exofs: convert io_state to use pages array instead of bio at input
* inode.c operations are full-pages based, and not actually
true scatter-gather
* Lets us use more pages at once upto 512 (from 249) in 64 bit
* Brings us much much closer to be able to use exofs's io_state engine
from objlayout driver. (Once I decide where to put the common code)
After RAID0 patch the outer (input) bio was never used as a bio, but
was simply a page carrier into the raid engine. Even in the simple
mirror/single-dev arrangement pages info was copied into a second bio.
It is now easer to just pass a pages array into the io_state and prepare
bio(s) once.
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
diff --git a/fs/exofs/exofs.h b/fs/exofs/exofs.h
index 0d8a34b..acfebd3 100644
--- a/fs/exofs/exofs.h
+++ b/fs/exofs/exofs.h
@@ -128,7 +128,10 @@
loff_t offset;
unsigned long length;
void *kern_buff;
- struct bio *bio;
+
+ struct page **pages;
+ unsigned nr_pages;
+ unsigned pgbase;
/* Attributes */
unsigned in_attr_len;
diff --git a/fs/exofs/inode.c b/fs/exofs/inode.c
index 2b3163e..6ca0b01 100644
--- a/fs/exofs/inode.c
+++ b/fs/exofs/inode.c
@@ -41,16 +41,18 @@
enum { BIO_MAX_PAGES_KMALLOC =
(PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
+ MAX_PAGES_KMALLOC =
+ PAGE_SIZE / sizeof(struct page *),
};
struct page_collect {
struct exofs_sb_info *sbi;
- struct request_queue *req_q;
struct inode *inode;
unsigned expected_pages;
struct exofs_io_state *ios;
- struct bio *bio;
+ struct page **pages;
+ unsigned alloc_pages;
unsigned nr_pages;
unsigned long length;
loff_t pg_first; /* keep 64bit also in 32-arches */
@@ -62,15 +64,12 @@
struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
pcol->sbi = sbi;
- /* Create master bios on first Q, later on cloning, each clone will be
- * allocated on it's destination Q
- */
- pcol->req_q = osd_request_queue(sbi->layout.s_ods[0]);
pcol->inode = inode;
pcol->expected_pages = expected_pages;
pcol->ios = NULL;
- pcol->bio = NULL;
+ pcol->pages = NULL;
+ pcol->alloc_pages = 0;
pcol->nr_pages = 0;
pcol->length = 0;
pcol->pg_first = -1;
@@ -80,7 +79,8 @@
{
pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
- pcol->bio = NULL;
+ pcol->pages = NULL;
+ pcol->alloc_pages = 0;
pcol->nr_pages = 0;
pcol->length = 0;
pcol->pg_first = -1;
@@ -90,13 +90,13 @@
* it might not end here. don't be left with nothing
*/
if (!pcol->expected_pages)
- pcol->expected_pages = BIO_MAX_PAGES_KMALLOC;
+ pcol->expected_pages = MAX_PAGES_KMALLOC;
}
static int pcol_try_alloc(struct page_collect *pcol)
{
- int pages = min_t(unsigned, pcol->expected_pages,
- BIO_MAX_PAGES_KMALLOC);
+ unsigned pages = min_t(unsigned, pcol->expected_pages,
+ MAX_PAGES_KMALLOC);
if (!pcol->ios) { /* First time allocate io_state */
int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
@@ -105,23 +105,28 @@
return ret;
}
+ /* TODO: easily support bio chaining */
+ pages = min_t(unsigned, pages,
+ pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
+
for (; pages; pages >>= 1) {
- pcol->bio = bio_kmalloc(GFP_KERNEL, pages);
- if (likely(pcol->bio))
+ pcol->pages = kmalloc(pages * sizeof(struct page *),
+ GFP_KERNEL);
+ if (likely(pcol->pages)) {
+ pcol->alloc_pages = pages;
return 0;
+ }
}
- EXOFS_ERR("Failed to bio_kmalloc expected_pages=%u\n",
+ EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
pcol->expected_pages);
return -ENOMEM;
}
static void pcol_free(struct page_collect *pcol)
{
- if (pcol->bio) {
- bio_put(pcol->bio);
- pcol->bio = NULL;
- }
+ kfree(pcol->pages);
+ pcol->pages = NULL;
if (pcol->ios) {
exofs_put_io_state(pcol->ios);
@@ -132,11 +137,10 @@
static int pcol_add_page(struct page_collect *pcol, struct page *page,
unsigned len)
{
- int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0);
- if (unlikely(len != added_len))
+ if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
return -ENOMEM;
- ++pcol->nr_pages;
+ pcol->pages[pcol->nr_pages++] = page;
pcol->length += len;
return 0;
}
@@ -181,7 +185,6 @@
*/
static int __readpages_done(struct page_collect *pcol, bool do_unlock)
{
- struct bio_vec *bvec;
int i;
u64 resid;
u64 good_bytes;
@@ -198,8 +201,8 @@
pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
pcol->nr_pages);
- __bio_for_each_segment(bvec, pcol->bio, i, 0) {
- struct page *page = bvec->bv_page;
+ for (i = 0; i < pcol->nr_pages; i++) {
+ struct page *page = pcol->pages[i];
struct inode *inode = page->mapping->host;
int page_stat;
@@ -218,7 +221,7 @@
ret = update_read_page(page, page_stat);
if (do_unlock)
unlock_page(page);
- length += bvec->bv_len;
+ length += PAGE_SIZE;
}
pcol_free(pcol);
@@ -238,11 +241,10 @@
static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
{
- struct bio_vec *bvec;
int i;
- __bio_for_each_segment(bvec, pcol->bio, i, 0) {
- struct page *page = bvec->bv_page;
+ for (i = 0; i < pcol->nr_pages; i++) {
+ struct page *page = pcol->pages[i];
if (rw == READ)
update_read_page(page, ret);
@@ -260,13 +262,14 @@
struct page_collect *pcol_copy = NULL;
int ret;
- if (!pcol->bio)
+ if (!pcol->pages)
return 0;
/* see comment in _readpage() about sync reads */
WARN_ON(is_sync && (pcol->nr_pages != 1));
- ios->bio = pcol->bio;
+ ios->pages = pcol->pages;
+ ios->nr_pages = pcol->nr_pages;
ios->length = pcol->length;
ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
@@ -366,7 +369,7 @@
goto try_again;
}
- if (!pcol->bio) {
+ if (!pcol->pages) {
ret = pcol_try_alloc(pcol);
if (unlikely(ret))
goto fail;
@@ -448,7 +451,6 @@
static void writepages_done(struct exofs_io_state *ios, void *p)
{
struct page_collect *pcol = p;
- struct bio_vec *bvec;
int i;
u64 resid;
u64 good_bytes;
@@ -467,8 +469,8 @@
pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
pcol->nr_pages);
- __bio_for_each_segment(bvec, pcol->bio, i, 0) {
- struct page *page = bvec->bv_page;
+ for (i = 0; i < pcol->nr_pages; i++) {
+ struct page *page = pcol->pages[i];
struct inode *inode = page->mapping->host;
int page_stat;
@@ -485,7 +487,7 @@
EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
inode->i_ino, page->index, page_stat);
- length += bvec->bv_len;
+ length += PAGE_SIZE;
}
pcol_free(pcol);
@@ -500,7 +502,7 @@
struct page_collect *pcol_copy = NULL;
int ret;
- if (!pcol->bio)
+ if (!pcol->pages)
return 0;
pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
@@ -512,9 +514,8 @@
*pcol_copy = *pcol;
- pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */
-
- ios->bio = pcol_copy->bio;
+ ios->pages = pcol_copy->pages;
+ ios->nr_pages = pcol_copy->nr_pages;
ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
ios->length = pcol_copy->length;
ios->done = writepages_done;
@@ -605,7 +606,7 @@
goto try_again;
}
- if (!pcol->bio) {
+ if (!pcol->pages) {
ret = pcol_try_alloc(pcol);
if (unlikely(ret))
goto fail;
diff --git a/fs/exofs/ios.c b/fs/exofs/ios.c
index 6e446b2..263052c 100644
--- a/fs/exofs/ios.c
+++ b/fs/exofs/ios.c
@@ -283,10 +283,11 @@
*dev = stripe_mod / stripe_unit * ios->layout->mirrors_p1;
}
-static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_bvec,
- struct exofs_per_dev_state *per_dev, int cur_len)
+static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
+ unsigned pgbase, struct exofs_per_dev_state *per_dev,
+ int cur_len)
{
- unsigned bv = *cur_bvec;
+ unsigned pg = *cur_pg;
struct request_queue *q =
osd_request_queue(exofs_ios_od(ios, per_dev->dev));
@@ -295,7 +296,7 @@
if (per_dev->bio == NULL) {
unsigned pages_in_stripe = ios->layout->group_width *
(ios->layout->stripe_unit / PAGE_SIZE);
- unsigned bio_size = (ios->bio->bi_vcnt + pages_in_stripe) /
+ unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
ios->layout->group_width;
per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
@@ -307,21 +308,22 @@
}
while (cur_len > 0) {
- int added_len;
- struct bio_vec *bvec = &ios->bio->bi_io_vec[bv];
+ unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
+ unsigned added_len;
- BUG_ON(ios->bio->bi_vcnt <= bv);
- cur_len -= bvec->bv_len;
+ BUG_ON(ios->nr_pages <= pg);
+ cur_len -= pglen;
- added_len = bio_add_pc_page(q, per_dev->bio, bvec->bv_page,
- bvec->bv_len, bvec->bv_offset);
- if (unlikely(bvec->bv_len != added_len))
+ added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
+ pglen, pgbase);
+ if (unlikely(pglen != added_len))
return -ENOMEM;
- ++bv;
+ pgbase = 0;
+ ++pg;
}
BUG_ON(cur_len);
- *cur_bvec = bv;
+ *cur_pg = pg;
return 0;
}
@@ -332,10 +334,10 @@
unsigned stripe_unit = ios->layout->stripe_unit;
unsigned comp = 0;
unsigned stripes = 0;
- unsigned cur_bvec = 0;
- int ret;
+ unsigned cur_pg = 0;
+ int ret = 0;
- if (!ios->bio) {
+ if (!ios->pages) {
if (ios->kern_buff) {
struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
unsigned unit_off;
@@ -352,7 +354,7 @@
while (length) {
struct exofs_per_dev_state *per_dev = &ios->per_dev[comp];
- unsigned cur_len;
+ unsigned cur_len, page_off;
if (!per_dev->length) {
unsigned unit_off;
@@ -362,11 +364,15 @@
stripes++;
cur_len = min_t(u64, stripe_unit - unit_off, length);
offset += cur_len;
+ page_off = unit_off & ~PAGE_MASK;
+ BUG_ON(page_off != ios->pgbase);
} else {
cur_len = min_t(u64, stripe_unit, length);
+ page_off = 0;
}
- ret = _add_stripe_unit(ios, &cur_bvec, per_dev, cur_len);
+ ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
+ cur_len);
if (unlikely(ret))
goto out;
@@ -448,7 +454,7 @@
per_dev->or = or;
per_dev->offset = master_dev->offset;
- if (ios->bio) {
+ if (ios->pages) {
struct bio *bio;
if (per_dev != master_dev) {
@@ -541,7 +547,7 @@
}
per_dev->or = or;
- if (ios->bio) {
+ if (ios->pages) {
osd_req_read(or, &ios->obj, per_dev->offset,
per_dev->bio, per_dev->length);
EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
