/* ext_depth() */
#include <ldiskfs/ldiskfs_extents.h>
+static inline bool osd_use_page_cache(struct osd_device *d)
+{
+ /* do not use pagecache if write and read caching are disabled */
+ if (d->od_writethrough_cache + d->od_read_cache == 0)
+ return false;
+ /* use pagecache by default */
+ return true;
+}
+
static int __osd_init_iobuf(struct osd_device *d, struct osd_iobuf *iobuf,
int rw, int line, int pages)
{
if (unlikely(iobuf->dr_pages == NULL))
return -ENOMEM;
+ lu_buf_realloc(&iobuf->dr_lnb_buf,
+ pages * sizeof(iobuf->dr_lnbs[0]));
+ iobuf->dr_lnbs = iobuf->dr_lnb_buf.lb_buf;
+ if (unlikely(iobuf->dr_lnbs == NULL))
+ return -ENOMEM;
+
iobuf->dr_max_pages = pages;
return 0;
#define osd_init_iobuf(dev, iobuf, rw, pages) \
__osd_init_iobuf(dev, iobuf, rw, __LINE__, pages)
-static void osd_iobuf_add_page(struct osd_iobuf *iobuf, struct page *page)
+static void osd_iobuf_add_page(struct osd_iobuf *iobuf,
+ struct niobuf_local *lnb)
{
- LASSERT(iobuf->dr_npages < iobuf->dr_max_pages);
- iobuf->dr_pages[iobuf->dr_npages++] = page;
+ LASSERT(iobuf->dr_npages < iobuf->dr_max_pages);
+ iobuf->dr_pages[iobuf->dr_npages] = lnb->lnb_page;
+ iobuf->dr_lnbs[iobuf->dr_npages] = lnb;
+ iobuf->dr_npages++;
}
void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf)
CERROR("***** bio->bi_private is NULL! This should never "
"happen. Normally, I would crash here, but instead I "
"will dump the bio contents to the console. Please "
- "report this to <https://jira.hpdd.intel.com/> , along "
+ "report this to <https://jira.whamcloud.com/> , along "
"with any interesting messages leading up to this point "
"(like SCSI errors, perhaps). Because bi_private is "
"NULL, I can't wake up the thread that initiated this "
}
}
+static int bio_dif_compare(__u16 *expected_guard_buf, void *bio_prot_buf,
+ unsigned int sectors, int tuple_size)
+{
+ __u16 *expected_guard;
+ __u16 *bio_guard;
+ int i;
+
+ expected_guard = expected_guard_buf;
+ for (i = 0; i < sectors; i++) {
+ bio_guard = (__u16 *)bio_prot_buf;
+ if (*bio_guard != *expected_guard) {
+ CERROR("unexpected guard tags on sector %d "
+ "expected guard %u, bio guard "
+ "%u, sectors %u, tuple size %d\n",
+ i, *expected_guard, *bio_guard, sectors,
+ tuple_size);
+ return -EIO;
+ }
+ expected_guard++;
+ bio_prot_buf += tuple_size;
+ }
+ return 0;
+}
+
+static int osd_bio_integrity_compare(struct bio *bio, struct block_device *bdev,
+ struct osd_iobuf *iobuf, int index)
+{
+ struct blk_integrity *bi = bdev_get_integrity(bdev);
+ struct bio_integrity_payload *bip = bio->bi_integrity;
+ struct niobuf_local *lnb;
+ unsigned short sector_size = blk_integrity_interval(bi);
+ void *bio_prot_buf = page_address(bip->bip_vec->bv_page) +
+ bip->bip_vec->bv_offset;
+ struct bio_vec *bv;
+ sector_t sector = bio_start_sector(bio);
+ unsigned int i, sectors, total;
+ __u16 *expected_guard;
+ int rc;
+
+ total = 0;
+ bio_for_each_segment_all(bv, bio, i) {
+ lnb = iobuf->dr_lnbs[index];
+ expected_guard = lnb->lnb_guards;
+ sectors = bv->bv_len / sector_size;
+ if (lnb->lnb_guard_rpc) {
+ rc = bio_dif_compare(expected_guard, bio_prot_buf,
+ sectors, bi->tuple_size);
+ if (rc)
+ return rc;
+ }
+
+ sector += sectors;
+ bio_prot_buf += sectors * bi->tuple_size;
+ total += sectors * bi->tuple_size;
+ LASSERT(total <= bip_size(bio->bi_integrity));
+ index++;
+ }
+ return 0;
+}
+
+static int osd_bio_integrity_handle(struct osd_device *osd, struct bio *bio,
+ struct osd_iobuf *iobuf,
+ int start_page_idx, bool fault_inject,
+ bool integrity_enabled)
+{
+ struct super_block *sb = osd_sb(osd);
+ int rc;
+#ifdef HAVE_BIO_INTEGRITY_PREP_FN
+ integrity_gen_fn *generate_fn = NULL;
+ integrity_vrfy_fn *verify_fn = NULL;
+#endif
+
+ ENTRY;
+
+ if (!integrity_enabled)
+ RETURN(0);
+
+#ifdef HAVE_BIO_INTEGRITY_PREP_FN
+ rc = osd_get_integrity_profile(osd, &generate_fn, &verify_fn);
+ if (rc)
+ RETURN(rc);
+
+ rc = bio_integrity_prep_fn(bio, generate_fn, verify_fn);
+#else
+ rc = bio_integrity_prep(bio);
+#endif
+ if (rc)
+ RETURN(rc);
+
+ /* Verify and inject fault only when writing */
+ if (iobuf->dr_rw == 1) {
+ if (unlikely(OBD_FAIL_CHECK(OBD_FAIL_OST_INTEGRITY_CMP))) {
+ rc = osd_bio_integrity_compare(bio, sb->s_bdev, iobuf,
+ start_page_idx);
+ if (rc)
+ RETURN(rc);
+ }
+
+ if (unlikely(fault_inject))
+ bio_integrity_fault_inject(bio);
+ }
+
+ RETURN(0);
+}
+
+#ifdef HAVE_BIO_INTEGRITY_PREP_FN
+# ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
+static void dio_integrity_complete_routine(struct bio *bio)
+{
+# else
+static void dio_integrity_complete_routine(struct bio *bio, int error)
+{
+# endif
+ struct osd_bio_private *bio_private = bio->bi_private;
+
+ bio->bi_private = bio_private->obp_iobuf;
+# ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
+ dio_complete_routine(bio);
+# else
+ dio_complete_routine(bio, error);
+# endif
+
+ OBD_FREE_PTR(bio_private);
+}
+#endif
+
+static int osd_bio_init(struct bio *bio, struct osd_iobuf *iobuf,
+ bool integrity_enabled, int start_page_idx,
+ struct osd_bio_private **pprivate)
+{
+#ifdef HAVE_BIO_INTEGRITY_PREP_FN
+ struct osd_bio_private *bio_private;
+
+ ENTRY;
+
+ *pprivate = NULL;
+ if (integrity_enabled) {
+ OBD_ALLOC_GFP(bio_private, sizeof(*bio_private), GFP_NOIO);
+ if (bio_private == NULL)
+ RETURN(-ENOMEM);
+ bio->bi_end_io = dio_integrity_complete_routine;
+ bio->bi_private = bio_private;
+ bio_private->obp_start_page_idx = start_page_idx;
+ bio_private->obp_iobuf = iobuf;
+ *pprivate = bio_private;
+ } else {
+ bio->bi_end_io = dio_complete_routine;
+ bio->bi_private = iobuf;
+ }
+ RETURN(0);
+#else
+ ENTRY;
+
+ bio->bi_end_io = dio_complete_routine;
+ bio->bi_private = iobuf;
+ RETURN(0);
+#endif
+}
+
static int osd_do_bio(struct osd_device *osd, struct inode *inode,
struct osd_iobuf *iobuf)
{
int npages = iobuf->dr_npages;
sector_t *blocks = iobuf->dr_blocks;
int total_blocks = npages * blocks_per_page;
- int sector_bits = inode->i_sb->s_blocksize_bits - 9;
- unsigned int blocksize = inode->i_sb->s_blocksize;
+ struct super_block *sb = inode->i_sb;
+ int sector_bits = sb->s_blocksize_bits - 9;
+ unsigned int blocksize = sb->s_blocksize;
+ struct block_device *bdev = sb->s_bdev;
+ struct osd_bio_private *bio_private = NULL;
struct bio *bio = NULL;
+ int bio_start_page_idx;
struct page *page;
unsigned int page_offset;
sector_t sector;
int i;
int rc = 0;
bool fault_inject;
+ bool integrity_enabled;
DECLARE_PLUG(plug);
ENTRY;
fault_inject = OBD_FAIL_CHECK(OBD_FAIL_OST_INTEGRITY_FAULT);
LASSERT(iobuf->dr_npages == npages);
+ integrity_enabled = bdev_integrity_enabled(bdev, iobuf->dr_rw);
+
osd_brw_stats_update(osd, iobuf);
iobuf->dr_start_time = ktime_get();
bio_phys_segments(q, bio),
queue_max_phys_segments(q),
0, queue_max_hw_segments(q));
- if (bio_integrity_enabled(bio)) {
- if (bio_integrity_prep(bio)) {
- bio_put(bio);
- rc = -EIO;
- goto out;
- }
- if (unlikely(fault_inject))
- bio_integrity_fault_inject(bio);
+ rc = osd_bio_integrity_handle(osd, bio,
+ iobuf, bio_start_page_idx,
+ fault_inject, integrity_enabled);
+ if (rc) {
+ bio_put(bio);
+ goto out;
}
record_start_io(iobuf, bi_size);
osd_submit_bio(iobuf->dr_rw, bio);
}
+ bio_start_page_idx = page_idx;
/* allocate new bio */
bio = bio_alloc(GFP_NOIO, min(BIO_MAX_PAGES,
(npages - page_idx) *
goto out;
}
- bio_set_dev(bio, inode->i_sb->s_bdev);
+ bio_set_dev(bio, bdev);
bio_set_sector(bio, sector);
#ifdef HAVE_BI_RW
bio->bi_rw = (iobuf->dr_rw == 0) ? READ : WRITE;
#else
bio->bi_opf = (iobuf->dr_rw == 0) ? READ : WRITE;
#endif
- bio->bi_end_io = dio_complete_routine;
- bio->bi_private = iobuf;
+ rc = osd_bio_init(bio, iobuf, integrity_enabled,
+ bio_start_page_idx, &bio_private);
+ if (rc) {
+ bio_put(bio);
+ goto out;
+ }
rc = bio_add_page(bio, page,
blocksize * nblocks, page_offset);
}
if (bio != NULL) {
- if (bio_integrity_enabled(bio)) {
- if (bio_integrity_prep(bio)) {
- bio_put(bio);
- rc = -EIO;
- goto out;
- }
- if (unlikely(fault_inject))
- bio_integrity_fault_inject(bio);
+ rc = osd_bio_integrity_handle(osd, bio, iobuf,
+ bio_start_page_idx,
+ fault_inject,
+ integrity_enabled);
+ if (rc) {
+ bio_put(bio);
+ goto out;
}
record_start_io(iobuf, bio_sectors(bio) << 9);
osd_fini_iobuf(osd, iobuf);
}
- if (rc == 0)
+ if (rc == 0) {
rc = iobuf->dr_error;
+ } else {
+ if (bio_private)
+ OBD_FREE_PTR(bio_private);
+ }
+
RETURN(rc);
}
lnb->lnb_flags = 0;
lnb->lnb_page = NULL;
lnb->lnb_rc = 0;
+ lnb->lnb_guard_rpc = 0;
+ lnb->lnb_guard_disk = 0;
LASSERTF(plen <= len, "plen %u, len %lld\n", plen,
(long long) len);
RETURN(0);
}
-static struct page *osd_get_page(struct dt_object *dt, loff_t offset,
- gfp_t gfp_mask)
+static struct page *osd_get_page(const struct lu_env *env, struct dt_object *dt,
+ loff_t offset, gfp_t gfp_mask)
{
+ struct osd_thread_info *oti = osd_oti_get(env);
struct inode *inode = osd_dt_obj(dt)->oo_inode;
struct osd_device *d = osd_obj2dev(osd_dt_obj(dt));
struct page *page;
+ int cur = oti->oti_dio_pages_used;
LASSERT(inode);
- page = find_or_create_page(inode->i_mapping, offset >> PAGE_SHIFT,
- gfp_mask);
+ if (osd_use_page_cache(d)) {
+ page = find_or_create_page(inode->i_mapping,
+ offset >> PAGE_SHIFT,
+ gfp_mask);
+
+ if (likely(page))
+ LASSERT(!test_bit(PG_private_2, &page->flags));
+ else
+ lprocfs_counter_add(d->od_stats, LPROC_OSD_NO_PAGE, 1);
+ } else {
+
+ LASSERT(oti->oti_dio_pages);
- if (unlikely(page == NULL))
- lprocfs_counter_add(d->od_stats, LPROC_OSD_NO_PAGE, 1);
+ if (unlikely(!oti->oti_dio_pages[cur])) {
+ LASSERT(cur < PTLRPC_MAX_BRW_PAGES);
+ page = alloc_page(gfp_mask);
+ if (!page)
+ return NULL;
+ oti->oti_dio_pages[cur] = page;
+ }
+
+ page = oti->oti_dio_pages[cur];
+ LASSERT(!test_bit(PG_private_2, &page->flags));
+ set_bit(PG_private_2, &page->flags);
+ oti->oti_dio_pages_used++;
+
+ LASSERT(!PageLocked(page));
+ lock_page(page);
+
+ LASSERT(!page->mapping);
+ LASSERT(!PageWriteback(page));
+ ClearPageUptodate(page);
+
+ page->index = offset >> PAGE_SHIFT;
+ }
- return page;
+ return page;
}
/*
static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt,
struct niobuf_local *lnb, int npages)
{
+ struct osd_thread_info *oti = osd_oti_get(env);
struct pagevec pvec;
int i;
#endif
for (i = 0; i < npages; i++) {
- if (lnb[i].lnb_page == NULL)
+ struct page *page = lnb[i].lnb_page;
+
+ if (page == NULL)
continue;
- LASSERT(PageLocked(lnb[i].lnb_page));
- unlock_page(lnb[i].lnb_page);
- if (pagevec_add(&pvec, lnb[i].lnb_page) == 0)
- pagevec_release(&pvec);
+ LASSERT(PageLocked(page));
+
+ /* if the page isn't cached, then reset uptodate
+ * to prevent reuse */
+ if (test_bit(PG_private_2, &page->flags)) {
+ clear_bit(PG_private_2, &page->flags);
+ ClearPageUptodate(page);
+ unlock_page(page);
+ oti->oti_dio_pages_used--;
+ } else {
+ unlock_page(page);
+ if (pagevec_add(&pvec, page) == 0)
+ pagevec_release(&pvec);
+ }
dt_object_put(env, dt);
+
lnb[i].lnb_page = NULL;
}
+ LASSERTF(oti->oti_dio_pages_used == 0, "%d\n", oti->oti_dio_pages_used);
+
/* Release any partial pagevec */
pagevec_release(&pvec);
loff_t pos, ssize_t len, struct niobuf_local *lnb,
enum dt_bufs_type rw)
{
+ struct osd_thread_info *oti = osd_oti_get(env);
struct osd_object *obj = osd_dt_obj(dt);
int npages, i, rc = 0;
gfp_t gfp_mask;
LASSERT(obj->oo_inode);
+ if (!osd_use_page_cache(osd_obj2dev(obj))) {
+ if (unlikely(!oti->oti_dio_pages)) {
+ OBD_ALLOC(oti->oti_dio_pages,
+ sizeof(struct page *) * PTLRPC_MAX_BRW_PAGES);
+ if (!oti->oti_dio_pages)
+ return -ENOMEM;
+ }
+ }
+
osd_map_remote_to_local(pos, len, &npages, lnb);
/* this could also try less hard for DT_BUFS_TYPE_READAHEAD pages */
gfp_mask = rw & DT_BUFS_TYPE_LOCAL ? (GFP_NOFS | __GFP_HIGHMEM) :
GFP_HIGHUSER;
for (i = 0; i < npages; i++, lnb++) {
- lnb->lnb_page = osd_get_page(dt, lnb->lnb_file_offset,
+ lnb->lnb_page = osd_get_page(env, dt, lnb->lnb_file_offset,
gfp_mask);
if (lnb->lnb_page == NULL)
GOTO(cleanup, rc = -ENOMEM);
continue;
if (maxidx >= lnb[i].lnb_page->index) {
- osd_iobuf_add_page(iobuf, lnb[i].lnb_page);
+ osd_iobuf_add_page(iobuf, &lnb[i]);
} else {
long off;
char *p = kmap(lnb[i].lnb_page);
SetPageUptodate(lnb[i].lnb_page);
- osd_iobuf_add_page(iobuf, lnb[i].lnb_page);
+ osd_iobuf_add_page(iobuf, &lnb[i]);
}
osd_trans_exec_op(env, thandle, OSD_OT_WRITE);
cache_hits++;
} else {
cache_misses++;
- osd_iobuf_add_page(iobuf, lnb[i].lnb_page);
+ osd_iobuf_add_page(iobuf, &lnb[i]);
}
if (cache == 0)
static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt,
const struct lu_buf *buf, loff_t *pos,
- struct thandle *handle, int ignore_quota)
+ struct thandle *handle)
{
struct inode *inode = osd_dt_obj(dt)->oo_inode;
struct osd_thandle *oh;
void osd_execute_truncate(struct osd_object *obj)
{
+ struct osd_device *d = osd_obj2dev(obj);
struct inode *inode = obj->oo_inode;
__u64 size;
* avoid data corruption during direct disk write. b=17397
*/
size = i_size_read(inode);
- if ((size & ~PAGE_MASK) != 0)
+ if ((size & ~PAGE_MASK) == 0)
+ return;
+ if (osd_use_page_cache(d)) {
filemap_fdatawrite_range(inode->i_mapping, size, size + 1);
+ } else {
+ /* Notice we use "wait" version to ensure I/O is complete */
+ filemap_write_and_wait_range(inode->i_mapping, size, size + 1);
+ invalidate_mapping_pages(inode->i_mapping, size >> PAGE_SHIFT,
+ size >> PAGE_SHIFT);
+ }
}
void osd_process_truncates(struct list_head *list)
git://git.whamcloud.com / fs / lustre-release.git / blobdiff