1200 lines · c
1// SPDX-License-Identifier: GPL-2.02/*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_trans.h"15#include "xfs_inode_item.h"16#include "xfs_trace.h"17#include "xfs_trans_priv.h"18#include "xfs_buf_item.h"19#include "xfs_log.h"20#include "xfs_log_priv.h"21#include "xfs_error.h"22#include "xfs_rtbitmap.h"23 24#include <linux/iversion.h>25 26struct kmem_cache *xfs_ili_cache; /* inode log item */27 28static inline struct xfs_inode_log_item *INODE_ITEM(struct xfs_log_item *lip)29{30 return container_of(lip, struct xfs_inode_log_item, ili_item);31}32 33static uint64_t34xfs_inode_item_sort(35 struct xfs_log_item *lip)36{37 return INODE_ITEM(lip)->ili_inode->i_ino;38}39 40#ifdef DEBUG_EXPENSIVE41static void42xfs_inode_item_precommit_check(43 struct xfs_inode *ip)44{45 struct xfs_mount *mp = ip->i_mount;46 struct xfs_dinode *dip;47 xfs_failaddr_t fa;48 49 dip = kzalloc(mp->m_sb.sb_inodesize, GFP_KERNEL | GFP_NOFS);50 if (!dip) {51 ASSERT(dip != NULL);52 return;53 }54 55 xfs_inode_to_disk(ip, dip, 0);56 xfs_dinode_calc_crc(mp, dip);57 fa = xfs_dinode_verify(mp, ip->i_ino, dip);58 if (fa) {59 xfs_inode_verifier_error(ip, -EFSCORRUPTED, __func__, dip,60 sizeof(*dip), fa);61 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);62 ASSERT(fa == NULL);63 }64 kfree(dip);65}66#else67# define xfs_inode_item_precommit_check(ip) ((void)0)68#endif69 70/*71 * Prior to finally logging the inode, we have to ensure that all the72 * per-modification inode state changes are applied. This includes VFS inode73 * state updates, format conversions, verifier state synchronisation and74 * ensuring the inode buffer remains in memory whilst the inode is dirty.75 *76 * We have to be careful when we grab the inode cluster buffer due to lock77 * ordering constraints. The unlinked inode modifications (xfs_iunlink_item)78 * require AGI -> inode cluster buffer lock order. The inode cluster buffer is79 * not locked until ->precommit, so it happens after everything else has been80 * modified.81 *82 * Further, we have AGI -> AGF lock ordering, and with O_TMPFILE handling we83 * have AGI -> AGF -> iunlink item -> inode cluster buffer lock order. Hence we84 * cannot safely lock the inode cluster buffer in xfs_trans_log_inode() because85 * it can be called on a inode (e.g. via bumplink/droplink) before we take the86 * AGF lock modifying directory blocks.87 *88 * Rather than force a complete rework of all the transactions to call89 * xfs_trans_log_inode() once and once only at the end of every transaction, we90 * move the pinning of the inode cluster buffer to a ->precommit operation. This91 * matches how the xfs_iunlink_item locks the inode cluster buffer, and it92 * ensures that the inode cluster buffer locking is always done last in a93 * transaction. i.e. we ensure the lock order is always AGI -> AGF -> inode94 * cluster buffer.95 *96 * If we return the inode number as the precommit sort key then we'll also97 * guarantee that the order all inode cluster buffer locking is the same all the98 * inodes and unlink items in the transaction.99 */100static int101xfs_inode_item_precommit(102 struct xfs_trans *tp,103 struct xfs_log_item *lip)104{105 struct xfs_inode_log_item *iip = INODE_ITEM(lip);106 struct xfs_inode *ip = iip->ili_inode;107 struct inode *inode = VFS_I(ip);108 unsigned int flags = iip->ili_dirty_flags;109 110 /*111 * Don't bother with i_lock for the I_DIRTY_TIME check here, as races112 * don't matter - we either will need an extra transaction in 24 hours113 * to log the timestamps, or will clear already cleared fields in the114 * worst case.115 */116 if (inode->i_state & I_DIRTY_TIME) {117 spin_lock(&inode->i_lock);118 inode->i_state &= ~I_DIRTY_TIME;119 spin_unlock(&inode->i_lock);120 }121 122 /*123 * If we're updating the inode core or the timestamps and it's possible124 * to upgrade this inode to bigtime format, do so now.125 */126 if ((flags & (XFS_ILOG_CORE | XFS_ILOG_TIMESTAMP)) &&127 xfs_has_bigtime(ip->i_mount) &&128 !xfs_inode_has_bigtime(ip)) {129 ip->i_diflags2 |= XFS_DIFLAG2_BIGTIME;130 flags |= XFS_ILOG_CORE;131 }132 133 /*134 * Inode verifiers do not check that the extent size hint is an integer135 * multiple of the rt extent size on a directory with both rtinherit136 * and extszinherit flags set. If we're logging a directory that is137 * misconfigured in this way, clear the hint.138 */139 if ((ip->i_diflags & XFS_DIFLAG_RTINHERIT) &&140 (ip->i_diflags & XFS_DIFLAG_EXTSZINHERIT) &&141 xfs_extlen_to_rtxmod(ip->i_mount, ip->i_extsize) > 0) {142 ip->i_diflags &= ~(XFS_DIFLAG_EXTSIZE |143 XFS_DIFLAG_EXTSZINHERIT);144 ip->i_extsize = 0;145 flags |= XFS_ILOG_CORE;146 }147 148 /*149 * Record the specific change for fdatasync optimisation. This allows150 * fdatasync to skip log forces for inodes that are only timestamp151 * dirty. Once we've processed the XFS_ILOG_IVERSION flag, convert it152 * to XFS_ILOG_CORE so that the actual on-disk dirty tracking153 * (ili_fields) correctly tracks that the version has changed.154 */155 spin_lock(&iip->ili_lock);156 iip->ili_fsync_fields |= (flags & ~XFS_ILOG_IVERSION);157 if (flags & XFS_ILOG_IVERSION)158 flags = ((flags & ~XFS_ILOG_IVERSION) | XFS_ILOG_CORE);159 160 if (!iip->ili_item.li_buf) {161 struct xfs_buf *bp;162 int error;163 164 /*165 * We hold the ILOCK here, so this inode is not going to be166 * flushed while we are here. Further, because there is no167 * buffer attached to the item, we know that there is no IO in168 * progress, so nothing will clear the ili_fields while we read169 * in the buffer. Hence we can safely drop the spin lock and170 * read the buffer knowing that the state will not change from171 * here.172 */173 spin_unlock(&iip->ili_lock);174 error = xfs_imap_to_bp(ip->i_mount, tp, &ip->i_imap, &bp);175 if (error)176 return error;177 178 /*179 * We need an explicit buffer reference for the log item but180 * don't want the buffer to remain attached to the transaction.181 * Hold the buffer but release the transaction reference once182 * we've attached the inode log item to the buffer log item183 * list.184 */185 xfs_buf_hold(bp);186 spin_lock(&iip->ili_lock);187 iip->ili_item.li_buf = bp;188 bp->b_flags |= _XBF_INODES;189 list_add_tail(&iip->ili_item.li_bio_list, &bp->b_li_list);190 xfs_trans_brelse(tp, bp);191 }192 193 /*194 * Always OR in the bits from the ili_last_fields field. This is to195 * coordinate with the xfs_iflush() and xfs_buf_inode_iodone() routines196 * in the eventual clearing of the ili_fields bits. See the big comment197 * in xfs_iflush() for an explanation of this coordination mechanism.198 */199 iip->ili_fields |= (flags | iip->ili_last_fields);200 spin_unlock(&iip->ili_lock);201 202 xfs_inode_item_precommit_check(ip);203 204 /*205 * We are done with the log item transaction dirty state, so clear it so206 * that it doesn't pollute future transactions.207 */208 iip->ili_dirty_flags = 0;209 return 0;210}211 212/*213 * The logged size of an inode fork is always the current size of the inode214 * fork. This means that when an inode fork is relogged, the size of the logged215 * region is determined by the current state, not the combination of the216 * previously logged state + the current state. This is different relogging217 * behaviour to most other log items which will retain the size of the218 * previously logged changes when smaller regions are relogged.219 *220 * Hence operations that remove data from the inode fork (e.g. shortform221 * dir/attr remove, extent form extent removal, etc), the size of the relogged222 * inode gets -smaller- rather than stays the same size as the previously logged223 * size and this can result in the committing transaction reducing the amount of224 * space being consumed by the CIL.225 */226STATIC void227xfs_inode_item_data_fork_size(228 struct xfs_inode_log_item *iip,229 int *nvecs,230 int *nbytes)231{232 struct xfs_inode *ip = iip->ili_inode;233 234 switch (ip->i_df.if_format) {235 case XFS_DINODE_FMT_EXTENTS:236 if ((iip->ili_fields & XFS_ILOG_DEXT) &&237 ip->i_df.if_nextents > 0 &&238 ip->i_df.if_bytes > 0) {239 /* worst case, doesn't subtract delalloc extents */240 *nbytes += xfs_inode_data_fork_size(ip);241 *nvecs += 1;242 }243 break;244 case XFS_DINODE_FMT_BTREE:245 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&246 ip->i_df.if_broot_bytes > 0) {247 *nbytes += ip->i_df.if_broot_bytes;248 *nvecs += 1;249 }250 break;251 case XFS_DINODE_FMT_LOCAL:252 if ((iip->ili_fields & XFS_ILOG_DDATA) &&253 ip->i_df.if_bytes > 0) {254 *nbytes += xlog_calc_iovec_len(ip->i_df.if_bytes);255 *nvecs += 1;256 }257 break;258 259 case XFS_DINODE_FMT_DEV:260 break;261 default:262 ASSERT(0);263 break;264 }265}266 267STATIC void268xfs_inode_item_attr_fork_size(269 struct xfs_inode_log_item *iip,270 int *nvecs,271 int *nbytes)272{273 struct xfs_inode *ip = iip->ili_inode;274 275 switch (ip->i_af.if_format) {276 case XFS_DINODE_FMT_EXTENTS:277 if ((iip->ili_fields & XFS_ILOG_AEXT) &&278 ip->i_af.if_nextents > 0 &&279 ip->i_af.if_bytes > 0) {280 /* worst case, doesn't subtract unused space */281 *nbytes += xfs_inode_attr_fork_size(ip);282 *nvecs += 1;283 }284 break;285 case XFS_DINODE_FMT_BTREE:286 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&287 ip->i_af.if_broot_bytes > 0) {288 *nbytes += ip->i_af.if_broot_bytes;289 *nvecs += 1;290 }291 break;292 case XFS_DINODE_FMT_LOCAL:293 if ((iip->ili_fields & XFS_ILOG_ADATA) &&294 ip->i_af.if_bytes > 0) {295 *nbytes += xlog_calc_iovec_len(ip->i_af.if_bytes);296 *nvecs += 1;297 }298 break;299 default:300 ASSERT(0);301 break;302 }303}304 305/*306 * This returns the number of iovecs needed to log the given inode item.307 *308 * We need one iovec for the inode log format structure, one for the309 * inode core, and possibly one for the inode data/extents/b-tree root310 * and one for the inode attribute data/extents/b-tree root.311 */312STATIC void313xfs_inode_item_size(314 struct xfs_log_item *lip,315 int *nvecs,316 int *nbytes)317{318 struct xfs_inode_log_item *iip = INODE_ITEM(lip);319 struct xfs_inode *ip = iip->ili_inode;320 321 *nvecs += 2;322 *nbytes += sizeof(struct xfs_inode_log_format) +323 xfs_log_dinode_size(ip->i_mount);324 325 xfs_inode_item_data_fork_size(iip, nvecs, nbytes);326 if (xfs_inode_has_attr_fork(ip))327 xfs_inode_item_attr_fork_size(iip, nvecs, nbytes);328}329 330STATIC void331xfs_inode_item_format_data_fork(332 struct xfs_inode_log_item *iip,333 struct xfs_inode_log_format *ilf,334 struct xfs_log_vec *lv,335 struct xfs_log_iovec **vecp)336{337 struct xfs_inode *ip = iip->ili_inode;338 size_t data_bytes;339 340 switch (ip->i_df.if_format) {341 case XFS_DINODE_FMT_EXTENTS:342 iip->ili_fields &=343 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEV);344 345 if ((iip->ili_fields & XFS_ILOG_DEXT) &&346 ip->i_df.if_nextents > 0 &&347 ip->i_df.if_bytes > 0) {348 struct xfs_bmbt_rec *p;349 350 ASSERT(xfs_iext_count(&ip->i_df) > 0);351 352 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IEXT);353 data_bytes = xfs_iextents_copy(ip, p, XFS_DATA_FORK);354 xlog_finish_iovec(lv, *vecp, data_bytes);355 356 ASSERT(data_bytes <= ip->i_df.if_bytes);357 358 ilf->ilf_dsize = data_bytes;359 ilf->ilf_size++;360 } else {361 iip->ili_fields &= ~XFS_ILOG_DEXT;362 }363 break;364 case XFS_DINODE_FMT_BTREE:365 iip->ili_fields &=366 ~(XFS_ILOG_DDATA | XFS_ILOG_DEXT | XFS_ILOG_DEV);367 368 if ((iip->ili_fields & XFS_ILOG_DBROOT) &&369 ip->i_df.if_broot_bytes > 0) {370 ASSERT(ip->i_df.if_broot != NULL);371 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IBROOT,372 ip->i_df.if_broot,373 ip->i_df.if_broot_bytes);374 ilf->ilf_dsize = ip->i_df.if_broot_bytes;375 ilf->ilf_size++;376 } else {377 ASSERT(!(iip->ili_fields &378 XFS_ILOG_DBROOT));379 iip->ili_fields &= ~XFS_ILOG_DBROOT;380 }381 break;382 case XFS_DINODE_FMT_LOCAL:383 iip->ili_fields &=384 ~(XFS_ILOG_DEXT | XFS_ILOG_DBROOT | XFS_ILOG_DEV);385 if ((iip->ili_fields & XFS_ILOG_DDATA) &&386 ip->i_df.if_bytes > 0) {387 ASSERT(ip->i_df.if_data != NULL);388 ASSERT(ip->i_disk_size > 0);389 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_ILOCAL,390 ip->i_df.if_data, ip->i_df.if_bytes);391 ilf->ilf_dsize = (unsigned)ip->i_df.if_bytes;392 ilf->ilf_size++;393 } else {394 iip->ili_fields &= ~XFS_ILOG_DDATA;395 }396 break;397 case XFS_DINODE_FMT_DEV:398 iip->ili_fields &=399 ~(XFS_ILOG_DDATA | XFS_ILOG_DBROOT | XFS_ILOG_DEXT);400 if (iip->ili_fields & XFS_ILOG_DEV)401 ilf->ilf_u.ilfu_rdev = sysv_encode_dev(VFS_I(ip)->i_rdev);402 break;403 default:404 ASSERT(0);405 break;406 }407}408 409STATIC void410xfs_inode_item_format_attr_fork(411 struct xfs_inode_log_item *iip,412 struct xfs_inode_log_format *ilf,413 struct xfs_log_vec *lv,414 struct xfs_log_iovec **vecp)415{416 struct xfs_inode *ip = iip->ili_inode;417 size_t data_bytes;418 419 switch (ip->i_af.if_format) {420 case XFS_DINODE_FMT_EXTENTS:421 iip->ili_fields &=422 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT);423 424 if ((iip->ili_fields & XFS_ILOG_AEXT) &&425 ip->i_af.if_nextents > 0 &&426 ip->i_af.if_bytes > 0) {427 struct xfs_bmbt_rec *p;428 429 ASSERT(xfs_iext_count(&ip->i_af) ==430 ip->i_af.if_nextents);431 432 p = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_EXT);433 data_bytes = xfs_iextents_copy(ip, p, XFS_ATTR_FORK);434 xlog_finish_iovec(lv, *vecp, data_bytes);435 436 ilf->ilf_asize = data_bytes;437 ilf->ilf_size++;438 } else {439 iip->ili_fields &= ~XFS_ILOG_AEXT;440 }441 break;442 case XFS_DINODE_FMT_BTREE:443 iip->ili_fields &=444 ~(XFS_ILOG_ADATA | XFS_ILOG_AEXT);445 446 if ((iip->ili_fields & XFS_ILOG_ABROOT) &&447 ip->i_af.if_broot_bytes > 0) {448 ASSERT(ip->i_af.if_broot != NULL);449 450 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_BROOT,451 ip->i_af.if_broot,452 ip->i_af.if_broot_bytes);453 ilf->ilf_asize = ip->i_af.if_broot_bytes;454 ilf->ilf_size++;455 } else {456 iip->ili_fields &= ~XFS_ILOG_ABROOT;457 }458 break;459 case XFS_DINODE_FMT_LOCAL:460 iip->ili_fields &=461 ~(XFS_ILOG_AEXT | XFS_ILOG_ABROOT);462 463 if ((iip->ili_fields & XFS_ILOG_ADATA) &&464 ip->i_af.if_bytes > 0) {465 ASSERT(ip->i_af.if_data != NULL);466 xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_IATTR_LOCAL,467 ip->i_af.if_data, ip->i_af.if_bytes);468 ilf->ilf_asize = (unsigned)ip->i_af.if_bytes;469 ilf->ilf_size++;470 } else {471 iip->ili_fields &= ~XFS_ILOG_ADATA;472 }473 break;474 default:475 ASSERT(0);476 break;477 }478}479 480/*481 * Convert an incore timestamp to a log timestamp. Note that the log format482 * specifies host endian format!483 */484static inline xfs_log_timestamp_t485xfs_inode_to_log_dinode_ts(486 struct xfs_inode *ip,487 const struct timespec64 tv)488{489 struct xfs_log_legacy_timestamp *lits;490 xfs_log_timestamp_t its;491 492 if (xfs_inode_has_bigtime(ip))493 return xfs_inode_encode_bigtime(tv);494 495 lits = (struct xfs_log_legacy_timestamp *)&its;496 lits->t_sec = tv.tv_sec;497 lits->t_nsec = tv.tv_nsec;498 499 return its;500}501 502/*503 * The legacy DMAPI fields are only present in the on-disk and in-log inodes,504 * but not in the in-memory one. But we are guaranteed to have an inode buffer505 * in memory when logging an inode, so we can just copy it from the on-disk506 * inode to the in-log inode here so that recovery of file system with these507 * fields set to non-zero values doesn't lose them. For all other cases we zero508 * the fields.509 */510static void511xfs_copy_dm_fields_to_log_dinode(512 struct xfs_inode *ip,513 struct xfs_log_dinode *to)514{515 struct xfs_dinode *dip;516 517 dip = xfs_buf_offset(ip->i_itemp->ili_item.li_buf,518 ip->i_imap.im_boffset);519 520 if (xfs_iflags_test(ip, XFS_IPRESERVE_DM_FIELDS)) {521 to->di_dmevmask = be32_to_cpu(dip->di_dmevmask);522 to->di_dmstate = be16_to_cpu(dip->di_dmstate);523 } else {524 to->di_dmevmask = 0;525 to->di_dmstate = 0;526 }527}528 529static inline void530xfs_inode_to_log_dinode_iext_counters(531 struct xfs_inode *ip,532 struct xfs_log_dinode *to)533{534 if (xfs_inode_has_large_extent_counts(ip)) {535 to->di_big_nextents = xfs_ifork_nextents(&ip->i_df);536 to->di_big_anextents = xfs_ifork_nextents(&ip->i_af);537 to->di_nrext64_pad = 0;538 } else {539 to->di_nextents = xfs_ifork_nextents(&ip->i_df);540 to->di_anextents = xfs_ifork_nextents(&ip->i_af);541 }542}543 544static void545xfs_inode_to_log_dinode(546 struct xfs_inode *ip,547 struct xfs_log_dinode *to,548 xfs_lsn_t lsn)549{550 struct inode *inode = VFS_I(ip);551 552 to->di_magic = XFS_DINODE_MAGIC;553 to->di_format = xfs_ifork_format(&ip->i_df);554 to->di_uid = i_uid_read(inode);555 to->di_gid = i_gid_read(inode);556 to->di_projid_lo = ip->i_projid & 0xffff;557 to->di_projid_hi = ip->i_projid >> 16;558 559 memset(to->di_pad3, 0, sizeof(to->di_pad3));560 to->di_atime = xfs_inode_to_log_dinode_ts(ip, inode_get_atime(inode));561 to->di_mtime = xfs_inode_to_log_dinode_ts(ip, inode_get_mtime(inode));562 to->di_ctime = xfs_inode_to_log_dinode_ts(ip, inode_get_ctime(inode));563 to->di_nlink = inode->i_nlink;564 to->di_gen = inode->i_generation;565 to->di_mode = inode->i_mode;566 567 to->di_size = ip->i_disk_size;568 to->di_nblocks = ip->i_nblocks;569 to->di_extsize = ip->i_extsize;570 to->di_forkoff = ip->i_forkoff;571 to->di_aformat = xfs_ifork_format(&ip->i_af);572 to->di_flags = ip->i_diflags;573 574 xfs_copy_dm_fields_to_log_dinode(ip, to);575 576 /* log a dummy value to ensure log structure is fully initialised */577 to->di_next_unlinked = NULLAGINO;578 579 if (xfs_has_v3inodes(ip->i_mount)) {580 to->di_version = 3;581 to->di_changecount = inode_peek_iversion(inode);582 to->di_crtime = xfs_inode_to_log_dinode_ts(ip, ip->i_crtime);583 to->di_flags2 = ip->i_diflags2;584 to->di_cowextsize = ip->i_cowextsize;585 to->di_ino = ip->i_ino;586 to->di_lsn = lsn;587 memset(to->di_pad2, 0, sizeof(to->di_pad2));588 uuid_copy(&to->di_uuid, &ip->i_mount->m_sb.sb_meta_uuid);589 to->di_v3_pad = 0;590 591 /* dummy value for initialisation */592 to->di_crc = 0;593 } else {594 to->di_version = 2;595 to->di_flushiter = ip->i_flushiter;596 memset(to->di_v2_pad, 0, sizeof(to->di_v2_pad));597 }598 599 xfs_inode_to_log_dinode_iext_counters(ip, to);600}601 602/*603 * Format the inode core. Current timestamp data is only in the VFS inode604 * fields, so we need to grab them from there. Hence rather than just copying605 * the XFS inode core structure, format the fields directly into the iovec.606 */607static void608xfs_inode_item_format_core(609 struct xfs_inode *ip,610 struct xfs_log_vec *lv,611 struct xfs_log_iovec **vecp)612{613 struct xfs_log_dinode *dic;614 615 dic = xlog_prepare_iovec(lv, vecp, XLOG_REG_TYPE_ICORE);616 xfs_inode_to_log_dinode(ip, dic, ip->i_itemp->ili_item.li_lsn);617 xlog_finish_iovec(lv, *vecp, xfs_log_dinode_size(ip->i_mount));618}619 620/*621 * This is called to fill in the vector of log iovecs for the given inode622 * log item. It fills the first item with an inode log format structure,623 * the second with the on-disk inode structure, and a possible third and/or624 * fourth with the inode data/extents/b-tree root and inode attributes625 * data/extents/b-tree root.626 *627 * Note: Always use the 64 bit inode log format structure so we don't628 * leave an uninitialised hole in the format item on 64 bit systems. Log629 * recovery on 32 bit systems handles this just fine, so there's no reason630 * for not using an initialising the properly padded structure all the time.631 */632STATIC void633xfs_inode_item_format(634 struct xfs_log_item *lip,635 struct xfs_log_vec *lv)636{637 struct xfs_inode_log_item *iip = INODE_ITEM(lip);638 struct xfs_inode *ip = iip->ili_inode;639 struct xfs_log_iovec *vecp = NULL;640 struct xfs_inode_log_format *ilf;641 642 ilf = xlog_prepare_iovec(lv, &vecp, XLOG_REG_TYPE_IFORMAT);643 ilf->ilf_type = XFS_LI_INODE;644 ilf->ilf_ino = ip->i_ino;645 ilf->ilf_blkno = ip->i_imap.im_blkno;646 ilf->ilf_len = ip->i_imap.im_len;647 ilf->ilf_boffset = ip->i_imap.im_boffset;648 ilf->ilf_fields = XFS_ILOG_CORE;649 ilf->ilf_size = 2; /* format + core */650 651 /*652 * make sure we don't leak uninitialised data into the log in the case653 * when we don't log every field in the inode.654 */655 ilf->ilf_dsize = 0;656 ilf->ilf_asize = 0;657 ilf->ilf_pad = 0;658 memset(&ilf->ilf_u, 0, sizeof(ilf->ilf_u));659 660 xlog_finish_iovec(lv, vecp, sizeof(*ilf));661 662 xfs_inode_item_format_core(ip, lv, &vecp);663 xfs_inode_item_format_data_fork(iip, ilf, lv, &vecp);664 if (xfs_inode_has_attr_fork(ip)) {665 xfs_inode_item_format_attr_fork(iip, ilf, lv, &vecp);666 } else {667 iip->ili_fields &=668 ~(XFS_ILOG_ADATA | XFS_ILOG_ABROOT | XFS_ILOG_AEXT);669 }670 671 /* update the format with the exact fields we actually logged */672 ilf->ilf_fields |= (iip->ili_fields & ~XFS_ILOG_TIMESTAMP);673}674 675/*676 * This is called to pin the inode associated with the inode log677 * item in memory so it cannot be written out.678 */679STATIC void680xfs_inode_item_pin(681 struct xfs_log_item *lip)682{683 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;684 685 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);686 ASSERT(lip->li_buf);687 688 trace_xfs_inode_pin(ip, _RET_IP_);689 atomic_inc(&ip->i_pincount);690}691 692 693/*694 * This is called to unpin the inode associated with the inode log695 * item which was previously pinned with a call to xfs_inode_item_pin().696 *697 * Also wake up anyone in xfs_iunpin_wait() if the count goes to 0.698 *699 * Note that unpin can race with inode cluster buffer freeing marking the buffer700 * stale. In that case, flush completions are run from the buffer unpin call,701 * which may happen before the inode is unpinned. If we lose the race, there702 * will be no buffer attached to the log item, but the inode will be marked703 * XFS_ISTALE.704 */705STATIC void706xfs_inode_item_unpin(707 struct xfs_log_item *lip,708 int remove)709{710 struct xfs_inode *ip = INODE_ITEM(lip)->ili_inode;711 712 trace_xfs_inode_unpin(ip, _RET_IP_);713 ASSERT(lip->li_buf || xfs_iflags_test(ip, XFS_ISTALE));714 ASSERT(atomic_read(&ip->i_pincount) > 0);715 if (atomic_dec_and_test(&ip->i_pincount))716 wake_up_bit(&ip->i_flags, __XFS_IPINNED_BIT);717}718 719STATIC uint720xfs_inode_item_push(721 struct xfs_log_item *lip,722 struct list_head *buffer_list)723 __releases(&lip->li_ailp->ail_lock)724 __acquires(&lip->li_ailp->ail_lock)725{726 struct xfs_inode_log_item *iip = INODE_ITEM(lip);727 struct xfs_inode *ip = iip->ili_inode;728 struct xfs_buf *bp = lip->li_buf;729 uint rval = XFS_ITEM_SUCCESS;730 int error;731 732 if (!bp || (ip->i_flags & XFS_ISTALE)) {733 /*734 * Inode item/buffer is being aborted due to cluster735 * buffer deletion. Trigger a log force to have that operation736 * completed and items removed from the AIL before the next push737 * attempt.738 */739 return XFS_ITEM_PINNED;740 }741 742 if (xfs_ipincount(ip) > 0 || xfs_buf_ispinned(bp))743 return XFS_ITEM_PINNED;744 745 if (xfs_iflags_test(ip, XFS_IFLUSHING))746 return XFS_ITEM_FLUSHING;747 748 if (!xfs_buf_trylock(bp))749 return XFS_ITEM_LOCKED;750 751 spin_unlock(&lip->li_ailp->ail_lock);752 753 /*754 * We need to hold a reference for flushing the cluster buffer as it may755 * fail the buffer without IO submission. In which case, we better get a756 * reference for that completion because otherwise we don't get a757 * reference for IO until we queue the buffer for delwri submission.758 */759 xfs_buf_hold(bp);760 error = xfs_iflush_cluster(bp);761 if (!error) {762 if (!xfs_buf_delwri_queue(bp, buffer_list))763 rval = XFS_ITEM_FLUSHING;764 xfs_buf_relse(bp);765 } else {766 /*767 * Release the buffer if we were unable to flush anything. On768 * any other error, the buffer has already been released.769 */770 if (error == -EAGAIN)771 xfs_buf_relse(bp);772 rval = XFS_ITEM_LOCKED;773 }774 775 spin_lock(&lip->li_ailp->ail_lock);776 return rval;777}778 779/*780 * Unlock the inode associated with the inode log item.781 */782STATIC void783xfs_inode_item_release(784 struct xfs_log_item *lip)785{786 struct xfs_inode_log_item *iip = INODE_ITEM(lip);787 struct xfs_inode *ip = iip->ili_inode;788 unsigned short lock_flags;789 790 ASSERT(ip->i_itemp != NULL);791 xfs_assert_ilocked(ip, XFS_ILOCK_EXCL);792 793 lock_flags = iip->ili_lock_flags;794 iip->ili_lock_flags = 0;795 if (lock_flags)796 xfs_iunlock(ip, lock_flags);797}798 799/*800 * This is called to find out where the oldest active copy of the inode log801 * item in the on disk log resides now that the last log write of it completed802 * at the given lsn. Since we always re-log all dirty data in an inode, the803 * latest copy in the on disk log is the only one that matters. Therefore,804 * simply return the given lsn.805 *806 * If the inode has been marked stale because the cluster is being freed, we807 * don't want to (re-)insert this inode into the AIL. There is a race condition808 * where the cluster buffer may be unpinned before the inode is inserted into809 * the AIL during transaction committed processing. If the buffer is unpinned810 * before the inode item has been committed and inserted, then it is possible811 * for the buffer to be written and IO completes before the inode is inserted812 * into the AIL. In that case, we'd be inserting a clean, stale inode into the813 * AIL which will never get removed. It will, however, get reclaimed which814 * triggers an assert in xfs_inode_free() complaining about freein an inode815 * still in the AIL.816 *817 * To avoid this, just unpin the inode directly and return a LSN of -1 so the818 * transaction committed code knows that it does not need to do any further819 * processing on the item.820 */821STATIC xfs_lsn_t822xfs_inode_item_committed(823 struct xfs_log_item *lip,824 xfs_lsn_t lsn)825{826 struct xfs_inode_log_item *iip = INODE_ITEM(lip);827 struct xfs_inode *ip = iip->ili_inode;828 829 if (xfs_iflags_test(ip, XFS_ISTALE)) {830 xfs_inode_item_unpin(lip, 0);831 return -1;832 }833 return lsn;834}835 836STATIC void837xfs_inode_item_committing(838 struct xfs_log_item *lip,839 xfs_csn_t seq)840{841 INODE_ITEM(lip)->ili_commit_seq = seq;842 return xfs_inode_item_release(lip);843}844 845static const struct xfs_item_ops xfs_inode_item_ops = {846 .iop_sort = xfs_inode_item_sort,847 .iop_precommit = xfs_inode_item_precommit,848 .iop_size = xfs_inode_item_size,849 .iop_format = xfs_inode_item_format,850 .iop_pin = xfs_inode_item_pin,851 .iop_unpin = xfs_inode_item_unpin,852 .iop_release = xfs_inode_item_release,853 .iop_committed = xfs_inode_item_committed,854 .iop_push = xfs_inode_item_push,855 .iop_committing = xfs_inode_item_committing,856};857 858 859/*860 * Initialize the inode log item for a newly allocated (in-core) inode.861 */862void863xfs_inode_item_init(864 struct xfs_inode *ip,865 struct xfs_mount *mp)866{867 struct xfs_inode_log_item *iip;868 869 ASSERT(ip->i_itemp == NULL);870 iip = ip->i_itemp = kmem_cache_zalloc(xfs_ili_cache,871 GFP_KERNEL | __GFP_NOFAIL);872 873 iip->ili_inode = ip;874 spin_lock_init(&iip->ili_lock);875 xfs_log_item_init(mp, &iip->ili_item, XFS_LI_INODE,876 &xfs_inode_item_ops);877}878 879/*880 * Free the inode log item and any memory hanging off of it.881 */882void883xfs_inode_item_destroy(884 struct xfs_inode *ip)885{886 struct xfs_inode_log_item *iip = ip->i_itemp;887 888 ASSERT(iip->ili_item.li_buf == NULL);889 890 ip->i_itemp = NULL;891 kvfree(iip->ili_item.li_lv_shadow);892 kmem_cache_free(xfs_ili_cache, iip);893}894 895 896/*897 * We only want to pull the item from the AIL if it is actually there898 * and its location in the log has not changed since we started the899 * flush. Thus, we only bother if the inode's lsn has not changed.900 */901static void902xfs_iflush_ail_updates(903 struct xfs_ail *ailp,904 struct list_head *list)905{906 struct xfs_log_item *lip;907 xfs_lsn_t tail_lsn = 0;908 909 /* this is an opencoded batch version of xfs_trans_ail_delete */910 spin_lock(&ailp->ail_lock);911 list_for_each_entry(lip, list, li_bio_list) {912 xfs_lsn_t lsn;913 914 clear_bit(XFS_LI_FAILED, &lip->li_flags);915 if (INODE_ITEM(lip)->ili_flush_lsn != lip->li_lsn)916 continue;917 918 /*919 * dgc: Not sure how this happens, but it happens very920 * occassionaly via generic/388. xfs_iflush_abort() also921 * silently handles this same "under writeback but not in AIL at922 * shutdown" condition via xfs_trans_ail_delete().923 */924 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {925 ASSERT(xlog_is_shutdown(lip->li_log));926 continue;927 }928 929 lsn = xfs_ail_delete_one(ailp, lip);930 if (!tail_lsn && lsn)931 tail_lsn = lsn;932 }933 xfs_ail_update_finish(ailp, tail_lsn);934}935 936/*937 * Walk the list of inodes that have completed their IOs. If they are clean938 * remove them from the list and dissociate them from the buffer. Buffers that939 * are still dirty remain linked to the buffer and on the list. Caller must940 * handle them appropriately.941 */942static void943xfs_iflush_finish(944 struct xfs_buf *bp,945 struct list_head *list)946{947 struct xfs_log_item *lip, *n;948 949 list_for_each_entry_safe(lip, n, list, li_bio_list) {950 struct xfs_inode_log_item *iip = INODE_ITEM(lip);951 bool drop_buffer = false;952 953 spin_lock(&iip->ili_lock);954 955 /*956 * Remove the reference to the cluster buffer if the inode is957 * clean in memory and drop the buffer reference once we've958 * dropped the locks we hold.959 */960 ASSERT(iip->ili_item.li_buf == bp);961 if (!iip->ili_fields) {962 iip->ili_item.li_buf = NULL;963 list_del_init(&lip->li_bio_list);964 drop_buffer = true;965 }966 iip->ili_last_fields = 0;967 iip->ili_flush_lsn = 0;968 clear_bit(XFS_LI_FLUSHING, &lip->li_flags);969 spin_unlock(&iip->ili_lock);970 xfs_iflags_clear(iip->ili_inode, XFS_IFLUSHING);971 if (drop_buffer)972 xfs_buf_rele(bp);973 }974}975 976/*977 * Inode buffer IO completion routine. It is responsible for removing inodes978 * attached to the buffer from the AIL if they have not been re-logged and979 * completing the inode flush.980 */981void982xfs_buf_inode_iodone(983 struct xfs_buf *bp)984{985 struct xfs_log_item *lip, *n;986 LIST_HEAD(flushed_inodes);987 LIST_HEAD(ail_updates);988 989 /*990 * Pull the attached inodes from the buffer one at a time and take the991 * appropriate action on them.992 */993 list_for_each_entry_safe(lip, n, &bp->b_li_list, li_bio_list) {994 struct xfs_inode_log_item *iip = INODE_ITEM(lip);995 996 if (xfs_iflags_test(iip->ili_inode, XFS_ISTALE)) {997 xfs_iflush_abort(iip->ili_inode);998 continue;999 }1000 if (!iip->ili_last_fields)1001 continue;1002 1003 /* Do an unlocked check for needing the AIL lock. */1004 if (iip->ili_flush_lsn == lip->li_lsn ||1005 test_bit(XFS_LI_FAILED, &lip->li_flags))1006 list_move_tail(&lip->li_bio_list, &ail_updates);1007 else1008 list_move_tail(&lip->li_bio_list, &flushed_inodes);1009 }1010 1011 if (!list_empty(&ail_updates)) {1012 xfs_iflush_ail_updates(bp->b_mount->m_ail, &ail_updates);1013 list_splice_tail(&ail_updates, &flushed_inodes);1014 }1015 1016 xfs_iflush_finish(bp, &flushed_inodes);1017 if (!list_empty(&flushed_inodes))1018 list_splice_tail(&flushed_inodes, &bp->b_li_list);1019}1020 1021void1022xfs_buf_inode_io_fail(1023 struct xfs_buf *bp)1024{1025 struct xfs_log_item *lip;1026 1027 list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {1028 set_bit(XFS_LI_FAILED, &lip->li_flags);1029 clear_bit(XFS_LI_FLUSHING, &lip->li_flags);1030 }1031}1032 1033/*1034 * Clear the inode logging fields so no more flushes are attempted. If we are1035 * on a buffer list, it is now safe to remove it because the buffer is1036 * guaranteed to be locked. The caller will drop the reference to the buffer1037 * the log item held.1038 */1039static void1040xfs_iflush_abort_clean(1041 struct xfs_inode_log_item *iip)1042{1043 iip->ili_last_fields = 0;1044 iip->ili_fields = 0;1045 iip->ili_fsync_fields = 0;1046 iip->ili_flush_lsn = 0;1047 iip->ili_item.li_buf = NULL;1048 list_del_init(&iip->ili_item.li_bio_list);1049 clear_bit(XFS_LI_FLUSHING, &iip->ili_item.li_flags);1050}1051 1052/*1053 * Abort flushing the inode from a context holding the cluster buffer locked.1054 *1055 * This is the normal runtime method of aborting writeback of an inode that is1056 * attached to a cluster buffer. It occurs when the inode and the backing1057 * cluster buffer have been freed (i.e. inode is XFS_ISTALE), or when cluster1058 * flushing or buffer IO completion encounters a log shutdown situation.1059 *1060 * If we need to abort inode writeback and we don't already hold the buffer1061 * locked, call xfs_iflush_shutdown_abort() instead as this should only ever be1062 * necessary in a shutdown situation.1063 */1064void1065xfs_iflush_abort(1066 struct xfs_inode *ip)1067{1068 struct xfs_inode_log_item *iip = ip->i_itemp;1069 struct xfs_buf *bp;1070 1071 if (!iip) {1072 /* clean inode, nothing to do */1073 xfs_iflags_clear(ip, XFS_IFLUSHING);1074 return;1075 }1076 1077 /*1078 * Remove the inode item from the AIL before we clear its internal1079 * state. Whilst the inode is in the AIL, it should have a valid buffer1080 * pointer for push operations to access - it is only safe to remove the1081 * inode from the buffer once it has been removed from the AIL.1082 *1083 * We also clear the failed bit before removing the item from the AIL1084 * as xfs_trans_ail_delete()->xfs_clear_li_failed() will release buffer1085 * references the inode item owns and needs to hold until we've fully1086 * aborted the inode log item and detached it from the buffer.1087 */1088 clear_bit(XFS_LI_FAILED, &iip->ili_item.li_flags);1089 xfs_trans_ail_delete(&iip->ili_item, 0);1090 1091 /*1092 * Grab the inode buffer so can we release the reference the inode log1093 * item holds on it.1094 */1095 spin_lock(&iip->ili_lock);1096 bp = iip->ili_item.li_buf;1097 xfs_iflush_abort_clean(iip);1098 spin_unlock(&iip->ili_lock);1099 1100 xfs_iflags_clear(ip, XFS_IFLUSHING);1101 if (bp)1102 xfs_buf_rele(bp);1103}1104 1105/*1106 * Abort an inode flush in the case of a shutdown filesystem. This can be called1107 * from anywhere with just an inode reference and does not require holding the1108 * inode cluster buffer locked. If the inode is attached to a cluster buffer,1109 * it will grab and lock it safely, then abort the inode flush.1110 */1111void1112xfs_iflush_shutdown_abort(1113 struct xfs_inode *ip)1114{1115 struct xfs_inode_log_item *iip = ip->i_itemp;1116 struct xfs_buf *bp;1117 1118 if (!iip) {1119 /* clean inode, nothing to do */1120 xfs_iflags_clear(ip, XFS_IFLUSHING);1121 return;1122 }1123 1124 spin_lock(&iip->ili_lock);1125 bp = iip->ili_item.li_buf;1126 if (!bp) {1127 spin_unlock(&iip->ili_lock);1128 xfs_iflush_abort(ip);1129 return;1130 }1131 1132 /*1133 * We have to take a reference to the buffer so that it doesn't get1134 * freed when we drop the ili_lock and then wait to lock the buffer.1135 * We'll clean up the extra reference after we pick up the ili_lock1136 * again.1137 */1138 xfs_buf_hold(bp);1139 spin_unlock(&iip->ili_lock);1140 xfs_buf_lock(bp);1141 1142 spin_lock(&iip->ili_lock);1143 if (!iip->ili_item.li_buf) {1144 /*1145 * Raced with another removal, hold the only reference1146 * to bp now. Inode should not be in the AIL now, so just clean1147 * up and return;1148 */1149 ASSERT(list_empty(&iip->ili_item.li_bio_list));1150 ASSERT(!test_bit(XFS_LI_IN_AIL, &iip->ili_item.li_flags));1151 xfs_iflush_abort_clean(iip);1152 spin_unlock(&iip->ili_lock);1153 xfs_iflags_clear(ip, XFS_IFLUSHING);1154 xfs_buf_relse(bp);1155 return;1156 }1157 1158 /*1159 * Got two references to bp. The first will get dropped by1160 * xfs_iflush_abort() when the item is removed from the buffer list, but1161 * we can't drop our reference until _abort() returns because we have to1162 * unlock the buffer as well. Hence we abort and then unlock and release1163 * our reference to the buffer.1164 */1165 ASSERT(iip->ili_item.li_buf == bp);1166 spin_unlock(&iip->ili_lock);1167 xfs_iflush_abort(ip);1168 xfs_buf_relse(bp);1169}1170 1171 1172/*1173 * convert an xfs_inode_log_format struct from the old 32 bit version1174 * (which can have different field alignments) to the native 64 bit version1175 */1176int1177xfs_inode_item_format_convert(1178 struct xfs_log_iovec *buf,1179 struct xfs_inode_log_format *in_f)1180{1181 struct xfs_inode_log_format_32 *in_f32 = buf->i_addr;1182 1183 if (buf->i_len != sizeof(*in_f32)) {1184 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, NULL);1185 return -EFSCORRUPTED;1186 }1187 1188 in_f->ilf_type = in_f32->ilf_type;1189 in_f->ilf_size = in_f32->ilf_size;1190 in_f->ilf_fields = in_f32->ilf_fields;1191 in_f->ilf_asize = in_f32->ilf_asize;1192 in_f->ilf_dsize = in_f32->ilf_dsize;1193 in_f->ilf_ino = in_f32->ilf_ino;1194 memcpy(&in_f->ilf_u, &in_f32->ilf_u, sizeof(in_f->ilf_u));1195 in_f->ilf_blkno = in_f32->ilf_blkno;1196 in_f->ilf_len = in_f32->ilf_len;1197 in_f->ilf_boffset = in_f32->ilf_boffset;1198 return 0;1199}1200