1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc. 4 * All Rights Reserved. 5 */ 6 #include "xfs.h" 7 #include "xfs_fs.h" 8 #include "xfs_shared.h" 9 #include "xfs_format.h" 10 #include "xfs_log_format.h" 11 #include "xfs_trans_resv.h" 12 #include "xfs_mount.h" 13 #include "xfs_inode.h" 14 #include "xfs_btree.h" 15 #include "xfs_ialloc.h" 16 #include "xfs_ialloc_btree.h" 17 #include "xfs_iwalk.h" 18 #include "xfs_itable.h" 19 #include "xfs_error.h" 20 #include "xfs_icache.h" 21 #include "xfs_health.h" 22 #include "xfs_trans.h" 23 24 /* 25 * Bulk Stat 26 * ========= 27 * 28 * Use the inode walking functions to fill out struct xfs_bulkstat for every 29 * allocated inode, then pass the stat information to some externally provided 30 * iteration function. 31 */ 32 33 struct xfs_bstat_chunk { 34 bulkstat_one_fmt_pf formatter; 35 struct xfs_ibulk *breq; 36 struct xfs_bulkstat *buf; 37 }; 38 39 /* 40 * Fill out the bulkstat info for a single inode and report it somewhere. 41 * 42 * bc->breq->lastino is effectively the inode cursor as we walk through the 43 * filesystem. Therefore, we update it any time we need to move the cursor 44 * forward, regardless of whether or not we're sending any bstat information 45 * back to userspace. If the inode is internal metadata or, has been freed 46 * out from under us, we just simply keep going. 47 * 48 * However, if any other type of error happens we want to stop right where we 49 * are so that userspace will call back with exact number of the bad inode and 50 * we can send back an error code. 51 * 52 * Note that if the formatter tells us there's no space left in the buffer we 53 * move the cursor forward and abort the walk. 54 */ 55 STATIC int 56 xfs_bulkstat_one_int( 57 struct xfs_mount *mp, 58 struct user_namespace *mnt_userns, 59 struct xfs_trans *tp, 60 xfs_ino_t ino, 61 struct xfs_bstat_chunk *bc) 62 { 63 struct user_namespace *sb_userns = mp->m_super->s_user_ns; 64 struct xfs_inode *ip; /* incore inode pointer */ 65 struct inode *inode; 66 struct xfs_bulkstat *buf = bc->buf; 67 xfs_extnum_t nextents; 68 int error = -EINVAL; 69 70 if (xfs_internal_inum(mp, ino)) 71 goto out_advance; 72 73 error = xfs_iget(mp, tp, ino, 74 (XFS_IGET_DONTCACHE | XFS_IGET_UNTRUSTED), 75 XFS_ILOCK_SHARED, &ip); 76 if (error == -ENOENT || error == -EINVAL) 77 goto out_advance; 78 if (error) 79 goto out; 80 81 ASSERT(ip != NULL); 82 ASSERT(ip->i_imap.im_blkno != 0); 83 inode = VFS_I(ip); 84 85 /* xfs_iget returns the following without needing 86 * further change. 87 */ 88 buf->bs_projectid = ip->i_projid; 89 buf->bs_ino = ino; 90 buf->bs_uid = from_kuid(sb_userns, i_uid_into_mnt(mnt_userns, inode)); 91 buf->bs_gid = from_kgid(sb_userns, i_gid_into_mnt(mnt_userns, inode)); 92 buf->bs_size = ip->i_disk_size; 93 94 buf->bs_nlink = inode->i_nlink; 95 buf->bs_atime = inode->i_atime.tv_sec; 96 buf->bs_atime_nsec = inode->i_atime.tv_nsec; 97 buf->bs_mtime = inode->i_mtime.tv_sec; 98 buf->bs_mtime_nsec = inode->i_mtime.tv_nsec; 99 buf->bs_ctime = inode->i_ctime.tv_sec; 100 buf->bs_ctime_nsec = inode->i_ctime.tv_nsec; 101 buf->bs_gen = inode->i_generation; 102 buf->bs_mode = inode->i_mode; 103 104 buf->bs_xflags = xfs_ip2xflags(ip); 105 buf->bs_extsize_blks = ip->i_extsize; 106 107 nextents = xfs_ifork_nextents(&ip->i_df); 108 if (!(bc->breq->flags & XFS_IBULK_NREXT64)) 109 buf->bs_extents = min(nextents, XFS_MAX_EXTCNT_DATA_FORK_SMALL); 110 else 111 buf->bs_extents64 = nextents; 112 113 xfs_bulkstat_health(ip, buf); 114 buf->bs_aextents = xfs_ifork_nextents(ip->i_afp); 115 buf->bs_forkoff = XFS_IFORK_BOFF(ip); 116 buf->bs_version = XFS_BULKSTAT_VERSION_V5; 117 118 if (xfs_has_v3inodes(mp)) { 119 buf->bs_btime = ip->i_crtime.tv_sec; 120 buf->bs_btime_nsec = ip->i_crtime.tv_nsec; 121 if (ip->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) 122 buf->bs_cowextsize_blks = ip->i_cowextsize; 123 } 124 125 switch (ip->i_df.if_format) { 126 case XFS_DINODE_FMT_DEV: 127 buf->bs_rdev = sysv_encode_dev(inode->i_rdev); 128 buf->bs_blksize = BLKDEV_IOSIZE; 129 buf->bs_blocks = 0; 130 break; 131 case XFS_DINODE_FMT_LOCAL: 132 buf->bs_rdev = 0; 133 buf->bs_blksize = mp->m_sb.sb_blocksize; 134 buf->bs_blocks = 0; 135 break; 136 case XFS_DINODE_FMT_EXTENTS: 137 case XFS_DINODE_FMT_BTREE: 138 buf->bs_rdev = 0; 139 buf->bs_blksize = mp->m_sb.sb_blocksize; 140 buf->bs_blocks = ip->i_nblocks + ip->i_delayed_blks; 141 break; 142 } 143 xfs_iunlock(ip, XFS_ILOCK_SHARED); 144 xfs_irele(ip); 145 146 error = bc->formatter(bc->breq, buf); 147 if (error == -ECANCELED) 148 goto out_advance; 149 if (error) 150 goto out; 151 152 out_advance: 153 /* 154 * Advance the cursor to the inode that comes after the one we just 155 * looked at. We want the caller to move along if the bulkstat 156 * information was copied successfully; if we tried to grab the inode 157 * but it's no longer allocated; or if it's internal metadata. 158 */ 159 bc->breq->startino = ino + 1; 160 out: 161 return error; 162 } 163 164 /* Bulkstat a single inode. */ 165 int 166 xfs_bulkstat_one( 167 struct xfs_ibulk *breq, 168 bulkstat_one_fmt_pf formatter) 169 { 170 struct xfs_bstat_chunk bc = { 171 .formatter = formatter, 172 .breq = breq, 173 }; 174 struct xfs_trans *tp; 175 int error; 176 177 if (breq->mnt_userns != &init_user_ns) { 178 xfs_warn_ratelimited(breq->mp, 179 "bulkstat not supported inside of idmapped mounts."); 180 return -EINVAL; 181 } 182 183 ASSERT(breq->icount == 1); 184 185 bc.buf = kmem_zalloc(sizeof(struct xfs_bulkstat), 186 KM_MAYFAIL); 187 if (!bc.buf) 188 return -ENOMEM; 189 190 /* 191 * Grab an empty transaction so that we can use its recursive buffer 192 * locking abilities to detect cycles in the inobt without deadlocking. 193 */ 194 error = xfs_trans_alloc_empty(breq->mp, &tp); 195 if (error) 196 goto out; 197 198 error = xfs_bulkstat_one_int(breq->mp, breq->mnt_userns, tp, 199 breq->startino, &bc); 200 xfs_trans_cancel(tp); 201 out: 202 kmem_free(bc.buf); 203 204 /* 205 * If we reported one inode to userspace then we abort because we hit 206 * the end of the buffer. Don't leak that back to userspace. 207 */ 208 if (error == -ECANCELED) 209 error = 0; 210 211 return error; 212 } 213 214 static int 215 xfs_bulkstat_iwalk( 216 struct xfs_mount *mp, 217 struct xfs_trans *tp, 218 xfs_ino_t ino, 219 void *data) 220 { 221 struct xfs_bstat_chunk *bc = data; 222 int error; 223 224 error = xfs_bulkstat_one_int(mp, bc->breq->mnt_userns, tp, ino, data); 225 /* bulkstat just skips over missing inodes */ 226 if (error == -ENOENT || error == -EINVAL) 227 return 0; 228 return error; 229 } 230 231 /* 232 * Check the incoming lastino parameter. 233 * 234 * We allow any inode value that could map to physical space inside the 235 * filesystem because if there are no inodes there, bulkstat moves on to the 236 * next chunk. In other words, the magic agino value of zero takes us to the 237 * first chunk in the AG, and an agino value past the end of the AG takes us to 238 * the first chunk in the next AG. 239 * 240 * Therefore we can end early if the requested inode is beyond the end of the 241 * filesystem or doesn't map properly. 242 */ 243 static inline bool 244 xfs_bulkstat_already_done( 245 struct xfs_mount *mp, 246 xfs_ino_t startino) 247 { 248 xfs_agnumber_t agno = XFS_INO_TO_AGNO(mp, startino); 249 xfs_agino_t agino = XFS_INO_TO_AGINO(mp, startino); 250 251 return agno >= mp->m_sb.sb_agcount || 252 startino != XFS_AGINO_TO_INO(mp, agno, agino); 253 } 254 255 /* Return stat information in bulk (by-inode) for the filesystem. */ 256 int 257 xfs_bulkstat( 258 struct xfs_ibulk *breq, 259 bulkstat_one_fmt_pf formatter) 260 { 261 struct xfs_bstat_chunk bc = { 262 .formatter = formatter, 263 .breq = breq, 264 }; 265 struct xfs_trans *tp; 266 unsigned int iwalk_flags = 0; 267 int error; 268 269 if (breq->mnt_userns != &init_user_ns) { 270 xfs_warn_ratelimited(breq->mp, 271 "bulkstat not supported inside of idmapped mounts."); 272 return -EINVAL; 273 } 274 if (xfs_bulkstat_already_done(breq->mp, breq->startino)) 275 return 0; 276 277 bc.buf = kmem_zalloc(sizeof(struct xfs_bulkstat), 278 KM_MAYFAIL); 279 if (!bc.buf) 280 return -ENOMEM; 281 282 /* 283 * Grab an empty transaction so that we can use its recursive buffer 284 * locking abilities to detect cycles in the inobt without deadlocking. 285 */ 286 error = xfs_trans_alloc_empty(breq->mp, &tp); 287 if (error) 288 goto out; 289 290 if (breq->flags & XFS_IBULK_SAME_AG) 291 iwalk_flags |= XFS_IWALK_SAME_AG; 292 293 error = xfs_iwalk(breq->mp, tp, breq->startino, iwalk_flags, 294 xfs_bulkstat_iwalk, breq->icount, &bc); 295 xfs_trans_cancel(tp); 296 out: 297 kmem_free(bc.buf); 298 299 /* 300 * We found some inodes, so clear the error status and return them. 301 * The lastino pointer will point directly at the inode that triggered 302 * any error that occurred, so on the next call the error will be 303 * triggered again and propagated to userspace as there will be no 304 * formatted inodes in the buffer. 305 */ 306 if (breq->ocount > 0) 307 error = 0; 308 309 return error; 310 } 311 312 /* Convert bulkstat (v5) to bstat (v1). */ 313 void 314 xfs_bulkstat_to_bstat( 315 struct xfs_mount *mp, 316 struct xfs_bstat *bs1, 317 const struct xfs_bulkstat *bstat) 318 { 319 /* memset is needed here because of padding holes in the structure. */ 320 memset(bs1, 0, sizeof(struct xfs_bstat)); 321 bs1->bs_ino = bstat->bs_ino; 322 bs1->bs_mode = bstat->bs_mode; 323 bs1->bs_nlink = bstat->bs_nlink; 324 bs1->bs_uid = bstat->bs_uid; 325 bs1->bs_gid = bstat->bs_gid; 326 bs1->bs_rdev = bstat->bs_rdev; 327 bs1->bs_blksize = bstat->bs_blksize; 328 bs1->bs_size = bstat->bs_size; 329 bs1->bs_atime.tv_sec = bstat->bs_atime; 330 bs1->bs_mtime.tv_sec = bstat->bs_mtime; 331 bs1->bs_ctime.tv_sec = bstat->bs_ctime; 332 bs1->bs_atime.tv_nsec = bstat->bs_atime_nsec; 333 bs1->bs_mtime.tv_nsec = bstat->bs_mtime_nsec; 334 bs1->bs_ctime.tv_nsec = bstat->bs_ctime_nsec; 335 bs1->bs_blocks = bstat->bs_blocks; 336 bs1->bs_xflags = bstat->bs_xflags; 337 bs1->bs_extsize = XFS_FSB_TO_B(mp, bstat->bs_extsize_blks); 338 bs1->bs_extents = bstat->bs_extents; 339 bs1->bs_gen = bstat->bs_gen; 340 bs1->bs_projid_lo = bstat->bs_projectid & 0xFFFF; 341 bs1->bs_forkoff = bstat->bs_forkoff; 342 bs1->bs_projid_hi = bstat->bs_projectid >> 16; 343 bs1->bs_sick = bstat->bs_sick; 344 bs1->bs_checked = bstat->bs_checked; 345 bs1->bs_cowextsize = XFS_FSB_TO_B(mp, bstat->bs_cowextsize_blks); 346 bs1->bs_dmevmask = 0; 347 bs1->bs_dmstate = 0; 348 bs1->bs_aextents = bstat->bs_aextents; 349 } 350 351 struct xfs_inumbers_chunk { 352 inumbers_fmt_pf formatter; 353 struct xfs_ibulk *breq; 354 }; 355 356 /* 357 * INUMBERS 358 * ======== 359 * This is how we export inode btree records to userspace, so that XFS tools 360 * can figure out where inodes are allocated. 361 */ 362 363 /* 364 * Format the inode group structure and report it somewhere. 365 * 366 * Similar to xfs_bulkstat_one_int, lastino is the inode cursor as we walk 367 * through the filesystem so we move it forward unless there was a runtime 368 * error. If the formatter tells us the buffer is now full we also move the 369 * cursor forward and abort the walk. 370 */ 371 STATIC int 372 xfs_inumbers_walk( 373 struct xfs_mount *mp, 374 struct xfs_trans *tp, 375 xfs_agnumber_t agno, 376 const struct xfs_inobt_rec_incore *irec, 377 void *data) 378 { 379 struct xfs_inumbers inogrp = { 380 .xi_startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino), 381 .xi_alloccount = irec->ir_count - irec->ir_freecount, 382 .xi_allocmask = ~irec->ir_free, 383 .xi_version = XFS_INUMBERS_VERSION_V5, 384 }; 385 struct xfs_inumbers_chunk *ic = data; 386 int error; 387 388 error = ic->formatter(ic->breq, &inogrp); 389 if (error && error != -ECANCELED) 390 return error; 391 392 ic->breq->startino = XFS_AGINO_TO_INO(mp, agno, irec->ir_startino) + 393 XFS_INODES_PER_CHUNK; 394 return error; 395 } 396 397 /* 398 * Return inode number table for the filesystem. 399 */ 400 int 401 xfs_inumbers( 402 struct xfs_ibulk *breq, 403 inumbers_fmt_pf formatter) 404 { 405 struct xfs_inumbers_chunk ic = { 406 .formatter = formatter, 407 .breq = breq, 408 }; 409 struct xfs_trans *tp; 410 int error = 0; 411 412 if (xfs_bulkstat_already_done(breq->mp, breq->startino)) 413 return 0; 414 415 /* 416 * Grab an empty transaction so that we can use its recursive buffer 417 * locking abilities to detect cycles in the inobt without deadlocking. 418 */ 419 error = xfs_trans_alloc_empty(breq->mp, &tp); 420 if (error) 421 goto out; 422 423 error = xfs_inobt_walk(breq->mp, tp, breq->startino, breq->flags, 424 xfs_inumbers_walk, breq->icount, &ic); 425 xfs_trans_cancel(tp); 426 out: 427 428 /* 429 * We found some inode groups, so clear the error status and return 430 * them. The lastino pointer will point directly at the inode that 431 * triggered any error that occurred, so on the next call the error 432 * will be triggered again and propagated to userspace as there will be 433 * no formatted inode groups in the buffer. 434 */ 435 if (breq->ocount > 0) 436 error = 0; 437 438 return error; 439 } 440 441 /* Convert an inumbers (v5) struct to a inogrp (v1) struct. */ 442 void 443 xfs_inumbers_to_inogrp( 444 struct xfs_inogrp *ig1, 445 const struct xfs_inumbers *ig) 446 { 447 /* memset is needed here because of padding holes in the structure. */ 448 memset(ig1, 0, sizeof(struct xfs_inogrp)); 449 ig1->xi_startino = ig->xi_startino; 450 ig1->xi_alloccount = ig->xi_alloccount; 451 ig1->xi_allocmask = ig->xi_allocmask; 452 } 453