13418d036SArtur Paszkiewicz /* 23418d036SArtur Paszkiewicz * Partial Parity Log for closing the RAID5 write hole 33418d036SArtur Paszkiewicz * Copyright (c) 2017, Intel Corporation. 43418d036SArtur Paszkiewicz * 53418d036SArtur Paszkiewicz * This program is free software; you can redistribute it and/or modify it 63418d036SArtur Paszkiewicz * under the terms and conditions of the GNU General Public License, 73418d036SArtur Paszkiewicz * version 2, as published by the Free Software Foundation. 83418d036SArtur Paszkiewicz * 93418d036SArtur Paszkiewicz * This program is distributed in the hope it will be useful, but WITHOUT 103418d036SArtur Paszkiewicz * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 113418d036SArtur Paszkiewicz * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 123418d036SArtur Paszkiewicz * more details. 133418d036SArtur Paszkiewicz */ 143418d036SArtur Paszkiewicz 153418d036SArtur Paszkiewicz #include <linux/kernel.h> 163418d036SArtur Paszkiewicz #include <linux/blkdev.h> 173418d036SArtur Paszkiewicz #include <linux/slab.h> 183418d036SArtur Paszkiewicz #include <linux/crc32c.h> 193418d036SArtur Paszkiewicz #include <linux/flex_array.h> 203418d036SArtur Paszkiewicz #include <linux/async_tx.h> 213418d036SArtur Paszkiewicz #include <linux/raid/md_p.h> 223418d036SArtur Paszkiewicz #include "md.h" 233418d036SArtur Paszkiewicz #include "raid5.h" 243418d036SArtur Paszkiewicz 253418d036SArtur Paszkiewicz /* 263418d036SArtur Paszkiewicz * PPL consists of a 4KB header (struct ppl_header) and at least 128KB for 273418d036SArtur Paszkiewicz * partial parity data. The header contains an array of entries 283418d036SArtur Paszkiewicz * (struct ppl_header_entry) which describe the logged write requests. 293418d036SArtur Paszkiewicz * Partial parity for the entries comes after the header, written in the same 303418d036SArtur Paszkiewicz * sequence as the entries: 313418d036SArtur Paszkiewicz * 323418d036SArtur Paszkiewicz * Header 333418d036SArtur Paszkiewicz * entry0 343418d036SArtur Paszkiewicz * ... 353418d036SArtur Paszkiewicz * entryN 363418d036SArtur Paszkiewicz * PP data 373418d036SArtur Paszkiewicz * PP for entry0 383418d036SArtur Paszkiewicz * ... 393418d036SArtur Paszkiewicz * PP for entryN 403418d036SArtur Paszkiewicz * 413418d036SArtur Paszkiewicz * An entry describes one or more consecutive stripe_heads, up to a full 423418d036SArtur Paszkiewicz * stripe. The modifed raid data chunks form an m-by-n matrix, where m is the 433418d036SArtur Paszkiewicz * number of stripe_heads in the entry and n is the number of modified data 443418d036SArtur Paszkiewicz * disks. Every stripe_head in the entry must write to the same data disks. 453418d036SArtur Paszkiewicz * An example of a valid case described by a single entry (writes to the first 463418d036SArtur Paszkiewicz * stripe of a 4 disk array, 16k chunk size): 473418d036SArtur Paszkiewicz * 483418d036SArtur Paszkiewicz * sh->sector dd0 dd1 dd2 ppl 493418d036SArtur Paszkiewicz * +-----+-----+-----+ 503418d036SArtur Paszkiewicz * 0 | --- | --- | --- | +----+ 513418d036SArtur Paszkiewicz * 8 | -W- | -W- | --- | | pp | data_sector = 8 523418d036SArtur Paszkiewicz * 16 | -W- | -W- | --- | | pp | data_size = 3 * 2 * 4k 533418d036SArtur Paszkiewicz * 24 | -W- | -W- | --- | | pp | pp_size = 3 * 4k 543418d036SArtur Paszkiewicz * +-----+-----+-----+ +----+ 553418d036SArtur Paszkiewicz * 563418d036SArtur Paszkiewicz * data_sector is the first raid sector of the modified data, data_size is the 573418d036SArtur Paszkiewicz * total size of modified data and pp_size is the size of partial parity for 583418d036SArtur Paszkiewicz * this entry. Entries for full stripe writes contain no partial parity 593418d036SArtur Paszkiewicz * (pp_size = 0), they only mark the stripes for which parity should be 603418d036SArtur Paszkiewicz * recalculated after an unclean shutdown. Every entry holds a checksum of its 613418d036SArtur Paszkiewicz * partial parity, the header also has a checksum of the header itself. 623418d036SArtur Paszkiewicz * 633418d036SArtur Paszkiewicz * A write request is always logged to the PPL instance stored on the parity 643418d036SArtur Paszkiewicz * disk of the corresponding stripe. For each member disk there is one ppl_log 653418d036SArtur Paszkiewicz * used to handle logging for this disk, independently from others. They are 663418d036SArtur Paszkiewicz * grouped in child_logs array in struct ppl_conf, which is assigned to 673418d036SArtur Paszkiewicz * r5conf->log_private. 683418d036SArtur Paszkiewicz * 693418d036SArtur Paszkiewicz * ppl_io_unit represents a full PPL write, header_page contains the ppl_header. 703418d036SArtur Paszkiewicz * PPL entries for logged stripes are added in ppl_log_stripe(). A stripe_head 713418d036SArtur Paszkiewicz * can be appended to the last entry if it meets the conditions for a valid 723418d036SArtur Paszkiewicz * entry described above, otherwise a new entry is added. Checksums of entries 733418d036SArtur Paszkiewicz * are calculated incrementally as stripes containing partial parity are being 743418d036SArtur Paszkiewicz * added. ppl_submit_iounit() calculates the checksum of the header and submits 753418d036SArtur Paszkiewicz * a bio containing the header page and partial parity pages (sh->ppl_page) for 763418d036SArtur Paszkiewicz * all stripes of the io_unit. When the PPL write completes, the stripes 773418d036SArtur Paszkiewicz * associated with the io_unit are released and raid5d starts writing their data 783418d036SArtur Paszkiewicz * and parity. When all stripes are written, the io_unit is freed and the next 793418d036SArtur Paszkiewicz * can be submitted. 803418d036SArtur Paszkiewicz * 813418d036SArtur Paszkiewicz * An io_unit is used to gather stripes until it is submitted or becomes full 823418d036SArtur Paszkiewicz * (if the maximum number of entries or size of PPL is reached). Another io_unit 833418d036SArtur Paszkiewicz * can't be submitted until the previous has completed (PPL and stripe 843418d036SArtur Paszkiewicz * data+parity is written). The log->io_list tracks all io_units of a log 853418d036SArtur Paszkiewicz * (for a single member disk). New io_units are added to the end of the list 863418d036SArtur Paszkiewicz * and the first io_unit is submitted, if it is not submitted already. 873418d036SArtur Paszkiewicz * The current io_unit accepting new stripes is always at the end of the list. 883418d036SArtur Paszkiewicz */ 893418d036SArtur Paszkiewicz 903418d036SArtur Paszkiewicz struct ppl_conf { 913418d036SArtur Paszkiewicz struct mddev *mddev; 923418d036SArtur Paszkiewicz 933418d036SArtur Paszkiewicz /* array of child logs, one for each raid disk */ 943418d036SArtur Paszkiewicz struct ppl_log *child_logs; 953418d036SArtur Paszkiewicz int count; 963418d036SArtur Paszkiewicz 973418d036SArtur Paszkiewicz int block_size; /* the logical block size used for data_sector 983418d036SArtur Paszkiewicz * in ppl_header_entry */ 993418d036SArtur Paszkiewicz u32 signature; /* raid array identifier */ 1003418d036SArtur Paszkiewicz atomic64_t seq; /* current log write sequence number */ 1013418d036SArtur Paszkiewicz 1023418d036SArtur Paszkiewicz struct kmem_cache *io_kc; 1033418d036SArtur Paszkiewicz mempool_t *io_pool; 1043418d036SArtur Paszkiewicz struct bio_set *bs; 1054536bf9bSArtur Paszkiewicz 1064536bf9bSArtur Paszkiewicz /* used only for recovery */ 1074536bf9bSArtur Paszkiewicz int recovered_entries; 1084536bf9bSArtur Paszkiewicz int mismatch_count; 10994568f64SArtur Paszkiewicz 11094568f64SArtur Paszkiewicz /* stripes to retry if failed to allocate io_unit */ 11194568f64SArtur Paszkiewicz struct list_head no_mem_stripes; 11294568f64SArtur Paszkiewicz spinlock_t no_mem_stripes_lock; 1133418d036SArtur Paszkiewicz }; 1143418d036SArtur Paszkiewicz 1153418d036SArtur Paszkiewicz struct ppl_log { 1163418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf; /* shared between all log instances */ 1173418d036SArtur Paszkiewicz 1183418d036SArtur Paszkiewicz struct md_rdev *rdev; /* array member disk associated with 1193418d036SArtur Paszkiewicz * this log instance */ 1203418d036SArtur Paszkiewicz struct mutex io_mutex; 1213418d036SArtur Paszkiewicz struct ppl_io_unit *current_io; /* current io_unit accepting new data 1223418d036SArtur Paszkiewicz * always at the end of io_list */ 1233418d036SArtur Paszkiewicz spinlock_t io_list_lock; 1243418d036SArtur Paszkiewicz struct list_head io_list; /* all io_units of this log */ 1253418d036SArtur Paszkiewicz }; 1263418d036SArtur Paszkiewicz 1273418d036SArtur Paszkiewicz #define PPL_IO_INLINE_BVECS 32 1283418d036SArtur Paszkiewicz 1293418d036SArtur Paszkiewicz struct ppl_io_unit { 1303418d036SArtur Paszkiewicz struct ppl_log *log; 1313418d036SArtur Paszkiewicz 1323418d036SArtur Paszkiewicz struct page *header_page; /* for ppl_header */ 1333418d036SArtur Paszkiewicz 1343418d036SArtur Paszkiewicz unsigned int entries_count; /* number of entries in ppl_header */ 1353418d036SArtur Paszkiewicz unsigned int pp_size; /* total size current of partial parity */ 1363418d036SArtur Paszkiewicz 1373418d036SArtur Paszkiewicz u64 seq; /* sequence number of this log write */ 1383418d036SArtur Paszkiewicz struct list_head log_sibling; /* log->io_list */ 1393418d036SArtur Paszkiewicz 1403418d036SArtur Paszkiewicz struct list_head stripe_list; /* stripes added to the io_unit */ 1413418d036SArtur Paszkiewicz atomic_t pending_stripes; /* how many stripes not written to raid */ 1423418d036SArtur Paszkiewicz 1433418d036SArtur Paszkiewicz bool submitted; /* true if write to log started */ 1443418d036SArtur Paszkiewicz 1453418d036SArtur Paszkiewicz /* inline bio and its biovec for submitting the iounit */ 1463418d036SArtur Paszkiewicz struct bio bio; 1473418d036SArtur Paszkiewicz struct bio_vec biovec[PPL_IO_INLINE_BVECS]; 1483418d036SArtur Paszkiewicz }; 1493418d036SArtur Paszkiewicz 1503418d036SArtur Paszkiewicz struct dma_async_tx_descriptor * 1513418d036SArtur Paszkiewicz ops_run_partial_parity(struct stripe_head *sh, struct raid5_percpu *percpu, 1523418d036SArtur Paszkiewicz struct dma_async_tx_descriptor *tx) 1533418d036SArtur Paszkiewicz { 1543418d036SArtur Paszkiewicz int disks = sh->disks; 155ae1713e2SArtur Paszkiewicz struct page **srcs = flex_array_get(percpu->scribble, 0); 1563418d036SArtur Paszkiewicz int count = 0, pd_idx = sh->pd_idx, i; 1573418d036SArtur Paszkiewicz struct async_submit_ctl submit; 1583418d036SArtur Paszkiewicz 1593418d036SArtur Paszkiewicz pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); 1603418d036SArtur Paszkiewicz 1613418d036SArtur Paszkiewicz /* 1623418d036SArtur Paszkiewicz * Partial parity is the XOR of stripe data chunks that are not changed 1633418d036SArtur Paszkiewicz * during the write request. Depending on available data 1643418d036SArtur Paszkiewicz * (read-modify-write vs. reconstruct-write case) we calculate it 1653418d036SArtur Paszkiewicz * differently. 1663418d036SArtur Paszkiewicz */ 1673418d036SArtur Paszkiewicz if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { 168ae1713e2SArtur Paszkiewicz /* 169ae1713e2SArtur Paszkiewicz * rmw: xor old data and parity from updated disks 170ae1713e2SArtur Paszkiewicz * This is calculated earlier by ops_run_prexor5() so just copy 171ae1713e2SArtur Paszkiewicz * the parity dev page. 172ae1713e2SArtur Paszkiewicz */ 173ae1713e2SArtur Paszkiewicz srcs[count++] = sh->dev[pd_idx].page; 1743418d036SArtur Paszkiewicz } else if (sh->reconstruct_state == reconstruct_state_drain_run) { 1753418d036SArtur Paszkiewicz /* rcw: xor data from all not updated disks */ 1763418d036SArtur Paszkiewicz for (i = disks; i--;) { 1773418d036SArtur Paszkiewicz struct r5dev *dev = &sh->dev[i]; 1783418d036SArtur Paszkiewicz if (test_bit(R5_UPTODATE, &dev->flags)) 179ae1713e2SArtur Paszkiewicz srcs[count++] = dev->page; 1803418d036SArtur Paszkiewicz } 1813418d036SArtur Paszkiewicz } else { 1823418d036SArtur Paszkiewicz return tx; 1833418d036SArtur Paszkiewicz } 1843418d036SArtur Paszkiewicz 1853418d036SArtur Paszkiewicz init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, tx, 1863418d036SArtur Paszkiewicz NULL, sh, flex_array_get(percpu->scribble, 0) 1873418d036SArtur Paszkiewicz + sizeof(struct page *) * (sh->disks + 2)); 1883418d036SArtur Paszkiewicz 1893418d036SArtur Paszkiewicz if (count == 1) 190ae1713e2SArtur Paszkiewicz tx = async_memcpy(sh->ppl_page, srcs[0], 0, 0, PAGE_SIZE, 1913418d036SArtur Paszkiewicz &submit); 1923418d036SArtur Paszkiewicz else 193ae1713e2SArtur Paszkiewicz tx = async_xor(sh->ppl_page, srcs, 0, count, PAGE_SIZE, 1943418d036SArtur Paszkiewicz &submit); 1953418d036SArtur Paszkiewicz 1963418d036SArtur Paszkiewicz return tx; 1973418d036SArtur Paszkiewicz } 1983418d036SArtur Paszkiewicz 199fcd403afSArtur Paszkiewicz static void *ppl_io_pool_alloc(gfp_t gfp_mask, void *pool_data) 200fcd403afSArtur Paszkiewicz { 201fcd403afSArtur Paszkiewicz struct kmem_cache *kc = pool_data; 202fcd403afSArtur Paszkiewicz struct ppl_io_unit *io; 203fcd403afSArtur Paszkiewicz 204fcd403afSArtur Paszkiewicz io = kmem_cache_alloc(kc, gfp_mask); 205fcd403afSArtur Paszkiewicz if (!io) 206fcd403afSArtur Paszkiewicz return NULL; 207fcd403afSArtur Paszkiewicz 208fcd403afSArtur Paszkiewicz io->header_page = alloc_page(gfp_mask); 209fcd403afSArtur Paszkiewicz if (!io->header_page) { 210fcd403afSArtur Paszkiewicz kmem_cache_free(kc, io); 211fcd403afSArtur Paszkiewicz return NULL; 212fcd403afSArtur Paszkiewicz } 213fcd403afSArtur Paszkiewicz 214fcd403afSArtur Paszkiewicz return io; 215fcd403afSArtur Paszkiewicz } 216fcd403afSArtur Paszkiewicz 217fcd403afSArtur Paszkiewicz static void ppl_io_pool_free(void *element, void *pool_data) 218fcd403afSArtur Paszkiewicz { 219fcd403afSArtur Paszkiewicz struct kmem_cache *kc = pool_data; 220fcd403afSArtur Paszkiewicz struct ppl_io_unit *io = element; 221fcd403afSArtur Paszkiewicz 222fcd403afSArtur Paszkiewicz __free_page(io->header_page); 223fcd403afSArtur Paszkiewicz kmem_cache_free(kc, io); 224fcd403afSArtur Paszkiewicz } 225fcd403afSArtur Paszkiewicz 2263418d036SArtur Paszkiewicz static struct ppl_io_unit *ppl_new_iounit(struct ppl_log *log, 2273418d036SArtur Paszkiewicz struct stripe_head *sh) 2283418d036SArtur Paszkiewicz { 2293418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 2303418d036SArtur Paszkiewicz struct ppl_io_unit *io; 2313418d036SArtur Paszkiewicz struct ppl_header *pplhdr; 232fcd403afSArtur Paszkiewicz struct page *header_page; 2333418d036SArtur Paszkiewicz 234fcd403afSArtur Paszkiewicz io = mempool_alloc(ppl_conf->io_pool, GFP_NOWAIT); 2353418d036SArtur Paszkiewicz if (!io) 2363418d036SArtur Paszkiewicz return NULL; 2373418d036SArtur Paszkiewicz 238fcd403afSArtur Paszkiewicz header_page = io->header_page; 2393418d036SArtur Paszkiewicz memset(io, 0, sizeof(*io)); 240fcd403afSArtur Paszkiewicz io->header_page = header_page; 241fcd403afSArtur Paszkiewicz 2423418d036SArtur Paszkiewicz io->log = log; 2433418d036SArtur Paszkiewicz INIT_LIST_HEAD(&io->log_sibling); 2443418d036SArtur Paszkiewicz INIT_LIST_HEAD(&io->stripe_list); 2453418d036SArtur Paszkiewicz atomic_set(&io->pending_stripes, 0); 2463418d036SArtur Paszkiewicz bio_init(&io->bio, io->biovec, PPL_IO_INLINE_BVECS); 2473418d036SArtur Paszkiewicz 2483418d036SArtur Paszkiewicz pplhdr = page_address(io->header_page); 2493418d036SArtur Paszkiewicz clear_page(pplhdr); 2503418d036SArtur Paszkiewicz memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED); 2513418d036SArtur Paszkiewicz pplhdr->signature = cpu_to_le32(ppl_conf->signature); 2523418d036SArtur Paszkiewicz 2533418d036SArtur Paszkiewicz io->seq = atomic64_add_return(1, &ppl_conf->seq); 2543418d036SArtur Paszkiewicz pplhdr->generation = cpu_to_le64(io->seq); 2553418d036SArtur Paszkiewicz 2563418d036SArtur Paszkiewicz return io; 2573418d036SArtur Paszkiewicz } 2583418d036SArtur Paszkiewicz 2593418d036SArtur Paszkiewicz static int ppl_log_stripe(struct ppl_log *log, struct stripe_head *sh) 2603418d036SArtur Paszkiewicz { 2613418d036SArtur Paszkiewicz struct ppl_io_unit *io = log->current_io; 2623418d036SArtur Paszkiewicz struct ppl_header_entry *e = NULL; 2633418d036SArtur Paszkiewicz struct ppl_header *pplhdr; 2643418d036SArtur Paszkiewicz int i; 2653418d036SArtur Paszkiewicz sector_t data_sector = 0; 2663418d036SArtur Paszkiewicz int data_disks = 0; 2673418d036SArtur Paszkiewicz unsigned int entry_space = (log->rdev->ppl.size << 9) - PPL_HEADER_SIZE; 2683418d036SArtur Paszkiewicz struct r5conf *conf = sh->raid_conf; 2693418d036SArtur Paszkiewicz 2703418d036SArtur Paszkiewicz pr_debug("%s: stripe: %llu\n", __func__, (unsigned long long)sh->sector); 2713418d036SArtur Paszkiewicz 2723418d036SArtur Paszkiewicz /* check if current io_unit is full */ 2733418d036SArtur Paszkiewicz if (io && (io->pp_size == entry_space || 2743418d036SArtur Paszkiewicz io->entries_count == PPL_HDR_MAX_ENTRIES)) { 2753418d036SArtur Paszkiewicz pr_debug("%s: add io_unit blocked by seq: %llu\n", 2763418d036SArtur Paszkiewicz __func__, io->seq); 2773418d036SArtur Paszkiewicz io = NULL; 2783418d036SArtur Paszkiewicz } 2793418d036SArtur Paszkiewicz 2803418d036SArtur Paszkiewicz /* add a new unit if there is none or the current is full */ 2813418d036SArtur Paszkiewicz if (!io) { 2823418d036SArtur Paszkiewicz io = ppl_new_iounit(log, sh); 2833418d036SArtur Paszkiewicz if (!io) 2843418d036SArtur Paszkiewicz return -ENOMEM; 2853418d036SArtur Paszkiewicz spin_lock_irq(&log->io_list_lock); 2863418d036SArtur Paszkiewicz list_add_tail(&io->log_sibling, &log->io_list); 2873418d036SArtur Paszkiewicz spin_unlock_irq(&log->io_list_lock); 2883418d036SArtur Paszkiewicz 2893418d036SArtur Paszkiewicz log->current_io = io; 2903418d036SArtur Paszkiewicz } 2913418d036SArtur Paszkiewicz 2923418d036SArtur Paszkiewicz for (i = 0; i < sh->disks; i++) { 2933418d036SArtur Paszkiewicz struct r5dev *dev = &sh->dev[i]; 2943418d036SArtur Paszkiewicz 2953418d036SArtur Paszkiewicz if (i != sh->pd_idx && test_bit(R5_Wantwrite, &dev->flags)) { 2963418d036SArtur Paszkiewicz if (!data_disks || dev->sector < data_sector) 2973418d036SArtur Paszkiewicz data_sector = dev->sector; 2983418d036SArtur Paszkiewicz data_disks++; 2993418d036SArtur Paszkiewicz } 3003418d036SArtur Paszkiewicz } 3013418d036SArtur Paszkiewicz BUG_ON(!data_disks); 3023418d036SArtur Paszkiewicz 3033418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu data_sector: %llu data_disks: %d\n", __func__, 3043418d036SArtur Paszkiewicz io->seq, (unsigned long long)data_sector, data_disks); 3053418d036SArtur Paszkiewicz 3063418d036SArtur Paszkiewicz pplhdr = page_address(io->header_page); 3073418d036SArtur Paszkiewicz 3083418d036SArtur Paszkiewicz if (io->entries_count > 0) { 3093418d036SArtur Paszkiewicz struct ppl_header_entry *last = 3103418d036SArtur Paszkiewicz &pplhdr->entries[io->entries_count - 1]; 3113418d036SArtur Paszkiewicz struct stripe_head *sh_last = list_last_entry( 3123418d036SArtur Paszkiewicz &io->stripe_list, struct stripe_head, log_list); 3133418d036SArtur Paszkiewicz u64 data_sector_last = le64_to_cpu(last->data_sector); 3143418d036SArtur Paszkiewicz u32 data_size_last = le32_to_cpu(last->data_size); 3153418d036SArtur Paszkiewicz 3163418d036SArtur Paszkiewicz /* 3173418d036SArtur Paszkiewicz * Check if we can append the stripe to the last entry. It must 3183418d036SArtur Paszkiewicz * be just after the last logged stripe and write to the same 3193418d036SArtur Paszkiewicz * disks. Use bit shift and logarithm to avoid 64-bit division. 3203418d036SArtur Paszkiewicz */ 3213418d036SArtur Paszkiewicz if ((sh->sector == sh_last->sector + STRIPE_SECTORS) && 3223418d036SArtur Paszkiewicz (data_sector >> ilog2(conf->chunk_sectors) == 3233418d036SArtur Paszkiewicz data_sector_last >> ilog2(conf->chunk_sectors)) && 3243418d036SArtur Paszkiewicz ((data_sector - data_sector_last) * data_disks == 3253418d036SArtur Paszkiewicz data_size_last >> 9)) 3263418d036SArtur Paszkiewicz e = last; 3273418d036SArtur Paszkiewicz } 3283418d036SArtur Paszkiewicz 3293418d036SArtur Paszkiewicz if (!e) { 3303418d036SArtur Paszkiewicz e = &pplhdr->entries[io->entries_count++]; 3313418d036SArtur Paszkiewicz e->data_sector = cpu_to_le64(data_sector); 3323418d036SArtur Paszkiewicz e->parity_disk = cpu_to_le32(sh->pd_idx); 3333418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(~0); 3343418d036SArtur Paszkiewicz } 3353418d036SArtur Paszkiewicz 3363418d036SArtur Paszkiewicz le32_add_cpu(&e->data_size, data_disks << PAGE_SHIFT); 3373418d036SArtur Paszkiewicz 3383418d036SArtur Paszkiewicz /* don't write any PP if full stripe write */ 3393418d036SArtur Paszkiewicz if (!test_bit(STRIPE_FULL_WRITE, &sh->state)) { 3403418d036SArtur Paszkiewicz le32_add_cpu(&e->pp_size, PAGE_SIZE); 3413418d036SArtur Paszkiewicz io->pp_size += PAGE_SIZE; 3423418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(crc32c_le(le32_to_cpu(e->checksum), 3433418d036SArtur Paszkiewicz page_address(sh->ppl_page), 3443418d036SArtur Paszkiewicz PAGE_SIZE)); 3453418d036SArtur Paszkiewicz } 3463418d036SArtur Paszkiewicz 3473418d036SArtur Paszkiewicz list_add_tail(&sh->log_list, &io->stripe_list); 3483418d036SArtur Paszkiewicz atomic_inc(&io->pending_stripes); 3493418d036SArtur Paszkiewicz sh->ppl_io = io; 3503418d036SArtur Paszkiewicz 3513418d036SArtur Paszkiewicz return 0; 3523418d036SArtur Paszkiewicz } 3533418d036SArtur Paszkiewicz 3543418d036SArtur Paszkiewicz int ppl_write_stripe(struct r5conf *conf, struct stripe_head *sh) 3553418d036SArtur Paszkiewicz { 3563418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 3573418d036SArtur Paszkiewicz struct ppl_io_unit *io = sh->ppl_io; 3583418d036SArtur Paszkiewicz struct ppl_log *log; 3593418d036SArtur Paszkiewicz 360845b9e22SArtur Paszkiewicz if (io || test_bit(STRIPE_SYNCING, &sh->state) || !sh->ppl_page || 3613418d036SArtur Paszkiewicz !test_bit(R5_Wantwrite, &sh->dev[sh->pd_idx].flags) || 3623418d036SArtur Paszkiewicz !test_bit(R5_Insync, &sh->dev[sh->pd_idx].flags)) { 3633418d036SArtur Paszkiewicz clear_bit(STRIPE_LOG_TRAPPED, &sh->state); 3643418d036SArtur Paszkiewicz return -EAGAIN; 3653418d036SArtur Paszkiewicz } 3663418d036SArtur Paszkiewicz 3673418d036SArtur Paszkiewicz log = &ppl_conf->child_logs[sh->pd_idx]; 3683418d036SArtur Paszkiewicz 3693418d036SArtur Paszkiewicz mutex_lock(&log->io_mutex); 3703418d036SArtur Paszkiewicz 3713418d036SArtur Paszkiewicz if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) { 3723418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 3733418d036SArtur Paszkiewicz return -EAGAIN; 3743418d036SArtur Paszkiewicz } 3753418d036SArtur Paszkiewicz 3763418d036SArtur Paszkiewicz set_bit(STRIPE_LOG_TRAPPED, &sh->state); 3773418d036SArtur Paszkiewicz clear_bit(STRIPE_DELAYED, &sh->state); 3783418d036SArtur Paszkiewicz atomic_inc(&sh->count); 3793418d036SArtur Paszkiewicz 3803418d036SArtur Paszkiewicz if (ppl_log_stripe(log, sh)) { 38194568f64SArtur Paszkiewicz spin_lock_irq(&ppl_conf->no_mem_stripes_lock); 38294568f64SArtur Paszkiewicz list_add_tail(&sh->log_list, &ppl_conf->no_mem_stripes); 38394568f64SArtur Paszkiewicz spin_unlock_irq(&ppl_conf->no_mem_stripes_lock); 3843418d036SArtur Paszkiewicz } 3853418d036SArtur Paszkiewicz 3863418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 3873418d036SArtur Paszkiewicz 3883418d036SArtur Paszkiewicz return 0; 3893418d036SArtur Paszkiewicz } 3903418d036SArtur Paszkiewicz 3913418d036SArtur Paszkiewicz static void ppl_log_endio(struct bio *bio) 3923418d036SArtur Paszkiewicz { 3933418d036SArtur Paszkiewicz struct ppl_io_unit *io = bio->bi_private; 3943418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 3953418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 3963418d036SArtur Paszkiewicz struct stripe_head *sh, *next; 3973418d036SArtur Paszkiewicz 3983418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu\n", __func__, io->seq); 3993418d036SArtur Paszkiewicz 4004e4cbee9SChristoph Hellwig if (bio->bi_status) 4013418d036SArtur Paszkiewicz md_error(ppl_conf->mddev, log->rdev); 4023418d036SArtur Paszkiewicz 4033418d036SArtur Paszkiewicz list_for_each_entry_safe(sh, next, &io->stripe_list, log_list) { 4043418d036SArtur Paszkiewicz list_del_init(&sh->log_list); 4053418d036SArtur Paszkiewicz 4063418d036SArtur Paszkiewicz set_bit(STRIPE_HANDLE, &sh->state); 4073418d036SArtur Paszkiewicz raid5_release_stripe(sh); 4083418d036SArtur Paszkiewicz } 4093418d036SArtur Paszkiewicz } 4103418d036SArtur Paszkiewicz 4113418d036SArtur Paszkiewicz static void ppl_submit_iounit_bio(struct ppl_io_unit *io, struct bio *bio) 4123418d036SArtur Paszkiewicz { 4133418d036SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 4143418d036SArtur Paszkiewicz 4153418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu size: %u sector: %llu dev: %s\n", 4163418d036SArtur Paszkiewicz __func__, io->seq, bio->bi_iter.bi_size, 4173418d036SArtur Paszkiewicz (unsigned long long)bio->bi_iter.bi_sector, 41874d46992SChristoph Hellwig bio_devname(bio, b)); 4193418d036SArtur Paszkiewicz 4203418d036SArtur Paszkiewicz submit_bio(bio); 4213418d036SArtur Paszkiewicz } 4223418d036SArtur Paszkiewicz 4233418d036SArtur Paszkiewicz static void ppl_submit_iounit(struct ppl_io_unit *io) 4243418d036SArtur Paszkiewicz { 4253418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 4263418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 4273418d036SArtur Paszkiewicz struct ppl_header *pplhdr = page_address(io->header_page); 4283418d036SArtur Paszkiewicz struct bio *bio = &io->bio; 4293418d036SArtur Paszkiewicz struct stripe_head *sh; 4303418d036SArtur Paszkiewicz int i; 4313418d036SArtur Paszkiewicz 4326358c239SArtur Paszkiewicz bio->bi_private = io; 4336358c239SArtur Paszkiewicz 4346358c239SArtur Paszkiewicz if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) { 4356358c239SArtur Paszkiewicz ppl_log_endio(bio); 4366358c239SArtur Paszkiewicz return; 4376358c239SArtur Paszkiewicz } 4386358c239SArtur Paszkiewicz 4393418d036SArtur Paszkiewicz for (i = 0; i < io->entries_count; i++) { 4403418d036SArtur Paszkiewicz struct ppl_header_entry *e = &pplhdr->entries[i]; 4413418d036SArtur Paszkiewicz 4423418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu entry: %d data_sector: %llu pp_size: %u data_size: %u\n", 4433418d036SArtur Paszkiewicz __func__, io->seq, i, le64_to_cpu(e->data_sector), 4443418d036SArtur Paszkiewicz le32_to_cpu(e->pp_size), le32_to_cpu(e->data_size)); 4453418d036SArtur Paszkiewicz 4463418d036SArtur Paszkiewicz e->data_sector = cpu_to_le64(le64_to_cpu(e->data_sector) >> 4473418d036SArtur Paszkiewicz ilog2(ppl_conf->block_size >> 9)); 4483418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(~le32_to_cpu(e->checksum)); 4493418d036SArtur Paszkiewicz } 4503418d036SArtur Paszkiewicz 4513418d036SArtur Paszkiewicz pplhdr->entries_count = cpu_to_le32(io->entries_count); 4523418d036SArtur Paszkiewicz pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PPL_HEADER_SIZE)); 4533418d036SArtur Paszkiewicz 4543418d036SArtur Paszkiewicz bio->bi_end_io = ppl_log_endio; 4553418d036SArtur Paszkiewicz bio->bi_opf = REQ_OP_WRITE | REQ_FUA; 45674d46992SChristoph Hellwig bio_set_dev(bio, log->rdev->bdev); 4573418d036SArtur Paszkiewicz bio->bi_iter.bi_sector = log->rdev->ppl.sector; 4583418d036SArtur Paszkiewicz bio_add_page(bio, io->header_page, PAGE_SIZE, 0); 4593418d036SArtur Paszkiewicz 4603418d036SArtur Paszkiewicz list_for_each_entry(sh, &io->stripe_list, log_list) { 4613418d036SArtur Paszkiewicz /* entries for full stripe writes have no partial parity */ 4623418d036SArtur Paszkiewicz if (test_bit(STRIPE_FULL_WRITE, &sh->state)) 4633418d036SArtur Paszkiewicz continue; 4643418d036SArtur Paszkiewicz 4653418d036SArtur Paszkiewicz if (!bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0)) { 4663418d036SArtur Paszkiewicz struct bio *prev = bio; 4673418d036SArtur Paszkiewicz 4683418d036SArtur Paszkiewicz bio = bio_alloc_bioset(GFP_NOIO, BIO_MAX_PAGES, 4693418d036SArtur Paszkiewicz ppl_conf->bs); 4703418d036SArtur Paszkiewicz bio->bi_opf = prev->bi_opf; 47174d46992SChristoph Hellwig bio_copy_dev(bio, prev); 4723418d036SArtur Paszkiewicz bio->bi_iter.bi_sector = bio_end_sector(prev); 4733418d036SArtur Paszkiewicz bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0); 4743418d036SArtur Paszkiewicz 4753418d036SArtur Paszkiewicz bio_chain(bio, prev); 4763418d036SArtur Paszkiewicz ppl_submit_iounit_bio(io, prev); 4773418d036SArtur Paszkiewicz } 4783418d036SArtur Paszkiewicz } 4793418d036SArtur Paszkiewicz 4803418d036SArtur Paszkiewicz ppl_submit_iounit_bio(io, bio); 4813418d036SArtur Paszkiewicz } 4823418d036SArtur Paszkiewicz 4833418d036SArtur Paszkiewicz static void ppl_submit_current_io(struct ppl_log *log) 4843418d036SArtur Paszkiewicz { 4853418d036SArtur Paszkiewicz struct ppl_io_unit *io; 4863418d036SArtur Paszkiewicz 4873418d036SArtur Paszkiewicz spin_lock_irq(&log->io_list_lock); 4883418d036SArtur Paszkiewicz 4893418d036SArtur Paszkiewicz io = list_first_entry_or_null(&log->io_list, struct ppl_io_unit, 4903418d036SArtur Paszkiewicz log_sibling); 4913418d036SArtur Paszkiewicz if (io && io->submitted) 4923418d036SArtur Paszkiewicz io = NULL; 4933418d036SArtur Paszkiewicz 4943418d036SArtur Paszkiewicz spin_unlock_irq(&log->io_list_lock); 4953418d036SArtur Paszkiewicz 4963418d036SArtur Paszkiewicz if (io) { 4973418d036SArtur Paszkiewicz io->submitted = true; 4983418d036SArtur Paszkiewicz 4993418d036SArtur Paszkiewicz if (io == log->current_io) 5003418d036SArtur Paszkiewicz log->current_io = NULL; 5013418d036SArtur Paszkiewicz 5023418d036SArtur Paszkiewicz ppl_submit_iounit(io); 5033418d036SArtur Paszkiewicz } 5043418d036SArtur Paszkiewicz } 5053418d036SArtur Paszkiewicz 5063418d036SArtur Paszkiewicz void ppl_write_stripe_run(struct r5conf *conf) 5073418d036SArtur Paszkiewicz { 5083418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 5093418d036SArtur Paszkiewicz struct ppl_log *log; 5103418d036SArtur Paszkiewicz int i; 5113418d036SArtur Paszkiewicz 5123418d036SArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 5133418d036SArtur Paszkiewicz log = &ppl_conf->child_logs[i]; 5143418d036SArtur Paszkiewicz 5153418d036SArtur Paszkiewicz mutex_lock(&log->io_mutex); 5163418d036SArtur Paszkiewicz ppl_submit_current_io(log); 5173418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 5183418d036SArtur Paszkiewicz } 5193418d036SArtur Paszkiewicz } 5203418d036SArtur Paszkiewicz 5213418d036SArtur Paszkiewicz static void ppl_io_unit_finished(struct ppl_io_unit *io) 5223418d036SArtur Paszkiewicz { 5233418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 52494568f64SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 5253418d036SArtur Paszkiewicz unsigned long flags; 5263418d036SArtur Paszkiewicz 5273418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu\n", __func__, io->seq); 5283418d036SArtur Paszkiewicz 52994568f64SArtur Paszkiewicz local_irq_save(flags); 5303418d036SArtur Paszkiewicz 53194568f64SArtur Paszkiewicz spin_lock(&log->io_list_lock); 5323418d036SArtur Paszkiewicz list_del(&io->log_sibling); 53394568f64SArtur Paszkiewicz spin_unlock(&log->io_list_lock); 5343418d036SArtur Paszkiewicz 53594568f64SArtur Paszkiewicz mempool_free(io, ppl_conf->io_pool); 53694568f64SArtur Paszkiewicz 53794568f64SArtur Paszkiewicz spin_lock(&ppl_conf->no_mem_stripes_lock); 53894568f64SArtur Paszkiewicz if (!list_empty(&ppl_conf->no_mem_stripes)) { 53994568f64SArtur Paszkiewicz struct stripe_head *sh; 54094568f64SArtur Paszkiewicz 54194568f64SArtur Paszkiewicz sh = list_first_entry(&ppl_conf->no_mem_stripes, 54294568f64SArtur Paszkiewicz struct stripe_head, log_list); 5433418d036SArtur Paszkiewicz list_del_init(&sh->log_list); 5443418d036SArtur Paszkiewicz set_bit(STRIPE_HANDLE, &sh->state); 5453418d036SArtur Paszkiewicz raid5_release_stripe(sh); 5463418d036SArtur Paszkiewicz } 54794568f64SArtur Paszkiewicz spin_unlock(&ppl_conf->no_mem_stripes_lock); 5483418d036SArtur Paszkiewicz 54994568f64SArtur Paszkiewicz local_irq_restore(flags); 5503418d036SArtur Paszkiewicz } 5513418d036SArtur Paszkiewicz 5523418d036SArtur Paszkiewicz void ppl_stripe_write_finished(struct stripe_head *sh) 5533418d036SArtur Paszkiewicz { 5543418d036SArtur Paszkiewicz struct ppl_io_unit *io; 5553418d036SArtur Paszkiewicz 5563418d036SArtur Paszkiewicz io = sh->ppl_io; 5573418d036SArtur Paszkiewicz sh->ppl_io = NULL; 5583418d036SArtur Paszkiewicz 5593418d036SArtur Paszkiewicz if (io && atomic_dec_and_test(&io->pending_stripes)) 5603418d036SArtur Paszkiewicz ppl_io_unit_finished(io); 5613418d036SArtur Paszkiewicz } 5623418d036SArtur Paszkiewicz 5634536bf9bSArtur Paszkiewicz static void ppl_xor(int size, struct page *page1, struct page *page2) 5644536bf9bSArtur Paszkiewicz { 5654536bf9bSArtur Paszkiewicz struct async_submit_ctl submit; 5664536bf9bSArtur Paszkiewicz struct dma_async_tx_descriptor *tx; 5674536bf9bSArtur Paszkiewicz struct page *xor_srcs[] = { page1, page2 }; 5684536bf9bSArtur Paszkiewicz 5694536bf9bSArtur Paszkiewicz init_async_submit(&submit, ASYNC_TX_ACK|ASYNC_TX_XOR_DROP_DST, 5704536bf9bSArtur Paszkiewicz NULL, NULL, NULL, NULL); 5714536bf9bSArtur Paszkiewicz tx = async_xor(page1, xor_srcs, 0, 2, size, &submit); 5724536bf9bSArtur Paszkiewicz 5734536bf9bSArtur Paszkiewicz async_tx_quiesce(&tx); 5744536bf9bSArtur Paszkiewicz } 5754536bf9bSArtur Paszkiewicz 5764536bf9bSArtur Paszkiewicz /* 5774536bf9bSArtur Paszkiewicz * PPL recovery strategy: xor partial parity and data from all modified data 5784536bf9bSArtur Paszkiewicz * disks within a stripe and write the result as the new stripe parity. If all 5794536bf9bSArtur Paszkiewicz * stripe data disks are modified (full stripe write), no partial parity is 5804536bf9bSArtur Paszkiewicz * available, so just xor the data disks. 5814536bf9bSArtur Paszkiewicz * 5824536bf9bSArtur Paszkiewicz * Recovery of a PPL entry shall occur only if all modified data disks are 5834536bf9bSArtur Paszkiewicz * available and read from all of them succeeds. 5844536bf9bSArtur Paszkiewicz * 5854536bf9bSArtur Paszkiewicz * A PPL entry applies to a stripe, partial parity size for an entry is at most 5864536bf9bSArtur Paszkiewicz * the size of the chunk. Examples of possible cases for a single entry: 5874536bf9bSArtur Paszkiewicz * 5884536bf9bSArtur Paszkiewicz * case 0: single data disk write: 5894536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 5904536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 5914536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 5924536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | -> | data1 ^ pp | 5934536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | -> | data1 ^ pp | 5944536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 5954536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 5964536bf9bSArtur Paszkiewicz * pp_size = data_size 5974536bf9bSArtur Paszkiewicz * 5984536bf9bSArtur Paszkiewicz * case 1: more than one data disk write: 5994536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 6004536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 6014536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 6024536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp | 6034536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp | 6044536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 6054536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 6064536bf9bSArtur Paszkiewicz * pp_size = data_size / modified_data_disks 6074536bf9bSArtur Paszkiewicz * 6084536bf9bSArtur Paszkiewicz * case 2: write to all data disks (also full stripe write): 6094536bf9bSArtur Paszkiewicz * data0 data1 data2 parity 6104536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 6114536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | (no change) | 6124536bf9bSArtur Paszkiewicz * | -data- | -data- | -data- | --------> | xor all data | 6134536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | --------> | (no change) | 6144536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | (no change) | 6154536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 6164536bf9bSArtur Paszkiewicz * pp_size = 0 6174536bf9bSArtur Paszkiewicz * 6184536bf9bSArtur Paszkiewicz * The following cases are possible only in other implementations. The recovery 6194536bf9bSArtur Paszkiewicz * code can handle them, but they are not generated at runtime because they can 6204536bf9bSArtur Paszkiewicz * be reduced to cases 0, 1 and 2: 6214536bf9bSArtur Paszkiewicz * 6224536bf9bSArtur Paszkiewicz * case 3: 6234536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 6244536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 6254536bf9bSArtur Paszkiewicz * | ------ | -data- | -data- | | pp | | data1 ^ data2 ^ pp | 6264536bf9bSArtur Paszkiewicz * | ------ | -data- | -data- | | pp | -> | data1 ^ data2 ^ pp | 6274536bf9bSArtur Paszkiewicz * | -data- | -data- | -data- | | -- | -> | xor all data | 6284536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | | data0 ^ data1 ^ pp | 6294536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 6304536bf9bSArtur Paszkiewicz * pp_size = chunk_size 6314536bf9bSArtur Paszkiewicz * 6324536bf9bSArtur Paszkiewicz * case 4: 6334536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 6344536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 6354536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | | data1 ^ pp | 6364536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | -- | -> | (no change) | 6374536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | -- | -> | (no change) | 6384536bf9bSArtur Paszkiewicz * | -data- | ------ | ------ | | pp | | data0 ^ pp | 6394536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 6404536bf9bSArtur Paszkiewicz * pp_size = chunk_size 6414536bf9bSArtur Paszkiewicz */ 6424536bf9bSArtur Paszkiewicz static int ppl_recover_entry(struct ppl_log *log, struct ppl_header_entry *e, 6434536bf9bSArtur Paszkiewicz sector_t ppl_sector) 6444536bf9bSArtur Paszkiewicz { 6454536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 6464536bf9bSArtur Paszkiewicz struct mddev *mddev = ppl_conf->mddev; 6474536bf9bSArtur Paszkiewicz struct r5conf *conf = mddev->private; 6484536bf9bSArtur Paszkiewicz int block_size = ppl_conf->block_size; 6494536bf9bSArtur Paszkiewicz struct page *page1; 6504536bf9bSArtur Paszkiewicz struct page *page2; 6514536bf9bSArtur Paszkiewicz sector_t r_sector_first; 6524536bf9bSArtur Paszkiewicz sector_t r_sector_last; 6534536bf9bSArtur Paszkiewicz int strip_sectors; 6544536bf9bSArtur Paszkiewicz int data_disks; 6554536bf9bSArtur Paszkiewicz int i; 6564536bf9bSArtur Paszkiewicz int ret = 0; 6574536bf9bSArtur Paszkiewicz char b[BDEVNAME_SIZE]; 6584536bf9bSArtur Paszkiewicz unsigned int pp_size = le32_to_cpu(e->pp_size); 6594536bf9bSArtur Paszkiewicz unsigned int data_size = le32_to_cpu(e->data_size); 6604536bf9bSArtur Paszkiewicz 6614536bf9bSArtur Paszkiewicz page1 = alloc_page(GFP_KERNEL); 6624536bf9bSArtur Paszkiewicz page2 = alloc_page(GFP_KERNEL); 6634536bf9bSArtur Paszkiewicz 6644536bf9bSArtur Paszkiewicz if (!page1 || !page2) { 6654536bf9bSArtur Paszkiewicz ret = -ENOMEM; 6664536bf9bSArtur Paszkiewicz goto out; 6674536bf9bSArtur Paszkiewicz } 6684536bf9bSArtur Paszkiewicz 6694536bf9bSArtur Paszkiewicz r_sector_first = le64_to_cpu(e->data_sector) * (block_size >> 9); 6704536bf9bSArtur Paszkiewicz 6714536bf9bSArtur Paszkiewicz if ((pp_size >> 9) < conf->chunk_sectors) { 6724536bf9bSArtur Paszkiewicz if (pp_size > 0) { 6734536bf9bSArtur Paszkiewicz data_disks = data_size / pp_size; 6744536bf9bSArtur Paszkiewicz strip_sectors = pp_size >> 9; 6754536bf9bSArtur Paszkiewicz } else { 6764536bf9bSArtur Paszkiewicz data_disks = conf->raid_disks - conf->max_degraded; 6774536bf9bSArtur Paszkiewicz strip_sectors = (data_size >> 9) / data_disks; 6784536bf9bSArtur Paszkiewicz } 6794536bf9bSArtur Paszkiewicz r_sector_last = r_sector_first + 6804536bf9bSArtur Paszkiewicz (data_disks - 1) * conf->chunk_sectors + 6814536bf9bSArtur Paszkiewicz strip_sectors; 6824536bf9bSArtur Paszkiewicz } else { 6834536bf9bSArtur Paszkiewicz data_disks = conf->raid_disks - conf->max_degraded; 6844536bf9bSArtur Paszkiewicz strip_sectors = conf->chunk_sectors; 6854536bf9bSArtur Paszkiewicz r_sector_last = r_sector_first + (data_size >> 9); 6864536bf9bSArtur Paszkiewicz } 6874536bf9bSArtur Paszkiewicz 6884536bf9bSArtur Paszkiewicz pr_debug("%s: array sector first: %llu last: %llu\n", __func__, 6894536bf9bSArtur Paszkiewicz (unsigned long long)r_sector_first, 6904536bf9bSArtur Paszkiewicz (unsigned long long)r_sector_last); 6914536bf9bSArtur Paszkiewicz 6924536bf9bSArtur Paszkiewicz /* if start and end is 4k aligned, use a 4k block */ 6934536bf9bSArtur Paszkiewicz if (block_size == 512 && 6944536bf9bSArtur Paszkiewicz (r_sector_first & (STRIPE_SECTORS - 1)) == 0 && 6954536bf9bSArtur Paszkiewicz (r_sector_last & (STRIPE_SECTORS - 1)) == 0) 6964536bf9bSArtur Paszkiewicz block_size = STRIPE_SIZE; 6974536bf9bSArtur Paszkiewicz 6984536bf9bSArtur Paszkiewicz /* iterate through blocks in strip */ 6994536bf9bSArtur Paszkiewicz for (i = 0; i < strip_sectors; i += (block_size >> 9)) { 7004536bf9bSArtur Paszkiewicz bool update_parity = false; 7014536bf9bSArtur Paszkiewicz sector_t parity_sector; 7024536bf9bSArtur Paszkiewicz struct md_rdev *parity_rdev; 7034536bf9bSArtur Paszkiewicz struct stripe_head sh; 7044536bf9bSArtur Paszkiewicz int disk; 7054536bf9bSArtur Paszkiewicz int indent = 0; 7064536bf9bSArtur Paszkiewicz 7074536bf9bSArtur Paszkiewicz pr_debug("%s:%*s iter %d start\n", __func__, indent, "", i); 7084536bf9bSArtur Paszkiewicz indent += 2; 7094536bf9bSArtur Paszkiewicz 7104536bf9bSArtur Paszkiewicz memset(page_address(page1), 0, PAGE_SIZE); 7114536bf9bSArtur Paszkiewicz 7124536bf9bSArtur Paszkiewicz /* iterate through data member disks */ 7134536bf9bSArtur Paszkiewicz for (disk = 0; disk < data_disks; disk++) { 7144536bf9bSArtur Paszkiewicz int dd_idx; 7154536bf9bSArtur Paszkiewicz struct md_rdev *rdev; 7164536bf9bSArtur Paszkiewicz sector_t sector; 7174536bf9bSArtur Paszkiewicz sector_t r_sector = r_sector_first + i + 7184536bf9bSArtur Paszkiewicz (disk * conf->chunk_sectors); 7194536bf9bSArtur Paszkiewicz 7204536bf9bSArtur Paszkiewicz pr_debug("%s:%*s data member disk %d start\n", 7214536bf9bSArtur Paszkiewicz __func__, indent, "", disk); 7224536bf9bSArtur Paszkiewicz indent += 2; 7234536bf9bSArtur Paszkiewicz 7244536bf9bSArtur Paszkiewicz if (r_sector >= r_sector_last) { 7254536bf9bSArtur Paszkiewicz pr_debug("%s:%*s array sector %llu doesn't need parity update\n", 7264536bf9bSArtur Paszkiewicz __func__, indent, "", 7274536bf9bSArtur Paszkiewicz (unsigned long long)r_sector); 7284536bf9bSArtur Paszkiewicz indent -= 2; 7294536bf9bSArtur Paszkiewicz continue; 7304536bf9bSArtur Paszkiewicz } 7314536bf9bSArtur Paszkiewicz 7324536bf9bSArtur Paszkiewicz update_parity = true; 7334536bf9bSArtur Paszkiewicz 7344536bf9bSArtur Paszkiewicz /* map raid sector to member disk */ 7354536bf9bSArtur Paszkiewicz sector = raid5_compute_sector(conf, r_sector, 0, 7364536bf9bSArtur Paszkiewicz &dd_idx, NULL); 7374536bf9bSArtur Paszkiewicz pr_debug("%s:%*s processing array sector %llu => data member disk %d, sector %llu\n", 7384536bf9bSArtur Paszkiewicz __func__, indent, "", 7394536bf9bSArtur Paszkiewicz (unsigned long long)r_sector, dd_idx, 7404536bf9bSArtur Paszkiewicz (unsigned long long)sector); 7414536bf9bSArtur Paszkiewicz 7424536bf9bSArtur Paszkiewicz rdev = conf->disks[dd_idx].rdev; 7434536bf9bSArtur Paszkiewicz if (!rdev) { 7444536bf9bSArtur Paszkiewicz pr_debug("%s:%*s data member disk %d missing\n", 7454536bf9bSArtur Paszkiewicz __func__, indent, "", dd_idx); 7464536bf9bSArtur Paszkiewicz update_parity = false; 7474536bf9bSArtur Paszkiewicz break; 7484536bf9bSArtur Paszkiewicz } 7494536bf9bSArtur Paszkiewicz 7504536bf9bSArtur Paszkiewicz pr_debug("%s:%*s reading data member disk %s sector %llu\n", 7514536bf9bSArtur Paszkiewicz __func__, indent, "", bdevname(rdev->bdev, b), 7524536bf9bSArtur Paszkiewicz (unsigned long long)sector); 7534536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, sector, block_size, page2, 7544536bf9bSArtur Paszkiewicz REQ_OP_READ, 0, false)) { 7554536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 7564536bf9bSArtur Paszkiewicz pr_debug("%s:%*s read failed!\n", __func__, 7574536bf9bSArtur Paszkiewicz indent, ""); 7584536bf9bSArtur Paszkiewicz ret = -EIO; 7594536bf9bSArtur Paszkiewicz goto out; 7604536bf9bSArtur Paszkiewicz } 7614536bf9bSArtur Paszkiewicz 7624536bf9bSArtur Paszkiewicz ppl_xor(block_size, page1, page2); 7634536bf9bSArtur Paszkiewicz 7644536bf9bSArtur Paszkiewicz indent -= 2; 7654536bf9bSArtur Paszkiewicz } 7664536bf9bSArtur Paszkiewicz 7674536bf9bSArtur Paszkiewicz if (!update_parity) 7684536bf9bSArtur Paszkiewicz continue; 7694536bf9bSArtur Paszkiewicz 7704536bf9bSArtur Paszkiewicz if (pp_size > 0) { 7714536bf9bSArtur Paszkiewicz pr_debug("%s:%*s reading pp disk sector %llu\n", 7724536bf9bSArtur Paszkiewicz __func__, indent, "", 7734536bf9bSArtur Paszkiewicz (unsigned long long)(ppl_sector + i)); 7744536bf9bSArtur Paszkiewicz if (!sync_page_io(log->rdev, 7754536bf9bSArtur Paszkiewicz ppl_sector - log->rdev->data_offset + i, 7764536bf9bSArtur Paszkiewicz block_size, page2, REQ_OP_READ, 0, 7774536bf9bSArtur Paszkiewicz false)) { 7784536bf9bSArtur Paszkiewicz pr_debug("%s:%*s read failed!\n", __func__, 7794536bf9bSArtur Paszkiewicz indent, ""); 7804536bf9bSArtur Paszkiewicz md_error(mddev, log->rdev); 7814536bf9bSArtur Paszkiewicz ret = -EIO; 7824536bf9bSArtur Paszkiewicz goto out; 7834536bf9bSArtur Paszkiewicz } 7844536bf9bSArtur Paszkiewicz 7854536bf9bSArtur Paszkiewicz ppl_xor(block_size, page1, page2); 7864536bf9bSArtur Paszkiewicz } 7874536bf9bSArtur Paszkiewicz 7884536bf9bSArtur Paszkiewicz /* map raid sector to parity disk */ 7894536bf9bSArtur Paszkiewicz parity_sector = raid5_compute_sector(conf, r_sector_first + i, 7904536bf9bSArtur Paszkiewicz 0, &disk, &sh); 7914536bf9bSArtur Paszkiewicz BUG_ON(sh.pd_idx != le32_to_cpu(e->parity_disk)); 7924536bf9bSArtur Paszkiewicz parity_rdev = conf->disks[sh.pd_idx].rdev; 7934536bf9bSArtur Paszkiewicz 7944536bf9bSArtur Paszkiewicz BUG_ON(parity_rdev->bdev->bd_dev != log->rdev->bdev->bd_dev); 7954536bf9bSArtur Paszkiewicz pr_debug("%s:%*s write parity at sector %llu, disk %s\n", 7964536bf9bSArtur Paszkiewicz __func__, indent, "", 7974536bf9bSArtur Paszkiewicz (unsigned long long)parity_sector, 7984536bf9bSArtur Paszkiewicz bdevname(parity_rdev->bdev, b)); 7994536bf9bSArtur Paszkiewicz if (!sync_page_io(parity_rdev, parity_sector, block_size, 8004536bf9bSArtur Paszkiewicz page1, REQ_OP_WRITE, 0, false)) { 8014536bf9bSArtur Paszkiewicz pr_debug("%s:%*s parity write error!\n", __func__, 8024536bf9bSArtur Paszkiewicz indent, ""); 8034536bf9bSArtur Paszkiewicz md_error(mddev, parity_rdev); 8044536bf9bSArtur Paszkiewicz ret = -EIO; 8054536bf9bSArtur Paszkiewicz goto out; 8064536bf9bSArtur Paszkiewicz } 8074536bf9bSArtur Paszkiewicz } 8084536bf9bSArtur Paszkiewicz out: 8094536bf9bSArtur Paszkiewicz if (page1) 8104536bf9bSArtur Paszkiewicz __free_page(page1); 8114536bf9bSArtur Paszkiewicz if (page2) 8124536bf9bSArtur Paszkiewicz __free_page(page2); 8134536bf9bSArtur Paszkiewicz return ret; 8144536bf9bSArtur Paszkiewicz } 8154536bf9bSArtur Paszkiewicz 8164536bf9bSArtur Paszkiewicz static int ppl_recover(struct ppl_log *log, struct ppl_header *pplhdr) 8174536bf9bSArtur Paszkiewicz { 8184536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 8194536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 8204536bf9bSArtur Paszkiewicz struct mddev *mddev = rdev->mddev; 8214536bf9bSArtur Paszkiewicz sector_t ppl_sector = rdev->ppl.sector + (PPL_HEADER_SIZE >> 9); 8224536bf9bSArtur Paszkiewicz struct page *page; 8234536bf9bSArtur Paszkiewicz int i; 8244536bf9bSArtur Paszkiewicz int ret = 0; 8254536bf9bSArtur Paszkiewicz 8264536bf9bSArtur Paszkiewicz page = alloc_page(GFP_KERNEL); 8274536bf9bSArtur Paszkiewicz if (!page) 8284536bf9bSArtur Paszkiewicz return -ENOMEM; 8294536bf9bSArtur Paszkiewicz 8304536bf9bSArtur Paszkiewicz /* iterate through all PPL entries saved */ 8314536bf9bSArtur Paszkiewicz for (i = 0; i < le32_to_cpu(pplhdr->entries_count); i++) { 8324536bf9bSArtur Paszkiewicz struct ppl_header_entry *e = &pplhdr->entries[i]; 8334536bf9bSArtur Paszkiewicz u32 pp_size = le32_to_cpu(e->pp_size); 8344536bf9bSArtur Paszkiewicz sector_t sector = ppl_sector; 8354536bf9bSArtur Paszkiewicz int ppl_entry_sectors = pp_size >> 9; 8364536bf9bSArtur Paszkiewicz u32 crc, crc_stored; 8374536bf9bSArtur Paszkiewicz 8384536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d entry: %d ppl_sector: %llu pp_size: %u\n", 8394536bf9bSArtur Paszkiewicz __func__, rdev->raid_disk, i, 8404536bf9bSArtur Paszkiewicz (unsigned long long)ppl_sector, pp_size); 8414536bf9bSArtur Paszkiewicz 8424536bf9bSArtur Paszkiewicz crc = ~0; 8434536bf9bSArtur Paszkiewicz crc_stored = le32_to_cpu(e->checksum); 8444536bf9bSArtur Paszkiewicz 8454536bf9bSArtur Paszkiewicz /* read parial parity for this entry and calculate its checksum */ 8464536bf9bSArtur Paszkiewicz while (pp_size) { 8474536bf9bSArtur Paszkiewicz int s = pp_size > PAGE_SIZE ? PAGE_SIZE : pp_size; 8484536bf9bSArtur Paszkiewicz 8494536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, sector - rdev->data_offset, 8504536bf9bSArtur Paszkiewicz s, page, REQ_OP_READ, 0, false)) { 8514536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 8524536bf9bSArtur Paszkiewicz ret = -EIO; 8534536bf9bSArtur Paszkiewicz goto out; 8544536bf9bSArtur Paszkiewicz } 8554536bf9bSArtur Paszkiewicz 8564536bf9bSArtur Paszkiewicz crc = crc32c_le(crc, page_address(page), s); 8574536bf9bSArtur Paszkiewicz 8584536bf9bSArtur Paszkiewicz pp_size -= s; 8594536bf9bSArtur Paszkiewicz sector += s >> 9; 8604536bf9bSArtur Paszkiewicz } 8614536bf9bSArtur Paszkiewicz 8624536bf9bSArtur Paszkiewicz crc = ~crc; 8634536bf9bSArtur Paszkiewicz 8644536bf9bSArtur Paszkiewicz if (crc != crc_stored) { 8654536bf9bSArtur Paszkiewicz /* 8664536bf9bSArtur Paszkiewicz * Don't recover this entry if the checksum does not 8674536bf9bSArtur Paszkiewicz * match, but keep going and try to recover other 8684536bf9bSArtur Paszkiewicz * entries. 8694536bf9bSArtur Paszkiewicz */ 8704536bf9bSArtur Paszkiewicz pr_debug("%s: ppl entry crc does not match: stored: 0x%x calculated: 0x%x\n", 8714536bf9bSArtur Paszkiewicz __func__, crc_stored, crc); 8724536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count++; 8734536bf9bSArtur Paszkiewicz } else { 8744536bf9bSArtur Paszkiewicz ret = ppl_recover_entry(log, e, ppl_sector); 8754536bf9bSArtur Paszkiewicz if (ret) 8764536bf9bSArtur Paszkiewicz goto out; 8774536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries++; 8784536bf9bSArtur Paszkiewicz } 8794536bf9bSArtur Paszkiewicz 8804536bf9bSArtur Paszkiewicz ppl_sector += ppl_entry_sectors; 8814536bf9bSArtur Paszkiewicz } 8824536bf9bSArtur Paszkiewicz 8834536bf9bSArtur Paszkiewicz /* flush the disk cache after recovery if necessary */ 8844536bf9bSArtur Paszkiewicz ret = blkdev_issue_flush(rdev->bdev, GFP_KERNEL, NULL); 8854536bf9bSArtur Paszkiewicz out: 8864536bf9bSArtur Paszkiewicz __free_page(page); 8874536bf9bSArtur Paszkiewicz return ret; 8884536bf9bSArtur Paszkiewicz } 8894536bf9bSArtur Paszkiewicz 8904536bf9bSArtur Paszkiewicz static int ppl_write_empty_header(struct ppl_log *log) 8914536bf9bSArtur Paszkiewicz { 8924536bf9bSArtur Paszkiewicz struct page *page; 8934536bf9bSArtur Paszkiewicz struct ppl_header *pplhdr; 8944536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 8954536bf9bSArtur Paszkiewicz int ret = 0; 8964536bf9bSArtur Paszkiewicz 8974536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d ppl_sector: %llu\n", __func__, 8984536bf9bSArtur Paszkiewicz rdev->raid_disk, (unsigned long long)rdev->ppl.sector); 8994536bf9bSArtur Paszkiewicz 9004536bf9bSArtur Paszkiewicz page = alloc_page(GFP_NOIO | __GFP_ZERO); 9014536bf9bSArtur Paszkiewicz if (!page) 9024536bf9bSArtur Paszkiewicz return -ENOMEM; 9034536bf9bSArtur Paszkiewicz 9044536bf9bSArtur Paszkiewicz pplhdr = page_address(page); 9054536bf9bSArtur Paszkiewicz memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED); 9064536bf9bSArtur Paszkiewicz pplhdr->signature = cpu_to_le32(log->ppl_conf->signature); 9074536bf9bSArtur Paszkiewicz pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PAGE_SIZE)); 9084536bf9bSArtur Paszkiewicz 9094536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, rdev->ppl.sector - rdev->data_offset, 9105a8948f8SJan Kara PPL_HEADER_SIZE, page, REQ_OP_WRITE | REQ_SYNC | 9115a8948f8SJan Kara REQ_FUA, 0, false)) { 9124536bf9bSArtur Paszkiewicz md_error(rdev->mddev, rdev); 9134536bf9bSArtur Paszkiewicz ret = -EIO; 9144536bf9bSArtur Paszkiewicz } 9154536bf9bSArtur Paszkiewicz 9164536bf9bSArtur Paszkiewicz __free_page(page); 9174536bf9bSArtur Paszkiewicz return ret; 9184536bf9bSArtur Paszkiewicz } 9194536bf9bSArtur Paszkiewicz 9204536bf9bSArtur Paszkiewicz static int ppl_load_distributed(struct ppl_log *log) 9214536bf9bSArtur Paszkiewicz { 9224536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 9234536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 9244536bf9bSArtur Paszkiewicz struct mddev *mddev = rdev->mddev; 9254536bf9bSArtur Paszkiewicz struct page *page; 9264536bf9bSArtur Paszkiewicz struct ppl_header *pplhdr; 9274536bf9bSArtur Paszkiewicz u32 crc, crc_stored; 9284536bf9bSArtur Paszkiewicz u32 signature; 9294536bf9bSArtur Paszkiewicz int ret = 0; 9304536bf9bSArtur Paszkiewicz 9314536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d\n", __func__, rdev->raid_disk); 9324536bf9bSArtur Paszkiewicz 9334536bf9bSArtur Paszkiewicz /* read PPL header */ 9344536bf9bSArtur Paszkiewicz page = alloc_page(GFP_KERNEL); 9354536bf9bSArtur Paszkiewicz if (!page) 9364536bf9bSArtur Paszkiewicz return -ENOMEM; 9374536bf9bSArtur Paszkiewicz 9384536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, rdev->ppl.sector - rdev->data_offset, 9394536bf9bSArtur Paszkiewicz PAGE_SIZE, page, REQ_OP_READ, 0, false)) { 9404536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 9414536bf9bSArtur Paszkiewicz ret = -EIO; 9424536bf9bSArtur Paszkiewicz goto out; 9434536bf9bSArtur Paszkiewicz } 9444536bf9bSArtur Paszkiewicz pplhdr = page_address(page); 9454536bf9bSArtur Paszkiewicz 9464536bf9bSArtur Paszkiewicz /* check header validity */ 9474536bf9bSArtur Paszkiewicz crc_stored = le32_to_cpu(pplhdr->checksum); 9484536bf9bSArtur Paszkiewicz pplhdr->checksum = 0; 9494536bf9bSArtur Paszkiewicz crc = ~crc32c_le(~0, pplhdr, PAGE_SIZE); 9504536bf9bSArtur Paszkiewicz 9514536bf9bSArtur Paszkiewicz if (crc_stored != crc) { 9524536bf9bSArtur Paszkiewicz pr_debug("%s: ppl header crc does not match: stored: 0x%x calculated: 0x%x\n", 9534536bf9bSArtur Paszkiewicz __func__, crc_stored, crc); 9544536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count++; 9554536bf9bSArtur Paszkiewicz goto out; 9564536bf9bSArtur Paszkiewicz } 9574536bf9bSArtur Paszkiewicz 9584536bf9bSArtur Paszkiewicz signature = le32_to_cpu(pplhdr->signature); 9594536bf9bSArtur Paszkiewicz 9604536bf9bSArtur Paszkiewicz if (mddev->external) { 9614536bf9bSArtur Paszkiewicz /* 9624536bf9bSArtur Paszkiewicz * For external metadata the header signature is set and 9634536bf9bSArtur Paszkiewicz * validated in userspace. 9644536bf9bSArtur Paszkiewicz */ 9654536bf9bSArtur Paszkiewicz ppl_conf->signature = signature; 9664536bf9bSArtur Paszkiewicz } else if (ppl_conf->signature != signature) { 9674536bf9bSArtur Paszkiewicz pr_debug("%s: ppl header signature does not match: stored: 0x%x configured: 0x%x\n", 9684536bf9bSArtur Paszkiewicz __func__, signature, ppl_conf->signature); 9694536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count++; 9704536bf9bSArtur Paszkiewicz goto out; 9714536bf9bSArtur Paszkiewicz } 9724536bf9bSArtur Paszkiewicz 9734536bf9bSArtur Paszkiewicz /* attempt to recover from log if we are starting a dirty array */ 9744536bf9bSArtur Paszkiewicz if (!mddev->pers && mddev->recovery_cp != MaxSector) 9754536bf9bSArtur Paszkiewicz ret = ppl_recover(log, pplhdr); 9764536bf9bSArtur Paszkiewicz out: 9774536bf9bSArtur Paszkiewicz /* write empty header if we are starting the array */ 9784536bf9bSArtur Paszkiewicz if (!ret && !mddev->pers) 9794536bf9bSArtur Paszkiewicz ret = ppl_write_empty_header(log); 9804536bf9bSArtur Paszkiewicz 9814536bf9bSArtur Paszkiewicz __free_page(page); 9824536bf9bSArtur Paszkiewicz 9834536bf9bSArtur Paszkiewicz pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n", 9844536bf9bSArtur Paszkiewicz __func__, ret, ppl_conf->mismatch_count, 9854536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries); 9864536bf9bSArtur Paszkiewicz return ret; 9874536bf9bSArtur Paszkiewicz } 9884536bf9bSArtur Paszkiewicz 9894536bf9bSArtur Paszkiewicz static int ppl_load(struct ppl_conf *ppl_conf) 9904536bf9bSArtur Paszkiewicz { 9914536bf9bSArtur Paszkiewicz int ret = 0; 9924536bf9bSArtur Paszkiewicz u32 signature = 0; 9934536bf9bSArtur Paszkiewicz bool signature_set = false; 9944536bf9bSArtur Paszkiewicz int i; 9954536bf9bSArtur Paszkiewicz 9964536bf9bSArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 9974536bf9bSArtur Paszkiewicz struct ppl_log *log = &ppl_conf->child_logs[i]; 9984536bf9bSArtur Paszkiewicz 9994536bf9bSArtur Paszkiewicz /* skip missing drive */ 10004536bf9bSArtur Paszkiewicz if (!log->rdev) 10014536bf9bSArtur Paszkiewicz continue; 10024536bf9bSArtur Paszkiewicz 10034536bf9bSArtur Paszkiewicz ret = ppl_load_distributed(log); 10044536bf9bSArtur Paszkiewicz if (ret) 10054536bf9bSArtur Paszkiewicz break; 10064536bf9bSArtur Paszkiewicz 10074536bf9bSArtur Paszkiewicz /* 10084536bf9bSArtur Paszkiewicz * For external metadata we can't check if the signature is 10094536bf9bSArtur Paszkiewicz * correct on a single drive, but we can check if it is the same 10104536bf9bSArtur Paszkiewicz * on all drives. 10114536bf9bSArtur Paszkiewicz */ 10124536bf9bSArtur Paszkiewicz if (ppl_conf->mddev->external) { 10134536bf9bSArtur Paszkiewicz if (!signature_set) { 10144536bf9bSArtur Paszkiewicz signature = ppl_conf->signature; 10154536bf9bSArtur Paszkiewicz signature_set = true; 10164536bf9bSArtur Paszkiewicz } else if (signature != ppl_conf->signature) { 10174536bf9bSArtur Paszkiewicz pr_warn("md/raid:%s: PPL header signature does not match on all member drives\n", 10184536bf9bSArtur Paszkiewicz mdname(ppl_conf->mddev)); 10194536bf9bSArtur Paszkiewicz ret = -EINVAL; 10204536bf9bSArtur Paszkiewicz break; 10214536bf9bSArtur Paszkiewicz } 10224536bf9bSArtur Paszkiewicz } 10234536bf9bSArtur Paszkiewicz } 10244536bf9bSArtur Paszkiewicz 10254536bf9bSArtur Paszkiewicz pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n", 10264536bf9bSArtur Paszkiewicz __func__, ret, ppl_conf->mismatch_count, 10274536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries); 10284536bf9bSArtur Paszkiewicz return ret; 10294536bf9bSArtur Paszkiewicz } 10304536bf9bSArtur Paszkiewicz 10313418d036SArtur Paszkiewicz static void __ppl_exit_log(struct ppl_conf *ppl_conf) 10323418d036SArtur Paszkiewicz { 10333418d036SArtur Paszkiewicz clear_bit(MD_HAS_PPL, &ppl_conf->mddev->flags); 10343418d036SArtur Paszkiewicz 10353418d036SArtur Paszkiewicz kfree(ppl_conf->child_logs); 10363418d036SArtur Paszkiewicz 10373418d036SArtur Paszkiewicz if (ppl_conf->bs) 10383418d036SArtur Paszkiewicz bioset_free(ppl_conf->bs); 10393418d036SArtur Paszkiewicz mempool_destroy(ppl_conf->io_pool); 10403418d036SArtur Paszkiewicz kmem_cache_destroy(ppl_conf->io_kc); 10413418d036SArtur Paszkiewicz 10423418d036SArtur Paszkiewicz kfree(ppl_conf); 10433418d036SArtur Paszkiewicz } 10443418d036SArtur Paszkiewicz 10453418d036SArtur Paszkiewicz void ppl_exit_log(struct r5conf *conf) 10463418d036SArtur Paszkiewicz { 10473418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 10483418d036SArtur Paszkiewicz 10493418d036SArtur Paszkiewicz if (ppl_conf) { 10503418d036SArtur Paszkiewicz __ppl_exit_log(ppl_conf); 10513418d036SArtur Paszkiewicz conf->log_private = NULL; 10523418d036SArtur Paszkiewicz } 10533418d036SArtur Paszkiewicz } 10543418d036SArtur Paszkiewicz 10553418d036SArtur Paszkiewicz static int ppl_validate_rdev(struct md_rdev *rdev) 10563418d036SArtur Paszkiewicz { 10573418d036SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 10583418d036SArtur Paszkiewicz int ppl_data_sectors; 10593418d036SArtur Paszkiewicz int ppl_size_new; 10603418d036SArtur Paszkiewicz 10613418d036SArtur Paszkiewicz /* 10623418d036SArtur Paszkiewicz * The configured PPL size must be enough to store 10633418d036SArtur Paszkiewicz * the header and (at the very least) partial parity 10643418d036SArtur Paszkiewicz * for one stripe. Round it down to ensure the data 10653418d036SArtur Paszkiewicz * space is cleanly divisible by stripe size. 10663418d036SArtur Paszkiewicz */ 10673418d036SArtur Paszkiewicz ppl_data_sectors = rdev->ppl.size - (PPL_HEADER_SIZE >> 9); 10683418d036SArtur Paszkiewicz 10693418d036SArtur Paszkiewicz if (ppl_data_sectors > 0) 10703418d036SArtur Paszkiewicz ppl_data_sectors = rounddown(ppl_data_sectors, STRIPE_SECTORS); 10713418d036SArtur Paszkiewicz 10723418d036SArtur Paszkiewicz if (ppl_data_sectors <= 0) { 10733418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space too small on %s\n", 10743418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 10753418d036SArtur Paszkiewicz return -ENOSPC; 10763418d036SArtur Paszkiewicz } 10773418d036SArtur Paszkiewicz 10783418d036SArtur Paszkiewicz ppl_size_new = ppl_data_sectors + (PPL_HEADER_SIZE >> 9); 10793418d036SArtur Paszkiewicz 10803418d036SArtur Paszkiewicz if ((rdev->ppl.sector < rdev->data_offset && 10813418d036SArtur Paszkiewicz rdev->ppl.sector + ppl_size_new > rdev->data_offset) || 10823418d036SArtur Paszkiewicz (rdev->ppl.sector >= rdev->data_offset && 10833418d036SArtur Paszkiewicz rdev->data_offset + rdev->sectors > rdev->ppl.sector)) { 10843418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space overlaps with data on %s\n", 10853418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 10863418d036SArtur Paszkiewicz return -EINVAL; 10873418d036SArtur Paszkiewicz } 10883418d036SArtur Paszkiewicz 10893418d036SArtur Paszkiewicz if (!rdev->mddev->external && 10903418d036SArtur Paszkiewicz ((rdev->ppl.offset > 0 && rdev->ppl.offset < (rdev->sb_size >> 9)) || 10913418d036SArtur Paszkiewicz (rdev->ppl.offset <= 0 && rdev->ppl.offset + ppl_size_new > 0))) { 10923418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space overlaps with superblock on %s\n", 10933418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 10943418d036SArtur Paszkiewicz return -EINVAL; 10953418d036SArtur Paszkiewicz } 10963418d036SArtur Paszkiewicz 10973418d036SArtur Paszkiewicz rdev->ppl.size = ppl_size_new; 10983418d036SArtur Paszkiewicz 10993418d036SArtur Paszkiewicz return 0; 11003418d036SArtur Paszkiewicz } 11013418d036SArtur Paszkiewicz 11023418d036SArtur Paszkiewicz int ppl_init_log(struct r5conf *conf) 11033418d036SArtur Paszkiewicz { 11043418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf; 11053418d036SArtur Paszkiewicz struct mddev *mddev = conf->mddev; 11063418d036SArtur Paszkiewicz int ret = 0; 11073418d036SArtur Paszkiewicz int i; 11080b408bafSDan Carpenter bool need_cache_flush = false; 11093418d036SArtur Paszkiewicz 11103418d036SArtur Paszkiewicz pr_debug("md/raid:%s: enabling distributed Partial Parity Log\n", 11113418d036SArtur Paszkiewicz mdname(conf->mddev)); 11123418d036SArtur Paszkiewicz 11133418d036SArtur Paszkiewicz if (PAGE_SIZE != 4096) 11143418d036SArtur Paszkiewicz return -EINVAL; 11153418d036SArtur Paszkiewicz 11163418d036SArtur Paszkiewicz if (mddev->level != 5) { 11173418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with raid level %d\n", 11183418d036SArtur Paszkiewicz mdname(mddev), mddev->level); 11193418d036SArtur Paszkiewicz return -EINVAL; 11203418d036SArtur Paszkiewicz } 11213418d036SArtur Paszkiewicz 11223418d036SArtur Paszkiewicz if (mddev->bitmap_info.file || mddev->bitmap_info.offset) { 11233418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with bitmap\n", 11243418d036SArtur Paszkiewicz mdname(mddev)); 11253418d036SArtur Paszkiewicz return -EINVAL; 11263418d036SArtur Paszkiewicz } 11273418d036SArtur Paszkiewicz 11283418d036SArtur Paszkiewicz if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { 11293418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with journal\n", 11303418d036SArtur Paszkiewicz mdname(mddev)); 11313418d036SArtur Paszkiewicz return -EINVAL; 11323418d036SArtur Paszkiewicz } 11333418d036SArtur Paszkiewicz 11343418d036SArtur Paszkiewicz ppl_conf = kzalloc(sizeof(struct ppl_conf), GFP_KERNEL); 11353418d036SArtur Paszkiewicz if (!ppl_conf) 11363418d036SArtur Paszkiewicz return -ENOMEM; 11373418d036SArtur Paszkiewicz 11383418d036SArtur Paszkiewicz ppl_conf->mddev = mddev; 11393418d036SArtur Paszkiewicz 11403418d036SArtur Paszkiewicz ppl_conf->io_kc = KMEM_CACHE(ppl_io_unit, 0); 11413418d036SArtur Paszkiewicz if (!ppl_conf->io_kc) { 1142fcd403afSArtur Paszkiewicz ret = -ENOMEM; 11433418d036SArtur Paszkiewicz goto err; 11443418d036SArtur Paszkiewicz } 11453418d036SArtur Paszkiewicz 1146fcd403afSArtur Paszkiewicz ppl_conf->io_pool = mempool_create(conf->raid_disks, ppl_io_pool_alloc, 1147fcd403afSArtur Paszkiewicz ppl_io_pool_free, ppl_conf->io_kc); 11483418d036SArtur Paszkiewicz if (!ppl_conf->io_pool) { 1149fcd403afSArtur Paszkiewicz ret = -ENOMEM; 11503418d036SArtur Paszkiewicz goto err; 11513418d036SArtur Paszkiewicz } 11523418d036SArtur Paszkiewicz 11536409e84eSArtur Paszkiewicz ppl_conf->bs = bioset_create(conf->raid_disks, 0, BIOSET_NEED_BVECS); 11543418d036SArtur Paszkiewicz if (!ppl_conf->bs) { 1155fcd403afSArtur Paszkiewicz ret = -ENOMEM; 11563418d036SArtur Paszkiewicz goto err; 11573418d036SArtur Paszkiewicz } 11583418d036SArtur Paszkiewicz 11593418d036SArtur Paszkiewicz ppl_conf->count = conf->raid_disks; 11603418d036SArtur Paszkiewicz ppl_conf->child_logs = kcalloc(ppl_conf->count, sizeof(struct ppl_log), 11613418d036SArtur Paszkiewicz GFP_KERNEL); 11623418d036SArtur Paszkiewicz if (!ppl_conf->child_logs) { 11633418d036SArtur Paszkiewicz ret = -ENOMEM; 11643418d036SArtur Paszkiewicz goto err; 11653418d036SArtur Paszkiewicz } 11663418d036SArtur Paszkiewicz 11673418d036SArtur Paszkiewicz atomic64_set(&ppl_conf->seq, 0); 116894568f64SArtur Paszkiewicz INIT_LIST_HEAD(&ppl_conf->no_mem_stripes); 116994568f64SArtur Paszkiewicz spin_lock_init(&ppl_conf->no_mem_stripes_lock); 11703418d036SArtur Paszkiewicz 11713418d036SArtur Paszkiewicz if (!mddev->external) { 11723418d036SArtur Paszkiewicz ppl_conf->signature = ~crc32c_le(~0, mddev->uuid, sizeof(mddev->uuid)); 11733418d036SArtur Paszkiewicz ppl_conf->block_size = 512; 11743418d036SArtur Paszkiewicz } else { 11753418d036SArtur Paszkiewicz ppl_conf->block_size = queue_logical_block_size(mddev->queue); 11763418d036SArtur Paszkiewicz } 11773418d036SArtur Paszkiewicz 11783418d036SArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 11793418d036SArtur Paszkiewicz struct ppl_log *log = &ppl_conf->child_logs[i]; 11803418d036SArtur Paszkiewicz struct md_rdev *rdev = conf->disks[i].rdev; 11813418d036SArtur Paszkiewicz 11823418d036SArtur Paszkiewicz mutex_init(&log->io_mutex); 11833418d036SArtur Paszkiewicz spin_lock_init(&log->io_list_lock); 11843418d036SArtur Paszkiewicz INIT_LIST_HEAD(&log->io_list); 11853418d036SArtur Paszkiewicz 11863418d036SArtur Paszkiewicz log->ppl_conf = ppl_conf; 11873418d036SArtur Paszkiewicz log->rdev = rdev; 11883418d036SArtur Paszkiewicz 11893418d036SArtur Paszkiewicz if (rdev) { 11903418d036SArtur Paszkiewicz struct request_queue *q; 11913418d036SArtur Paszkiewicz 11923418d036SArtur Paszkiewicz ret = ppl_validate_rdev(rdev); 11933418d036SArtur Paszkiewicz if (ret) 11943418d036SArtur Paszkiewicz goto err; 11953418d036SArtur Paszkiewicz 11963418d036SArtur Paszkiewicz q = bdev_get_queue(rdev->bdev); 11973418d036SArtur Paszkiewicz if (test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 11983418d036SArtur Paszkiewicz need_cache_flush = true; 11993418d036SArtur Paszkiewicz } 12003418d036SArtur Paszkiewicz } 12013418d036SArtur Paszkiewicz 12023418d036SArtur Paszkiewicz if (need_cache_flush) 12033418d036SArtur Paszkiewicz pr_warn("md/raid:%s: Volatile write-back cache should be disabled on all member drives when using PPL!\n", 12043418d036SArtur Paszkiewicz mdname(mddev)); 12053418d036SArtur Paszkiewicz 12064536bf9bSArtur Paszkiewicz /* load and possibly recover the logs from the member disks */ 12074536bf9bSArtur Paszkiewicz ret = ppl_load(ppl_conf); 12084536bf9bSArtur Paszkiewicz 12094536bf9bSArtur Paszkiewicz if (ret) { 12104536bf9bSArtur Paszkiewicz goto err; 12114536bf9bSArtur Paszkiewicz } else if (!mddev->pers && 12124536bf9bSArtur Paszkiewicz mddev->recovery_cp == 0 && !mddev->degraded && 12134536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries > 0 && 12144536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count == 0) { 12154536bf9bSArtur Paszkiewicz /* 12164536bf9bSArtur Paszkiewicz * If we are starting a dirty array and the recovery succeeds 12174536bf9bSArtur Paszkiewicz * without any issues, set the array as clean. 12184536bf9bSArtur Paszkiewicz */ 12194536bf9bSArtur Paszkiewicz mddev->recovery_cp = MaxSector; 12204536bf9bSArtur Paszkiewicz set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); 1221ba903a3eSArtur Paszkiewicz } else if (mddev->pers && ppl_conf->mismatch_count > 0) { 1222ba903a3eSArtur Paszkiewicz /* no mismatch allowed when enabling PPL for a running array */ 1223ba903a3eSArtur Paszkiewicz ret = -EINVAL; 1224ba903a3eSArtur Paszkiewicz goto err; 12254536bf9bSArtur Paszkiewicz } 12264536bf9bSArtur Paszkiewicz 12273418d036SArtur Paszkiewicz conf->log_private = ppl_conf; 1228845b9e22SArtur Paszkiewicz set_bit(MD_HAS_PPL, &ppl_conf->mddev->flags); 12293418d036SArtur Paszkiewicz 12303418d036SArtur Paszkiewicz return 0; 12313418d036SArtur Paszkiewicz err: 12323418d036SArtur Paszkiewicz __ppl_exit_log(ppl_conf); 12333418d036SArtur Paszkiewicz return ret; 12343418d036SArtur Paszkiewicz } 12356358c239SArtur Paszkiewicz 12366358c239SArtur Paszkiewicz int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add) 12376358c239SArtur Paszkiewicz { 12386358c239SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 12396358c239SArtur Paszkiewicz struct ppl_log *log; 12406358c239SArtur Paszkiewicz int ret = 0; 12416358c239SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 12426358c239SArtur Paszkiewicz 12436358c239SArtur Paszkiewicz if (!rdev) 12446358c239SArtur Paszkiewicz return -EINVAL; 12456358c239SArtur Paszkiewicz 12466358c239SArtur Paszkiewicz pr_debug("%s: disk: %d operation: %s dev: %s\n", 12476358c239SArtur Paszkiewicz __func__, rdev->raid_disk, add ? "add" : "remove", 12486358c239SArtur Paszkiewicz bdevname(rdev->bdev, b)); 12496358c239SArtur Paszkiewicz 12506358c239SArtur Paszkiewicz if (rdev->raid_disk < 0) 12516358c239SArtur Paszkiewicz return 0; 12526358c239SArtur Paszkiewicz 12536358c239SArtur Paszkiewicz if (rdev->raid_disk >= ppl_conf->count) 12546358c239SArtur Paszkiewicz return -ENODEV; 12556358c239SArtur Paszkiewicz 12566358c239SArtur Paszkiewicz log = &ppl_conf->child_logs[rdev->raid_disk]; 12576358c239SArtur Paszkiewicz 12586358c239SArtur Paszkiewicz mutex_lock(&log->io_mutex); 12596358c239SArtur Paszkiewicz if (add) { 12606358c239SArtur Paszkiewicz ret = ppl_validate_rdev(rdev); 12616358c239SArtur Paszkiewicz if (!ret) { 12626358c239SArtur Paszkiewicz log->rdev = rdev; 12636358c239SArtur Paszkiewicz ret = ppl_write_empty_header(log); 12646358c239SArtur Paszkiewicz } 12656358c239SArtur Paszkiewicz } else { 12666358c239SArtur Paszkiewicz log->rdev = NULL; 12676358c239SArtur Paszkiewicz } 12686358c239SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 12696358c239SArtur Paszkiewicz 12706358c239SArtur Paszkiewicz return ret; 12716358c239SArtur Paszkiewicz } 1272