On Wed, Jan 03, 2018 at 04:48:01PM +0800, Eryu Guan wrote:
> On Thu, Dec 14, 2017 at 06:07:31PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
> >
> > Mix it up a bit by reflinking and deduping data blocks when possible.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx>
>
> This looks fine overall, but I noticed a soft lockup bug in generic/083
> and generic/269 (both test exercise ENOSPC behavior), test config is
> reflink+rmapbt XFS with 4k block size. Not sure if the soft lockup is
> related to the clonerange/deduperange ops in fsstress yet, will confirm
> without clone/dedupe ops.
>
> [12968.100008] watchdog: BUG: soft lockup - CPU#2 stuck for 22s! [fsstress:6903]
> [12968.100038] Modules linked in: loop dm_flakey xfs ip6t_rpfilter ip6t_REJECT nf_reject_ipv6 xt_conntrack ip_set nfnetlink ebtable_nat ebtable_broute bridge stp llc ip6table_nat nf_conntrack_ipv6 nf_defrag_ipv6 nf_nat_ipv6 ip6table_mangle ip6table_raw ip6table_security iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack libcrc32c iptable_mangle iptable_raw iptable_security ebtable_filter ebtables ip6table_filter ip6_tables sunrpc 8139too 8139cp i2c_piix4 joydev mii pcspkr virtio_balloon virtio_pci serio_raw virtio_ring virtio floppy ata_generic pata_acpi
> [12968.104043] irq event stamp: 23222196
> [12968.104043] hardirqs last enabled at (23222195): [<000000007d0c2e75>] restore_regs_and_return_to_kernel+0x0/0x2e
> [12968.105111] hardirqs last disabled at (23222196): [<000000008f80dc57>] apic_timer_interrupt+0xa7/0xc0
> [12968.105111] softirqs last enabled at (877594): [<0000000034c53d5e>] __do_softirq+0x392/0x502
> [12968.105111] softirqs last disabled at (877585): [<000000003f4d9e0b>] irq_exit+0x102/0x110
> [12968.105111] CPU: 2 PID: 6903 Comm: fsstress Tainted: G W L 4.15.0-rc5 #10
> [12968.105111] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2007
> [12968.108043] RIP: 0010:xfs_bmapi_update_map+0xc/0xc0 [xfs]
Hmmm, I haven't seen such a hang; I wonder if we're doing something
we shouldn't be doing and looping in bmapi_write. In any case it's
a bug with xfs, not fsstress.
--D
> [12968.108043] RSP: 0018:ffffb8cbc2b8ba88 EFLAGS: 00000246 ORIG_RAX: ffffffffffffff11
> [12968.109028] RAX: ffffb8cbc2b8bc50 RBX: 0000000000000a40 RCX: 000000000000012b
> [12968.109111] RDX: ffffb8cbc2b8bb00 RSI: ffffb8cbc2b8bb08 RDI: ffffb8cbc2b8baf8
> [12968.109111] RBP: ffffb8cbc2b8bc10 R08: 000000000000012c R09: ffffb8cbc2b8bb14
> [12968.109111] R10: 0000000000000000 R11: 0000000000000000 R12: ffffb8cbc2b8bb28
> [12968.109111] R13: ffffb8cbc2b8bb68 R14: 000000000000012c R15: 0000000000000001
> [12968.109111] FS: 00007fed71507b80(0000) GS:ffff98f457200000(0000) knlGS:0000000000000000
> [12968.112047] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [12968.112047] CR2: 00007fed71503000 CR3: 000000020f50d000 CR4: 00000000000006e0
> [12968.113049] Call Trace:
> [12968.113049] xfs_bmapi_write+0x33e/0xcc0 [xfs]
> [12968.113049] xfs_reflink_convert_cow+0x8c/0xc0 [xfs]
> [12968.113049] ? xfs_vm_writepages+0x54/0xd0 [xfs]
> [12968.113049] xfs_submit_ioend+0x18f/0x1f0 [xfs]
> [12968.113049] xfs_vm_writepages+0xc5/0xd0 [xfs]
> [12968.113049] do_writepages+0x48/0xf0
> [12968.113049] ? __filemap_fdatawrite_range+0xb4/0x100
> [12968.116073] ? __filemap_fdatawrite_range+0xc1/0x100
> [12968.116073] __filemap_fdatawrite_range+0xc1/0x100
> [12968.116073] xfs_release+0x11c/0x160 [xfs]
> [12968.117049] __fput+0xe6/0x1f0
> [12968.117049] task_work_run+0x82/0xb0
> [12968.117049] exit_to_usermode_loop+0xa8/0xb0
> [12968.117049] syscall_return_slowpath+0x153/0x160
> [12968.117049] entry_SYSCALL_64_fastpath+0x94/0x96
> [12968.117049] RIP: 0033:0x7fed70cddcb1
> [12968.117049] RSP: 002b:00007ffd8d566118 EFLAGS: 00000246 ORIG_RAX: 0000000000000003
> [12968.117049] RAX: 0000000000000000 RBX: 00000000000002da RCX: 00007fed70cddcb1
> [12968.117049] RDX: 0000000000c1f440 RSI: 0000000000c1e010 RDI: 0000000000000003
> [12968.120048] RBP: 0000000000000003 R08: 0000000000000006 R09: 00007ffd8d56612c
> [12968.120048] R10: 0000000000000000 R11: 0000000000000246 R12: 000000000012bd3b
> [12968.121048] R13: 00000000004073c0 R14: 0000000000000000 R15: 0000000000000000
>
> > ---
> > v2: don't disable broken commands, just ignore them
> > ---
> > ltp/fsstress.c | 391 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 391 insertions(+)
> >
> > diff --git a/ltp/fsstress.c b/ltp/fsstress.c
> > index 96f48b1..b02cb0c 100644
> > --- a/ltp/fsstress.c
> > +++ b/ltp/fsstress.c
> > @@ -68,7 +68,9 @@ typedef enum {
> > OP_BULKSTAT,
> > OP_BULKSTAT1,
> > OP_CHOWN,
> > + OP_CLONERANGE,
> > OP_CREAT,
> > + OP_DEDUPERANGE,
> > OP_DREAD,
> > OP_DWRITE,
> > OP_FALLOCATE,
> > @@ -174,7 +176,9 @@ void awrite_f(int, long);
> > void bulkstat_f(int, long);
> > void bulkstat1_f(int, long);
> > void chown_f(int, long);
> > +void clonerange_f(int, long);
> > void creat_f(int, long);
> > +void deduperange_f(int, long);
> > void dread_f(int, long);
> > void dwrite_f(int, long);
> > void fallocate_f(int, long);
> > @@ -221,7 +225,9 @@ opdesc_t ops[] = {
> > { OP_BULKSTAT, "bulkstat", bulkstat_f, 1, 0 },
> > { OP_BULKSTAT1, "bulkstat1", bulkstat1_f, 1, 0 },
> > { OP_CHOWN, "chown", chown_f, 3, 1 },
> > + { OP_CLONERANGE, "clonerange", clonerange_f, 4, 1 },
> > { OP_CREAT, "creat", creat_f, 4, 1 },
> > + { OP_DEDUPERANGE, "deduperange", deduperange_f, 4, 1},
> > { OP_DREAD, "dread", dread_f, 4, 0 },
> > { OP_DWRITE, "dwrite", dwrite_f, 4, 1 },
> > { OP_FALLOCATE, "fallocate", fallocate_f, 1, 1 },
> > @@ -2189,6 +2195,391 @@ chown_f(int opno, long r)
> > free_pathname(&f);
> > }
> >
> > +/* reflink some arbitrary range of f1 to f2. */
> > +void
> > +clonerange_f(
> > + int opno,
> > + long r)
> > +{
> > +#ifdef FICLONERANGE
> > + struct file_clone_range fcr;
> > + struct pathname fpath1;
> > + struct pathname fpath2;
> > + struct stat64 stat1;
> > + struct stat64 stat2;
> > + char inoinfo1[1024];
> > + char inoinfo2[1024];
> > + off64_t lr;
> > + off64_t off1;
> > + off64_t off2;
> > + size_t len;
> > + int v1;
> > + int v2;
> > + int fd1;
> > + int fd2;
> > + int ret;
> > + int e;
> > +
> > + /* Load paths */
> > + init_pathname(&fpath1);
> > + if (!get_fname(FT_REGm, r, &fpath1, NULL, NULL, &v1)) {
> > + if (v1)
> > + printf("%d/%d: clonerange read - no filename\n",
> > + procid, opno);
> > + goto out_fpath1;
> > + }
> > +
> > + init_pathname(&fpath2);
> > + if (!get_fname(FT_REGm, random(), &fpath2, NULL, NULL, &v2)) {
> > + if (v2)
> > + printf("%d/%d: clonerange write - no filename\n",
> > + procid, opno);
> > + goto out_fpath2;
> > + }
> > +
> > + /* Open files */
> > + fd1 = open_path(&fpath1, O_RDONLY);
> > + e = fd1 < 0 ? errno : 0;
> > + check_cwd();
> > + if (fd1 < 0) {
> > + if (v1)
> > + printf("%d/%d: clonerange read - open %s failed %d\n",
> > + procid, opno, fpath1.path, e);
> > + goto out_fpath2;
> > + }
> > +
> > + fd2 = open_path(&fpath2, O_WRONLY);
> > + e = fd2 < 0 ? errno : 0;
> > + check_cwd();
> > + if (fd2 < 0) {
> > + if (v2)
> > + printf("%d/%d: clonerange write - open %s failed %d\n",
> > + procid, opno, fpath2.path, e);
> > + goto out_fd1;
> > + }
> > +
> > + /* Get file stats */
> > + if (fstat64(fd1, &stat1) < 0) {
> > + if (v1)
> > + printf("%d/%d: clonerange read - fstat64 %s failed %d\n",
> > + procid, opno, fpath1.path, errno);
> > + goto out_fd2;
> > + }
> > + inode_info(inoinfo1, sizeof(inoinfo1), &stat1, v1);
> > +
> > + if (fstat64(fd2, &stat2) < 0) {
> > + if (v2)
> > + printf("%d/%d: clonerange write - fstat64 %s failed %d\n",
> > + procid, opno, fpath2.path, errno);
> > + goto out_fd2;
> > + }
> > + inode_info(inoinfo2, sizeof(inoinfo2), &stat2, v1);
> ^^^^ should be v2?
> > +
> > + /* Calculate offsets */
> > + len = (random() % FILELEN_MAX) + 1;
> > + len &= ~(stat1.st_blksize - 1);
> > + if (len == 0)
> > + len = stat1.st_blksize;
> > + if (len > stat1.st_size)
> > + len = stat1.st_size;
> > +
> > + lr = ((__int64_t)random() << 32) + random();
> > + if (stat1.st_size == len)
> > + off1 = 0;
> > + else
> > + off1 = (off64_t)(lr % MIN(stat1.st_size - len, MAXFSIZE));
> > + off1 %= maxfsize;
> > + off1 &= ~(stat1.st_blksize - 1);
>
> Seems that the offset and len are not required to be block size aligned,
> mind adding some comments on the consideration on offset and len, in
> both clonerange and deduperange cases?
>
> Thanks,
> Eryu
>
> > +
> > + /*
> > + * If srcfile == destfile, randomly generate destination ranges
> > + * until we find one that doesn't overlap the source range.
> > + */
> > + do {
> > + lr = ((__int64_t)random() << 32) + random();
> > + off2 = (off64_t)(lr % MIN(stat2.st_size + (1024 * 1024), MAXFSIZE));
> > + off2 %= maxfsize;
> > + off2 &= ~(stat2.st_blksize - 1);
> > + } while (stat1.st_ino == stat2.st_ino && llabs(off2 - off1) < len);
> > +
> > + /* Clone data blocks */
> > + fcr.src_fd = fd1;
> > + fcr.src_offset = off1;
> > + fcr.src_length = len;
> > + fcr.dest_offset = off2;
> > +
> > + ret = ioctl(fd2, FICLONERANGE, &fcr);
> > + e = ret < 0 ? errno : 0;
> > + if (v1 || v2) {
> > + printf("%d/%d: clonerange %s%s [%lld,%lld] -> %s%s [%lld,%lld]",
> > + procid, opno,
> > + fpath1.path, inoinfo1, (long long)off1, (long long)len,
> > + fpath2.path, inoinfo2, (long long)off2, (long long)len);
> > +
> > + if (ret < 0)
> > + printf(" error %d", e);
> > + printf("\n");
> > + }
> > +
> > +out_fd2:
> > + close(fd2);
> > +out_fd1:
> > + close(fd1);
> > +out_fpath2:
> > + free_pathname(&fpath2);
> > +out_fpath1:
> > + free_pathname(&fpath1);
> > +#endif
> > +}
> > +
> > +/* dedupe some arbitrary range of f1 to f2...fn. */
> > +void
> > +deduperange_f(
> > + int opno,
> > + long r)
> > +{
> > +#ifdef FIDEDUPERANGE
> > +#define INFO_SZ 1024
> > + struct file_dedupe_range *fdr;
> > + struct pathname *fpath;
> > + struct stat64 *stat;
> > + char *info;
> > + off64_t *off;
> > + int *v;
> > + int *fd;
> > + int nr;
> > + off64_t lr;
> > + size_t len;
> > + int ret;
> > + int i;
> > + int e;
> > +
> > + if (flist[FT_REG].nfiles < 2)
> > + return;
> > +
> > + /* Pick somewhere between 2 and 128 files. */
> > + do {
> > + nr = random() % (flist[FT_REG].nfiles + 1);
> > + } while (nr < 2 || nr > 128);
> > +
> > + /* Alloc memory */
> > + fdr = malloc(nr * sizeof(struct file_dedupe_range_info) +
> > + sizeof(struct file_dedupe_range));
> > + if (!fdr) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + return;
> > + }
> > + memset(fdr, 0, (nr * sizeof(struct file_dedupe_range_info) +
> > + sizeof(struct file_dedupe_range)));
> > +
> > + fpath = calloc(nr, sizeof(struct pathname));
> > + if (!fpath) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_fdr;
> > + }
> > +
> > + stat = calloc(nr, sizeof(struct stat64));
> > + if (!stat) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_paths;
> > + }
> > +
> > + info = calloc(nr, INFO_SZ);
> > + if (!info) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_stats;
> > + }
> > +
> > + off = calloc(nr, sizeof(off64_t));
> > + if (!off) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_info;
> > + }
> > +
> > + v = calloc(nr, sizeof(int));
> > + if (!v) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_offsets;
> > + }
> > + fd = calloc(nr, sizeof(int));
> > + if (!fd) {
> > + printf("%d/%d: line %d error %d\n",
> > + procid, opno, __LINE__, errno);
> > + goto out_v;
> > + }
> > + memset(fd, 0xFF, nr * sizeof(int));
> > +
> > + /* Get paths for all files */
> > + for (i = 0; i < nr; i++)
> > + init_pathname(&fpath[i]);
> > +
> > + if (!get_fname(FT_REGm, r, &fpath[0], NULL, NULL, &v[0])) {
> > + if (v[0])
> > + printf("%d/%d: deduperange read - no filename\n",
> > + procid, opno);
> > + goto out_pathnames;
> > + }
> > +
> > + for (i = 1; i < nr; i++) {
> > + if (!get_fname(FT_REGm, random(), &fpath[i], NULL, NULL, &v[i])) {
> > + if (v[i])
> > + printf("%d/%d: deduperange write - no filename\n",
> > + procid, opno);
> > + goto out_pathnames;
> > + }
> > + }
> > +
> > + /* Open files */
> > + fd[0] = open_path(&fpath[0], O_RDONLY);
> > + e = fd[0] < 0 ? errno : 0;
> > + check_cwd();
> > + if (fd[0] < 0) {
> > + if (v[0])
> > + printf("%d/%d: deduperange read - open %s failed %d\n",
> > + procid, opno, fpath[0].path, e);
> > + goto out_pathnames;
> > + }
> > +
> > + for (i = 1; i < nr; i++) {
> > + fd[i] = open_path(&fpath[i], O_WRONLY);
> > + e = fd[i] < 0 ? errno : 0;
> > + check_cwd();
> > + if (fd[i] < 0) {
> > + if (v[i])
> > + printf("%d/%d: deduperange write - open %s failed %d\n",
> > + procid, opno, fpath[i].path, e);
> > + goto out_fds;
> > + }
> > + }
> > +
> > + /* Get file stats */
> > + if (fstat64(fd[0], &stat[0]) < 0) {
> > + if (v[0])
> > + printf("%d/%d: deduperange read - fstat64 %s failed %d\n",
> > + procid, opno, fpath[0].path, errno);
> > + goto out_fds;
> > + }
> > +
> > + inode_info(&info[0], INFO_SZ, &stat[0], v[0]);
> > +
> > + for (i = 1; i < nr; i++) {
> > + if (fstat64(fd[i], &stat[i]) < 0) {
> > + if (v[i])
> > + printf("%d/%d: deduperange write - fstat64 %s failed %d\n",
> > + procid, opno, fpath[i].path, errno);
> > + goto out_fds;
> > + }
> > + inode_info(&info[i * INFO_SZ], INFO_SZ, &stat[i], v[i]);
> > + }
> > +
> > + /* Never try to dedupe more than half of the src file. */
> > + len = (random() % FILELEN_MAX) + 1;
> > + len &= ~(stat[0].st_blksize - 1);
> > + if (len == 0)
> > + len = stat[0].st_blksize / 2;
> > + if (len > stat[0].st_size / 2)
> > + len = stat[0].st_size / 2;
> > +
> > + /* Calculate offsets */
> > + lr = ((__int64_t)random() << 32) + random();
> > + if (stat[0].st_size == len)
> > + off[0] = 0;
> > + else
> > + off[0] = (off64_t)(lr % MIN(stat[0].st_size - len, MAXFSIZE));
> > + off[0] %= maxfsize;
> > + off[0] &= ~(stat[0].st_blksize - 1);
> > +
> > + /*
> > + * If srcfile == destfile[i], randomly generate destination ranges
> > + * until we find one that doesn't overlap the source range.
> > + */
> > + for (i = 1; i < nr; i++) {
> > + int tries = 0;
> > +
> > + do {
> > + lr = ((__int64_t)random() << 32) + random();
> > + if (stat[i].st_size <= len)
> > + off[i] = 0;
> > + else
> > + off[i] = (off64_t)(lr % MIN(stat[i].st_size - len, MAXFSIZE));
> > + off[i] %= maxfsize;
> > + off[i] &= ~(stat[i].st_blksize - 1);
> > + } while (stat[0].st_ino == stat[i].st_ino &&
> > + llabs(off[i] - off[0]) < len &&
> > + tries++ < 10);
> > + }
> > +
> > + /* Clone data blocks */
> > + fdr->src_offset = off[0];
> > + fdr->src_length = len;
> > + fdr->dest_count = nr - 1;
> > + for (i = 1; i < nr; i++) {
> > + fdr->info[i - 1].dest_fd = fd[i];
> > + fdr->info[i - 1].dest_offset = off[i];
> > + }
> > +
> > + ret = ioctl(fd[0], FIDEDUPERANGE, fdr);
> > + e = ret < 0 ? errno : 0;
> > + if (v[0]) {
> > + printf("%d/%d: deduperange from %s%s [%lld,%lld]",
> > + procid, opno,
> > + fpath[0].path, &info[0], (long long)off[0],
> > + (long long)len);
> > + if (ret < 0)
> > + printf(" error %d", e);
> > + printf("\n");
> > + }
> > + if (ret < 0)
> > + goto out_fds;
> > +
> > + for (i = 1; i < nr; i++) {
> > + e = fdr->info[i - 1].status < 0 ? fdr->info[i - 1].status : 0;
> > + if (v[i]) {
> > + printf("%d/%d: ...to %s%s [%lld,%lld]",
> > + procid, opno,
> > + fpath[i].path, &info[i * INFO_SZ],
> > + (long long)off[i], (long long)len);
> > + if (fdr->info[i - 1].status < 0)
> > + printf(" error %d", e);
> > + if (fdr->info[i - 1].status == FILE_DEDUPE_RANGE_SAME)
> > + printf(" %llu bytes deduplicated",
> > + fdr->info[i - 1].bytes_deduped);
> > + if (fdr->info[i - 1].status == FILE_DEDUPE_RANGE_DIFFERS)
> > + printf(" differed");
> > + printf("\n");
> > + }
> > + }
> > +
> > +out_fds:
> > + for (i = 0; i < nr; i++)
> > + if (fd[i] >= 0)
> > + close(fd[i]);
> > +out_pathnames:
> > + for (i = 0; i < nr; i++)
> > + free_pathname(&fpath[i]);
> > +
> > + free(fd);
> > +out_v:
> > + free(v);
> > +out_offsets:
> > + free(off);
> > +out_info:
> > + free(info);
> > +out_stats:
> > + free(stat);
> > +out_paths:
> > + free(fpath);
> > +out_fdr:
> > + free(fdr);
> > +#endif
> > +}
> > +
> > void
> > setxattr_f(int opno, long r)
> > {
> --
> To unsubscribe from this list: send the line "unsubscribe linux-xfs" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe fstests" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
