Date: Mon, 14 Dec 2009 14:43:24 -0800 From: Matt Reimer <mattjreimer@gmail.com> To: freebsd-fs <freebsd-fs@freebsd.org> Cc: Pawel Jakub Dawidek <pjd@freebsd.org> Subject: PATCH: more efficient raidz memory usage for (gpt)zfsboot Message-ID: <f383264b0912141443k13d1df48jf5a93c6007f2f29b@mail.gmail.com>
next in thread | raw e-mail | index | archive | help
[-- Attachment #1 --]
Teach the (gpt)zfsboot and zfsloader raidz code to use its buffers
more efficiently.
Before this patch, in the worst case memory use would increase
exponentially on the number of drives in the raidz vdev.
Sponsored by: VPOP Technologies, Inc.
Matt Reimer
[-- Attachment #2 --]
--- /sys/cddl/boot/zfs/zfssubr.c.ORIG 2009-11-14 08:14:51.000000000 -0800
+++ /sys/cddl/boot/zfs/zfssubr.c 2009-12-07 15:27:49.000000000 -0800
@@ -454,7 +454,7 @@
static void
vdev_raidz_reconstruct_pq(raidz_col_t *cols, int nparity, int acols,
- int x, int y)
+ int x, int y, void *temp_p, void *temp_q)
{
uint8_t *p, *q, *pxy, *qxy, *xd, *yd, tmp, a, b, aexp, bexp;
void *pdata, *qdata;
@@ -478,10 +478,8 @@
xsize = cols[x].rc_size;
ysize = cols[y].rc_size;
- cols[VDEV_RAIDZ_P].rc_data =
- zfs_alloc_temp(cols[VDEV_RAIDZ_P].rc_size);
- cols[VDEV_RAIDZ_Q].rc_data =
- zfs_alloc_temp(cols[VDEV_RAIDZ_Q].rc_size);
+ cols[VDEV_RAIDZ_P].rc_data = temp_p;
+ cols[VDEV_RAIDZ_Q].rc_data = temp_q;
cols[x].rc_size = 0;
cols[y].rc_size = 0;
@@ -551,9 +549,12 @@
uint64_t f = b % dcols;
uint64_t o = (b / dcols) << unit_shift;
uint64_t q, r, coff;
- int c, c1, bc, col, acols, devidx, asize, n;
+ int c, c1, bc, col, acols, devidx, asize, n, max_rc_size;
static raidz_col_t cols[16];
raidz_col_t *rc, *rc1;
+ void *orig, *orig1, *temp_p, *temp_q;
+
+ orig = orig1 = temp_p = temp_q = NULL;
q = s / (dcols - nparity);
r = s - q * (dcols - nparity);
@@ -561,6 +562,7 @@
acols = (q == 0 ? bc : dcols);
asize = 0;
+ max_rc_size = 0;
for (c = 0; c < acols; c++) {
col = f + c;
@@ -577,6 +579,8 @@
cols[c].rc_tried = 0;
cols[c].rc_skipped = 0;
asize += cols[c].rc_size;
+ if (cols[c].rc_size > max_rc_size)
+ max_rc_size = cols[c].rc_size;
}
asize = roundup(asize, (nparity + 1) << unit_shift);
@@ -777,8 +781,13 @@
//ASSERT(c != acols);
//ASSERT(!rc->rc_skipped || rc->rc_error == ENXIO || rc->rc_error == ESTALE);
+ if (!temp_p)
+ temp_p = zfs_alloc_temp(max_rc_size);
+ if (!temp_q)
+ temp_q = zfs_alloc_temp(max_rc_size);
+
vdev_raidz_reconstruct_pq(cols, nparity, acols,
- c1, c);
+ c1, c, temp_p, temp_q);
if (zio_checksum_error(bp, buf) == 0)
return (0);
@@ -845,18 +854,12 @@
return (EIO);
}
- asize = 0;
- for (c = 0; c < acols; c++) {
- rc = &cols[c];
- if (rc->rc_size > asize)
- asize = rc->rc_size;
- }
if (cols[VDEV_RAIDZ_P].rc_error == 0) {
/*
* Attempt to reconstruct the data from parity P.
*/
- void *orig;
- orig = zfs_alloc_temp(asize);
+ if (!orig)
+ orig = zfs_alloc_temp(max_rc_size);
for (c = nparity; c < acols; c++) {
rc = &cols[c];
@@ -874,8 +877,8 @@
/*
* Attempt to reconstruct the data from parity Q.
*/
- void *orig;
- orig = zfs_alloc_temp(asize);
+ if (!orig)
+ orig = zfs_alloc_temp(max_rc_size);
for (c = nparity; c < acols; c++) {
rc = &cols[c];
@@ -895,9 +898,14 @@
/*
* Attempt to reconstruct the data from both P and Q.
*/
- void *orig, *orig1;
- orig = zfs_alloc_temp(asize);
- orig1 = zfs_alloc_temp(asize);
+ if (!orig)
+ orig = zfs_alloc_temp(max_rc_size);
+ if (!orig1)
+ orig1 = zfs_alloc_temp(max_rc_size);
+ if (!temp_p)
+ temp_p = zfs_alloc_temp(max_rc_size);
+ if (!temp_q)
+ temp_q = zfs_alloc_temp(max_rc_size);
for (c = nparity; c < acols - 1; c++) {
rc = &cols[c];
@@ -909,7 +917,7 @@
memcpy(orig1, rc1->rc_data, rc1->rc_size);
vdev_raidz_reconstruct_pq(cols, nparity,
- acols, c, c1);
+ acols, c, c1, temp_p, temp_q);
if (zio_checksum_error(bp, buf) == 0)
return (0);
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?f383264b0912141443k13d1df48jf5a93c6007f2f29b>