
| 1163 | * store the current valid flag. (It may have been changed |
| 1164 | * by fsck since we originally mounted the partition.) |
| 1165 | */ |
| 1166 | down_read(sem: &nilfs->ns_sem); |
| 1167 | features = le64_to_cpu(nilfs->ns_sbp[0]->s_feature_compat_ro) & |
| 1168 | ~NILFS_FEATURE_COMPAT_RO_SUPP; |
| 1169 | up_read(sem: &nilfs->ns_sem); |
| 1170 | if (features) { |
| 1171 | nilfs_warn(sb, |
| 1172 | "couldn't remount RDWR because of unsupported optional features (%llx)" , |
| 1173 | (unsigned long long)features); |
| 1174 | err = -EROFS; |
| 1175 | goto ignore_opts; |
| 1176 | } |
| 1177 | |
| 1178 | sb->s_flags &= ~SB_RDONLY; |
| 1179 | |
| 1180 | root = NILFS_I(inode: d_inode(dentry: sb->s_root))->i_root; |
| 1181 | err = nilfs_attach_log_writer(sb, root); |
| 1182 | if (err) { |
| 1183 | sb->s_flags |= SB_RDONLY; |
| 1184 | goto ignore_opts; |
| 1185 | } |
| 1186 | |
| 1187 | down_write(sem: &nilfs->ns_sem); |
| 1188 | nilfs_setup_super(sb, is_mount: true); |
| 1189 | up_write(sem: &nilfs->ns_sem); |
| 1190 | } |
| 1191 | out: |
| 1192 | sb->s_flags = (sb->s_flags & ~SB_POSIXACL); |
| 1193 | /* Copy over parsed remount options */ |
| 1194 | nilfs->ns_mount_opt = ctx->ns_mount_opt; |
| 1195 | |
| 1196 | return 0; |
| 1197 | |
| 1198 | ignore_opts: |
| 1199 | return err; |
| 1200 | } |
| 1201 | |
| 1202 | static int |
| 1203 | nilfs_get_tree(struct fs_context *fc) |
| 1204 | { |
| 1205 | struct nilfs_fs_context *ctx = fc->fs_private; |
| 1206 | struct super_block *s; |
| 1207 | dev_t dev; |
| 1208 | int err; |
| 1209 | |
| 1210 | if (ctx->cno && !(fc->sb_flags & SB_RDONLY)) { |
| 1211 | nilfs_err(NULL, |
| 1212 | "invalid option \"cp=%llu\": read-only option is not specified" , |
| 1213 | ctx->cno); |
| 1214 | return -EINVAL; |
| 1215 | } |
| 1216 | |
| 1217 | err = lookup_bdev(pathname: fc->source, dev: &dev); |
| 1218 | if (err) |
| 1219 | return err; |
| 1220 | |
| 1221 | s = sget_dev(fc, dev); |
| 1222 | if (IS_ERR(ptr: s)) |
| 1223 | return PTR_ERR(ptr: s); |
| 1224 | |
| 1225 | if (!s->s_root) { |
| 1226 | err = setup_bdev_super(sb: s, sb_flags: fc->sb_flags, fc); |
| 1227 | if (!err) |
| 1228 | err = nilfs_fill_super(sb: s, fc); |
| 1229 | if (err) |
| 1230 | goto failed_super; |
| 1231 | |
| 1232 | s->s_flags |= SB_ACTIVE; |
| 1233 | } else if (!ctx->cno) { |
| 1234 | if (nilfs_tree_is_busy(root_dentry: s->s_root)) { |
| 1235 | if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { |
| 1236 | nilfs_err(s, |
| 1237 | "the device already has a %s mount." , |
| 1238 | sb_rdonly(s) ? "read-only" : "read/write" ); |
| 1239 | err = -EBUSY; |
| 1240 | goto failed_super; |
| 1241 | } |
| 1242 | } else { |
| 1243 | /* |
| 1244 | * Try reconfigure to setup mount states if the current |
| 1245 | * tree is not mounted and only snapshots use this sb. |
| 1246 | * |
| 1247 | * Since nilfs_reconfigure() requires fc->root to be |
| 1248 | * set, set it first and release it on failure. |
| 1249 | */ |
| 1250 | fc->root = dget(dentry: s->s_root); |
| 1251 | err = nilfs_reconfigure(fc); |
| 1252 | if (err) { |
| 1253 | dput(fc->root); |
| 1254 | fc->root = NULL; /* prevent double release */ |
| 1255 | goto failed_super; |
| 1256 | } |
| 1257 | return 0; |
| 1258 | } |
| 1259 | } |
| 1260 | |
| 1261 | if (ctx->cno) { |
| 1262 | struct dentry *root_dentry; |
| 1263 | |
| 1264 | err = nilfs_attach_snapshot(s, cno: ctx->cno, root_dentry: &root_dentry); |
| 1265 | if (err) |
| 1266 | goto failed_super; |
| 1267 | fc->root = root_dentry; |
| 1268 | return 0; |
| 1269 | } |
| 1270 | |
| 1271 | fc->root = dget(dentry: s->s_root); |
| 1272 | return 0; |
| 1273 | |
| 1274 | failed_super: |
| 1275 | deactivate_locked_super(sb: s); |
| 1276 | return err; |
| 1277 | } |
| 1278 | |
| 1279 | static void nilfs_free_fc(struct fs_context *fc) |
| 1280 | { |
| 1281 | kfree(objp: fc->fs_private); |
| 1282 | } |
| 1283 | |
| 1284 | static const struct fs_context_operations nilfs_context_ops = { |
| 1285 | .parse_param = nilfs_parse_param, |
| 1286 | .get_tree = nilfs_get_tree, |
| 1287 | .reconfigure = nilfs_reconfigure, |
| 1288 | .free = nilfs_free_fc, |
| 1289 | }; |
| 1290 | |
| 1291 | static int nilfs_init_fs_context(struct fs_context *fc) |
| 1292 | { |
| 1293 | struct nilfs_fs_context *ctx; |
| 1294 | |
| 1295 | ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); |
| 1296 | if (!ctx) |
| 1297 | return -ENOMEM; |
| 1298 | |
| 1299 | ctx->ns_mount_opt = NILFS_MOUNT_ERRORS_RO | NILFS_MOUNT_BARRIER; |
| 1300 | fc->fs_private = ctx; |
| 1301 | fc->ops = &nilfs_context_ops; |
| 1302 | |
| 1303 | return 0; |
| 1304 | } |
| 1305 | |
| 1306 | struct file_system_type nilfs_fs_type = { |
| 1307 | .owner = THIS_MODULE, |
| 1308 | .name = "nilfs2" , |
| 1309 | .kill_sb = kill_block_super, |
| 1310 | .fs_flags = FS_REQUIRES_DEV, |
| 1311 | .init_fs_context = nilfs_init_fs_context, |
| 1312 | .parameters = nilfs_param_spec, |
| 1313 | }; |
| 1314 | MODULE_ALIAS_FS("nilfs2" ); |
| 1315 | |
| 1316 | static void nilfs_inode_init_once(void *obj) |
| 1317 | { |
| 1318 | struct nilfs_inode_info *ii = obj; |
| 1319 | |
| 1320 | INIT_LIST_HEAD(list: &ii->i_dirty); |
| 1321 | #ifdef CONFIG_NILFS_XATTR |
| 1322 | init_rwsem(&ii->xattr_sem); |
| 1323 | #endif |
| 1324 | inode_init_once(&ii->vfs_inode); |
| 1325 | } |
| 1326 | |
| 1327 | static void nilfs_segbuf_init_once(void *obj) |
| 1328 | { |
| 1329 | memset(obj, 0, sizeof(struct nilfs_segment_buffer)); |
| 1330 | } |
| 1331 | |
| 1332 | static void nilfs_destroy_cachep(void) |
| 1333 | { |
| 1334 | /* |
| 1335 | * Make sure all delayed rcu free inodes are flushed before we |
| 1336 | * destroy cache. |
| 1337 | */ |
| 1338 | rcu_barrier(); |
| 1339 | |
| 1340 | kmem_cache_destroy(s: nilfs_inode_cachep); |
| 1341 | kmem_cache_destroy(s: nilfs_transaction_cachep); |
| 1342 | kmem_cache_destroy(s: nilfs_segbuf_cachep); |
| 1343 | kmem_cache_destroy(s: nilfs_btree_path_cache); |
| 1344 | } |
| 1345 | |
| 1346 | static int __init nilfs_init_cachep(void) |
| 1347 | { |
| 1348 | nilfs_inode_cachep = kmem_cache_create("nilfs2_inode_cache" , |
| 1349 | sizeof(struct nilfs_inode_info), 0, |
| 1350 | SLAB_RECLAIM_ACCOUNT|SLAB_ACCOUNT, |
| 1351 | nilfs_inode_init_once); |
| 1352 | if (!nilfs_inode_cachep) |
| 1353 | goto fail; |
| 1354 | |
| 1355 | nilfs_transaction_cachep = kmem_cache_create("nilfs2_transaction_cache" , |
| 1356 | sizeof(struct nilfs_transaction_info), 0, |
| 1357 | SLAB_RECLAIM_ACCOUNT, NULL); |
| 1358 | if (!nilfs_transaction_cachep) |
| 1359 | goto fail; |
| 1360 | |
| 1361 | nilfs_segbuf_cachep = kmem_cache_create("nilfs2_segbuf_cache" , |
| 1362 | sizeof(struct nilfs_segment_buffer), 0, |
| 1363 | SLAB_RECLAIM_ACCOUNT, nilfs_segbuf_init_once); |
| 1364 | if (!nilfs_segbuf_cachep) |
| 1365 | goto fail; |
| 1366 | |
| 1367 | nilfs_btree_path_cache = kmem_cache_create("nilfs2_btree_path_cache" , |
| 1368 | sizeof(struct nilfs_btree_path) * NILFS_BTREE_LEVEL_MAX, |
| 1369 | 0, 0, NULL); |
| 1370 | if (!nilfs_btree_path_cache) |
| 1371 | goto fail; |
| 1372 | |
| 1373 | return 0; |
| 1374 | |
| 1375 | fail: |
| 1376 | nilfs_destroy_cachep(); |
| 1377 | return -ENOMEM; |
| 1378 | } |
| 1379 | |
| 1380 | static int __init init_nilfs_fs(void) |
| 1381 | { |
| 1382 | int err; |
| 1383 | |
| 1384 | err = nilfs_init_cachep(); |
| 1385 | if (err) |
| 1386 | goto fail; |
| 1387 | |
| 1388 | err = nilfs_sysfs_init(); |
| 1389 | if (err) |
| 1390 | goto free_cachep; |
| 1391 | |
| 1392 | err = register_filesystem(&nilfs_fs_type); |
| 1393 | if (err) |
| 1394 | goto deinit_sysfs_entry; |
| 1395 | |
| 1396 | printk(KERN_INFO "NILFS version 2 loaded\n" ); |
| 1397 | return 0; |
| 1398 | |
| 1399 | deinit_sysfs_entry: |
| 1400 | nilfs_sysfs_exit(); |
| 1401 | free_cachep: |
| 1402 | nilfs_destroy_cachep(); |
| 1403 | fail: |
| 1404 | return err; |
| 1405 | } |
| 1406 | |
| 1407 | static void __exit exit_nilfs_fs(void) |
| 1408 | { |
| 1409 | nilfs_destroy_cachep(); |
| 1410 | nilfs_sysfs_exit(); |
| 1411 | unregister_filesystem(&nilfs_fs_type); |
| 1412 | } |
| 1413 | |
| 1414 | module_init(init_nilfs_fs) |
| 1415 | module_exit(exit_nilfs_fs) |
| 1416 | |