| 1 | /* |
| 2 | * Copyright (c) 2000-2006 Silicon Graphics, Inc. |
| 3 | * All Rights Reserved. |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or |
| 6 | * modify it under the terms of the GNU General Public License as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope that it would be useful, |
| 10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | * GNU General Public License for more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License |
| 15 | * along with this program; if not, write the Free Software Foundation, |
| 16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 17 | */ |
| 18 | #include "xfs.h" |
| 19 | #include "xfs_bit.h" |
| 20 | #include "xfs_log.h" |
| 21 | #include "xfs_clnt.h" |
| 22 | #include "xfs_inum.h" |
| 23 | #include "xfs_trans.h" |
| 24 | #include "xfs_sb.h" |
| 25 | #include "xfs_ag.h" |
| 26 | #include "xfs_dir2.h" |
| 27 | #include "xfs_alloc.h" |
| 28 | #include "xfs_dmapi.h" |
| 29 | #include "xfs_quota.h" |
| 30 | #include "xfs_mount.h" |
| 31 | #include "xfs_bmap_btree.h" |
| 32 | #include "xfs_alloc_btree.h" |
| 33 | #include "xfs_ialloc_btree.h" |
| 34 | #include "xfs_dir2_sf.h" |
| 35 | #include "xfs_attr_sf.h" |
| 36 | #include "xfs_dinode.h" |
| 37 | #include "xfs_inode.h" |
| 38 | #include "xfs_btree.h" |
| 39 | #include "xfs_ialloc.h" |
| 40 | #include "xfs_bmap.h" |
| 41 | #include "xfs_rtalloc.h" |
| 42 | #include "xfs_error.h" |
| 43 | #include "xfs_itable.h" |
| 44 | #include "xfs_rw.h" |
| 45 | #include "xfs_acl.h" |
| 46 | #include "xfs_attr.h" |
| 47 | #include "xfs_buf_item.h" |
| 48 | #include "xfs_utils.h" |
| 49 | #include "xfs_vnodeops.h" |
| 50 | #include "xfs_vfsops.h" |
| 51 | #include "xfs_version.h" |
| 52 | |
| 53 | #include <linux/namei.h> |
| 54 | #include <linux/init.h> |
| 55 | #include <linux/mount.h> |
| 56 | #include <linux/mempool.h> |
| 57 | #include <linux/writeback.h> |
| 58 | #include <linux/kthread.h> |
| 59 | #include <linux/freezer.h> |
| 60 | |
| 61 | static struct quotactl_ops xfs_quotactl_operations; |
| 62 | static struct super_operations xfs_super_operations; |
| 63 | static kmem_zone_t *xfs_vnode_zone; |
| 64 | static kmem_zone_t *xfs_ioend_zone; |
| 65 | mempool_t *xfs_ioend_pool; |
| 66 | |
| 67 | STATIC struct xfs_mount_args * |
| 68 | xfs_args_allocate( |
| 69 | struct super_block *sb, |
| 70 | int silent) |
| 71 | { |
| 72 | struct xfs_mount_args *args; |
| 73 | |
| 74 | args = kmem_zalloc(sizeof(struct xfs_mount_args), KM_SLEEP); |
| 75 | args->logbufs = args->logbufsize = -1; |
| 76 | strncpy(args->fsname, sb->s_id, MAXNAMELEN); |
| 77 | |
| 78 | /* Copy the already-parsed mount(2) flags we're interested in */ |
| 79 | if (sb->s_flags & MS_DIRSYNC) |
| 80 | args->flags |= XFSMNT_DIRSYNC; |
| 81 | if (sb->s_flags & MS_SYNCHRONOUS) |
| 82 | args->flags |= XFSMNT_WSYNC; |
| 83 | if (silent) |
| 84 | args->flags |= XFSMNT_QUIET; |
| 85 | args->flags |= XFSMNT_32BITINODES; |
| 86 | |
| 87 | return args; |
| 88 | } |
| 89 | |
| 90 | __uint64_t |
| 91 | xfs_max_file_offset( |
| 92 | unsigned int blockshift) |
| 93 | { |
| 94 | unsigned int pagefactor = 1; |
| 95 | unsigned int bitshift = BITS_PER_LONG - 1; |
| 96 | |
| 97 | /* Figure out maximum filesize, on Linux this can depend on |
| 98 | * the filesystem blocksize (on 32 bit platforms). |
| 99 | * __block_prepare_write does this in an [unsigned] long... |
| 100 | * page->index << (PAGE_CACHE_SHIFT - bbits) |
| 101 | * So, for page sized blocks (4K on 32 bit platforms), |
| 102 | * this wraps at around 8Tb (hence MAX_LFS_FILESIZE which is |
| 103 | * (((u64)PAGE_CACHE_SIZE << (BITS_PER_LONG-1))-1) |
| 104 | * but for smaller blocksizes it is less (bbits = log2 bsize). |
| 105 | * Note1: get_block_t takes a long (implicit cast from above) |
| 106 | * Note2: The Large Block Device (LBD and HAVE_SECTOR_T) patch |
| 107 | * can optionally convert the [unsigned] long from above into |
| 108 | * an [unsigned] long long. |
| 109 | */ |
| 110 | |
| 111 | #if BITS_PER_LONG == 32 |
| 112 | # if defined(CONFIG_LBD) |
| 113 | ASSERT(sizeof(sector_t) == 8); |
| 114 | pagefactor = PAGE_CACHE_SIZE; |
| 115 | bitshift = BITS_PER_LONG; |
| 116 | # else |
| 117 | pagefactor = PAGE_CACHE_SIZE >> (PAGE_CACHE_SHIFT - blockshift); |
| 118 | # endif |
| 119 | #endif |
| 120 | |
| 121 | return (((__uint64_t)pagefactor) << bitshift) - 1; |
| 122 | } |
| 123 | |
| 124 | STATIC_INLINE void |
| 125 | xfs_set_inodeops( |
| 126 | struct inode *inode) |
| 127 | { |
| 128 | switch (inode->i_mode & S_IFMT) { |
| 129 | case S_IFREG: |
| 130 | inode->i_op = &xfs_inode_operations; |
| 131 | inode->i_fop = &xfs_file_operations; |
| 132 | inode->i_mapping->a_ops = &xfs_address_space_operations; |
| 133 | break; |
| 134 | case S_IFDIR: |
| 135 | inode->i_op = &xfs_dir_inode_operations; |
| 136 | inode->i_fop = &xfs_dir_file_operations; |
| 137 | break; |
| 138 | case S_IFLNK: |
| 139 | inode->i_op = &xfs_symlink_inode_operations; |
| 140 | if (inode->i_blocks) |
| 141 | inode->i_mapping->a_ops = &xfs_address_space_operations; |
| 142 | break; |
| 143 | default: |
| 144 | inode->i_op = &xfs_inode_operations; |
| 145 | init_special_inode(inode, inode->i_mode, inode->i_rdev); |
| 146 | break; |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | STATIC_INLINE void |
| 151 | xfs_revalidate_inode( |
| 152 | xfs_mount_t *mp, |
| 153 | bhv_vnode_t *vp, |
| 154 | xfs_inode_t *ip) |
| 155 | { |
| 156 | struct inode *inode = vn_to_inode(vp); |
| 157 | |
| 158 | inode->i_mode = ip->i_d.di_mode; |
| 159 | inode->i_nlink = ip->i_d.di_nlink; |
| 160 | inode->i_uid = ip->i_d.di_uid; |
| 161 | inode->i_gid = ip->i_d.di_gid; |
| 162 | |
| 163 | switch (inode->i_mode & S_IFMT) { |
| 164 | case S_IFBLK: |
| 165 | case S_IFCHR: |
| 166 | inode->i_rdev = |
| 167 | MKDEV(sysv_major(ip->i_df.if_u2.if_rdev) & 0x1ff, |
| 168 | sysv_minor(ip->i_df.if_u2.if_rdev)); |
| 169 | break; |
| 170 | default: |
| 171 | inode->i_rdev = 0; |
| 172 | break; |
| 173 | } |
| 174 | |
| 175 | inode->i_generation = ip->i_d.di_gen; |
| 176 | i_size_write(inode, ip->i_d.di_size); |
| 177 | inode->i_blocks = |
| 178 | XFS_FSB_TO_BB(mp, ip->i_d.di_nblocks + ip->i_delayed_blks); |
| 179 | inode->i_atime.tv_sec = ip->i_d.di_atime.t_sec; |
| 180 | inode->i_atime.tv_nsec = ip->i_d.di_atime.t_nsec; |
| 181 | inode->i_mtime.tv_sec = ip->i_d.di_mtime.t_sec; |
| 182 | inode->i_mtime.tv_nsec = ip->i_d.di_mtime.t_nsec; |
| 183 | inode->i_ctime.tv_sec = ip->i_d.di_ctime.t_sec; |
| 184 | inode->i_ctime.tv_nsec = ip->i_d.di_ctime.t_nsec; |
| 185 | if (ip->i_d.di_flags & XFS_DIFLAG_IMMUTABLE) |
| 186 | inode->i_flags |= S_IMMUTABLE; |
| 187 | else |
| 188 | inode->i_flags &= ~S_IMMUTABLE; |
| 189 | if (ip->i_d.di_flags & XFS_DIFLAG_APPEND) |
| 190 | inode->i_flags |= S_APPEND; |
| 191 | else |
| 192 | inode->i_flags &= ~S_APPEND; |
| 193 | if (ip->i_d.di_flags & XFS_DIFLAG_SYNC) |
| 194 | inode->i_flags |= S_SYNC; |
| 195 | else |
| 196 | inode->i_flags &= ~S_SYNC; |
| 197 | if (ip->i_d.di_flags & XFS_DIFLAG_NOATIME) |
| 198 | inode->i_flags |= S_NOATIME; |
| 199 | else |
| 200 | inode->i_flags &= ~S_NOATIME; |
| 201 | xfs_iflags_clear(ip, XFS_IMODIFIED); |
| 202 | } |
| 203 | |
| 204 | void |
| 205 | xfs_initialize_vnode( |
| 206 | struct xfs_mount *mp, |
| 207 | bhv_vnode_t *vp, |
| 208 | struct xfs_inode *ip) |
| 209 | { |
| 210 | struct inode *inode = vn_to_inode(vp); |
| 211 | |
| 212 | if (!ip->i_vnode) { |
| 213 | ip->i_vnode = vp; |
| 214 | inode->i_private = ip; |
| 215 | } |
| 216 | |
| 217 | /* |
| 218 | * We need to set the ops vectors, and unlock the inode, but if |
| 219 | * we have been called during the new inode create process, it is |
| 220 | * too early to fill in the Linux inode. We will get called a |
| 221 | * second time once the inode is properly set up, and then we can |
| 222 | * finish our work. |
| 223 | */ |
| 224 | if (ip->i_d.di_mode != 0 && (inode->i_state & I_NEW)) { |
| 225 | xfs_revalidate_inode(mp, vp, ip); |
| 226 | xfs_set_inodeops(inode); |
| 227 | |
| 228 | xfs_iflags_clear(ip, XFS_INEW); |
| 229 | barrier(); |
| 230 | |
| 231 | unlock_new_inode(inode); |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | int |
| 236 | xfs_blkdev_get( |
| 237 | xfs_mount_t *mp, |
| 238 | const char *name, |
| 239 | struct block_device **bdevp) |
| 240 | { |
| 241 | int error = 0; |
| 242 | |
| 243 | *bdevp = open_bdev_excl(name, 0, mp); |
| 244 | if (IS_ERR(*bdevp)) { |
| 245 | error = PTR_ERR(*bdevp); |
| 246 | printk("XFS: Invalid device [%s], error=%d\n", name, error); |
| 247 | } |
| 248 | |
| 249 | return -error; |
| 250 | } |
| 251 | |
| 252 | void |
| 253 | xfs_blkdev_put( |
| 254 | struct block_device *bdev) |
| 255 | { |
| 256 | if (bdev) |
| 257 | close_bdev_excl(bdev); |
| 258 | } |
| 259 | |
| 260 | /* |
| 261 | * Try to write out the superblock using barriers. |
| 262 | */ |
| 263 | STATIC int |
| 264 | xfs_barrier_test( |
| 265 | xfs_mount_t *mp) |
| 266 | { |
| 267 | xfs_buf_t *sbp = xfs_getsb(mp, 0); |
| 268 | int error; |
| 269 | |
| 270 | XFS_BUF_UNDONE(sbp); |
| 271 | XFS_BUF_UNREAD(sbp); |
| 272 | XFS_BUF_UNDELAYWRITE(sbp); |
| 273 | XFS_BUF_WRITE(sbp); |
| 274 | XFS_BUF_UNASYNC(sbp); |
| 275 | XFS_BUF_ORDERED(sbp); |
| 276 | |
| 277 | xfsbdstrat(mp, sbp); |
| 278 | error = xfs_iowait(sbp); |
| 279 | |
| 280 | /* |
| 281 | * Clear all the flags we set and possible error state in the |
| 282 | * buffer. We only did the write to try out whether barriers |
| 283 | * worked and shouldn't leave any traces in the superblock |
| 284 | * buffer. |
| 285 | */ |
| 286 | XFS_BUF_DONE(sbp); |
| 287 | XFS_BUF_ERROR(sbp, 0); |
| 288 | XFS_BUF_UNORDERED(sbp); |
| 289 | |
| 290 | xfs_buf_relse(sbp); |
| 291 | return error; |
| 292 | } |
| 293 | |
| 294 | void |
| 295 | xfs_mountfs_check_barriers(xfs_mount_t *mp) |
| 296 | { |
| 297 | int error; |
| 298 | |
| 299 | if (mp->m_logdev_targp != mp->m_ddev_targp) { |
| 300 | xfs_fs_cmn_err(CE_NOTE, mp, |
| 301 | "Disabling barriers, not supported with external log device"); |
| 302 | mp->m_flags &= ~XFS_MOUNT_BARRIER; |
| 303 | return; |
| 304 | } |
| 305 | |
| 306 | if (mp->m_ddev_targp->bt_bdev->bd_disk->queue->ordered == |
| 307 | QUEUE_ORDERED_NONE) { |
| 308 | xfs_fs_cmn_err(CE_NOTE, mp, |
| 309 | "Disabling barriers, not supported by the underlying device"); |
| 310 | mp->m_flags &= ~XFS_MOUNT_BARRIER; |
| 311 | return; |
| 312 | } |
| 313 | |
| 314 | if (xfs_readonly_buftarg(mp->m_ddev_targp)) { |
| 315 | xfs_fs_cmn_err(CE_NOTE, mp, |
| 316 | "Disabling barriers, underlying device is readonly"); |
| 317 | mp->m_flags &= ~XFS_MOUNT_BARRIER; |
| 318 | return; |
| 319 | } |
| 320 | |
| 321 | error = xfs_barrier_test(mp); |
| 322 | if (error) { |
| 323 | xfs_fs_cmn_err(CE_NOTE, mp, |
| 324 | "Disabling barriers, trial barrier write failed"); |
| 325 | mp->m_flags &= ~XFS_MOUNT_BARRIER; |
| 326 | return; |
| 327 | } |
| 328 | } |
| 329 | |
| 330 | void |
| 331 | xfs_blkdev_issue_flush( |
| 332 | xfs_buftarg_t *buftarg) |
| 333 | { |
| 334 | blkdev_issue_flush(buftarg->bt_bdev, NULL); |
| 335 | } |
| 336 | |
| 337 | STATIC struct inode * |
| 338 | xfs_fs_alloc_inode( |
| 339 | struct super_block *sb) |
| 340 | { |
| 341 | bhv_vnode_t *vp; |
| 342 | |
| 343 | vp = kmem_zone_alloc(xfs_vnode_zone, KM_SLEEP); |
| 344 | if (unlikely(!vp)) |
| 345 | return NULL; |
| 346 | return vn_to_inode(vp); |
| 347 | } |
| 348 | |
| 349 | STATIC void |
| 350 | xfs_fs_destroy_inode( |
| 351 | struct inode *inode) |
| 352 | { |
| 353 | kmem_zone_free(xfs_vnode_zone, vn_from_inode(inode)); |
| 354 | } |
| 355 | |
| 356 | STATIC void |
| 357 | xfs_fs_inode_init_once( |
| 358 | kmem_zone_t *zonep, |
| 359 | void *vnode) |
| 360 | { |
| 361 | inode_init_once(vn_to_inode((bhv_vnode_t *)vnode)); |
| 362 | } |
| 363 | |
| 364 | STATIC int |
| 365 | xfs_init_zones(void) |
| 366 | { |
| 367 | xfs_vnode_zone = kmem_zone_init_flags(sizeof(bhv_vnode_t), "xfs_vnode", |
| 368 | KM_ZONE_HWALIGN | KM_ZONE_RECLAIM | |
| 369 | KM_ZONE_SPREAD, |
| 370 | xfs_fs_inode_init_once); |
| 371 | if (!xfs_vnode_zone) |
| 372 | goto out; |
| 373 | |
| 374 | xfs_ioend_zone = kmem_zone_init(sizeof(xfs_ioend_t), "xfs_ioend"); |
| 375 | if (!xfs_ioend_zone) |
| 376 | goto out_destroy_vnode_zone; |
| 377 | |
| 378 | xfs_ioend_pool = mempool_create_slab_pool(4 * MAX_BUF_PER_PAGE, |
| 379 | xfs_ioend_zone); |
| 380 | if (!xfs_ioend_pool) |
| 381 | goto out_free_ioend_zone; |
| 382 | return 0; |
| 383 | |
| 384 | out_free_ioend_zone: |
| 385 | kmem_zone_destroy(xfs_ioend_zone); |
| 386 | out_destroy_vnode_zone: |
| 387 | kmem_zone_destroy(xfs_vnode_zone); |
| 388 | out: |
| 389 | return -ENOMEM; |
| 390 | } |
| 391 | |
| 392 | STATIC void |
| 393 | xfs_destroy_zones(void) |
| 394 | { |
| 395 | mempool_destroy(xfs_ioend_pool); |
| 396 | kmem_zone_destroy(xfs_vnode_zone); |
| 397 | kmem_zone_destroy(xfs_ioend_zone); |
| 398 | } |
| 399 | |
| 400 | /* |
| 401 | * Attempt to flush the inode, this will actually fail |
| 402 | * if the inode is pinned, but we dirty the inode again |
| 403 | * at the point when it is unpinned after a log write, |
| 404 | * since this is when the inode itself becomes flushable. |
| 405 | */ |
| 406 | STATIC int |
| 407 | xfs_fs_write_inode( |
| 408 | struct inode *inode, |
| 409 | int sync) |
| 410 | { |
| 411 | int error = 0, flags = FLUSH_INODE; |
| 412 | |
| 413 | xfs_itrace_entry(XFS_I(inode)); |
| 414 | if (sync) { |
| 415 | filemap_fdatawait(inode->i_mapping); |
| 416 | flags |= FLUSH_SYNC; |
| 417 | } |
| 418 | error = xfs_inode_flush(XFS_I(inode), flags); |
| 419 | /* |
| 420 | * if we failed to write out the inode then mark |
| 421 | * it dirty again so we'll try again later. |
| 422 | */ |
| 423 | if (error) |
| 424 | mark_inode_dirty_sync(inode); |
| 425 | |
| 426 | return -error; |
| 427 | } |
| 428 | |
| 429 | STATIC void |
| 430 | xfs_fs_clear_inode( |
| 431 | struct inode *inode) |
| 432 | { |
| 433 | xfs_inode_t *ip = XFS_I(inode); |
| 434 | |
| 435 | /* |
| 436 | * ip can be null when xfs_iget_core calls xfs_idestroy if we |
| 437 | * find an inode with di_mode == 0 but without IGET_CREATE set. |
| 438 | */ |
| 439 | if (ip) { |
| 440 | xfs_itrace_entry(ip); |
| 441 | XFS_STATS_INC(vn_rele); |
| 442 | XFS_STATS_INC(vn_remove); |
| 443 | XFS_STATS_INC(vn_reclaim); |
| 444 | XFS_STATS_DEC(vn_active); |
| 445 | |
| 446 | xfs_inactive(ip); |
| 447 | xfs_iflags_clear(ip, XFS_IMODIFIED); |
| 448 | if (xfs_reclaim(ip)) |
| 449 | panic("%s: cannot reclaim 0x%p\n", __FUNCTION__, inode); |
| 450 | } |
| 451 | |
| 452 | ASSERT(XFS_I(inode) == NULL); |
| 453 | } |
| 454 | |
| 455 | /* |
| 456 | * Enqueue a work item to be picked up by the vfs xfssyncd thread. |
| 457 | * Doing this has two advantages: |
| 458 | * - It saves on stack space, which is tight in certain situations |
| 459 | * - It can be used (with care) as a mechanism to avoid deadlocks. |
| 460 | * Flushing while allocating in a full filesystem requires both. |
| 461 | */ |
| 462 | STATIC void |
| 463 | xfs_syncd_queue_work( |
| 464 | struct xfs_mount *mp, |
| 465 | void *data, |
| 466 | void (*syncer)(struct xfs_mount *, void *)) |
| 467 | { |
| 468 | struct bhv_vfs_sync_work *work; |
| 469 | |
| 470 | work = kmem_alloc(sizeof(struct bhv_vfs_sync_work), KM_SLEEP); |
| 471 | INIT_LIST_HEAD(&work->w_list); |
| 472 | work->w_syncer = syncer; |
| 473 | work->w_data = data; |
| 474 | work->w_mount = mp; |
| 475 | spin_lock(&mp->m_sync_lock); |
| 476 | list_add_tail(&work->w_list, &mp->m_sync_list); |
| 477 | spin_unlock(&mp->m_sync_lock); |
| 478 | wake_up_process(mp->m_sync_task); |
| 479 | } |
| 480 | |
| 481 | /* |
| 482 | * Flush delayed allocate data, attempting to free up reserved space |
| 483 | * from existing allocations. At this point a new allocation attempt |
| 484 | * has failed with ENOSPC and we are in the process of scratching our |
| 485 | * heads, looking about for more room... |
| 486 | */ |
| 487 | STATIC void |
| 488 | xfs_flush_inode_work( |
| 489 | struct xfs_mount *mp, |
| 490 | void *arg) |
| 491 | { |
| 492 | struct inode *inode = arg; |
| 493 | filemap_flush(inode->i_mapping); |
| 494 | iput(inode); |
| 495 | } |
| 496 | |
| 497 | void |
| 498 | xfs_flush_inode( |
| 499 | xfs_inode_t *ip) |
| 500 | { |
| 501 | struct inode *inode = ip->i_vnode; |
| 502 | |
| 503 | igrab(inode); |
| 504 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_inode_work); |
| 505 | delay(msecs_to_jiffies(500)); |
| 506 | } |
| 507 | |
| 508 | /* |
| 509 | * This is the "bigger hammer" version of xfs_flush_inode_work... |
| 510 | * (IOW, "If at first you don't succeed, use a Bigger Hammer"). |
| 511 | */ |
| 512 | STATIC void |
| 513 | xfs_flush_device_work( |
| 514 | struct xfs_mount *mp, |
| 515 | void *arg) |
| 516 | { |
| 517 | struct inode *inode = arg; |
| 518 | sync_blockdev(mp->m_super->s_bdev); |
| 519 | iput(inode); |
| 520 | } |
| 521 | |
| 522 | void |
| 523 | xfs_flush_device( |
| 524 | xfs_inode_t *ip) |
| 525 | { |
| 526 | struct inode *inode = vn_to_inode(XFS_ITOV(ip)); |
| 527 | |
| 528 | igrab(inode); |
| 529 | xfs_syncd_queue_work(ip->i_mount, inode, xfs_flush_device_work); |
| 530 | delay(msecs_to_jiffies(500)); |
| 531 | xfs_log_force(ip->i_mount, (xfs_lsn_t)0, XFS_LOG_FORCE|XFS_LOG_SYNC); |
| 532 | } |
| 533 | |
| 534 | STATIC void |
| 535 | xfs_sync_worker( |
| 536 | struct xfs_mount *mp, |
| 537 | void *unused) |
| 538 | { |
| 539 | int error; |
| 540 | |
| 541 | if (!(mp->m_flags & XFS_MOUNT_RDONLY)) |
| 542 | error = xfs_sync(mp, SYNC_FSDATA | SYNC_BDFLUSH | SYNC_ATTR | |
| 543 | SYNC_REFCACHE | SYNC_SUPER); |
| 544 | mp->m_sync_seq++; |
| 545 | wake_up(&mp->m_wait_single_sync_task); |
| 546 | } |
| 547 | |
| 548 | STATIC int |
| 549 | xfssyncd( |
| 550 | void *arg) |
| 551 | { |
| 552 | struct xfs_mount *mp = arg; |
| 553 | long timeleft; |
| 554 | bhv_vfs_sync_work_t *work, *n; |
| 555 | LIST_HEAD (tmp); |
| 556 | |
| 557 | set_freezable(); |
| 558 | timeleft = xfs_syncd_centisecs * msecs_to_jiffies(10); |
| 559 | for (;;) { |
| 560 | timeleft = schedule_timeout_interruptible(timeleft); |
| 561 | /* swsusp */ |
| 562 | try_to_freeze(); |
| 563 | if (kthread_should_stop() && list_empty(&mp->m_sync_list)) |
| 564 | break; |
| 565 | |
| 566 | spin_lock(&mp->m_sync_lock); |
| 567 | /* |
| 568 | * We can get woken by laptop mode, to do a sync - |
| 569 | * that's the (only!) case where the list would be |
| 570 | * empty with time remaining. |
| 571 | */ |
| 572 | if (!timeleft || list_empty(&mp->m_sync_list)) { |
| 573 | if (!timeleft) |
| 574 | timeleft = xfs_syncd_centisecs * |
| 575 | msecs_to_jiffies(10); |
| 576 | INIT_LIST_HEAD(&mp->m_sync_work.w_list); |
| 577 | list_add_tail(&mp->m_sync_work.w_list, |
| 578 | &mp->m_sync_list); |
| 579 | } |
| 580 | list_for_each_entry_safe(work, n, &mp->m_sync_list, w_list) |
| 581 | list_move(&work->w_list, &tmp); |
| 582 | spin_unlock(&mp->m_sync_lock); |
| 583 | |
| 584 | list_for_each_entry_safe(work, n, &tmp, w_list) { |
| 585 | (*work->w_syncer)(mp, work->w_data); |
| 586 | list_del(&work->w_list); |
| 587 | if (work == &mp->m_sync_work) |
| 588 | continue; |
| 589 | kmem_free(work, sizeof(struct bhv_vfs_sync_work)); |
| 590 | } |
| 591 | } |
| 592 | |
| 593 | return 0; |
| 594 | } |
| 595 | |
| 596 | STATIC void |
| 597 | xfs_fs_put_super( |
| 598 | struct super_block *sb) |
| 599 | { |
| 600 | struct xfs_mount *mp = XFS_M(sb); |
| 601 | int error; |
| 602 | |
| 603 | kthread_stop(mp->m_sync_task); |
| 604 | |
| 605 | xfs_sync(mp, SYNC_ATTR | SYNC_DELWRI); |
| 606 | error = xfs_unmount(mp, 0, NULL); |
| 607 | if (error) |
| 608 | printk("XFS: unmount got error=%d\n", error); |
| 609 | } |
| 610 | |
| 611 | STATIC void |
| 612 | xfs_fs_write_super( |
| 613 | struct super_block *sb) |
| 614 | { |
| 615 | if (!(sb->s_flags & MS_RDONLY)) |
| 616 | xfs_sync(XFS_M(sb), SYNC_FSDATA); |
| 617 | sb->s_dirt = 0; |
| 618 | } |
| 619 | |
| 620 | STATIC int |
| 621 | xfs_fs_sync_super( |
| 622 | struct super_block *sb, |
| 623 | int wait) |
| 624 | { |
| 625 | struct xfs_mount *mp = XFS_M(sb); |
| 626 | int error; |
| 627 | int flags; |
| 628 | |
| 629 | /* |
| 630 | * Treat a sync operation like a freeze. This is to work |
| 631 | * around a race in sync_inodes() which works in two phases |
| 632 | * - an asynchronous flush, which can write out an inode |
| 633 | * without waiting for file size updates to complete, and a |
| 634 | * synchronous flush, which wont do anything because the |
| 635 | * async flush removed the inode's dirty flag. Also |
| 636 | * sync_inodes() will not see any files that just have |
| 637 | * outstanding transactions to be flushed because we don't |
| 638 | * dirty the Linux inode until after the transaction I/O |
| 639 | * completes. |
| 640 | */ |
| 641 | if (wait || unlikely(sb->s_frozen == SB_FREEZE_WRITE)) { |
| 642 | /* |
| 643 | * First stage of freeze - no more writers will make progress |
| 644 | * now we are here, so we flush delwri and delalloc buffers |
| 645 | * here, then wait for all I/O to complete. Data is frozen at |
| 646 | * that point. Metadata is not frozen, transactions can still |
| 647 | * occur here so don't bother flushing the buftarg (i.e |
| 648 | * SYNC_QUIESCE) because it'll just get dirty again. |
| 649 | */ |
| 650 | flags = SYNC_DATA_QUIESCE; |
| 651 | } else |
| 652 | flags = SYNC_FSDATA; |
| 653 | |
| 654 | error = xfs_sync(mp, flags); |
| 655 | sb->s_dirt = 0; |
| 656 | |
| 657 | if (unlikely(laptop_mode)) { |
| 658 | int prev_sync_seq = mp->m_sync_seq; |
| 659 | |
| 660 | /* |
| 661 | * The disk must be active because we're syncing. |
| 662 | * We schedule xfssyncd now (now that the disk is |
| 663 | * active) instead of later (when it might not be). |
| 664 | */ |
| 665 | wake_up_process(mp->m_sync_task); |
| 666 | /* |
| 667 | * We have to wait for the sync iteration to complete. |
| 668 | * If we don't, the disk activity caused by the sync |
| 669 | * will come after the sync is completed, and that |
| 670 | * triggers another sync from laptop mode. |
| 671 | */ |
| 672 | wait_event(mp->m_wait_single_sync_task, |
| 673 | mp->m_sync_seq != prev_sync_seq); |
| 674 | } |
| 675 | |
| 676 | return -error; |
| 677 | } |
| 678 | |
| 679 | STATIC int |
| 680 | xfs_fs_statfs( |
| 681 | struct dentry *dentry, |
| 682 | struct kstatfs *statp) |
| 683 | { |
| 684 | return -xfs_statvfs(XFS_M(dentry->d_sb), statp, |
| 685 | vn_from_inode(dentry->d_inode)); |
| 686 | } |
| 687 | |
| 688 | STATIC int |
| 689 | xfs_fs_remount( |
| 690 | struct super_block *sb, |
| 691 | int *flags, |
| 692 | char *options) |
| 693 | { |
| 694 | struct xfs_mount *mp = XFS_M(sb); |
| 695 | struct xfs_mount_args *args = xfs_args_allocate(sb, 0); |
| 696 | int error; |
| 697 | |
| 698 | error = xfs_parseargs(mp, options, args, 1); |
| 699 | if (!error) |
| 700 | error = xfs_mntupdate(mp, flags, args); |
| 701 | kmem_free(args, sizeof(*args)); |
| 702 | return -error; |
| 703 | } |
| 704 | |
| 705 | STATIC void |
| 706 | xfs_fs_lockfs( |
| 707 | struct super_block *sb) |
| 708 | { |
| 709 | xfs_freeze(XFS_M(sb)); |
| 710 | } |
| 711 | |
| 712 | STATIC int |
| 713 | xfs_fs_show_options( |
| 714 | struct seq_file *m, |
| 715 | struct vfsmount *mnt) |
| 716 | { |
| 717 | return -xfs_showargs(XFS_M(mnt->mnt_sb), m); |
| 718 | } |
| 719 | |
| 720 | STATIC int |
| 721 | xfs_fs_quotasync( |
| 722 | struct super_block *sb, |
| 723 | int type) |
| 724 | { |
| 725 | return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XQUOTASYNC, 0, NULL); |
| 726 | } |
| 727 | |
| 728 | STATIC int |
| 729 | xfs_fs_getxstate( |
| 730 | struct super_block *sb, |
| 731 | struct fs_quota_stat *fqs) |
| 732 | { |
| 733 | return -XFS_QM_QUOTACTL(XFS_M(sb), Q_XGETQSTAT, 0, (caddr_t)fqs); |
| 734 | } |
| 735 | |
| 736 | STATIC int |
| 737 | xfs_fs_setxstate( |
| 738 | struct super_block *sb, |
| 739 | unsigned int flags, |
| 740 | int op) |
| 741 | { |
| 742 | return -XFS_QM_QUOTACTL(XFS_M(sb), op, 0, (caddr_t)&flags); |
| 743 | } |
| 744 | |
| 745 | STATIC int |
| 746 | xfs_fs_getxquota( |
| 747 | struct super_block *sb, |
| 748 | int type, |
| 749 | qid_t id, |
| 750 | struct fs_disk_quota *fdq) |
| 751 | { |
| 752 | return -XFS_QM_QUOTACTL(XFS_M(sb), |
| 753 | (type == USRQUOTA) ? Q_XGETQUOTA : |
| 754 | ((type == GRPQUOTA) ? Q_XGETGQUOTA : |
| 755 | Q_XGETPQUOTA), id, (caddr_t)fdq); |
| 756 | } |
| 757 | |
| 758 | STATIC int |
| 759 | xfs_fs_setxquota( |
| 760 | struct super_block *sb, |
| 761 | int type, |
| 762 | qid_t id, |
| 763 | struct fs_disk_quota *fdq) |
| 764 | { |
| 765 | return -XFS_QM_QUOTACTL(XFS_M(sb), |
| 766 | (type == USRQUOTA) ? Q_XSETQLIM : |
| 767 | ((type == GRPQUOTA) ? Q_XSETGQLIM : |
| 768 | Q_XSETPQLIM), id, (caddr_t)fdq); |
| 769 | } |
| 770 | |
| 771 | STATIC int |
| 772 | xfs_fs_fill_super( |
| 773 | struct super_block *sb, |
| 774 | void *data, |
| 775 | int silent) |
| 776 | { |
| 777 | struct inode *rootvp; |
| 778 | struct xfs_mount *mp = NULL; |
| 779 | struct xfs_mount_args *args = xfs_args_allocate(sb, silent); |
| 780 | struct kstatfs statvfs; |
| 781 | int error; |
| 782 | |
| 783 | mp = xfs_mount_init(); |
| 784 | |
| 785 | INIT_LIST_HEAD(&mp->m_sync_list); |
| 786 | spin_lock_init(&mp->m_sync_lock); |
| 787 | init_waitqueue_head(&mp->m_wait_single_sync_task); |
| 788 | |
| 789 | mp->m_super = sb; |
| 790 | sb->s_fs_info = mp; |
| 791 | |
| 792 | if (sb->s_flags & MS_RDONLY) |
| 793 | mp->m_flags |= XFS_MOUNT_RDONLY; |
| 794 | |
| 795 | error = xfs_parseargs(mp, (char *)data, args, 0); |
| 796 | if (error) |
| 797 | goto fail_vfsop; |
| 798 | |
| 799 | sb_min_blocksize(sb, BBSIZE); |
| 800 | sb->s_export_op = &xfs_export_operations; |
| 801 | sb->s_qcop = &xfs_quotactl_operations; |
| 802 | sb->s_op = &xfs_super_operations; |
| 803 | |
| 804 | error = xfs_mount(mp, args, NULL); |
| 805 | if (error) |
| 806 | goto fail_vfsop; |
| 807 | |
| 808 | error = xfs_statvfs(mp, &statvfs, NULL); |
| 809 | if (error) |
| 810 | goto fail_unmount; |
| 811 | |
| 812 | sb->s_dirt = 1; |
| 813 | sb->s_magic = statvfs.f_type; |
| 814 | sb->s_blocksize = statvfs.f_bsize; |
| 815 | sb->s_blocksize_bits = ffs(statvfs.f_bsize) - 1; |
| 816 | sb->s_maxbytes = xfs_max_file_offset(sb->s_blocksize_bits); |
| 817 | sb->s_time_gran = 1; |
| 818 | set_posix_acl_flag(sb); |
| 819 | |
| 820 | error = xfs_root(mp, &rootvp); |
| 821 | if (error) |
| 822 | goto fail_unmount; |
| 823 | |
| 824 | sb->s_root = d_alloc_root(vn_to_inode(rootvp)); |
| 825 | if (!sb->s_root) { |
| 826 | error = ENOMEM; |
| 827 | goto fail_vnrele; |
| 828 | } |
| 829 | if (is_bad_inode(sb->s_root->d_inode)) { |
| 830 | error = EINVAL; |
| 831 | goto fail_vnrele; |
| 832 | } |
| 833 | |
| 834 | mp->m_sync_work.w_syncer = xfs_sync_worker; |
| 835 | mp->m_sync_work.w_mount = mp; |
| 836 | mp->m_sync_task = kthread_run(xfssyncd, mp, "xfssyncd"); |
| 837 | if (IS_ERR(mp->m_sync_task)) { |
| 838 | error = -PTR_ERR(mp->m_sync_task); |
| 839 | goto fail_vnrele; |
| 840 | } |
| 841 | |
| 842 | xfs_itrace_exit(XFS_I(sb->s_root->d_inode)); |
| 843 | |
| 844 | kmem_free(args, sizeof(*args)); |
| 845 | return 0; |
| 846 | |
| 847 | fail_vnrele: |
| 848 | if (sb->s_root) { |
| 849 | dput(sb->s_root); |
| 850 | sb->s_root = NULL; |
| 851 | } else { |
| 852 | VN_RELE(rootvp); |
| 853 | } |
| 854 | |
| 855 | fail_unmount: |
| 856 | xfs_unmount(mp, 0, NULL); |
| 857 | |
| 858 | fail_vfsop: |
| 859 | kmem_free(args, sizeof(*args)); |
| 860 | return -error; |
| 861 | } |
| 862 | |
| 863 | STATIC int |
| 864 | xfs_fs_get_sb( |
| 865 | struct file_system_type *fs_type, |
| 866 | int flags, |
| 867 | const char *dev_name, |
| 868 | void *data, |
| 869 | struct vfsmount *mnt) |
| 870 | { |
| 871 | return get_sb_bdev(fs_type, flags, dev_name, data, xfs_fs_fill_super, |
| 872 | mnt); |
| 873 | } |
| 874 | |
| 875 | static struct super_operations xfs_super_operations = { |
| 876 | .alloc_inode = xfs_fs_alloc_inode, |
| 877 | .destroy_inode = xfs_fs_destroy_inode, |
| 878 | .write_inode = xfs_fs_write_inode, |
| 879 | .clear_inode = xfs_fs_clear_inode, |
| 880 | .put_super = xfs_fs_put_super, |
| 881 | .write_super = xfs_fs_write_super, |
| 882 | .sync_fs = xfs_fs_sync_super, |
| 883 | .write_super_lockfs = xfs_fs_lockfs, |
| 884 | .statfs = xfs_fs_statfs, |
| 885 | .remount_fs = xfs_fs_remount, |
| 886 | .show_options = xfs_fs_show_options, |
| 887 | }; |
| 888 | |
| 889 | static struct quotactl_ops xfs_quotactl_operations = { |
| 890 | .quota_sync = xfs_fs_quotasync, |
| 891 | .get_xstate = xfs_fs_getxstate, |
| 892 | .set_xstate = xfs_fs_setxstate, |
| 893 | .get_xquota = xfs_fs_getxquota, |
| 894 | .set_xquota = xfs_fs_setxquota, |
| 895 | }; |
| 896 | |
| 897 | static struct file_system_type xfs_fs_type = { |
| 898 | .owner = THIS_MODULE, |
| 899 | .name = "xfs", |
| 900 | .get_sb = xfs_fs_get_sb, |
| 901 | .kill_sb = kill_block_super, |
| 902 | .fs_flags = FS_REQUIRES_DEV, |
| 903 | }; |
| 904 | |
| 905 | |
| 906 | STATIC int __init |
| 907 | init_xfs_fs( void ) |
| 908 | { |
| 909 | int error; |
| 910 | static char message[] __initdata = KERN_INFO \ |
| 911 | XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled\n"; |
| 912 | |
| 913 | printk(message); |
| 914 | |
| 915 | ktrace_init(64); |
| 916 | |
| 917 | error = xfs_init_zones(); |
| 918 | if (error < 0) |
| 919 | goto undo_zones; |
| 920 | |
| 921 | error = xfs_buf_init(); |
| 922 | if (error < 0) |
| 923 | goto undo_buffers; |
| 924 | |
| 925 | vn_init(); |
| 926 | xfs_init(); |
| 927 | uuid_init(); |
| 928 | vfs_initquota(); |
| 929 | |
| 930 | error = register_filesystem(&xfs_fs_type); |
| 931 | if (error) |
| 932 | goto undo_register; |
| 933 | return 0; |
| 934 | |
| 935 | undo_register: |
| 936 | xfs_buf_terminate(); |
| 937 | |
| 938 | undo_buffers: |
| 939 | xfs_destroy_zones(); |
| 940 | |
| 941 | undo_zones: |
| 942 | return error; |
| 943 | } |
| 944 | |
| 945 | STATIC void __exit |
| 946 | exit_xfs_fs( void ) |
| 947 | { |
| 948 | vfs_exitquota(); |
| 949 | unregister_filesystem(&xfs_fs_type); |
| 950 | xfs_cleanup(); |
| 951 | xfs_buf_terminate(); |
| 952 | xfs_destroy_zones(); |
| 953 | ktrace_uninit(); |
| 954 | } |
| 955 | |
| 956 | module_init(init_xfs_fs); |
| 957 | module_exit(exit_xfs_fs); |
| 958 | |
| 959 | MODULE_AUTHOR("Silicon Graphics, Inc."); |
| 960 | MODULE_DESCRIPTION(XFS_VERSION_STRING " with " XFS_BUILD_OPTIONS " enabled"); |
| 961 | MODULE_LICENSE("GPL"); |