X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fosd-ldiskfs%2Fosd_io.c;h=30190178ace92a87e4db6713fb4c96eefa599a88;hp=390582bad102446f8e0c0d222b2683d7287cf2cb;hb=605a139a193e4ef212560673b4bb93eec763d088;hpb=61d4357fa8c0c6371e3ab66a9d38aaa6cc7bf4d8 diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c index 390582b..3019017 100644 --- a/lustre/osd-ldiskfs/osd_io.c +++ b/lustre/osd-ldiskfs/osd_io.c @@ -15,11 +15,7 @@ * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see - * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf - * - * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, - * CA 95054 USA or visit www.sun.com if you need additional information or - * have any questions. + * http://www.gnu.org/licenses/gpl-2.0.html * * GPL HEADER END */ @@ -27,7 +23,7 @@ * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * - * Copyright (c) 2012, 2014, Intel Corporation. + * Copyright (c) 2012, 2017, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -42,12 +38,12 @@ * */ -/* LUSTRE_VERSION_CODE */ -#include /* prerequisite for linux/xattr.h */ #include /* prerequisite for linux/xattr.h */ #include +#include +#include /* * struct OBD_{ALLOC,FREE}*() @@ -77,12 +73,12 @@ static int __osd_init_iobuf(struct osd_device *d, struct osd_iobuf *iobuf, iobuf->dr_error = 0; iobuf->dr_dev = d; iobuf->dr_frags = 0; - iobuf->dr_elapsed = 0; + iobuf->dr_elapsed = ktime_set(0, 0); /* must be counted before, so assert */ iobuf->dr_rw = rw; iobuf->dr_init_at = line; - blocks = pages * (PAGE_CACHE_SIZE >> osd_sb(d)->s_blocksize_bits); + blocks = pages * (PAGE_SIZE >> osd_sb(d)->s_blocksize_bits); if (iobuf->dr_bl_buf.lb_len >= blocks * sizeof(iobuf->dr_blocks[0])) { LASSERT(iobuf->dr_pg_buf.lb_len >= pages * sizeof(iobuf->dr_pages[0])); @@ -97,7 +93,7 @@ static int __osd_init_iobuf(struct osd_device *d, struct osd_iobuf *iobuf, CDEBUG(D_OTHER, "realloc %u for %u (%u) pages\n", (unsigned)(pages * sizeof(iobuf->dr_pages[0])), i, pages); pages = i; - blocks = pages * (PAGE_CACHE_SIZE >> osd_sb(d)->s_blocksize_bits); + blocks = pages * (PAGE_SIZE >> osd_sb(d)->s_blocksize_bits); iobuf->dr_max_pages = 0; CDEBUG(D_OTHER, "realloc %u for %u blocks\n", (unsigned)(blocks * sizeof(iobuf->dr_blocks[0])), blocks); @@ -136,25 +132,26 @@ void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf) lprocfs_oh_tally(&d->od_brw_stats. hist[BRW_R_DIO_FRAGS+rw], iobuf->dr_frags); - lprocfs_oh_tally_log2(&d->od_brw_stats.hist[BRW_R_IO_TIME+rw], - iobuf->dr_elapsed); + lprocfs_oh_tally_log2(&d->od_brw_stats.hist[BRW_R_IO_TIME+rw], + ktime_to_ms(iobuf->dr_elapsed)); } } -#ifndef REQ_WRITE /* pre-2.6.35 */ -#define __REQ_WRITE BIO_RW -#endif - +#ifdef HAVE_BIO_ENDIO_USES_ONE_ARG +static void dio_complete_routine(struct bio *bio) +{ +# ifdef HAVE_BI_STATUS + int error = bio->bi_status; +# else + int error = bio->bi_error; +# endif +#else static void dio_complete_routine(struct bio *bio, int error) { +#endif struct osd_iobuf *iobuf = bio->bi_private; -#ifdef HAVE_BVEC_ITER - struct bvec_iter iter; - struct bio_vec bvl; -#else int iter; struct bio_vec *bvl; -#endif /* CAVEAT EMPTOR: possibly in IRQ context * DO NOT record procfs stats here!!! */ @@ -163,26 +160,42 @@ static void dio_complete_routine(struct bio *bio, int error) 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 , along " + "report this to , 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 " "IO - you will probably have to reboot this node.\n"); - CERROR("bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d, " - "bi_idx: %d, bi->size: %d, bi_end_io: %p, bi_cnt: %d, " - "bi_private: %p\n", bio->bi_next, bio->bi_flags, - bio->bi_rw, bio->bi_vcnt, bio_idx(bio), + CERROR("bi_next: %p, bi_flags: %lx, " +#ifdef HAVE_BI_RW + "bi_rw: %lu," +#else + "bi_opf: %u," +#endif + "bi_vcnt: %d, bi_idx: %d, bi->size: %d, bi_end_io: %p," + "bi_cnt: %d, bi_private: %p\n", bio->bi_next, + (unsigned long)bio->bi_flags, +#ifdef HAVE_BI_RW + bio->bi_rw, +#else + bio->bi_opf, +#endif + bio->bi_vcnt, bio_idx(bio), bio_sectors(bio) << 9, bio->bi_end_io, - atomic_read(&bio->bi_cnt), bio->bi_private); +#ifdef HAVE_BI_CNT + atomic_read(&bio->bi_cnt), +#else + atomic_read(&bio->__bi_cnt), +#endif + bio->bi_private); return; } /* the check is outside of the cycle for performance reason -bzzz */ - if (!test_bit(__REQ_WRITE, &bio->bi_rw)) { - bio_for_each_segment(bvl, bio, iter) { + if (!bio_data_dir(bio)) { + bio_for_each_segment_all(bvl, bio, iter) { if (likely(error == 0)) - SetPageUptodate(bvec_iter_page(&bvl, iter)); - LASSERT(PageLocked(bvec_iter_page(&bvl, iter))); + SetPageUptodate(bvl_to_page(bvl)); + LASSERT(PageLocked(bvl_to_page(bvl))); } atomic_dec(&iobuf->dr_dev->od_r_in_flight); } else { @@ -201,7 +214,9 @@ static void dio_complete_routine(struct bio *bio, int error) * call to OSD. */ if (atomic_read(&iobuf->dr_numreqs) == 1) { - iobuf->dr_elapsed = jiffies - iobuf->dr_start_time; + ktime_t now = ktime_get(); + + iobuf->dr_elapsed = ktime_sub(now, iobuf->dr_start_time); iobuf->dr_elapsed_valid = 1; } if (atomic_dec_and_test(&iobuf->dr_numreqs)) @@ -241,10 +256,15 @@ static void record_start_io(struct osd_iobuf *iobuf, int size) static void osd_submit_bio(int rw, struct bio *bio) { LASSERTF(rw == 0 || rw == 1, "%x\n", rw); +#ifdef HAVE_SUBMIT_BIO_2ARGS if (rw == 0) submit_bio(READ, bio); else submit_bio(WRITE, bio); +#else + bio->bi_opf |= rw; + submit_bio(bio); +#endif } static int can_be_merged(struct bio *bio, sector_t sector) @@ -255,32 +275,58 @@ static int can_be_merged(struct bio *bio, sector_t sector) return bio_end_sector(bio) == sector ? 1 : 0; } +/* + * This function will change the data written, thus it should only be + * used when checking data integrity feature + */ +static void bio_integrity_fault_inject(struct bio *bio) +{ + struct bio_vec *bvec; + int i; + void *kaddr; + char *addr; + + bio_for_each_segment_all(bvec, bio, i) { + struct page *page = bvec->bv_page; + + kaddr = kmap(page); + addr = kaddr; + *addr = ~(*addr); + kunmap(page); + break; + } +} + static int osd_do_bio(struct osd_device *osd, struct inode *inode, struct osd_iobuf *iobuf) { - int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits; - struct page **pages = iobuf->dr_pages; - int npages = iobuf->dr_npages; - unsigned long *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 bio *bio = NULL; - struct page *page; - unsigned int page_offset; - sector_t sector; - int nblocks; - int block_idx; - int page_idx; - int i; - int rc = 0; - ENTRY; + int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; + struct page **pages = iobuf->dr_pages; + 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 bio *bio = NULL; + struct page *page; + unsigned int page_offset; + sector_t sector; + int nblocks; + int block_idx; + int page_idx; + int i; + int rc = 0; + bool fault_inject; + DECLARE_PLUG(plug); + ENTRY; + fault_inject = OBD_FAIL_CHECK(OBD_FAIL_OST_INTEGRITY_FAULT); LASSERT(iobuf->dr_npages == npages); - osd_brw_stats_update(osd, iobuf); - iobuf->dr_start_time = cfs_time_current(); + osd_brw_stats_update(osd, iobuf); + iobuf->dr_start_time = ktime_get(); + blk_start_plug(&plug); for (page_idx = 0, block_idx = 0; page_idx < npages; page_idx++, block_idx += blocks_per_page) { @@ -319,8 +365,7 @@ static int osd_do_bio(struct osd_device *osd, struct inode *inode, continue; /* added this frag OK */ if (bio != NULL) { - struct request_queue *q = - bdev_get_queue(bio->bi_bdev); + struct request_queue *q = bio_get_queue(bio); unsigned int bi_size = bio_sectors(bio) << 9; /* Dang! I have to fragment this I/O */ @@ -332,6 +377,16 @@ static int osd_do_bio(struct osd_device *osd, struct inode *inode, 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); + } + record_start_io(iobuf, bi_size); osd_submit_bio(iobuf->dr_rw, bio); } @@ -348,9 +403,13 @@ static int osd_do_bio(struct osd_device *osd, struct inode *inode, goto out; } - bio->bi_bdev = inode->i_sb->s_bdev; + bio_set_dev(bio, inode->i_sb->s_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; @@ -361,17 +420,29 @@ static int osd_do_bio(struct osd_device *osd, struct inode *inode, } 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); + } + record_start_io(iobuf, bio_sectors(bio) << 9); osd_submit_bio(iobuf->dr_rw, bio); rc = 0; } out: + blk_finish_plug(&plug); + /* in order to achieve better IO throughput, we don't wait for writes * completion here. instead we proceed with transaction commit in * parallel and wait for IO completion once transaction is stopped * see osd_trans_stop() for more details -bzzz */ - if (iobuf->dr_rw == 0) { + if (iobuf->dr_rw == 0 || fault_inject) { wait_event(iobuf->dr_wait, atomic_read(&iobuf->dr_numreqs) == 0); osd_fini_iobuf(osd, iobuf); @@ -390,8 +461,8 @@ static int osd_map_remote_to_local(loff_t offset, ssize_t len, int *nrpages, *nrpages = 0; while (len > 0) { - int poff = offset & (PAGE_CACHE_SIZE - 1); - int plen = PAGE_CACHE_SIZE - poff; + int poff = offset & (PAGE_SIZE - 1); + int plen = PAGE_SIZE - poff; if (plen > len) plen = len; @@ -414,16 +485,18 @@ static int osd_map_remote_to_local(loff_t offset, ssize_t len, int *nrpages, RETURN(0); } -static struct page *osd_get_page(struct dt_object *dt, loff_t offset, int rw) +static struct page *osd_get_page(struct dt_object *dt, loff_t offset, + gfp_t gfp_mask) { - struct inode *inode = osd_dt_obj(dt)->oo_inode; - struct osd_device *d = osd_obj2dev(osd_dt_obj(dt)); - struct page *page; + struct inode *inode = osd_dt_obj(dt)->oo_inode; + struct osd_device *d = osd_obj2dev(osd_dt_obj(dt)); + struct page *page; LASSERT(inode); - page = find_or_create_page(inode->i_mapping, offset >> PAGE_CACHE_SHIFT, - GFP_NOFS | __GFP_HIGHMEM); + page = find_or_create_page(inode->i_mapping, offset >> PAGE_SHIFT, + gfp_mask); + if (unlikely(page == NULL)) lprocfs_counter_add(d->od_stats, LPROC_OSD_NO_PAGE, 1); @@ -464,18 +537,29 @@ static struct page *osd_get_page(struct dt_object *dt, loff_t offset, int rw) static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt, struct niobuf_local *lnb, int npages) { + struct pagevec pvec; int i; +#ifdef HAVE_PAGEVEC_INIT_ONE_PARAM + pagevec_init(&pvec); +#else + pagevec_init(&pvec, 0); +#endif + for (i = 0; i < npages; i++) { if (lnb[i].lnb_page == NULL) continue; LASSERT(PageLocked(lnb[i].lnb_page)); unlock_page(lnb[i].lnb_page); - page_cache_release(lnb[i].lnb_page); - lu_object_put(env, &dt->do_lu); + if (pagevec_add(&pvec, lnb[i].lnb_page) == 0) + pagevec_release(&pvec); + dt_object_put(env, dt); lnb[i].lnb_page = NULL; } + /* Release any partial pagevec */ + pagevec_release(&pvec); + RETURN(0); } @@ -497,7 +581,7 @@ static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt, * \param pos byte offset of IO start * \param len number of bytes of IO * \param lnb array of extents undergoing IO - * \param rw read or write operation? + * \param rw read or write operation, and other flags * \param capa capabilities * * \retval pages (zero or more) loaded successfully @@ -505,17 +589,22 @@ static int osd_bufs_put(const struct lu_env *env, struct dt_object *dt, */ static int osd_bufs_get(const struct lu_env *env, struct dt_object *dt, loff_t pos, ssize_t len, struct niobuf_local *lnb, - int rw, struct lustre_capa *capa) + enum dt_bufs_type rw) { - struct osd_object *obj = osd_dt_obj(dt); + struct osd_object *obj = osd_dt_obj(dt); int npages, i, rc = 0; + gfp_t gfp_mask; LASSERT(obj->oo_inode); 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, rw); + lnb->lnb_page = osd_get_page(dt, lnb->lnb_file_offset, + gfp_mask); if (lnb->lnb_page == NULL) GOTO(cleanup, rc = -ENOMEM); @@ -540,7 +629,7 @@ cleanup: #endif struct bpointers { - unsigned long *blocks; + sector_t *blocks; unsigned long start; int num; int init_num; @@ -619,7 +708,7 @@ static int ldiskfs_ext_new_extent_cb(struct inode *inode, { struct bpointers *bp = cbdata; struct ldiskfs_extent nex; - unsigned long pblock; + unsigned long pblock = 0; unsigned long tgen; int err, i; unsigned long count; @@ -745,15 +834,16 @@ map: i, cex->ec_len); for (; i < cex->ec_len && bp->num; i++) { *(bp->blocks) = cex->ec_start + i; -#ifdef LDISKFS_EXT_CACHE_EXTENT /* until kernel 2.6.37 */ - if (cex->ec_type != LDISKFS_EXT_CACHE_EXTENT) { -#else - if ((cex->ec_len == 0) || (cex->ec_start == 0)) { -#endif + if (pblock != 0) { /* unmap any possible underlying metadata from * the block device mapping. bug 6998. */ +#ifndef HAVE_CLEAN_BDEV_ALIASES unmap_underlying_metadata(inode->i_sb->s_bdev, *(bp->blocks)); +#else + clean_bdev_aliases(inode->i_sb->s_bdev, + *(bp->blocks), 1); +#endif } bp->blocks++; bp->num--; @@ -763,23 +853,26 @@ map: return err; } -static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long block, - unsigned long num, unsigned long *blocks, - int create) +static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long index, + int clen, sector_t *blocks, int create) { + int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; struct bpointers bp; int err; - CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n", - block, block + num - 1, (unsigned) inode->i_ino); + if (index + clen >= inode->i_sb->s_maxbytes >> PAGE_SHIFT) + return -EFBIG; bp.blocks = blocks; - bp.start = block; - bp.init_num = bp.num = num; + bp.start = index * blocks_per_page; + bp.init_num = bp.num = clen * blocks_per_page; bp.create = create; - err = ldiskfs_ext_walk_space(inode, block, num, - ldiskfs_ext_new_extent_cb, &bp); + CDEBUG(D_OTHER, "blocks %lu-%lu requested for inode %u\n", + bp.start, bp.start + bp.num - 1, (unsigned)inode->i_ino); + + err = ldiskfs_ext_walk_space(inode, bp.start, bp.num, + ldiskfs_ext_new_extent_cb, &bp); ldiskfs_ext_invalidate_cache(inode); return err; @@ -787,11 +880,11 @@ static int osd_ldiskfs_map_nblocks(struct inode *inode, unsigned long block, static int osd_ldiskfs_map_bm_inode_pages(struct inode *inode, struct page **page, int pages, - unsigned long *blocks, int create) + sector_t *blocks, int create) { - int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits; + int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; pgoff_t bitmap_max_page_index; - unsigned long *b; + sector_t *b; int rc = 0, i; bitmap_max_page_index = LDISKFS_SB(inode->i_sb)->s_bitmap_maxbytes >> @@ -803,8 +896,9 @@ static int osd_ldiskfs_map_bm_inode_pages(struct inode *inode, } rc = ldiskfs_map_inode_page(inode, *page, b, create); if (rc) { - CERROR("ino %lu, blk %lu create %d: rc %d\n", - inode->i_ino, *b, create, rc); + CERROR("ino %lu, blk %llu create %d: rc %d\n", + inode->i_ino, + (unsigned long long)*b, create, rc); break; } b += blocks_per_page; @@ -814,16 +908,11 @@ static int osd_ldiskfs_map_bm_inode_pages(struct inode *inode, static int osd_ldiskfs_map_ext_inode_pages(struct inode *inode, struct page **page, - int pages, unsigned long *blocks, + int pages, sector_t *blocks, int create) { - int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits; - int rc = 0, i = 0; + int rc = 0, i = 0, clen = 0; struct page *fp = NULL; - int clen = 0; - pgoff_t extent_max_page_index; - - extent_max_page_index = inode->i_sb->s_maxbytes >> PAGE_SHIFT; CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n", inode->i_ino, pages, (*page)->index); @@ -845,31 +934,27 @@ static int osd_ldiskfs_map_ext_inode_pages(struct inode *inode, continue; } - if (fp->index + i >= extent_max_page_index) - GOTO(cleanup, rc = -EFBIG); - /* process found extent */ - rc = osd_ldiskfs_map_nblocks(inode, fp->index * blocks_per_page, - clen * blocks_per_page, blocks, - create); + rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen, + blocks, create); if (rc) GOTO(cleanup, rc); /* look for next extent */ fp = NULL; - blocks += blocks_per_page * clen; + blocks += clen * (PAGE_SIZE >> inode->i_blkbits); } if (fp) - rc = osd_ldiskfs_map_nblocks(inode, fp->index * blocks_per_page, - clen * blocks_per_page, blocks, - create); + rc = osd_ldiskfs_map_nblocks(inode, fp->index, clen, + blocks, create); + cleanup: return rc; } static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page, - int pages, unsigned long *blocks, + int pages, sector_t *blocks, int create) { int rc; @@ -885,25 +970,33 @@ static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page, } #else static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page, - int pages, unsigned long *blocks, + int pages, sector_t *blocks, int create) { - int blocks_per_page = PAGE_CACHE_SIZE >> inode->i_blkbits; + int blocks_per_page = PAGE_SIZE >> inode->i_blkbits; int rc = 0, i = 0; struct page *fp = NULL; int clen = 0; pgoff_t max_page_index; + handle_t *handle = NULL; max_page_index = inode->i_sb->s_maxbytes >> PAGE_SHIFT; CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n", inode->i_ino, pages, (*page)->index); + if (create) { + create = LDISKFS_GET_BLOCKS_CREATE; + handle = ldiskfs_journal_current_handle(); + LASSERT(handle != NULL); + rc = osd_attach_jinode(inode); + if (rc) + return rc; + } /* pages are sorted already. so, we just have to find * contig. space and process them properly */ while (i < pages) { long blen, total = 0; - handle_t *handle = NULL; struct ldiskfs_map_blocks map = { 0 }; if (fp == NULL) { /* start new extent */ @@ -918,16 +1011,11 @@ static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page, if (++i != pages) continue; } - if (fp->index + i >= max_page_index) + if (fp->index + clen >= max_page_index) GOTO(cleanup, rc = -EFBIG); /* process found extent */ map.m_lblk = fp->index * blocks_per_page; map.m_len = blen = clen * blocks_per_page; - if (create) { - create = LDISKFS_GET_BLOCKS_CREATE; - handle = ldiskfs_journal_current_handle(); - LASSERT(handle != NULL); - } cont_map: rc = ldiskfs_map_blocks(handle, inode, &map, create); if (rc >= 0) { @@ -944,9 +1032,15 @@ cont_map: * mapping. bug 6998. */ if ((map.m_flags & LDISKFS_MAP_NEW) && create) +#ifndef HAVE_CLEAN_BDEV_ALIASES unmap_underlying_metadata( inode->i_sb->s_bdev, map.m_pblk + c); +#else + clean_bdev_aliases( + inode->i_sb->s_bdev, + map.m_pblk + c, 1); +#endif } } rc = 0; @@ -975,9 +1069,9 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt, struct osd_iobuf *iobuf = &oti->oti_iobuf; struct inode *inode = osd_dt_obj(dt)->oo_inode; struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt)); - struct timeval start; - struct timeval end; - unsigned long timediff; + ktime_t start; + ktime_t end; + s64 timediff; ssize_t isize; __s64 maxidx; int rc = 0; @@ -991,14 +1085,14 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt, RETURN(rc); isize = i_size_read(inode); - maxidx = ((isize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT) - 1; + maxidx = ((isize + PAGE_SIZE - 1) >> PAGE_SHIFT) - 1; if (osd->od_writethrough_cache) cache = 1; if (isize > osd->od_readcache_max_filesize) cache = 0; - do_gettimeofday(&start); + start = ktime_get(); for (i = 0; i < npages; i++) { if (cache == 0) @@ -1012,7 +1106,7 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt, */ ClearPageUptodate(lnb[i].lnb_page); - if (lnb[i].lnb_len == PAGE_CACHE_SIZE) + if (lnb[i].lnb_len == PAGE_SIZE) continue; if (maxidx >= lnb[i].lnb_page->index) { @@ -1025,14 +1119,14 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt, if (off) memset(p, 0, off); off = (lnb[i].lnb_page_offset + lnb[i].lnb_len) & - ~CFS_PAGE_MASK; + ~PAGE_MASK; if (off) - memset(p + off, 0, PAGE_CACHE_SIZE - off); + memset(p + off, 0, PAGE_SIZE - off); kunmap(lnb[i].lnb_page); } } - do_gettimeofday(&end); - timediff = cfs_timeval_sub(&end, &start, NULL); + end = ktime_get(); + timediff = ktime_us_delta(end, start); lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff); if (iobuf->dr_npages) { @@ -1048,29 +1142,57 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt, RETURN(rc); } -/* Check if a block is allocated or not */ -static int osd_is_mapped(struct inode *inode, u64 offset) -{ - sector_t (*fs_bmap)(struct address_space *, sector_t); +struct osd_fextent { + sector_t start; + sector_t end; + unsigned int mapped:1; +}; - fs_bmap = inode->i_mapping->a_ops->bmap; +static int osd_is_mapped(struct dt_object *dt, __u64 offset, + struct osd_fextent *cached_extent) +{ + struct inode *inode = osd_dt_obj(dt)->oo_inode; + sector_t block = offset >> inode->i_blkbits; + sector_t start; + struct fiemap_extent_info fei = { 0 }; + struct fiemap_extent fe = { 0 }; + mm_segment_t saved_fs; + int rc; - /* We can't know if we are overwriting or not */ - if (unlikely(fs_bmap == NULL)) - return 0; + if (block >= cached_extent->start && block < cached_extent->end) + return cached_extent->mapped; if (i_size_read(inode) == 0) return 0; /* Beyond EOF, must not be mapped */ - if (((i_size_read(inode) - 1) >> inode->i_blkbits) < - (offset >> inode->i_blkbits)) + if (((i_size_read(inode) - 1) >> inode->i_blkbits) < block) return 0; - if (fs_bmap(inode->i_mapping, offset >> inode->i_blkbits) == 0) + fei.fi_extents_max = 1; + fei.fi_extents_start = &fe; + + saved_fs = get_fs(); + set_fs(get_ds()); + rc = inode->i_op->fiemap(inode, &fei, offset, FIEMAP_MAX_OFFSET-offset); + set_fs(saved_fs); + if (rc != 0) return 0; - return 1; + start = fe.fe_logical >> inode->i_blkbits; + + if (start > block) { + cached_extent->start = block; + cached_extent->end = start; + cached_extent->mapped = 0; + } else { + cached_extent->start = start; + cached_extent->end = (fe.fe_logical + fe.fe_length) >> + inode->i_blkbits; + cached_extent->mapped = 1; + } + + return cached_extent->mapped; } static int osd_declare_write_commit(const struct lu_env *env, @@ -1078,17 +1200,19 @@ static int osd_declare_write_commit(const struct lu_env *env, struct niobuf_local *lnb, int npages, struct thandle *handle) { - const struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt)); - struct inode *inode = osd_dt_obj(dt)->oo_inode; - struct osd_thandle *oh; - int extents = 1; - int depth; - int i; - int newblocks; - int rc = 0; - int flags = 0; - bool ignore_quota = false; - long long quota_space = 0; + const struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt)); + struct inode *inode = osd_dt_obj(dt)->oo_inode; + struct osd_thandle *oh; + int extents = 1; + int depth; + int i; + int newblocks; + int rc = 0; + int flags = 0; + int credits = 0; + long long quota_space = 0; + struct osd_fextent extent = { 0 }; + enum osd_qid_declare_flags declare_flags = OSD_QID_BLK; ENTRY; LASSERT(handle != NULL); @@ -1103,8 +1227,10 @@ static int osd_declare_write_commit(const struct lu_env *env, lnb[i - 1].lnb_file_offset + lnb[i - 1].lnb_len) extents++; - if (!osd_is_mapped(inode, lnb[i].lnb_file_offset)) - quota_space += PAGE_CACHE_SIZE; + if (osd_is_mapped(dt, lnb[i].lnb_file_offset, &extent)) + lnb[i].lnb_flags |= OBD_BRW_MAPPED; + else + quota_space += PAGE_SIZE; /* ignore quota for the whole request if any page is from * client cache or written by root. @@ -1118,7 +1244,7 @@ static int osd_declare_write_commit(const struct lu_env *env, if ((lnb[i].lnb_flags & OBD_BRW_NOQUOTA) || (lnb[i].lnb_flags & (OBD_BRW_FROM_GRANT | OBD_BRW_SYNC)) == OBD_BRW_FROM_GRANT) - ignore_quota = true; + declare_flags |= OSD_QID_FORCE; } /* @@ -1134,14 +1260,14 @@ static int osd_declare_write_commit(const struct lu_env *env, depth = ext_depth(inode); depth = max(depth, 1) + 1; newblocks += depth; - oh->ot_credits++; /* inode */ - oh->ot_credits += depth * 2 * extents; - } else { - depth = 3; - newblocks += depth; - oh->ot_credits++; /* inode */ - oh->ot_credits += depth * extents; - } + credits++; /* inode */ + credits += depth * 2 * extents; + } else { + depth = 3; + newblocks += depth; + credits++; /* inode */ + credits += depth * extents; + } /* quota space for metadata blocks */ quota_space += depth * extents * LDISKFS_BLOCK_SIZE(osd_sb(osd)); @@ -1153,22 +1279,24 @@ static int osd_declare_write_commit(const struct lu_env *env, /* we can't dirty more bitmap blocks than exist */ if (newblocks > LDISKFS_SB(osd_sb(osd))->s_groups_count) - oh->ot_credits += LDISKFS_SB(osd_sb(osd))->s_groups_count; + credits += LDISKFS_SB(osd_sb(osd))->s_groups_count; else - oh->ot_credits += newblocks; + credits += newblocks; - /* we can't dirty more gd blocks than exist */ - if (newblocks > LDISKFS_SB(osd_sb(osd))->s_gdb_count) - oh->ot_credits += LDISKFS_SB(osd_sb(osd))->s_gdb_count; - else - oh->ot_credits += newblocks; + /* we can't dirty more gd blocks than exist */ + if (newblocks > LDISKFS_SB(osd_sb(osd))->s_gdb_count) + credits += LDISKFS_SB(osd_sb(osd))->s_gdb_count; + else + credits += newblocks; + + osd_trans_declare_op(env, oh, OSD_OT_WRITE, credits); /* make sure the over quota flags were not set */ - lnb[0].lnb_flags &= ~(OBD_BRW_OVER_USRQUOTA | OBD_BRW_OVER_GRPQUOTA); + lnb[0].lnb_flags &= ~OBD_BRW_OVER_ALLQUOTA; rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - quota_space, oh, osd_dt_obj(dt), true, - &flags, ignore_quota); + i_projid_read(inode), quota_space, oh, + osd_dt_obj(dt), &flags, declare_flags); /* we need only to store the overquota flags in the first lnb for * now, once we support multiple objects BRW, this code needs be @@ -1177,6 +1305,11 @@ static int osd_declare_write_commit(const struct lu_env *env, lnb[0].lnb_flags |= OBD_BRW_OVER_USRQUOTA; if (flags & QUOTA_FL_OVER_GRPQUOTA) lnb[0].lnb_flags |= OBD_BRW_OVER_GRPQUOTA; + if (flags & QUOTA_FL_OVER_PRJQUOTA) + lnb[0].lnb_flags |= OBD_BRW_OVER_PRJQUOTA; + + if (rc == 0) + rc = osd_trunc_lock(osd_dt_obj(dt), oh, true); RETURN(rc); } @@ -1204,7 +1337,7 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, for (i = 0; i < npages; i++) { if (lnb[i].lnb_rc == -ENOSPC && - osd_is_mapped(inode, lnb[i].lnb_file_offset)) { + (lnb[i].lnb_flags & OBD_BRW_MAPPED)) { /* Allow the write to proceed if overwriting an * existing block */ lnb[i].lnb_rc = 0; @@ -1237,6 +1370,8 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, osd_iobuf_add_page(iobuf, lnb[i].lnb_page); } + osd_trans_exec_op(env, thandle, OSD_OT_WRITE); + if (OBD_FAIL_CHECK(OBD_FAIL_OST_MAPBLK_ENOSPC)) { rc = -ENOSPC; } else if (iobuf->dr_npages > 0) { @@ -1248,23 +1383,29 @@ static int osd_write_commit(const struct lu_env *env, struct dt_object *dt, thandle->th_local = 1; } - if (likely(rc == 0)) { - if (isize > i_size_read(inode)) { - i_size_write(inode, isize); - LDISKFS_I(inode)->i_disksize = isize; + if (likely(rc == 0)) { + spin_lock(&inode->i_lock); + if (isize > i_size_read(inode)) { + i_size_write(inode, isize); + LDISKFS_I(inode)->i_disksize = isize; + spin_unlock(&inode->i_lock); ll_dirty_inode(inode, I_DIRTY_DATASYNC); - } + } else { + spin_unlock(&inode->i_lock); + } - rc = osd_do_bio(osd, inode, iobuf); - /* we don't do stats here as in read path because - * write is async: we'll do this in osd_put_bufs() */ + rc = osd_do_bio(osd, inode, iobuf); + /* we don't do stats here as in read path because + * write is async: we'll do this in osd_put_bufs() */ } else { osd_fini_iobuf(osd, iobuf); } - if (unlikely(rc != 0)) { - /* if write fails, we should drop pages from the cache */ - for (i = 0; i < npages; i++) { + osd_trans_exec_check(env, thandle, OSD_OT_WRITE); + + if (unlikely(rc != 0)) { + /* if write fails, we should drop pages from the cache */ + for (i = 0; i < npages; i++) { if (lnb[i].lnb_page == NULL) continue; LASSERT(PageLocked(lnb[i].lnb_page)); @@ -1283,9 +1424,9 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt, struct osd_iobuf *iobuf = &oti->oti_iobuf; struct inode *inode = osd_dt_obj(dt)->oo_inode; struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt)); - struct timeval start, end; - unsigned long timediff; int rc = 0, i, cache = 0, cache_hits = 0, cache_misses = 0; + ktime_t start, end; + s64 timediff; loff_t isize; LASSERT(inode); @@ -1301,7 +1442,7 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt, if (isize > osd->od_readcache_max_filesize) cache = 0; - do_gettimeofday(&start); + start = ktime_get(); for (i = 0; i < npages; i++) { if (isize <= lnb[i].lnb_file_offset) @@ -1309,11 +1450,15 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt, * lnb->lnb_rc == 0, so it's easy to detect later. */ break; - if (isize < lnb[i].lnb_file_offset + lnb[i].lnb_len - 1) + if (isize < lnb[i].lnb_file_offset + lnb[i].lnb_len) lnb[i].lnb_rc = isize - lnb[i].lnb_file_offset; else lnb[i].lnb_rc = lnb[i].lnb_len; + /* Bypass disk read if fail_loc is set properly */ + if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW)) + SetPageUptodate(lnb[i].lnb_page); + if (PageUptodate(lnb[i].lnb_page)) { cache_hits++; } else { @@ -1325,8 +1470,8 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt, generic_error_remove_page(inode->i_mapping, lnb[i].lnb_page); } - do_gettimeofday(&end); - timediff = cfs_timeval_sub(&end, &start, NULL); + end = ktime_get(); + timediff = ktime_us_delta(end, start); lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff); if (cache_hits != 0) @@ -1379,7 +1524,6 @@ int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs) int blocksize; int csize; int boffs; - int err; /* prevent reading after eof */ spin_lock(&inode->i_lock); @@ -1405,14 +1549,13 @@ int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs) block = *offs >> inode->i_blkbits; boffs = *offs & (blocksize - 1); csize = min(blocksize - boffs, size); - bh = ldiskfs_bread(NULL, inode, block, 0, &err); - if (err != 0) { - CERROR("%s: can't read %u@%llu on ino %lu: rc = %d\n", - LDISKFS_SB(inode->i_sb)->s_es->s_volume_name, - csize, *offs, inode->i_ino, err); - if (bh != NULL) - brelse(bh); - return err; + bh = __ldiskfs_bread(NULL, inode, block, 0); + if (IS_ERR(bh)) { + CERROR("%s: can't read %u@%llu on ino %lu: " + "rc = %ld\n", osd_ino2name(inode), + csize, *offs, inode->i_ino, + PTR_ERR(bh)); + return PTR_ERR(bh); } if (bh != NULL) { @@ -1430,15 +1573,11 @@ int osd_ldiskfs_read(struct inode *inode, void *buf, int size, loff_t *offs) } static ssize_t osd_read(const struct lu_env *env, struct dt_object *dt, - struct lu_buf *buf, loff_t *pos, - struct lustre_capa *capa) + struct lu_buf *buf, loff_t *pos) { struct inode *inode = osd_dt_obj(dt)->oo_inode; int rc; - if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_READ)) - RETURN(-EACCES); - /* Read small symlink from inode body as we need to maintain correct * on-disk symlinks for ldiskfs. */ @@ -1457,18 +1596,15 @@ static inline int osd_extents_enabled(struct super_block *sb, if (inode != NULL) { if (LDISKFS_I(inode)->i_flags & LDISKFS_EXTENTS_FL) return 1; - } else if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, - LDISKFS_FEATURE_INCOMPAT_EXTENTS)) { + } else if (ldiskfs_has_feature_extents(sb)) { return 1; } return 0; } -static inline int osd_calc_bkmap_credits(struct super_block *sb, - struct inode *inode, - const loff_t size, - const loff_t pos, - const int blocks) +int osd_calc_bkmap_credits(struct super_block *sb, struct inode *inode, + const loff_t size, const loff_t pos, + const int blocks) { int credits, bits, bs, i; @@ -1503,9 +1639,12 @@ static inline int osd_calc_bkmap_credits(struct super_block *sb, } else if (pos + size <= (LDISKFS_NDIR_BLOCKS + 1024) * bs) { /* single indirect */ credits = blocks * 3; - /* probably indirect block has been allocated already */ - if (!inode || LDISKFS_I(inode)->i_data[LDISKFS_IND_BLOCK]) + if (inode == NULL || + LDISKFS_I(inode)->i_data[LDISKFS_IND_BLOCK] == 0) credits += 3; + else + /* The indirect block may be modified. */ + credits += 1; } return credits; @@ -1573,7 +1712,7 @@ static ssize_t osd_declare_write(const struct lu_env *env, struct dt_object *dt, credits = depth; /* if not append, then split may need to modify * existing blocks moving entries into the new ones */ - if (_pos == -1) + if (_pos != -1) credits += depth; /* blocks to store data: bitmap,gd,itself */ credits += blocks * 3; @@ -1594,8 +1733,13 @@ out: * objects, so always set the lqi_space as 0. */ if (inode != NULL) rc = osd_declare_inode_qid(env, i_uid_read(inode), - i_gid_read(inode), 0, oh, obj, true, - NULL, false); + i_gid_read(inode), + i_projid_read(inode), 0, + oh, obj, NULL, OSD_QID_BLK); + + if (rc == 0) + rc = osd_trunc_lock(obj, oh, true); + RETURN(rc); } @@ -1604,12 +1748,14 @@ static int osd_ldiskfs_writelink(struct inode *inode, char *buffer, int buflen) /* LU-2634: clear the extent format for fast symlink */ ldiskfs_clear_inode_flag(inode, LDISKFS_INODE_EXTENTS); - memcpy((char *)&LDISKFS_I(inode)->i_data, (char *)buffer, buflen); - LDISKFS_I(inode)->i_disksize = buflen; - i_size_write(inode, buflen); + memcpy((char *)&LDISKFS_I(inode)->i_data, (char *)buffer, buflen); + spin_lock(&inode->i_lock); + LDISKFS_I(inode)->i_disksize = buflen; + i_size_write(inode, buflen); + spin_unlock(&inode->i_lock); ll_dirty_inode(inode, I_DIRTY_DATASYNC); - return 0; + return 0; } int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize, @@ -1634,18 +1780,29 @@ int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize, ((char *)buf)[bufsize] = '\0'; ++bufsize; } - while (bufsize > 0) { - if (bh != NULL) - brelse(bh); - - block = offset >> inode->i_blkbits; - boffs = offset & (blocksize - 1); - size = min(blocksize - boffs, bufsize); - bh = ldiskfs_bread(handle, inode, block, 1, &err); - if (!bh) { - CERROR("%s: error reading offset %llu (block %lu): " - "rc = %d\n", - inode->i_sb->s_id, offset, block, err); + + while (bufsize > 0) { + int credits = handle->h_buffer_credits; + + if (bh) + brelse(bh); + + block = offset >> inode->i_blkbits; + boffs = offset & (blocksize - 1); + size = min(blocksize - boffs, bufsize); + bh = __ldiskfs_bread(handle, inode, block, 1); + if (IS_ERR_OR_NULL(bh)) { + if (bh == NULL) { + err = -EIO; + } else { + err = PTR_ERR(bh); + bh = NULL; + } + + CERROR("%s: error reading offset %llu (block %lu, " + "size %d, offs %llu), credits %d/%d: rc = %d\n", + inode->i_sb->s_id, offset, block, bufsize, *offs, + credits, handle->h_buffer_credits, err); break; } @@ -1674,8 +1831,8 @@ int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize, if (write_NUL) --new_size; - /* correct in-core and on-disk sizes */ - if (new_size > i_size_read(inode)) { + /* correct in-core and on-disk sizes */ + if (new_size > i_size_read(inode)) { spin_lock(&inode->i_lock); if (new_size > i_size_read(inode)) i_size_write(inode, new_size); @@ -1694,9 +1851,8 @@ int osd_ldiskfs_write_record(struct inode *inode, void *buf, int bufsize, } 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, struct lustre_capa *capa, - int ignore_quota) + const struct lu_buf *buf, loff_t *pos, + struct thandle *handle, int ignore_quota) { struct inode *inode = osd_dt_obj(dt)->oo_inode; struct osd_thandle *oh; @@ -1705,9 +1861,6 @@ static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt, LASSERT(dt_object_exists(dt)); - if (osd_object_auth(env, dt, capa, CAPA_OPC_BODY_WRITE)) - return -EACCES; - LASSERT(handle != NULL); LASSERT(inode != NULL); ll_vfs_dq_init(inode); @@ -1717,6 +1870,8 @@ static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt, oh = container_of(handle, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); + osd_trans_exec_op(env, handle, OSD_OT_WRITE); + /* Write small symlink to inode body as we need to maintain correct * on-disk symlinks for ldiskfs. * Note: the buf->lb_buf contains a NUL terminator while buf->lb_len @@ -1729,9 +1884,12 @@ static ssize_t osd_write(const struct lu_env *env, struct dt_object *dt, result = osd_ldiskfs_write_record(inode, buf->lb_buf, buf->lb_len, is_link, pos, oh->ot_handle); - if (result == 0) - result = buf->lb_len; - return result; + if (result == 0) + result = buf->lb_len; + + osd_trans_exec_check(env, handle, OSD_OT_WRITE); + + return result; } static int osd_declare_punch(const struct lu_env *env, struct dt_object *dt, @@ -1760,20 +1918,25 @@ static int osd_declare_punch(const struct lu_env *env, struct dt_object *dt, LASSERT(inode); rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode), - 0, oh, osd_dt_obj(dt), true, NULL, false); + i_projid_read(inode), 0, oh, osd_dt_obj(dt), + NULL, OSD_QID_BLK); + + if (rc == 0) + rc = osd_trunc_lock(osd_dt_obj(dt), oh, false); + RETURN(rc); } static int osd_punch(const struct lu_env *env, struct dt_object *dt, - __u64 start, __u64 end, struct thandle *th, - struct lustre_capa *capa) + __u64 start, __u64 end, struct thandle *th) { + struct osd_object *obj = osd_dt_obj(dt); + struct osd_device *osd = osd_obj2dev(obj); + struct inode *inode = obj->oo_inode; + struct osd_access_lock *al; struct osd_thandle *oh; - struct osd_object *obj = osd_dt_obj(dt); - struct inode *inode = obj->oo_inode; - handle_t *h; - tid_t tid; - int rc = 0, rc2 = 0; + int rc = 0, found = 0; + bool grow = false; ENTRY; LASSERT(end == OBD_OBJECT_EOF); @@ -1786,43 +1949,51 @@ static int osd_punch(const struct lu_env *env, struct dt_object *dt, oh = container_of(th, struct osd_thandle, ot_super); LASSERT(oh->ot_handle->h_transaction != NULL); - osd_trans_exec_op(env, th, OSD_OT_PUNCH); + /* we used to skip truncate to current size to + * optimize truncates on OST. with DoM we can + * get attr_set to set specific size (MDS_REINT) + * and then get truncate RPC which essentially + * would be skipped. this is bad.. so, disable + * this optimization on MDS till the client stop + * to sent MDS_REINT (LU-11033) -bzzz */ + if (osd->od_is_ost && i_size_read(inode) == start) + RETURN(0); - tid = oh->ot_handle->h_transaction->t_tid; + osd_trans_exec_op(env, th, OSD_OT_PUNCH); + spin_lock(&inode->i_lock); + if (i_size_read(inode) < start) + grow = true; i_size_write(inode, start); + spin_unlock(&inode->i_lock); ll_truncate_pagecache(inode, start); -#ifdef HAVE_INODEOPS_TRUNCATE - if (inode->i_op->truncate) { - inode->i_op->truncate(inode); - } else -#endif - ldiskfs_truncate(inode); - /* - * For a partial-page truncate, flush the page to disk immediately to - * avoid data corruption during direct disk write. b=17397 - */ - if ((start & ~CFS_PAGE_MASK) != 0) - rc = filemap_fdatawrite_range(inode->i_mapping, start, start+1); + /* optimize grow case */ + if (grow) { + osd_execute_truncate(obj); + GOTO(out, rc); + } - h = journal_current_handle(); - LASSERT(h != NULL); - LASSERT(h == oh->ot_handle); + /* add to orphan list to ensure truncate completion + * if this transaction succeed. ldiskfs_truncate() + * will take the inode out of the list */ + rc = ldiskfs_orphan_add(oh->ot_handle, inode); + if (rc != 0) + GOTO(out, rc); - if (tid != h->h_transaction->t_tid) { - int credits = oh->ot_credits; - /* - * transaction has changed during truncate - * we need to restart the handle with our credits - */ - if (h->h_buffer_credits < credits) { - if (ldiskfs_journal_extend(h, credits)) - rc2 = ldiskfs_journal_restart(h, credits); - } - } + list_for_each_entry(al, &oh->ot_trunc_locks, tl_list) { + if (obj != al->tl_obj) + continue; + LASSERT(al->tl_shared == 0); + found = 1; + /* do actual truncate in osd_trans_stop() */ + al->tl_truncate = 1; + break; + } + LASSERT(found); - RETURN(rc == 0 ? rc2 : rc); +out: + RETURN(rc); } static int fiemap_check_ranges(struct inode *inode, @@ -1856,7 +2027,7 @@ static int fiemap_check_ranges(struct inode *inode, #define FIEMAP_MAX_EXTENTS (UINT_MAX / sizeof(struct fiemap_extent)) static int osd_fiemap_get(const struct lu_env *env, struct dt_object *dt, - struct ll_user_fiemap *fm) + struct fiemap *fm) { struct fiemap_extent_info fieinfo = {0, }; struct inode *inode = osd_dt_obj(dt)->oo_inode; @@ -1889,6 +2060,29 @@ static int osd_fiemap_get(const struct lu_env *env, struct dt_object *dt, return rc; } +static int osd_ladvise(const struct lu_env *env, struct dt_object *dt, + __u64 start, __u64 end, enum lu_ladvise_type advice) +{ + int rc = 0; + struct inode *inode = osd_dt_obj(dt)->oo_inode; + ENTRY; + + switch (advice) { + case LU_LADVISE_DONTNEED: + if (end == 0) + break; + invalidate_mapping_pages(inode->i_mapping, + start >> PAGE_SHIFT, + (end - 1) >> PAGE_SHIFT); + break; + default: + rc = -ENOTSUPP; + break; + } + + RETURN(rc); +} + /* * in some cases we may need declare methods for objects being created * e.g., when we create symlink @@ -1898,17 +2092,125 @@ const struct dt_body_operations osd_body_ops_new = { }; const struct dt_body_operations osd_body_ops = { - .dbo_read = osd_read, - .dbo_declare_write = osd_declare_write, - .dbo_write = osd_write, - .dbo_bufs_get = osd_bufs_get, - .dbo_bufs_put = osd_bufs_put, - .dbo_write_prep = osd_write_prep, - .dbo_declare_write_commit = osd_declare_write_commit, - .dbo_write_commit = osd_write_commit, - .dbo_read_prep = osd_read_prep, - .dbo_declare_punch = osd_declare_punch, - .dbo_punch = osd_punch, - .dbo_fiemap_get = osd_fiemap_get, + .dbo_read = osd_read, + .dbo_declare_write = osd_declare_write, + .dbo_write = osd_write, + .dbo_bufs_get = osd_bufs_get, + .dbo_bufs_put = osd_bufs_put, + .dbo_write_prep = osd_write_prep, + .dbo_declare_write_commit = osd_declare_write_commit, + .dbo_write_commit = osd_write_commit, + .dbo_read_prep = osd_read_prep, + .dbo_declare_punch = osd_declare_punch, + .dbo_punch = osd_punch, + .dbo_fiemap_get = osd_fiemap_get, + .dbo_ladvise = osd_ladvise, }; +/** + * Get a truncate lock + * + * In order to take multi-transaction truncate out of main transaction we let + * the caller grab a lock on the object passed. the lock can be shared (for + * writes) and exclusive (for truncate). It's not allowed to mix truncate + * and write in the same transaction handle (do not confuse with big ldiskfs + * transaction containing lots of handles). + * The lock must be taken at declaration. + * + * \param obj object to lock + * \oh transaction + * \shared shared or exclusive + * + * \retval 0 lock is granted + * \retval -NOMEM no memory to allocate lock + */ +int osd_trunc_lock(struct osd_object *obj, struct osd_thandle *oh, bool shared) +{ + struct osd_access_lock *al, *tmp; + + LASSERT(obj); + LASSERT(oh); + + list_for_each_entry(tmp, &oh->ot_trunc_locks, tl_list) { + if (tmp->tl_obj != obj) + continue; + LASSERT(tmp->tl_shared == shared); + /* found same lock */ + return 0; + } + + OBD_ALLOC_PTR(al); + if (unlikely(al == NULL)) + return -ENOMEM; + al->tl_obj = obj; + al->tl_truncate = false; + if (shared) + down_read(&obj->oo_ext_idx_sem); + else + down_write(&obj->oo_ext_idx_sem); + al->tl_shared = shared; + + list_add(&al->tl_list, &oh->ot_trunc_locks); + + return 0; +} + +void osd_trunc_unlock_all(struct list_head *list) +{ + struct osd_access_lock *al, *tmp; + list_for_each_entry_safe(al, tmp, list, tl_list) { + if (al->tl_shared) + up_read(&al->tl_obj->oo_ext_idx_sem); + else + up_write(&al->tl_obj->oo_ext_idx_sem); + list_del(&al->tl_list); + OBD_FREE_PTR(al); + } +} + +void osd_execute_truncate(struct osd_object *obj) +{ + struct inode *inode = obj->oo_inode; + __u64 size; + + /* simulate crash before (in the middle) of delayed truncate */ + if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FAIL_AT_TRUNCATE)) { + struct ldiskfs_inode_info *ei = LDISKFS_I(inode); + struct ldiskfs_sb_info *sbi = LDISKFS_SB(inode->i_sb); + + mutex_lock(&sbi->s_orphan_lock); + list_del_init(&ei->i_orphan); + mutex_unlock(&sbi->s_orphan_lock); + return; + } + +#ifdef HAVE_INODEOPS_TRUNCATE + if (inode->i_op->truncate) + inode->i_op->truncate(inode); + else +#endif + ldiskfs_truncate(inode); + + /* + * For a partial-page truncate, flush the page to disk immediately to + * avoid data corruption during direct disk write. b=17397 + */ + size = i_size_read(inode); + if ((size & ~PAGE_MASK) != 0) + filemap_fdatawrite_range(inode->i_mapping, size, size + 1); +} + +void osd_process_truncates(struct list_head *list) +{ + struct osd_access_lock *al; + + LASSERT(journal_current_handle() == NULL); + + list_for_each_entry(al, list, tl_list) { + if (al->tl_shared) + continue; + if (!al->tl_truncate) + continue; + osd_execute_truncate(al->tl_obj); + } +}