Author: scottl
Date: Tue Jan 22 07:40:38 2013
New Revision: 245779
URL: http://svnweb.freebsd.org/changeset/base/245779
Log:
MFC r243018:
- Fix a truncation bug with softdep journaling that could leak blocks on
crash. When truncating a file that never made it to disk we use the
canceled allocation dependencies to hold the journal records until
the truncation completes. Previously allocdirect dependencies on
the id_bufwait list were not considered and their journal space
could expire before the bitmaps were written. Cancel them and attach
them to the freeblks as we do for other allocdirects.
- Add KTR traces that were used to debug this problem.
- When adding jsegdeps, always use jwork_insert() so we don't have more
than one segdep on a given jwork list.
Modified:
stable/9/sys/ufs/ffs/ffs_softdep.c
Directory Properties:
stable/9/sys/ (props changed)
Modified: stable/9/sys/ufs/ffs/ffs_softdep.c
==============================================================================
--- stable/9/sys/ufs/ffs/ffs_softdep.c Tue Jan 22 07:38:43 2013
(r245778)
+++ stable/9/sys/ufs/ffs/ffs_softdep.c Tue Jan 22 07:40:38 2013
(r245779)
@@ -60,6 +60,7 @@ __FBSDID("$FreeBSD$");
#include <sys/buf.h>
#include <sys/kdb.h>
#include <sys/kthread.h>
+#include <sys/ktr.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/malloc.h>
@@ -92,6 +93,8 @@ __FBSDID("$FreeBSD$");
#include <ddb/ddb.h>
+#define KTR_SUJ 0 /* Define to KTR_SPARE. */
+
#ifndef SOFTUPDATES
int
@@ -769,6 +772,34 @@ struct pagedep_hashhead;
struct bmsafemap_hashhead;
/*
+ * Private journaling structures.
+ */
+struct jblocks {
+ struct jseglst jb_segs; /* TAILQ of current segments. */
+ struct jseg *jb_writeseg; /* Next write to complete. */
+ struct jseg *jb_oldestseg; /* Oldest segment with valid entries. */
+ struct jextent *jb_extent; /* Extent array. */
+ uint64_t jb_nextseq; /* Next sequence number. */
+ uint64_t jb_oldestwrseq; /* Oldest written sequence number. */
+ uint8_t jb_needseg; /* Need a forced segment. */
+ uint8_t jb_suspended; /* Did journal suspend writes? */
+ int jb_avail; /* Available extents. */
+ int jb_used; /* Last used extent. */
+ int jb_head; /* Allocator head. */
+ int jb_off; /* Allocator extent offset. */
+ int jb_blocks; /* Total disk blocks covered. */
+ int jb_free; /* Total disk blocks free. */
+ int jb_min; /* Minimum free space. */
+ int jb_low; /* Low on space. */
+ int jb_age; /* Insertion time of oldest rec. */
+};
+
+struct jextent {
+ ufs2_daddr_t je_daddr; /* Disk block address. */
+ int je_blocks; /* Disk block count. */
+};
+
+/*
* Internal function prototypes.
*/
static void softdep_error(char *, int);
@@ -2273,19 +2304,15 @@ static void
indirblk_insert(freework)
struct freework *freework;
{
- struct freeblks *freeblks;
- struct jsegdep *jsegdep;
- struct worklist *wk;
+ struct jblocks *jblocks;
+ struct jseg *jseg;
- freeblks = freework->fw_freeblks;
- LIST_FOREACH(wk, &freeblks->fb_jwork, wk_list)
- if (wk->wk_type == D_JSEGDEP)
- break;
- if (wk == NULL)
+ jblocks = VFSTOUFS(freework->fw_list.wk_mp)->softdep_jblocks;
+ jseg = TAILQ_LAST(&jblocks->jb_segs, jseglst);
+ if (jseg == NULL)
return;
- jsegdep = WK_JSEGDEP(wk);
- LIST_INSERT_HEAD(&jsegdep->jd_seg->js_indirs, freework, fw_segs);
+ LIST_INSERT_HEAD(&jseg->js_indirs, freework, fw_segs);
TAILQ_INSERT_HEAD(INDIR_HASH(freework->fw_list.wk_mp,
freework->fw_blkno), freework, fw_next);
freework->fw_state &= ~DEPCOMPLETE;
@@ -2438,31 +2465,6 @@ softdep_unmount(mp)
journal_unmount(mp);
}
-struct jblocks {
- struct jseglst jb_segs; /* TAILQ of current segments. */
- struct jseg *jb_writeseg; /* Next write to complete. */
- struct jseg *jb_oldestseg; /* Oldest segment with valid entries. */
- struct jextent *jb_extent; /* Extent array. */
- uint64_t jb_nextseq; /* Next sequence number. */
- uint64_t jb_oldestwrseq; /* Oldest written sequence number. */
- uint8_t jb_needseg; /* Need a forced segment. */
- uint8_t jb_suspended; /* Did journal suspend writes? */
- int jb_avail; /* Available extents. */
- int jb_used; /* Last used extent. */
- int jb_head; /* Allocator head. */
- int jb_off; /* Allocator extent offset. */
- int jb_blocks; /* Total disk blocks covered. */
- int jb_free; /* Total disk blocks free. */
- int jb_min; /* Minimum free space. */
- int jb_low; /* Low on space. */
- int jb_age; /* Insertion time of oldest rec. */
-};
-
-struct jextent {
- ufs2_daddr_t je_daddr; /* Disk block address. */
- int je_blocks; /* Disk block count. */
-};
-
static struct jblocks *
jblocks_create(void)
{
@@ -3668,7 +3670,7 @@ handle_written_jnewblk(jnewblk)
*/
freefrag = WK_FREEFRAG(jnewblk->jn_dep);
freefrag->ff_jdep = NULL;
- WORKLIST_INSERT(&freefrag->ff_jwork, &jsegdep->jd_list);
+ jwork_insert(&freefrag->ff_jwork, jsegdep);
break;
case D_FREEWORK:
/*
@@ -3676,8 +3678,7 @@ handle_written_jnewblk(jnewblk)
*/
freework = WK_FREEWORK(jnewblk->jn_dep);
freework->fw_jnewblk = NULL;
- WORKLIST_INSERT(&freework->fw_freeblks->fb_jwork,
- &jsegdep->jd_list);
+ jwork_insert(&freework->fw_freeblks->fb_jwork, jsegdep);
break;
default:
panic("handle_written_jnewblk: Unknown type %d.",
@@ -3707,6 +3708,7 @@ cancel_jfreefrag(jfreefrag)
jfreefrag->fr_freefrag = NULL;
free_jfreefrag(jfreefrag);
freefrag->ff_state |= DEPCOMPLETE;
+ CTR1(KTR_SUJ, "cancel_jfreefrag: blkno %jd", freefrag->ff_blkno);
}
/*
@@ -3770,7 +3772,7 @@ handle_written_jblkdep(jblkdep)
jblkdep->jb_jsegdep = NULL;
freeblks = jblkdep->jb_freeblks;
LIST_REMOVE(jblkdep, jb_deps);
- WORKLIST_INSERT(&freeblks->fb_jwork, &jsegdep->jd_list);
+ jwork_insert(&freeblks->fb_jwork, jsegdep);
/*
* If the freeblks is all journaled, we can add it to the worklist.
*/
@@ -3973,6 +3975,7 @@ cancel_jfreeblk(freeblks, blkno)
}
if (jblkdep == NULL)
return;
+ CTR1(KTR_SUJ, "cancel_jfreeblk: blkno %jd", blkno);
free_jsegdep(jblkdep->jb_jsegdep);
LIST_REMOVE(jblkdep, jb_deps);
WORKITEM_FREE(jfreeblk, D_JFREEBLK);
@@ -4213,6 +4216,7 @@ cancel_jnewblk(jnewblk, wkhd)
{
struct jsegdep *jsegdep;
+ CTR1(KTR_SUJ, "cancel_jnewblk: blkno %jd", jnewblk->jn_blkno);
jsegdep = jnewblk->jn_jsegdep;
if (jnewblk->jn_jsegdep == NULL || jnewblk->jn_dep == NULL)
panic("cancel_jnewblk: Invalid state");
@@ -4904,6 +4908,10 @@ softdep_setup_blkmapdep(bp, mp, newblkno
}
#endif
}
+
+ CTR3(KTR_SUJ,
+ "softdep_setup_blkmapdep: blkno %jd frags %d oldfrags %d",
+ newblkno, frags, oldfrags);
ACQUIRE_LOCK(&lk);
if (newblk_lookup(mp, newblkno, DEPALLOC, &newblk) != 0)
panic("softdep_setup_blkmapdep: found block");
@@ -5065,6 +5073,10 @@ softdep_setup_allocdirect(ip, off, newbl
else
freefrag = NULL;
+ CTR6(KTR_SUJ,
+ "softdep_setup_allocdirect: ino %d blkno %jd oldblkno %jd "
+ "off %jd newsize %ld oldsize %d",
+ ip->i_number, newblkno, oldblkno, off, newsize, oldsize);
ACQUIRE_LOCK(&lk);
if (off >= NDADDR) {
if (lbn > 0)
@@ -5343,6 +5355,8 @@ newfreefrag(ip, blkno, size, lbn)
struct freefrag *freefrag;
struct fs *fs;
+ CTR4(KTR_SUJ, "newfreefrag: ino %d blkno %jd size %ld lbn %jd",
+ ip->i_number, blkno, size, lbn);
fs = ip->i_fs;
if (fragnum(fs, blkno) + numfrags(fs, size) > fs->fs_frag)
panic("newfreefrag: frag size");
@@ -5378,6 +5392,9 @@ handle_workitem_freefrag(freefrag)
struct ufsmount *ump = VFSTOUFS(freefrag->ff_list.wk_mp);
struct workhead wkhd;
+ CTR3(KTR_SUJ,
+ "handle_workitem_freefrag: ino %d blkno %jd size %ld",
+ freefrag->ff_inum, freefrag->ff_blkno, freefrag->ff_fragsize);
/*
* It would be illegal to add new completion items to the
* freefrag after it was schedule to be done so it must be
@@ -5596,6 +5613,9 @@ softdep_setup_allocindir_page(ip, lbn, b
if (lbn != nbp->b_lblkno)
panic("softdep_setup_allocindir_page: lbn %jd != lblkno %jd",
lbn, bp->b_lblkno);
+ CTR4(KTR_SUJ,
+ "softdep_setup_allocindir_page: ino %d blkno %jd oldblkno %jd "
+ "lbn %jd", ip->i_number, newblkno, oldblkno, lbn);
ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_page");
mp = UFSTOVFS(ip->i_ump);
aip = newallocindir(ip, ptrno, newblkno, oldblkno, lbn);
@@ -5634,6 +5654,9 @@ softdep_setup_allocindir_meta(nbp, ip, b
ufs_lbn_t lbn;
int dflags;
+ CTR3(KTR_SUJ,
+ "softdep_setup_allocindir_meta: ino %d blkno %jd ptrno %d",
+ ip->i_number, newblkno, ptrno);
lbn = nbp->b_lblkno;
ASSERT_VOP_LOCKED(ITOV(ip), "softdep_setup_allocindir_meta");
aip = newallocindir(ip, ptrno, newblkno, 0, lbn);
@@ -6238,6 +6261,7 @@ softdep_journal_freeblocks(ip, cred, len
int flags; /* IO_EXT and/or IO_NORMAL */
{
struct freeblks *freeblks, *fbn;
+ struct worklist *wk, *wkn;
struct inodedep *inodedep;
struct jblkdep *jblkdep;
struct allocdirect *adp, *adpn;
@@ -6272,6 +6296,8 @@ softdep_journal_freeblocks(ip, cred, len
if ((inodedep->id_state & (UNLINKED | DEPCOMPLETE)) == UNLINKED &&
length == 0)
needj = 0;
+ CTR3(KTR_SUJ, "softdep_journal_freeblks: ip %d length %ld needj %d",
+ ip->i_number, length, needj);
FREE_LOCK(&lk);
/*
* Calculate the lbn that we are truncating to. This results in -1
@@ -6425,6 +6451,21 @@ softdep_journal_freeblocks(ip, cred, len
cancel_allocdirect(&inodedep->id_extupdt, adp,
freeblks);
/*
+ * Scan the bufwait list for newblock dependencies that will never
+ * make it to disk.
+ */
+ LIST_FOREACH_SAFE(wk, &inodedep->id_bufwait, wk_list, wkn) {
+ if (wk->wk_type != D_ALLOCDIRECT)
+ continue;
+ adp = WK_ALLOCDIRECT(wk);
+ if (((flags & IO_NORMAL) != 0 && (adp->ad_offset > iboff)) ||
+ ((flags & IO_EXT) != 0 && (adp->ad_state & EXTDATA))) {
+ cancel_jfreeblk(freeblks, adp->ad_newblkno);
+ cancel_newblk(WK_NEWBLK(wk), NULL, &freeblks->fb_jwork);
+ WORKLIST_INSERT(&freeblks->fb_freeworkhd, wk);
+ }
+ }
+ /*
* Add journal work.
*/
LIST_FOREACH(jblkdep, &freeblks->fb_jblkdephd, jb_deps)
@@ -6563,6 +6604,8 @@ softdep_setup_freeblocks(ip, length, fla
ufs_lbn_t tmpval;
ufs_lbn_t lbn;
+ CTR2(KTR_SUJ, "softdep_setup_freeblks: ip %d length %ld",
+ ip->i_number, length);
fs = ip->i_fs;
mp = UFSTOVFS(ip->i_ump);
if (length != 0)
@@ -7088,6 +7131,8 @@ cancel_newblk(newblk, wk, wkhd)
{
struct jnewblk *jnewblk;
+ CTR1(KTR_SUJ, "cancel_newblk: blkno %jd", newblk->nb_newblkno);
+
newblk->nb_state |= GOINGAWAY;
/*
* Previously we traversed the completedhd on each indirdep
@@ -7456,6 +7501,9 @@ freework_freeblock(freework)
}
FREE_LOCK(&lk);
freeblks_free(ump, freeblks, btodb(bsize));
+ CTR4(KTR_SUJ,
+ "freework_freeblock: ino %d blkno %jd lbn %jd size %ld",
+ freeblks->fb_inum, freework->fw_blkno, freework->fw_lbn, bsize);
ffs_blkfree(ump, fs, freeblks->fb_devvp, freework->fw_blkno, bsize,
freeblks->fb_inum, freeblks->fb_vtype, &wkhd);
ACQUIRE_LOCK(&lk);
@@ -7889,6 +7937,9 @@ indir_trunc(freework, dbn, lbn)
&freedep->fd_list);
freedeps++;
}
+ CTR3(KTR_SUJ,
+ "indir_trunc: ino %d blkno %jd size %ld",
+ freeblks->fb_inum, nb, fs->fs_bsize);
ffs_blkfree(ump, fs, freeblks->fb_devvp, nb,
fs->fs_bsize, freeblks->fb_inum,
freeblks->fb_vtype, &wkhd);
@@ -7924,6 +7975,9 @@ indir_trunc(freework, dbn, lbn)
* If we're not journaling we can free the indirect now.
*/
dbn = dbtofsb(fs, dbn);
+ CTR3(KTR_SUJ,
+ "indir_trunc 2: ino %d blkno %jd size %ld",
+ freeblks->fb_inum, dbn, fs->fs_bsize);
ffs_blkfree(ump, fs, freeblks->fb_devvp, dbn, fs->fs_bsize,
freeblks->fb_inum, freeblks->fb_vtype, NULL);
/* Non SUJ softdep does single-threaded truncations. */
@@ -10359,6 +10413,10 @@ softdep_setup_blkfree(mp, bp, blkno, fra
int i;
#endif
+ CTR3(KTR_SUJ,
+ "softdep_setup_blkfree: blkno %jd frags %d wk head %p",
+ blkno, frags, wkhd);
+
ACQUIRE_LOCK(&lk);
/* Lookup the bmsafemap so we track when it is dirty. */
fs = VFSTOUFS(mp)->um_fs;
@@ -10370,6 +10428,9 @@ softdep_setup_blkfree(mp, bp, blkno, fra
*/
if (wkhd) {
while ((wk = LIST_FIRST(wkhd)) != NULL) {
+ CTR2(KTR_SUJ,
+ "softdep_setup_blkfree: blkno %jd wk type %d",
+ blkno, wk->wk_type);
WORKLIST_REMOVE(wk);
if (wk->wk_type != D_JNEWBLK) {
WORKLIST_INSERT(&bmsafemap->sm_freehd, wk);
_______________________________________________
svn-src-stable-9@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/svn-src-stable-9
To unsubscribe, send any mail to "svn-src-stable-9-unsubscr...@freebsd.org"
