On Mon, Jun 01, 2020 at 08:03:38AM -0400, Brian Foster wrote: > On Fri, May 29, 2020 at 02:01:59PM -0700, Darrick J. Wong wrote: > > On Thu, May 28, 2020 at 11:08:36AM -0400, Brian Foster wrote: > > > On Wed, May 27, 2020 at 03:34:24PM -0700, Darrick J. Wong wrote: > > > > On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote: > > > > > On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote: > > > > > > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > > > > > > > > > > > Port the new btree staging context and related block reservation helper > > > > > > code from the kernel to repair. We'll use this in subsequent patches to > > > > > > implement btree bulk loading. > > > > > > > > > > > > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > > > > > --- > > > > > > include/libxfs.h | 1 > > > > > > libxfs/libxfs_api_defs.h | 2 > > > > > > repair/Makefile | 4 - > > > > > > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++ > > > > > > repair/bload.h | 77 ++++++++++++ > > > > > > repair/xfs_repair.c | 17 +++ > > > > > > 6 files changed, 402 insertions(+), 2 deletions(-) > > > > > > create mode 100644 repair/bload.c > > > > > > create mode 100644 repair/bload.h > > > > > > > > > > > > > > > > > ... > > > > > > diff --git a/repair/bload.c b/repair/bload.c > > > > > > new file mode 100644 > > > > > > index 00000000..9bc17468 > > > > > > --- /dev/null > > > > > > +++ b/repair/bload.c > > > > > > @@ -0,0 +1,303 @@ > > > > > > +// SPDX-License-Identifier: GPL-2.0-or-later > > > > > > +/* > > > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved. > > > > > > + * Author: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > > > > > + */ > > > > > > +#include <libxfs.h> > > > > > > +#include "bload.h" > > > > > > + > > > > > > +#define trace_xrep_newbt_claim_block(...) ((void) 0) > > > > > > +#define trace_xrep_newbt_free_blocks(...) ((void) 0) > > > > > > + > > > > > > +int bload_leaf_slack = -1; > > > > > > +int bload_node_slack = -1; > > > > > > + > > > > > > +/* Ported routines from fs/xfs/scrub/repair.c */ > > > > > > + > > > > > > > > > > Looks mostly straightforward, but I'll have to come back to this as I > > > > > get to the code that uses it later in the series. In the meantime, I see > > > > > some of these helpers in scrub/repair.c while not others. Are there > > > > > references to other routines that are intended to be copies from kernel > > > > > code? > > > > > > > > Hm. I might not understand the question, but in general the code should > > > > be fairly similar to the kernel functions. The biggest differences are > > > > (a) that whole libxfs error code mess, (b) the much simpler repair_ctx > > > > structure, and (c) the fact that repair doesn't bother with EFIs to > > > > automatically reap blocks. > > > > > > > > So... the ten functions you see here do the same things as their kernel > > > > counterparts, but they get to do it in the much simpler userspace > > > > environment. > > > > > > > > > > Right.. I was able to find the first function (xrep_roll_ag_trans()) > > > easily in the kernel because it has the same name. The next one or two > > > (i.e., xrep_newbt_*()) I couldn't find and then gave up. Are they > > > renamed? Unmerged? > > > > Ahh, silly me, the rest are as yet unmerged components of the online > > repair code. Maybe it makes more sense to drop the "Ported routines > > from XXXX" comment, particularly since it probably will be easier to > > merge the xfs_repair series first, which is /much/ smaller in scope. > > > > Probably so. It's just going to confuse if this is not intended to land > at the same time as the kernel code. <nod> I also realized while re-examining the two codebases that the xrep_newbt code *isn't* going to converge. The kernel needs extra bits for smuggling in EFIs, and userspace doesn't ever need the ability to bhold the AG header buffers to roll a transaction. So in effect I dropped a bunch of code and re-prefixed the surviving code so that this patch will get quite a bit smaller. I'll try to send a new revision tomorrow. --D > > > > The other functions in scrub/repair.c that didn't get ported are either > > > > for other types of repairs or exist to support the in-kernel code and > > > > aren't needed here. > > > > > > > > > > Sure, I'm just curious how to identify the source of the ones that are. > > > > https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/tree/fs/xfs/scrub/repair.c?h=djwong-wtf > > > > Is a fairly recent snapshot of what the kernel version looks like. > > > > Thanks. > > Brian > > > --D > > > > > > > > Brian > > > > > > > --D > > > > > > > > > Brian > > > > > > > > > > > +/* > > > > > > + * Roll a transaction, keeping the AG headers locked and reinitializing > > > > > > + * the btree cursors. > > > > > > + */ > > > > > > +int > > > > > > +xrep_roll_ag_trans( > > > > > > + struct repair_ctx *sc) > > > > > > +{ > > > > > > + int error; > > > > > > + > > > > > > + /* Keep the AG header buffers locked so we can keep going. */ > > > > > > + if (sc->agi_bp) > > > > > > + libxfs_trans_bhold(sc->tp, sc->agi_bp); > > > > > > + if (sc->agf_bp) > > > > > > + libxfs_trans_bhold(sc->tp, sc->agf_bp); > > > > > > + if (sc->agfl_bp) > > > > > > + libxfs_trans_bhold(sc->tp, sc->agfl_bp); > > > > > > + > > > > > > + /* > > > > > > + * Roll the transaction. We still own the buffer and the buffer lock > > > > > > + * regardless of whether or not the roll succeeds. If the roll fails, > > > > > > + * the buffers will be released during teardown on our way out of the > > > > > > + * kernel. If it succeeds, we join them to the new transaction and > > > > > > + * move on. > > > > > > + */ > > > > > > + error = -libxfs_trans_roll(&sc->tp); > > > > > > + if (error) > > > > > > + return error; > > > > > > + > > > > > > + /* Join AG headers to the new transaction. */ > > > > > > + if (sc->agi_bp) > > > > > > + libxfs_trans_bjoin(sc->tp, sc->agi_bp); > > > > > > + if (sc->agf_bp) > > > > > > + libxfs_trans_bjoin(sc->tp, sc->agf_bp); > > > > > > + if (sc->agfl_bp) > > > > > > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp); > > > > > > + > > > > > > + return 0; > > > > > > +} > > > > > > + > > > > > > +/* Initialize accounting resources for staging a new AG btree. */ > > > > > > +void > > > > > > +xrep_newbt_init_ag( > > > > > > + struct xrep_newbt *xnr, > > > > > > + struct repair_ctx *sc, > > > > > > + const struct xfs_owner_info *oinfo, > > > > > > + xfs_fsblock_t alloc_hint, > > > > > > + enum xfs_ag_resv_type resv) > > > > > > +{ > > > > > > + memset(xnr, 0, sizeof(struct xrep_newbt)); > > > > > > + xnr->sc = sc; > > > > > > + xnr->oinfo = *oinfo; /* structure copy */ > > > > > > + xnr->alloc_hint = alloc_hint; > > > > > > + xnr->resv = resv; > > > > > > + INIT_LIST_HEAD(&xnr->resv_list); > > > > > > +} > > > > > > + > > > > > > +/* Initialize accounting resources for staging a new inode fork btree. */ > > > > > > +void > > > > > > +xrep_newbt_init_inode( > > > > > > + struct xrep_newbt *xnr, > > > > > > + struct repair_ctx *sc, > > > > > > + int whichfork, > > > > > > + const struct xfs_owner_info *oinfo) > > > > > > +{ > > > > > > + xrep_newbt_init_ag(xnr, sc, oinfo, > > > > > > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino), > > > > > > + XFS_AG_RESV_NONE); > > > > > > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0); > > > > > > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork); > > > > > > +} > > > > > > + > > > > > > +/* > > > > > > + * Initialize accounting resources for staging a new btree. Callers are > > > > > > + * expected to add their own reservations (and clean them up) manually. > > > > > > + */ > > > > > > +void > > > > > > +xrep_newbt_init_bare( > > > > > > + struct xrep_newbt *xnr, > > > > > > + struct repair_ctx *sc) > > > > > > +{ > > > > > > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK, > > > > > > + XFS_AG_RESV_NONE); > > > > > > +} > > > > > > + > > > > > > +/* Designate specific blocks to be used to build our new btree. */ > > > > > > +int > > > > > > +xrep_newbt_add_blocks( > > > > > > + struct xrep_newbt *xnr, > > > > > > + xfs_fsblock_t fsbno, > > > > > > + xfs_extlen_t len) > > > > > > +{ > > > > > > + struct xrep_newbt_resv *resv; > > > > > > + > > > > > > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL); > > > > > > + if (!resv) > > > > > > + return ENOMEM; > > > > > > + > > > > > > + INIT_LIST_HEAD(&resv->list); > > > > > > + resv->fsbno = fsbno; > > > > > > + resv->len = len; > > > > > > + resv->used = 0; > > > > > > + list_add_tail(&resv->list, &xnr->resv_list); > > > > > > + return 0; > > > > > > +} > > > > > > + > > > > > > +/* Reserve disk space for our new btree. */ > > > > > > +int > > > > > > +xrep_newbt_alloc_blocks( > > > > > > + struct xrep_newbt *xnr, > > > > > > + uint64_t nr_blocks) > > > > > > +{ > > > > > > + struct repair_ctx *sc = xnr->sc; > > > > > > + xfs_alloctype_t type; > > > > > > + xfs_fsblock_t alloc_hint = xnr->alloc_hint; > > > > > > + int error = 0; > > > > > > + > > > > > > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO; > > > > > > + > > > > > > + while (nr_blocks > 0 && !error) { > > > > > > + struct xfs_alloc_arg args = { > > > > > > + .tp = sc->tp, > > > > > > + .mp = sc->mp, > > > > > > + .type = type, > > > > > > + .fsbno = alloc_hint, > > > > > > + .oinfo = xnr->oinfo, > > > > > > + .minlen = 1, > > > > > > + .maxlen = nr_blocks, > > > > > > + .prod = 1, > > > > > > + .resv = xnr->resv, > > > > > > + }; > > > > > > + > > > > > > + error = -libxfs_alloc_vextent(&args); > > > > > > + if (error) > > > > > > + return error; > > > > > > + if (args.fsbno == NULLFSBLOCK) > > > > > > + return ENOSPC; > > > > > > + > > > > > > + /* We don't have real EFIs here so skip that. */ > > > > > > + > > > > > > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len); > > > > > > + if (error) > > > > > > + break; > > > > > > + > > > > > > + nr_blocks -= args.len; > > > > > > + alloc_hint = args.fsbno + args.len - 1; > > > > > > + > > > > > > + if (sc->ip) > > > > > > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip); > > > > > > + else > > > > > > + error = xrep_roll_ag_trans(sc); > > > > > > + } > > > > > > + > > > > > > + return error; > > > > > > +} > > > > > > + > > > > > > +/* > > > > > > + * Release blocks that were reserved for a btree repair. If the repair > > > > > > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try > > > > > > + * to free the extents immediately to roll the filesystem back to where it was > > > > > > + * before we started. > > > > > > + */ > > > > > > +static inline int > > > > > > +xrep_newbt_destroy_reservation( > > > > > > + struct xrep_newbt *xnr, > > > > > > + struct xrep_newbt_resv *resv, > > > > > > + bool cancel_repair) > > > > > > +{ > > > > > > + struct repair_ctx *sc = xnr->sc; > > > > > > + > > > > > > + if (cancel_repair) { > > > > > > + int error; > > > > > > + > > > > > > + /* Free the extent then roll the transaction. */ > > > > > > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len, > > > > > > + &xnr->oinfo, xnr->resv); > > > > > > + if (error) > > > > > > + return error; > > > > > > + > > > > > > + if (sc->ip) > > > > > > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip); > > > > > > + return xrep_roll_ag_trans(sc); > > > > > > + } > > > > > > + > > > > > > + /* We don't have EFIs here so skip the EFD. */ > > > > > > + > > > > > > + /* > > > > > > + * Use the deferred freeing mechanism to schedule for deletion any > > > > > > + * blocks we didn't use to rebuild the tree. This enables us to log > > > > > > + * them all in the same transaction as the root change. > > > > > > + */ > > > > > > + resv->fsbno += resv->used; > > > > > > + resv->len -= resv->used; > > > > > > + resv->used = 0; > > > > > > + > > > > > > + if (resv->len == 0) > > > > > > + return 0; > > > > > > + > > > > > > + trace_xrep_newbt_free_blocks(sc->mp, > > > > > > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno), > > > > > > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno), > > > > > > + resv->len, xnr->oinfo.oi_owner); > > > > > > + > > > > > > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true); > > > > > > + > > > > > > + return 0; > > > > > > +} > > > > > > + > > > > > > +/* Free all the accounting info and disk space we reserved for a new btree. */ > > > > > > +void > > > > > > +xrep_newbt_destroy( > > > > > > + struct xrep_newbt *xnr, > > > > > > + int error) > > > > > > +{ > > > > > > + struct repair_ctx *sc = xnr->sc; > > > > > > + struct xrep_newbt_resv *resv, *n; > > > > > > + int err2; > > > > > > + > > > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) { > > > > > > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0); > > > > > > + if (err2) > > > > > > + goto junkit; > > > > > > + > > > > > > + list_del(&resv->list); > > > > > > + kmem_free(resv); > > > > > > + } > > > > > > + > > > > > > +junkit: > > > > > > + /* > > > > > > + * If we still have reservations attached to @newbt, cleanup must have > > > > > > + * failed and the filesystem is about to go down. Clean up the incore > > > > > > + * reservations. > > > > > > + */ > > > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) { > > > > > > + list_del(&resv->list); > > > > > > + kmem_free(resv); > > > > > > + } > > > > > > + > > > > > > + if (sc->ip) { > > > > > > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork); > > > > > > + xnr->ifake.if_fork = NULL; > > > > > > + } > > > > > > +} > > > > > > + > > > > > > +/* Feed one of the reserved btree blocks to the bulk loader. */ > > > > > > +int > > > > > > +xrep_newbt_claim_block( > > > > > > + struct xfs_btree_cur *cur, > > > > > > + struct xrep_newbt *xnr, > > > > > > + union xfs_btree_ptr *ptr) > > > > > > +{ > > > > > > + struct xrep_newbt_resv *resv; > > > > > > + xfs_fsblock_t fsb; > > > > > > + > > > > > > + /* > > > > > > + * The first item in the list should always have a free block unless > > > > > > + * we're completely out. > > > > > > + */ > > > > > > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list); > > > > > > + if (resv->used == resv->len) > > > > > > + return ENOSPC; > > > > > > + > > > > > > + /* > > > > > > + * Peel off a block from the start of the reservation. We allocate > > > > > > + * blocks in order to place blocks on disk in increasing record or key > > > > > > + * order. The block reservations tend to end up on the list in > > > > > > + * decreasing order, which hopefully results in leaf blocks ending up > > > > > > + * together. > > > > > > + */ > > > > > > + fsb = resv->fsbno + resv->used; > > > > > > + resv->used++; > > > > > > + > > > > > > + /* If we used all the blocks in this reservation, move it to the end. */ > > > > > > + if (resv->used == resv->len) > > > > > > + list_move_tail(&resv->list, &xnr->resv_list); > > > > > > + > > > > > > + trace_xrep_newbt_claim_block(cur->bc_mp, > > > > > > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb), > > > > > > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb), > > > > > > + 1, xnr->oinfo.oi_owner); > > > > > > + > > > > > > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS) > > > > > > + ptr->l = cpu_to_be64(fsb); > > > > > > + else > > > > > > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb)); > > > > > > + return 0; > > > > > > +} > > > > > > diff --git a/repair/bload.h b/repair/bload.h > > > > > > new file mode 100644 > > > > > > index 00000000..020c4834 > > > > > > --- /dev/null > > > > > > +++ b/repair/bload.h > > > > > > @@ -0,0 +1,77 @@ > > > > > > +// SPDX-License-Identifier: GPL-2.0-or-later > > > > > > +/* > > > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved. > > > > > > + * Author: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > > > > > + */ > > > > > > +#ifndef __XFS_REPAIR_BLOAD_H__ > > > > > > +#define __XFS_REPAIR_BLOAD_H__ > > > > > > + > > > > > > +extern int bload_leaf_slack; > > > > > > +extern int bload_node_slack; > > > > > > + > > > > > > +struct repair_ctx { > > > > > > + struct xfs_mount *mp; > > > > > > + struct xfs_inode *ip; > > > > > > + struct xfs_trans *tp; > > > > > > + > > > > > > + struct xfs_buf *agi_bp; > > > > > > + struct xfs_buf *agf_bp; > > > > > > + struct xfs_buf *agfl_bp; > > > > > > +}; > > > > > > + > > > > > > +struct xrep_newbt_resv { > > > > > > + /* Link to list of extents that we've reserved. */ > > > > > > + struct list_head list; > > > > > > + > > > > > > + /* FSB of the block we reserved. */ > > > > > > + xfs_fsblock_t fsbno; > > > > > > + > > > > > > + /* Length of the reservation. */ > > > > > > + xfs_extlen_t len; > > > > > > + > > > > > > + /* How much of this reservation we've used. */ > > > > > > + xfs_extlen_t used; > > > > > > +}; > > > > > > + > > > > > > +struct xrep_newbt { > > > > > > + struct repair_ctx *sc; > > > > > > + > > > > > > + /* List of extents that we've reserved. */ > > > > > > + struct list_head resv_list; > > > > > > + > > > > > > + /* Fake root for new btree. */ > > > > > > + union { > > > > > > + struct xbtree_afakeroot afake; > > > > > > + struct xbtree_ifakeroot ifake; > > > > > > + }; > > > > > > + > > > > > > + /* rmap owner of these blocks */ > > > > > > + struct xfs_owner_info oinfo; > > > > > > + > > > > > > + /* The last reservation we allocated from. */ > > > > > > + struct xrep_newbt_resv *last_resv; > > > > > > + > > > > > > + /* Allocation hint */ > > > > > > + xfs_fsblock_t alloc_hint; > > > > > > + > > > > > > + /* per-ag reservation type */ > > > > > > + enum xfs_ag_resv_type resv; > > > > > > +}; > > > > > > + > > > > > > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \ > > > > > > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list) > > > > > > + > > > > > > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc); > > > > > > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc, > > > > > > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint, > > > > > > + enum xfs_ag_resv_type resv); > > > > > > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc, > > > > > > + int whichfork, const struct xfs_owner_info *oinfo); > > > > > > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno, > > > > > > + xfs_extlen_t len); > > > > > > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks); > > > > > > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error); > > > > > > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr, > > > > > > + union xfs_btree_ptr *ptr); > > > > > > + > > > > > > +#endif /* __XFS_REPAIR_BLOAD_H__ */ > > > > > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c > > > > > > index 9d72fa8e..8fbd3649 100644 > > > > > > --- a/repair/xfs_repair.c > > > > > > +++ b/repair/xfs_repair.c > > > > > > @@ -24,6 +24,7 @@ > > > > > > #include "rmap.h" > > > > > > #include "libfrog/fsgeom.h" > > > > > > #include "libfrog/platform.h" > > > > > > +#include "bload.h" > > > > > > > > > > > > /* > > > > > > * option tables for getsubopt calls > > > > > > @@ -39,6 +40,8 @@ enum o_opt_nums { > > > > > > AG_STRIDE, > > > > > > FORCE_GEO, > > > > > > PHASE2_THREADS, > > > > > > + BLOAD_LEAF_SLACK, > > > > > > + BLOAD_NODE_SLACK, > > > > > > O_MAX_OPTS, > > > > > > }; > > > > > > > > > > > > @@ -49,6 +52,8 @@ static char *o_opts[] = { > > > > > > [AG_STRIDE] = "ag_stride", > > > > > > [FORCE_GEO] = "force_geometry", > > > > > > [PHASE2_THREADS] = "phase2_threads", > > > > > > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack", > > > > > > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack", > > > > > > [O_MAX_OPTS] = NULL, > > > > > > }; > > > > > > > > > > > > @@ -260,6 +265,18 @@ process_args(int argc, char **argv) > > > > > > _("-o phase2_threads requires a parameter\n")); > > > > > > phase2_threads = (int)strtol(val, NULL, 0); > > > > > > break; > > > > > > + case BLOAD_LEAF_SLACK: > > > > > > + if (!val) > > > > > > + do_abort( > > > > > > + _("-o debug_bload_leaf_slack requires a parameter\n")); > > > > > > + bload_leaf_slack = (int)strtol(val, NULL, 0); > > > > > > + break; > > > > > > + case BLOAD_NODE_SLACK: > > > > > > + if (!val) > > > > > > + do_abort( > > > > > > + _("-o debug_bload_node_slack requires a parameter\n")); > > > > > > + bload_node_slack = (int)strtol(val, NULL, 0); > > > > > > + break; > > > > > > default: > > > > > > unknown('o', val); > > > > > > break; > > > > > > > > > > > > > > > > > > > > >
