12025cf9eSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 23418d036SArtur Paszkiewicz /* 33418d036SArtur Paszkiewicz * Partial Parity Log for closing the RAID5 write hole 43418d036SArtur Paszkiewicz * Copyright (c) 2017, Intel Corporation. 53418d036SArtur Paszkiewicz */ 63418d036SArtur Paszkiewicz 73418d036SArtur Paszkiewicz #include <linux/kernel.h> 83418d036SArtur Paszkiewicz #include <linux/blkdev.h> 93418d036SArtur Paszkiewicz #include <linux/slab.h> 103418d036SArtur Paszkiewicz #include <linux/crc32c.h> 113418d036SArtur Paszkiewicz #include <linux/async_tx.h> 123418d036SArtur Paszkiewicz #include <linux/raid/md_p.h> 133418d036SArtur Paszkiewicz #include "md.h" 143418d036SArtur Paszkiewicz #include "raid5.h" 15a596d086SMariusz Dabrowski #include "raid5-log.h" 163418d036SArtur Paszkiewicz 173418d036SArtur Paszkiewicz /* 183418d036SArtur Paszkiewicz * PPL consists of a 4KB header (struct ppl_header) and at least 128KB for 193418d036SArtur Paszkiewicz * partial parity data. The header contains an array of entries 203418d036SArtur Paszkiewicz * (struct ppl_header_entry) which describe the logged write requests. 213418d036SArtur Paszkiewicz * Partial parity for the entries comes after the header, written in the same 223418d036SArtur Paszkiewicz * sequence as the entries: 233418d036SArtur Paszkiewicz * 243418d036SArtur Paszkiewicz * Header 253418d036SArtur Paszkiewicz * entry0 263418d036SArtur Paszkiewicz * ... 273418d036SArtur Paszkiewicz * entryN 283418d036SArtur Paszkiewicz * PP data 293418d036SArtur Paszkiewicz * PP for entry0 303418d036SArtur Paszkiewicz * ... 313418d036SArtur Paszkiewicz * PP for entryN 323418d036SArtur Paszkiewicz * 333418d036SArtur Paszkiewicz * An entry describes one or more consecutive stripe_heads, up to a full 343418d036SArtur Paszkiewicz * stripe. The modifed raid data chunks form an m-by-n matrix, where m is the 353418d036SArtur Paszkiewicz * number of stripe_heads in the entry and n is the number of modified data 363418d036SArtur Paszkiewicz * disks. Every stripe_head in the entry must write to the same data disks. 373418d036SArtur Paszkiewicz * An example of a valid case described by a single entry (writes to the first 383418d036SArtur Paszkiewicz * stripe of a 4 disk array, 16k chunk size): 393418d036SArtur Paszkiewicz * 403418d036SArtur Paszkiewicz * sh->sector dd0 dd1 dd2 ppl 413418d036SArtur Paszkiewicz * +-----+-----+-----+ 423418d036SArtur Paszkiewicz * 0 | --- | --- | --- | +----+ 433418d036SArtur Paszkiewicz * 8 | -W- | -W- | --- | | pp | data_sector = 8 443418d036SArtur Paszkiewicz * 16 | -W- | -W- | --- | | pp | data_size = 3 * 2 * 4k 453418d036SArtur Paszkiewicz * 24 | -W- | -W- | --- | | pp | pp_size = 3 * 4k 463418d036SArtur Paszkiewicz * +-----+-----+-----+ +----+ 473418d036SArtur Paszkiewicz * 483418d036SArtur Paszkiewicz * data_sector is the first raid sector of the modified data, data_size is the 493418d036SArtur Paszkiewicz * total size of modified data and pp_size is the size of partial parity for 503418d036SArtur Paszkiewicz * this entry. Entries for full stripe writes contain no partial parity 513418d036SArtur Paszkiewicz * (pp_size = 0), they only mark the stripes for which parity should be 523418d036SArtur Paszkiewicz * recalculated after an unclean shutdown. Every entry holds a checksum of its 533418d036SArtur Paszkiewicz * partial parity, the header also has a checksum of the header itself. 543418d036SArtur Paszkiewicz * 553418d036SArtur Paszkiewicz * A write request is always logged to the PPL instance stored on the parity 563418d036SArtur Paszkiewicz * disk of the corresponding stripe. For each member disk there is one ppl_log 573418d036SArtur Paszkiewicz * used to handle logging for this disk, independently from others. They are 583418d036SArtur Paszkiewicz * grouped in child_logs array in struct ppl_conf, which is assigned to 593418d036SArtur Paszkiewicz * r5conf->log_private. 603418d036SArtur Paszkiewicz * 613418d036SArtur Paszkiewicz * ppl_io_unit represents a full PPL write, header_page contains the ppl_header. 623418d036SArtur Paszkiewicz * PPL entries for logged stripes are added in ppl_log_stripe(). A stripe_head 633418d036SArtur Paszkiewicz * can be appended to the last entry if it meets the conditions for a valid 643418d036SArtur Paszkiewicz * entry described above, otherwise a new entry is added. Checksums of entries 653418d036SArtur Paszkiewicz * are calculated incrementally as stripes containing partial parity are being 663418d036SArtur Paszkiewicz * added. ppl_submit_iounit() calculates the checksum of the header and submits 673418d036SArtur Paszkiewicz * a bio containing the header page and partial parity pages (sh->ppl_page) for 683418d036SArtur Paszkiewicz * all stripes of the io_unit. When the PPL write completes, the stripes 693418d036SArtur Paszkiewicz * associated with the io_unit are released and raid5d starts writing their data 703418d036SArtur Paszkiewicz * and parity. When all stripes are written, the io_unit is freed and the next 713418d036SArtur Paszkiewicz * can be submitted. 723418d036SArtur Paszkiewicz * 733418d036SArtur Paszkiewicz * An io_unit is used to gather stripes until it is submitted or becomes full 743418d036SArtur Paszkiewicz * (if the maximum number of entries or size of PPL is reached). Another io_unit 753418d036SArtur Paszkiewicz * can't be submitted until the previous has completed (PPL and stripe 763418d036SArtur Paszkiewicz * data+parity is written). The log->io_list tracks all io_units of a log 773418d036SArtur Paszkiewicz * (for a single member disk). New io_units are added to the end of the list 783418d036SArtur Paszkiewicz * and the first io_unit is submitted, if it is not submitted already. 793418d036SArtur Paszkiewicz * The current io_unit accepting new stripes is always at the end of the list. 801532d9e8STomasz Majchrzak * 811532d9e8STomasz Majchrzak * If write-back cache is enabled for any of the disks in the array, its data 821532d9e8STomasz Majchrzak * must be flushed before next io_unit is submitted. 833418d036SArtur Paszkiewicz */ 843418d036SArtur Paszkiewicz 85ddc08823SPawel Baldysiak #define PPL_SPACE_SIZE (128 * 1024) 86ddc08823SPawel Baldysiak 873418d036SArtur Paszkiewicz struct ppl_conf { 883418d036SArtur Paszkiewicz struct mddev *mddev; 893418d036SArtur Paszkiewicz 903418d036SArtur Paszkiewicz /* array of child logs, one for each raid disk */ 913418d036SArtur Paszkiewicz struct ppl_log *child_logs; 923418d036SArtur Paszkiewicz int count; 933418d036SArtur Paszkiewicz 943418d036SArtur Paszkiewicz int block_size; /* the logical block size used for data_sector 953418d036SArtur Paszkiewicz * in ppl_header_entry */ 963418d036SArtur Paszkiewicz u32 signature; /* raid array identifier */ 973418d036SArtur Paszkiewicz atomic64_t seq; /* current log write sequence number */ 983418d036SArtur Paszkiewicz 993418d036SArtur Paszkiewicz struct kmem_cache *io_kc; 100afeee514SKent Overstreet mempool_t io_pool; 101afeee514SKent Overstreet struct bio_set bs; 102afeee514SKent Overstreet struct bio_set flush_bs; 1034536bf9bSArtur Paszkiewicz 1044536bf9bSArtur Paszkiewicz /* used only for recovery */ 1054536bf9bSArtur Paszkiewicz int recovered_entries; 1064536bf9bSArtur Paszkiewicz int mismatch_count; 10794568f64SArtur Paszkiewicz 10894568f64SArtur Paszkiewicz /* stripes to retry if failed to allocate io_unit */ 10994568f64SArtur Paszkiewicz struct list_head no_mem_stripes; 11094568f64SArtur Paszkiewicz spinlock_t no_mem_stripes_lock; 111a596d086SMariusz Dabrowski 112a596d086SMariusz Dabrowski unsigned short write_hint; 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 */ 125ddc08823SPawel Baldysiak 126ddc08823SPawel Baldysiak sector_t next_io_sector; 127ddc08823SPawel Baldysiak unsigned int entry_space; 128ddc08823SPawel Baldysiak bool use_multippl; 1291532d9e8STomasz Majchrzak bool wb_cache_on; 1301532d9e8STomasz Majchrzak unsigned long disk_flush_bitmap; 1313418d036SArtur Paszkiewicz }; 1323418d036SArtur Paszkiewicz 1333418d036SArtur Paszkiewicz #define PPL_IO_INLINE_BVECS 32 1343418d036SArtur Paszkiewicz 1353418d036SArtur Paszkiewicz struct ppl_io_unit { 1363418d036SArtur Paszkiewicz struct ppl_log *log; 1373418d036SArtur Paszkiewicz 1383418d036SArtur Paszkiewicz struct page *header_page; /* for ppl_header */ 1393418d036SArtur Paszkiewicz 1403418d036SArtur Paszkiewicz unsigned int entries_count; /* number of entries in ppl_header */ 1413418d036SArtur Paszkiewicz unsigned int pp_size; /* total size current of partial parity */ 1423418d036SArtur Paszkiewicz 1433418d036SArtur Paszkiewicz u64 seq; /* sequence number of this log write */ 1443418d036SArtur Paszkiewicz struct list_head log_sibling; /* log->io_list */ 1453418d036SArtur Paszkiewicz 1463418d036SArtur Paszkiewicz struct list_head stripe_list; /* stripes added to the io_unit */ 1473418d036SArtur Paszkiewicz atomic_t pending_stripes; /* how many stripes not written to raid */ 1481532d9e8STomasz Majchrzak atomic_t pending_flushes; /* how many disk flushes are in progress */ 1493418d036SArtur Paszkiewicz 1503418d036SArtur Paszkiewicz bool submitted; /* true if write to log started */ 1513418d036SArtur Paszkiewicz 1523418d036SArtur Paszkiewicz /* inline bio and its biovec for submitting the iounit */ 1533418d036SArtur Paszkiewicz struct bio bio; 1543418d036SArtur Paszkiewicz struct bio_vec biovec[PPL_IO_INLINE_BVECS]; 1553418d036SArtur Paszkiewicz }; 1563418d036SArtur Paszkiewicz 1573418d036SArtur Paszkiewicz struct dma_async_tx_descriptor * 1583418d036SArtur Paszkiewicz ops_run_partial_parity(struct stripe_head *sh, struct raid5_percpu *percpu, 1593418d036SArtur Paszkiewicz struct dma_async_tx_descriptor *tx) 1603418d036SArtur Paszkiewicz { 1613418d036SArtur Paszkiewicz int disks = sh->disks; 162b330e6a4SKent Overstreet struct page **srcs = percpu->scribble; 1633418d036SArtur Paszkiewicz int count = 0, pd_idx = sh->pd_idx, i; 1643418d036SArtur Paszkiewicz struct async_submit_ctl submit; 1653418d036SArtur Paszkiewicz 1663418d036SArtur Paszkiewicz pr_debug("%s: stripe %llu\n", __func__, (unsigned long long)sh->sector); 1673418d036SArtur Paszkiewicz 1683418d036SArtur Paszkiewicz /* 1693418d036SArtur Paszkiewicz * Partial parity is the XOR of stripe data chunks that are not changed 1703418d036SArtur Paszkiewicz * during the write request. Depending on available data 1713418d036SArtur Paszkiewicz * (read-modify-write vs. reconstruct-write case) we calculate it 1723418d036SArtur Paszkiewicz * differently. 1733418d036SArtur Paszkiewicz */ 1743418d036SArtur Paszkiewicz if (sh->reconstruct_state == reconstruct_state_prexor_drain_run) { 175ae1713e2SArtur Paszkiewicz /* 176ae1713e2SArtur Paszkiewicz * rmw: xor old data and parity from updated disks 177ae1713e2SArtur Paszkiewicz * This is calculated earlier by ops_run_prexor5() so just copy 178ae1713e2SArtur Paszkiewicz * the parity dev page. 179ae1713e2SArtur Paszkiewicz */ 180ae1713e2SArtur Paszkiewicz srcs[count++] = sh->dev[pd_idx].page; 1813418d036SArtur Paszkiewicz } else if (sh->reconstruct_state == reconstruct_state_drain_run) { 1823418d036SArtur Paszkiewicz /* rcw: xor data from all not updated disks */ 1833418d036SArtur Paszkiewicz for (i = disks; i--;) { 1843418d036SArtur Paszkiewicz struct r5dev *dev = &sh->dev[i]; 1853418d036SArtur Paszkiewicz if (test_bit(R5_UPTODATE, &dev->flags)) 186ae1713e2SArtur Paszkiewicz srcs[count++] = dev->page; 1873418d036SArtur Paszkiewicz } 1883418d036SArtur Paszkiewicz } else { 1893418d036SArtur Paszkiewicz return tx; 1903418d036SArtur Paszkiewicz } 1913418d036SArtur Paszkiewicz 1923418d036SArtur Paszkiewicz init_async_submit(&submit, ASYNC_TX_FENCE|ASYNC_TX_XOR_ZERO_DST, tx, 193b330e6a4SKent Overstreet NULL, sh, (void *) (srcs + sh->disks + 2)); 1943418d036SArtur Paszkiewicz 1953418d036SArtur Paszkiewicz if (count == 1) 196ae1713e2SArtur Paszkiewicz tx = async_memcpy(sh->ppl_page, srcs[0], 0, 0, PAGE_SIZE, 1973418d036SArtur Paszkiewicz &submit); 1983418d036SArtur Paszkiewicz else 199ae1713e2SArtur Paszkiewicz tx = async_xor(sh->ppl_page, srcs, 0, count, PAGE_SIZE, 2003418d036SArtur Paszkiewicz &submit); 2013418d036SArtur Paszkiewicz 2023418d036SArtur Paszkiewicz return tx; 2033418d036SArtur Paszkiewicz } 2043418d036SArtur Paszkiewicz 205fcd403afSArtur Paszkiewicz static void *ppl_io_pool_alloc(gfp_t gfp_mask, void *pool_data) 206fcd403afSArtur Paszkiewicz { 207fcd403afSArtur Paszkiewicz struct kmem_cache *kc = pool_data; 208fcd403afSArtur Paszkiewicz struct ppl_io_unit *io; 209fcd403afSArtur Paszkiewicz 210fcd403afSArtur Paszkiewicz io = kmem_cache_alloc(kc, gfp_mask); 211fcd403afSArtur Paszkiewicz if (!io) 212fcd403afSArtur Paszkiewicz return NULL; 213fcd403afSArtur Paszkiewicz 214fcd403afSArtur Paszkiewicz io->header_page = alloc_page(gfp_mask); 215fcd403afSArtur Paszkiewicz if (!io->header_page) { 216fcd403afSArtur Paszkiewicz kmem_cache_free(kc, io); 217fcd403afSArtur Paszkiewicz return NULL; 218fcd403afSArtur Paszkiewicz } 219fcd403afSArtur Paszkiewicz 220fcd403afSArtur Paszkiewicz return io; 221fcd403afSArtur Paszkiewicz } 222fcd403afSArtur Paszkiewicz 223fcd403afSArtur Paszkiewicz static void ppl_io_pool_free(void *element, void *pool_data) 224fcd403afSArtur Paszkiewicz { 225fcd403afSArtur Paszkiewicz struct kmem_cache *kc = pool_data; 226fcd403afSArtur Paszkiewicz struct ppl_io_unit *io = element; 227fcd403afSArtur Paszkiewicz 228fcd403afSArtur Paszkiewicz __free_page(io->header_page); 229fcd403afSArtur Paszkiewicz kmem_cache_free(kc, io); 230fcd403afSArtur Paszkiewicz } 231fcd403afSArtur Paszkiewicz 2323418d036SArtur Paszkiewicz static struct ppl_io_unit *ppl_new_iounit(struct ppl_log *log, 2333418d036SArtur Paszkiewicz struct stripe_head *sh) 2343418d036SArtur Paszkiewicz { 2353418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 2363418d036SArtur Paszkiewicz struct ppl_io_unit *io; 2373418d036SArtur Paszkiewicz struct ppl_header *pplhdr; 238fcd403afSArtur Paszkiewicz struct page *header_page; 2393418d036SArtur Paszkiewicz 240afeee514SKent Overstreet io = mempool_alloc(&ppl_conf->io_pool, GFP_NOWAIT); 2413418d036SArtur Paszkiewicz if (!io) 2423418d036SArtur Paszkiewicz return NULL; 2433418d036SArtur Paszkiewicz 244fcd403afSArtur Paszkiewicz header_page = io->header_page; 2453418d036SArtur Paszkiewicz memset(io, 0, sizeof(*io)); 246fcd403afSArtur Paszkiewicz io->header_page = header_page; 247fcd403afSArtur Paszkiewicz 2483418d036SArtur Paszkiewicz io->log = log; 2493418d036SArtur Paszkiewicz INIT_LIST_HEAD(&io->log_sibling); 2503418d036SArtur Paszkiewicz INIT_LIST_HEAD(&io->stripe_list); 2513418d036SArtur Paszkiewicz atomic_set(&io->pending_stripes, 0); 2521532d9e8STomasz Majchrzak atomic_set(&io->pending_flushes, 0); 2533418d036SArtur Paszkiewicz bio_init(&io->bio, io->biovec, PPL_IO_INLINE_BVECS); 2543418d036SArtur Paszkiewicz 2553418d036SArtur Paszkiewicz pplhdr = page_address(io->header_page); 2563418d036SArtur Paszkiewicz clear_page(pplhdr); 2573418d036SArtur Paszkiewicz memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED); 2583418d036SArtur Paszkiewicz pplhdr->signature = cpu_to_le32(ppl_conf->signature); 2593418d036SArtur Paszkiewicz 2603418d036SArtur Paszkiewicz io->seq = atomic64_add_return(1, &ppl_conf->seq); 2613418d036SArtur Paszkiewicz pplhdr->generation = cpu_to_le64(io->seq); 2623418d036SArtur Paszkiewicz 2633418d036SArtur Paszkiewicz return io; 2643418d036SArtur Paszkiewicz } 2653418d036SArtur Paszkiewicz 2663418d036SArtur Paszkiewicz static int ppl_log_stripe(struct ppl_log *log, struct stripe_head *sh) 2673418d036SArtur Paszkiewicz { 2683418d036SArtur Paszkiewicz struct ppl_io_unit *io = log->current_io; 2693418d036SArtur Paszkiewicz struct ppl_header_entry *e = NULL; 2703418d036SArtur Paszkiewicz struct ppl_header *pplhdr; 2713418d036SArtur Paszkiewicz int i; 2723418d036SArtur Paszkiewicz sector_t data_sector = 0; 2733418d036SArtur Paszkiewicz int data_disks = 0; 2743418d036SArtur Paszkiewicz struct r5conf *conf = sh->raid_conf; 2753418d036SArtur Paszkiewicz 2763418d036SArtur Paszkiewicz pr_debug("%s: stripe: %llu\n", __func__, (unsigned long long)sh->sector); 2773418d036SArtur Paszkiewicz 2783418d036SArtur Paszkiewicz /* check if current io_unit is full */ 279ddc08823SPawel Baldysiak if (io && (io->pp_size == log->entry_space || 2803418d036SArtur Paszkiewicz io->entries_count == PPL_HDR_MAX_ENTRIES)) { 2813418d036SArtur Paszkiewicz pr_debug("%s: add io_unit blocked by seq: %llu\n", 2823418d036SArtur Paszkiewicz __func__, io->seq); 2833418d036SArtur Paszkiewicz io = NULL; 2843418d036SArtur Paszkiewicz } 2853418d036SArtur Paszkiewicz 2863418d036SArtur Paszkiewicz /* add a new unit if there is none or the current is full */ 2873418d036SArtur Paszkiewicz if (!io) { 2883418d036SArtur Paszkiewicz io = ppl_new_iounit(log, sh); 2893418d036SArtur Paszkiewicz if (!io) 2903418d036SArtur Paszkiewicz return -ENOMEM; 2913418d036SArtur Paszkiewicz spin_lock_irq(&log->io_list_lock); 2923418d036SArtur Paszkiewicz list_add_tail(&io->log_sibling, &log->io_list); 2933418d036SArtur Paszkiewicz spin_unlock_irq(&log->io_list_lock); 2943418d036SArtur Paszkiewicz 2953418d036SArtur Paszkiewicz log->current_io = io; 2963418d036SArtur Paszkiewicz } 2973418d036SArtur Paszkiewicz 2983418d036SArtur Paszkiewicz for (i = 0; i < sh->disks; i++) { 2993418d036SArtur Paszkiewicz struct r5dev *dev = &sh->dev[i]; 3003418d036SArtur Paszkiewicz 3013418d036SArtur Paszkiewicz if (i != sh->pd_idx && test_bit(R5_Wantwrite, &dev->flags)) { 3023418d036SArtur Paszkiewicz if (!data_disks || dev->sector < data_sector) 3033418d036SArtur Paszkiewicz data_sector = dev->sector; 3043418d036SArtur Paszkiewicz data_disks++; 3053418d036SArtur Paszkiewicz } 3063418d036SArtur Paszkiewicz } 3073418d036SArtur Paszkiewicz BUG_ON(!data_disks); 3083418d036SArtur Paszkiewicz 3093418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu data_sector: %llu data_disks: %d\n", __func__, 3103418d036SArtur Paszkiewicz io->seq, (unsigned long long)data_sector, data_disks); 3113418d036SArtur Paszkiewicz 3123418d036SArtur Paszkiewicz pplhdr = page_address(io->header_page); 3133418d036SArtur Paszkiewicz 3143418d036SArtur Paszkiewicz if (io->entries_count > 0) { 3153418d036SArtur Paszkiewicz struct ppl_header_entry *last = 3163418d036SArtur Paszkiewicz &pplhdr->entries[io->entries_count - 1]; 3173418d036SArtur Paszkiewicz struct stripe_head *sh_last = list_last_entry( 3183418d036SArtur Paszkiewicz &io->stripe_list, struct stripe_head, log_list); 3193418d036SArtur Paszkiewicz u64 data_sector_last = le64_to_cpu(last->data_sector); 3203418d036SArtur Paszkiewicz u32 data_size_last = le32_to_cpu(last->data_size); 3213418d036SArtur Paszkiewicz 3223418d036SArtur Paszkiewicz /* 3233418d036SArtur Paszkiewicz * Check if we can append the stripe to the last entry. It must 3243418d036SArtur Paszkiewicz * be just after the last logged stripe and write to the same 3253418d036SArtur Paszkiewicz * disks. Use bit shift and logarithm to avoid 64-bit division. 3263418d036SArtur Paszkiewicz */ 3273418d036SArtur Paszkiewicz if ((sh->sector == sh_last->sector + STRIPE_SECTORS) && 3283418d036SArtur Paszkiewicz (data_sector >> ilog2(conf->chunk_sectors) == 3293418d036SArtur Paszkiewicz data_sector_last >> ilog2(conf->chunk_sectors)) && 3303418d036SArtur Paszkiewicz ((data_sector - data_sector_last) * data_disks == 3313418d036SArtur Paszkiewicz data_size_last >> 9)) 3323418d036SArtur Paszkiewicz e = last; 3333418d036SArtur Paszkiewicz } 3343418d036SArtur Paszkiewicz 3353418d036SArtur Paszkiewicz if (!e) { 3363418d036SArtur Paszkiewicz e = &pplhdr->entries[io->entries_count++]; 3373418d036SArtur Paszkiewicz e->data_sector = cpu_to_le64(data_sector); 3383418d036SArtur Paszkiewicz e->parity_disk = cpu_to_le32(sh->pd_idx); 3393418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(~0); 3403418d036SArtur Paszkiewicz } 3413418d036SArtur Paszkiewicz 3423418d036SArtur Paszkiewicz le32_add_cpu(&e->data_size, data_disks << PAGE_SHIFT); 3433418d036SArtur Paszkiewicz 3443418d036SArtur Paszkiewicz /* don't write any PP if full stripe write */ 3453418d036SArtur Paszkiewicz if (!test_bit(STRIPE_FULL_WRITE, &sh->state)) { 3463418d036SArtur Paszkiewicz le32_add_cpu(&e->pp_size, PAGE_SIZE); 3473418d036SArtur Paszkiewicz io->pp_size += PAGE_SIZE; 3483418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(crc32c_le(le32_to_cpu(e->checksum), 3493418d036SArtur Paszkiewicz page_address(sh->ppl_page), 3503418d036SArtur Paszkiewicz PAGE_SIZE)); 3513418d036SArtur Paszkiewicz } 3523418d036SArtur Paszkiewicz 3533418d036SArtur Paszkiewicz list_add_tail(&sh->log_list, &io->stripe_list); 3543418d036SArtur Paszkiewicz atomic_inc(&io->pending_stripes); 3553418d036SArtur Paszkiewicz sh->ppl_io = io; 3563418d036SArtur Paszkiewicz 3573418d036SArtur Paszkiewicz return 0; 3583418d036SArtur Paszkiewicz } 3593418d036SArtur Paszkiewicz 3603418d036SArtur Paszkiewicz int ppl_write_stripe(struct r5conf *conf, struct stripe_head *sh) 3613418d036SArtur Paszkiewicz { 3623418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 3633418d036SArtur Paszkiewicz struct ppl_io_unit *io = sh->ppl_io; 3643418d036SArtur Paszkiewicz struct ppl_log *log; 3653418d036SArtur Paszkiewicz 366845b9e22SArtur Paszkiewicz if (io || test_bit(STRIPE_SYNCING, &sh->state) || !sh->ppl_page || 3673418d036SArtur Paszkiewicz !test_bit(R5_Wantwrite, &sh->dev[sh->pd_idx].flags) || 3683418d036SArtur Paszkiewicz !test_bit(R5_Insync, &sh->dev[sh->pd_idx].flags)) { 3693418d036SArtur Paszkiewicz clear_bit(STRIPE_LOG_TRAPPED, &sh->state); 3703418d036SArtur Paszkiewicz return -EAGAIN; 3713418d036SArtur Paszkiewicz } 3723418d036SArtur Paszkiewicz 3733418d036SArtur Paszkiewicz log = &ppl_conf->child_logs[sh->pd_idx]; 3743418d036SArtur Paszkiewicz 3753418d036SArtur Paszkiewicz mutex_lock(&log->io_mutex); 3763418d036SArtur Paszkiewicz 3773418d036SArtur Paszkiewicz if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) { 3783418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 3793418d036SArtur Paszkiewicz return -EAGAIN; 3803418d036SArtur Paszkiewicz } 3813418d036SArtur Paszkiewicz 3823418d036SArtur Paszkiewicz set_bit(STRIPE_LOG_TRAPPED, &sh->state); 3833418d036SArtur Paszkiewicz clear_bit(STRIPE_DELAYED, &sh->state); 3843418d036SArtur Paszkiewicz atomic_inc(&sh->count); 3853418d036SArtur Paszkiewicz 3863418d036SArtur Paszkiewicz if (ppl_log_stripe(log, sh)) { 38794568f64SArtur Paszkiewicz spin_lock_irq(&ppl_conf->no_mem_stripes_lock); 38894568f64SArtur Paszkiewicz list_add_tail(&sh->log_list, &ppl_conf->no_mem_stripes); 38994568f64SArtur Paszkiewicz spin_unlock_irq(&ppl_conf->no_mem_stripes_lock); 3903418d036SArtur Paszkiewicz } 3913418d036SArtur Paszkiewicz 3923418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 3933418d036SArtur Paszkiewicz 3943418d036SArtur Paszkiewicz return 0; 3953418d036SArtur Paszkiewicz } 3963418d036SArtur Paszkiewicz 3973418d036SArtur Paszkiewicz static void ppl_log_endio(struct bio *bio) 3983418d036SArtur Paszkiewicz { 3993418d036SArtur Paszkiewicz struct ppl_io_unit *io = bio->bi_private; 4003418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 4013418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 4023418d036SArtur Paszkiewicz struct stripe_head *sh, *next; 4033418d036SArtur Paszkiewicz 4043418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu\n", __func__, io->seq); 4053418d036SArtur Paszkiewicz 4064e4cbee9SChristoph Hellwig if (bio->bi_status) 4073418d036SArtur Paszkiewicz md_error(ppl_conf->mddev, log->rdev); 4083418d036SArtur Paszkiewicz 4093418d036SArtur Paszkiewicz list_for_each_entry_safe(sh, next, &io->stripe_list, log_list) { 4103418d036SArtur Paszkiewicz list_del_init(&sh->log_list); 4113418d036SArtur Paszkiewicz 4123418d036SArtur Paszkiewicz set_bit(STRIPE_HANDLE, &sh->state); 4133418d036SArtur Paszkiewicz raid5_release_stripe(sh); 4143418d036SArtur Paszkiewicz } 4153418d036SArtur Paszkiewicz } 4163418d036SArtur Paszkiewicz 4173418d036SArtur Paszkiewicz static void ppl_submit_iounit_bio(struct ppl_io_unit *io, struct bio *bio) 4183418d036SArtur Paszkiewicz { 4193418d036SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 4203418d036SArtur Paszkiewicz 4213418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu size: %u sector: %llu dev: %s\n", 4223418d036SArtur Paszkiewicz __func__, io->seq, bio->bi_iter.bi_size, 4233418d036SArtur Paszkiewicz (unsigned long long)bio->bi_iter.bi_sector, 42474d46992SChristoph Hellwig bio_devname(bio, b)); 4253418d036SArtur Paszkiewicz 4263418d036SArtur Paszkiewicz submit_bio(bio); 4273418d036SArtur Paszkiewicz } 4283418d036SArtur Paszkiewicz 4293418d036SArtur Paszkiewicz static void ppl_submit_iounit(struct ppl_io_unit *io) 4303418d036SArtur Paszkiewicz { 4313418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 4323418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 4333418d036SArtur Paszkiewicz struct ppl_header *pplhdr = page_address(io->header_page); 4343418d036SArtur Paszkiewicz struct bio *bio = &io->bio; 4353418d036SArtur Paszkiewicz struct stripe_head *sh; 4363418d036SArtur Paszkiewicz int i; 4373418d036SArtur Paszkiewicz 4386358c239SArtur Paszkiewicz bio->bi_private = io; 4396358c239SArtur Paszkiewicz 4406358c239SArtur Paszkiewicz if (!log->rdev || test_bit(Faulty, &log->rdev->flags)) { 4416358c239SArtur Paszkiewicz ppl_log_endio(bio); 4426358c239SArtur Paszkiewicz return; 4436358c239SArtur Paszkiewicz } 4446358c239SArtur Paszkiewicz 4453418d036SArtur Paszkiewicz for (i = 0; i < io->entries_count; i++) { 4463418d036SArtur Paszkiewicz struct ppl_header_entry *e = &pplhdr->entries[i]; 4473418d036SArtur Paszkiewicz 4483418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu entry: %d data_sector: %llu pp_size: %u data_size: %u\n", 4493418d036SArtur Paszkiewicz __func__, io->seq, i, le64_to_cpu(e->data_sector), 4503418d036SArtur Paszkiewicz le32_to_cpu(e->pp_size), le32_to_cpu(e->data_size)); 4513418d036SArtur Paszkiewicz 4523418d036SArtur Paszkiewicz e->data_sector = cpu_to_le64(le64_to_cpu(e->data_sector) >> 4533418d036SArtur Paszkiewicz ilog2(ppl_conf->block_size >> 9)); 4543418d036SArtur Paszkiewicz e->checksum = cpu_to_le32(~le32_to_cpu(e->checksum)); 4553418d036SArtur Paszkiewicz } 4563418d036SArtur Paszkiewicz 4573418d036SArtur Paszkiewicz pplhdr->entries_count = cpu_to_le32(io->entries_count); 4583418d036SArtur Paszkiewicz pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PPL_HEADER_SIZE)); 4593418d036SArtur Paszkiewicz 460ddc08823SPawel Baldysiak /* Rewind the buffer if current PPL is larger then remaining space */ 461ddc08823SPawel Baldysiak if (log->use_multippl && 462ddc08823SPawel Baldysiak log->rdev->ppl.sector + log->rdev->ppl.size - log->next_io_sector < 463ddc08823SPawel Baldysiak (PPL_HEADER_SIZE + io->pp_size) >> 9) 464ddc08823SPawel Baldysiak log->next_io_sector = log->rdev->ppl.sector; 465ddc08823SPawel Baldysiak 466ddc08823SPawel Baldysiak 4673418d036SArtur Paszkiewicz bio->bi_end_io = ppl_log_endio; 4683418d036SArtur Paszkiewicz bio->bi_opf = REQ_OP_WRITE | REQ_FUA; 46974d46992SChristoph Hellwig bio_set_dev(bio, log->rdev->bdev); 470ddc08823SPawel Baldysiak bio->bi_iter.bi_sector = log->next_io_sector; 4713418d036SArtur Paszkiewicz bio_add_page(bio, io->header_page, PAGE_SIZE, 0); 472a596d086SMariusz Dabrowski bio->bi_write_hint = ppl_conf->write_hint; 4733418d036SArtur Paszkiewicz 474ddc08823SPawel Baldysiak pr_debug("%s: log->current_io_sector: %llu\n", __func__, 475ddc08823SPawel Baldysiak (unsigned long long)log->next_io_sector); 476ddc08823SPawel Baldysiak 477ddc08823SPawel Baldysiak if (log->use_multippl) 478ddc08823SPawel Baldysiak log->next_io_sector += (PPL_HEADER_SIZE + io->pp_size) >> 9; 479ddc08823SPawel Baldysiak 4801532d9e8STomasz Majchrzak WARN_ON(log->disk_flush_bitmap != 0); 4811532d9e8STomasz Majchrzak 4823418d036SArtur Paszkiewicz list_for_each_entry(sh, &io->stripe_list, log_list) { 4831532d9e8STomasz Majchrzak for (i = 0; i < sh->disks; i++) { 4841532d9e8STomasz Majchrzak struct r5dev *dev = &sh->dev[i]; 4851532d9e8STomasz Majchrzak 4861532d9e8STomasz Majchrzak if ((ppl_conf->child_logs[i].wb_cache_on) && 4871532d9e8STomasz Majchrzak (test_bit(R5_Wantwrite, &dev->flags))) { 4881532d9e8STomasz Majchrzak set_bit(i, &log->disk_flush_bitmap); 4891532d9e8STomasz Majchrzak } 4901532d9e8STomasz Majchrzak } 4911532d9e8STomasz Majchrzak 4923418d036SArtur Paszkiewicz /* entries for full stripe writes have no partial parity */ 4933418d036SArtur Paszkiewicz if (test_bit(STRIPE_FULL_WRITE, &sh->state)) 4943418d036SArtur Paszkiewicz continue; 4953418d036SArtur Paszkiewicz 4963418d036SArtur Paszkiewicz if (!bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0)) { 4973418d036SArtur Paszkiewicz struct bio *prev = bio; 4983418d036SArtur Paszkiewicz 4993418d036SArtur Paszkiewicz bio = bio_alloc_bioset(GFP_NOIO, BIO_MAX_PAGES, 500afeee514SKent Overstreet &ppl_conf->bs); 5013418d036SArtur Paszkiewicz bio->bi_opf = prev->bi_opf; 502a596d086SMariusz Dabrowski bio->bi_write_hint = prev->bi_write_hint; 50374d46992SChristoph Hellwig bio_copy_dev(bio, prev); 5043418d036SArtur Paszkiewicz bio->bi_iter.bi_sector = bio_end_sector(prev); 5053418d036SArtur Paszkiewicz bio_add_page(bio, sh->ppl_page, PAGE_SIZE, 0); 5063418d036SArtur Paszkiewicz 5073418d036SArtur Paszkiewicz bio_chain(bio, prev); 5083418d036SArtur Paszkiewicz ppl_submit_iounit_bio(io, prev); 5093418d036SArtur Paszkiewicz } 5103418d036SArtur Paszkiewicz } 5113418d036SArtur Paszkiewicz 5123418d036SArtur Paszkiewicz ppl_submit_iounit_bio(io, bio); 5133418d036SArtur Paszkiewicz } 5143418d036SArtur Paszkiewicz 5153418d036SArtur Paszkiewicz static void ppl_submit_current_io(struct ppl_log *log) 5163418d036SArtur Paszkiewicz { 5173418d036SArtur Paszkiewicz struct ppl_io_unit *io; 5183418d036SArtur Paszkiewicz 5193418d036SArtur Paszkiewicz spin_lock_irq(&log->io_list_lock); 5203418d036SArtur Paszkiewicz 5213418d036SArtur Paszkiewicz io = list_first_entry_or_null(&log->io_list, struct ppl_io_unit, 5223418d036SArtur Paszkiewicz log_sibling); 5233418d036SArtur Paszkiewicz if (io && io->submitted) 5243418d036SArtur Paszkiewicz io = NULL; 5253418d036SArtur Paszkiewicz 5263418d036SArtur Paszkiewicz spin_unlock_irq(&log->io_list_lock); 5273418d036SArtur Paszkiewicz 5283418d036SArtur Paszkiewicz if (io) { 5293418d036SArtur Paszkiewicz io->submitted = true; 5303418d036SArtur Paszkiewicz 5313418d036SArtur Paszkiewicz if (io == log->current_io) 5323418d036SArtur Paszkiewicz log->current_io = NULL; 5333418d036SArtur Paszkiewicz 5343418d036SArtur Paszkiewicz ppl_submit_iounit(io); 5353418d036SArtur Paszkiewicz } 5363418d036SArtur Paszkiewicz } 5373418d036SArtur Paszkiewicz 5383418d036SArtur Paszkiewicz void ppl_write_stripe_run(struct r5conf *conf) 5393418d036SArtur Paszkiewicz { 5403418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 5413418d036SArtur Paszkiewicz struct ppl_log *log; 5423418d036SArtur Paszkiewicz int i; 5433418d036SArtur Paszkiewicz 5443418d036SArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 5453418d036SArtur Paszkiewicz log = &ppl_conf->child_logs[i]; 5463418d036SArtur Paszkiewicz 5473418d036SArtur Paszkiewicz mutex_lock(&log->io_mutex); 5483418d036SArtur Paszkiewicz ppl_submit_current_io(log); 5493418d036SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 5503418d036SArtur Paszkiewicz } 5513418d036SArtur Paszkiewicz } 5523418d036SArtur Paszkiewicz 5533418d036SArtur Paszkiewicz static void ppl_io_unit_finished(struct ppl_io_unit *io) 5543418d036SArtur Paszkiewicz { 5553418d036SArtur Paszkiewicz struct ppl_log *log = io->log; 55694568f64SArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 5571532d9e8STomasz Majchrzak struct r5conf *conf = ppl_conf->mddev->private; 5583418d036SArtur Paszkiewicz unsigned long flags; 5593418d036SArtur Paszkiewicz 5603418d036SArtur Paszkiewicz pr_debug("%s: seq: %llu\n", __func__, io->seq); 5613418d036SArtur Paszkiewicz 56294568f64SArtur Paszkiewicz local_irq_save(flags); 5633418d036SArtur Paszkiewicz 56494568f64SArtur Paszkiewicz spin_lock(&log->io_list_lock); 5653418d036SArtur Paszkiewicz list_del(&io->log_sibling); 56694568f64SArtur Paszkiewicz spin_unlock(&log->io_list_lock); 5673418d036SArtur Paszkiewicz 568afeee514SKent Overstreet mempool_free(io, &ppl_conf->io_pool); 56994568f64SArtur Paszkiewicz 57094568f64SArtur Paszkiewicz spin_lock(&ppl_conf->no_mem_stripes_lock); 57194568f64SArtur Paszkiewicz if (!list_empty(&ppl_conf->no_mem_stripes)) { 57294568f64SArtur Paszkiewicz struct stripe_head *sh; 57394568f64SArtur Paszkiewicz 57494568f64SArtur Paszkiewicz sh = list_first_entry(&ppl_conf->no_mem_stripes, 57594568f64SArtur Paszkiewicz struct stripe_head, log_list); 5763418d036SArtur Paszkiewicz list_del_init(&sh->log_list); 5773418d036SArtur Paszkiewicz set_bit(STRIPE_HANDLE, &sh->state); 5783418d036SArtur Paszkiewicz raid5_release_stripe(sh); 5793418d036SArtur Paszkiewicz } 58094568f64SArtur Paszkiewicz spin_unlock(&ppl_conf->no_mem_stripes_lock); 5813418d036SArtur Paszkiewicz 58294568f64SArtur Paszkiewicz local_irq_restore(flags); 5831532d9e8STomasz Majchrzak 5841532d9e8STomasz Majchrzak wake_up(&conf->wait_for_quiescent); 5851532d9e8STomasz Majchrzak } 5861532d9e8STomasz Majchrzak 5871532d9e8STomasz Majchrzak static void ppl_flush_endio(struct bio *bio) 5881532d9e8STomasz Majchrzak { 5891532d9e8STomasz Majchrzak struct ppl_io_unit *io = bio->bi_private; 5901532d9e8STomasz Majchrzak struct ppl_log *log = io->log; 5911532d9e8STomasz Majchrzak struct ppl_conf *ppl_conf = log->ppl_conf; 5921532d9e8STomasz Majchrzak struct r5conf *conf = ppl_conf->mddev->private; 5931532d9e8STomasz Majchrzak char b[BDEVNAME_SIZE]; 5941532d9e8STomasz Majchrzak 5951532d9e8STomasz Majchrzak pr_debug("%s: dev: %s\n", __func__, bio_devname(bio, b)); 5961532d9e8STomasz Majchrzak 5971532d9e8STomasz Majchrzak if (bio->bi_status) { 5981532d9e8STomasz Majchrzak struct md_rdev *rdev; 5991532d9e8STomasz Majchrzak 6001532d9e8STomasz Majchrzak rcu_read_lock(); 6011532d9e8STomasz Majchrzak rdev = md_find_rdev_rcu(conf->mddev, bio_dev(bio)); 6021532d9e8STomasz Majchrzak if (rdev) 6031532d9e8STomasz Majchrzak md_error(rdev->mddev, rdev); 6041532d9e8STomasz Majchrzak rcu_read_unlock(); 6051532d9e8STomasz Majchrzak } 6061532d9e8STomasz Majchrzak 6071532d9e8STomasz Majchrzak bio_put(bio); 6081532d9e8STomasz Majchrzak 6091532d9e8STomasz Majchrzak if (atomic_dec_and_test(&io->pending_flushes)) { 6101532d9e8STomasz Majchrzak ppl_io_unit_finished(io); 6111532d9e8STomasz Majchrzak md_wakeup_thread(conf->mddev->thread); 6121532d9e8STomasz Majchrzak } 6131532d9e8STomasz Majchrzak } 6141532d9e8STomasz Majchrzak 6151532d9e8STomasz Majchrzak static void ppl_do_flush(struct ppl_io_unit *io) 6161532d9e8STomasz Majchrzak { 6171532d9e8STomasz Majchrzak struct ppl_log *log = io->log; 6181532d9e8STomasz Majchrzak struct ppl_conf *ppl_conf = log->ppl_conf; 6191532d9e8STomasz Majchrzak struct r5conf *conf = ppl_conf->mddev->private; 6201532d9e8STomasz Majchrzak int raid_disks = conf->raid_disks; 6211532d9e8STomasz Majchrzak int flushed_disks = 0; 6221532d9e8STomasz Majchrzak int i; 6231532d9e8STomasz Majchrzak 6241532d9e8STomasz Majchrzak atomic_set(&io->pending_flushes, raid_disks); 6251532d9e8STomasz Majchrzak 6261532d9e8STomasz Majchrzak for_each_set_bit(i, &log->disk_flush_bitmap, raid_disks) { 6271532d9e8STomasz Majchrzak struct md_rdev *rdev; 6281532d9e8STomasz Majchrzak struct block_device *bdev = NULL; 6291532d9e8STomasz Majchrzak 6301532d9e8STomasz Majchrzak rcu_read_lock(); 6311532d9e8STomasz Majchrzak rdev = rcu_dereference(conf->disks[i].rdev); 6321532d9e8STomasz Majchrzak if (rdev && !test_bit(Faulty, &rdev->flags)) 6331532d9e8STomasz Majchrzak bdev = rdev->bdev; 6341532d9e8STomasz Majchrzak rcu_read_unlock(); 6351532d9e8STomasz Majchrzak 6361532d9e8STomasz Majchrzak if (bdev) { 6371532d9e8STomasz Majchrzak struct bio *bio; 6381532d9e8STomasz Majchrzak char b[BDEVNAME_SIZE]; 6391532d9e8STomasz Majchrzak 640afeee514SKent Overstreet bio = bio_alloc_bioset(GFP_NOIO, 0, &ppl_conf->flush_bs); 6411532d9e8STomasz Majchrzak bio_set_dev(bio, bdev); 6421532d9e8STomasz Majchrzak bio->bi_private = io; 6431532d9e8STomasz Majchrzak bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; 6441532d9e8STomasz Majchrzak bio->bi_end_io = ppl_flush_endio; 6451532d9e8STomasz Majchrzak 6461532d9e8STomasz Majchrzak pr_debug("%s: dev: %s\n", __func__, 6471532d9e8STomasz Majchrzak bio_devname(bio, b)); 6481532d9e8STomasz Majchrzak 6491532d9e8STomasz Majchrzak submit_bio(bio); 6501532d9e8STomasz Majchrzak flushed_disks++; 6511532d9e8STomasz Majchrzak } 6521532d9e8STomasz Majchrzak } 6531532d9e8STomasz Majchrzak 6541532d9e8STomasz Majchrzak log->disk_flush_bitmap = 0; 6551532d9e8STomasz Majchrzak 6561532d9e8STomasz Majchrzak for (i = flushed_disks ; i < raid_disks; i++) { 6571532d9e8STomasz Majchrzak if (atomic_dec_and_test(&io->pending_flushes)) 6581532d9e8STomasz Majchrzak ppl_io_unit_finished(io); 6591532d9e8STomasz Majchrzak } 6601532d9e8STomasz Majchrzak } 6611532d9e8STomasz Majchrzak 6621532d9e8STomasz Majchrzak static inline bool ppl_no_io_unit_submitted(struct r5conf *conf, 6631532d9e8STomasz Majchrzak struct ppl_log *log) 6641532d9e8STomasz Majchrzak { 6651532d9e8STomasz Majchrzak struct ppl_io_unit *io; 6661532d9e8STomasz Majchrzak 6671532d9e8STomasz Majchrzak io = list_first_entry_or_null(&log->io_list, struct ppl_io_unit, 6681532d9e8STomasz Majchrzak log_sibling); 6691532d9e8STomasz Majchrzak 6701532d9e8STomasz Majchrzak return !io || !io->submitted; 6711532d9e8STomasz Majchrzak } 6721532d9e8STomasz Majchrzak 6731532d9e8STomasz Majchrzak void ppl_quiesce(struct r5conf *conf, int quiesce) 6741532d9e8STomasz Majchrzak { 6751532d9e8STomasz Majchrzak struct ppl_conf *ppl_conf = conf->log_private; 6761532d9e8STomasz Majchrzak int i; 6771532d9e8STomasz Majchrzak 6781532d9e8STomasz Majchrzak if (quiesce) { 6791532d9e8STomasz Majchrzak for (i = 0; i < ppl_conf->count; i++) { 6801532d9e8STomasz Majchrzak struct ppl_log *log = &ppl_conf->child_logs[i]; 6811532d9e8STomasz Majchrzak 6821532d9e8STomasz Majchrzak spin_lock_irq(&log->io_list_lock); 6831532d9e8STomasz Majchrzak wait_event_lock_irq(conf->wait_for_quiescent, 6841532d9e8STomasz Majchrzak ppl_no_io_unit_submitted(conf, log), 6851532d9e8STomasz Majchrzak log->io_list_lock); 6861532d9e8STomasz Majchrzak spin_unlock_irq(&log->io_list_lock); 6871532d9e8STomasz Majchrzak } 6881532d9e8STomasz Majchrzak } 6893418d036SArtur Paszkiewicz } 6903418d036SArtur Paszkiewicz 691f4bc0c81SArtur Paszkiewicz int ppl_handle_flush_request(struct r5l_log *log, struct bio *bio) 692f4bc0c81SArtur Paszkiewicz { 693f4bc0c81SArtur Paszkiewicz if (bio->bi_iter.bi_size == 0) { 694f4bc0c81SArtur Paszkiewicz bio_endio(bio); 695f4bc0c81SArtur Paszkiewicz return 0; 696f4bc0c81SArtur Paszkiewicz } 697f4bc0c81SArtur Paszkiewicz bio->bi_opf &= ~REQ_PREFLUSH; 698f4bc0c81SArtur Paszkiewicz return -EAGAIN; 699f4bc0c81SArtur Paszkiewicz } 700f4bc0c81SArtur Paszkiewicz 7013418d036SArtur Paszkiewicz void ppl_stripe_write_finished(struct stripe_head *sh) 7023418d036SArtur Paszkiewicz { 7033418d036SArtur Paszkiewicz struct ppl_io_unit *io; 7043418d036SArtur Paszkiewicz 7053418d036SArtur Paszkiewicz io = sh->ppl_io; 7063418d036SArtur Paszkiewicz sh->ppl_io = NULL; 7073418d036SArtur Paszkiewicz 7081532d9e8STomasz Majchrzak if (io && atomic_dec_and_test(&io->pending_stripes)) { 7091532d9e8STomasz Majchrzak if (io->log->disk_flush_bitmap) 7101532d9e8STomasz Majchrzak ppl_do_flush(io); 7111532d9e8STomasz Majchrzak else 7123418d036SArtur Paszkiewicz ppl_io_unit_finished(io); 7133418d036SArtur Paszkiewicz } 7141532d9e8STomasz Majchrzak } 7153418d036SArtur Paszkiewicz 7164536bf9bSArtur Paszkiewicz static void ppl_xor(int size, struct page *page1, struct page *page2) 7174536bf9bSArtur Paszkiewicz { 7184536bf9bSArtur Paszkiewicz struct async_submit_ctl submit; 7194536bf9bSArtur Paszkiewicz struct dma_async_tx_descriptor *tx; 7204536bf9bSArtur Paszkiewicz struct page *xor_srcs[] = { page1, page2 }; 7214536bf9bSArtur Paszkiewicz 7224536bf9bSArtur Paszkiewicz init_async_submit(&submit, ASYNC_TX_ACK|ASYNC_TX_XOR_DROP_DST, 7234536bf9bSArtur Paszkiewicz NULL, NULL, NULL, NULL); 7244536bf9bSArtur Paszkiewicz tx = async_xor(page1, xor_srcs, 0, 2, size, &submit); 7254536bf9bSArtur Paszkiewicz 7264536bf9bSArtur Paszkiewicz async_tx_quiesce(&tx); 7274536bf9bSArtur Paszkiewicz } 7284536bf9bSArtur Paszkiewicz 7294536bf9bSArtur Paszkiewicz /* 7304536bf9bSArtur Paszkiewicz * PPL recovery strategy: xor partial parity and data from all modified data 7314536bf9bSArtur Paszkiewicz * disks within a stripe and write the result as the new stripe parity. If all 7324536bf9bSArtur Paszkiewicz * stripe data disks are modified (full stripe write), no partial parity is 7334536bf9bSArtur Paszkiewicz * available, so just xor the data disks. 7344536bf9bSArtur Paszkiewicz * 7354536bf9bSArtur Paszkiewicz * Recovery of a PPL entry shall occur only if all modified data disks are 7364536bf9bSArtur Paszkiewicz * available and read from all of them succeeds. 7374536bf9bSArtur Paszkiewicz * 7384536bf9bSArtur Paszkiewicz * A PPL entry applies to a stripe, partial parity size for an entry is at most 7394536bf9bSArtur Paszkiewicz * the size of the chunk. Examples of possible cases for a single entry: 7404536bf9bSArtur Paszkiewicz * 7414536bf9bSArtur Paszkiewicz * case 0: single data disk write: 7424536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 7434536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7444536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 7454536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | -> | data1 ^ pp | 7464536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | -> | data1 ^ pp | 7474536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 7484536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7494536bf9bSArtur Paszkiewicz * pp_size = data_size 7504536bf9bSArtur Paszkiewicz * 7514536bf9bSArtur Paszkiewicz * case 1: more than one data disk write: 7524536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 7534536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7544536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 7554536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp | 7564536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | -> | data0 ^ data1 ^ pp | 7574536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | +----+ | (no change) | 7584536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7594536bf9bSArtur Paszkiewicz * pp_size = data_size / modified_data_disks 7604536bf9bSArtur Paszkiewicz * 7614536bf9bSArtur Paszkiewicz * case 2: write to all data disks (also full stripe write): 7624536bf9bSArtur Paszkiewicz * data0 data1 data2 parity 7634536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7644536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | (no change) | 7654536bf9bSArtur Paszkiewicz * | -data- | -data- | -data- | --------> | xor all data | 7664536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | --------> | (no change) | 7674536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | (no change) | 7684536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +--------------------+ 7694536bf9bSArtur Paszkiewicz * pp_size = 0 7704536bf9bSArtur Paszkiewicz * 7714536bf9bSArtur Paszkiewicz * The following cases are possible only in other implementations. The recovery 7724536bf9bSArtur Paszkiewicz * code can handle them, but they are not generated at runtime because they can 7734536bf9bSArtur Paszkiewicz * be reduced to cases 0, 1 and 2: 7744536bf9bSArtur Paszkiewicz * 7754536bf9bSArtur Paszkiewicz * case 3: 7764536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 7774536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 7784536bf9bSArtur Paszkiewicz * | ------ | -data- | -data- | | pp | | data1 ^ data2 ^ pp | 7794536bf9bSArtur Paszkiewicz * | ------ | -data- | -data- | | pp | -> | data1 ^ data2 ^ pp | 7804536bf9bSArtur Paszkiewicz * | -data- | -data- | -data- | | -- | -> | xor all data | 7814536bf9bSArtur Paszkiewicz * | -data- | -data- | ------ | | pp | | data0 ^ data1 ^ pp | 7824536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 7834536bf9bSArtur Paszkiewicz * pp_size = chunk_size 7844536bf9bSArtur Paszkiewicz * 7854536bf9bSArtur Paszkiewicz * case 4: 7864536bf9bSArtur Paszkiewicz * data0 data1 data2 ppl parity 7874536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 7884536bf9bSArtur Paszkiewicz * | ------ | -data- | ------ | | pp | | data1 ^ pp | 7894536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | -- | -> | (no change) | 7904536bf9bSArtur Paszkiewicz * | ------ | ------ | ------ | | -- | -> | (no change) | 7914536bf9bSArtur Paszkiewicz * | -data- | ------ | ------ | | pp | | data0 ^ pp | 7924536bf9bSArtur Paszkiewicz * +--------+--------+--------+ +----+ +--------------------+ 7934536bf9bSArtur Paszkiewicz * pp_size = chunk_size 7944536bf9bSArtur Paszkiewicz */ 7954536bf9bSArtur Paszkiewicz static int ppl_recover_entry(struct ppl_log *log, struct ppl_header_entry *e, 7964536bf9bSArtur Paszkiewicz sector_t ppl_sector) 7974536bf9bSArtur Paszkiewicz { 7984536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 7994536bf9bSArtur Paszkiewicz struct mddev *mddev = ppl_conf->mddev; 8004536bf9bSArtur Paszkiewicz struct r5conf *conf = mddev->private; 8014536bf9bSArtur Paszkiewicz int block_size = ppl_conf->block_size; 8024536bf9bSArtur Paszkiewicz struct page *page1; 8034536bf9bSArtur Paszkiewicz struct page *page2; 8044536bf9bSArtur Paszkiewicz sector_t r_sector_first; 8054536bf9bSArtur Paszkiewicz sector_t r_sector_last; 8064536bf9bSArtur Paszkiewicz int strip_sectors; 8074536bf9bSArtur Paszkiewicz int data_disks; 8084536bf9bSArtur Paszkiewicz int i; 8094536bf9bSArtur Paszkiewicz int ret = 0; 8104536bf9bSArtur Paszkiewicz char b[BDEVNAME_SIZE]; 8114536bf9bSArtur Paszkiewicz unsigned int pp_size = le32_to_cpu(e->pp_size); 8124536bf9bSArtur Paszkiewicz unsigned int data_size = le32_to_cpu(e->data_size); 8134536bf9bSArtur Paszkiewicz 8144536bf9bSArtur Paszkiewicz page1 = alloc_page(GFP_KERNEL); 8154536bf9bSArtur Paszkiewicz page2 = alloc_page(GFP_KERNEL); 8164536bf9bSArtur Paszkiewicz 8174536bf9bSArtur Paszkiewicz if (!page1 || !page2) { 8184536bf9bSArtur Paszkiewicz ret = -ENOMEM; 8194536bf9bSArtur Paszkiewicz goto out; 8204536bf9bSArtur Paszkiewicz } 8214536bf9bSArtur Paszkiewicz 8224536bf9bSArtur Paszkiewicz r_sector_first = le64_to_cpu(e->data_sector) * (block_size >> 9); 8234536bf9bSArtur Paszkiewicz 8244536bf9bSArtur Paszkiewicz if ((pp_size >> 9) < conf->chunk_sectors) { 8254536bf9bSArtur Paszkiewicz if (pp_size > 0) { 8264536bf9bSArtur Paszkiewicz data_disks = data_size / pp_size; 8274536bf9bSArtur Paszkiewicz strip_sectors = pp_size >> 9; 8284536bf9bSArtur Paszkiewicz } else { 8294536bf9bSArtur Paszkiewicz data_disks = conf->raid_disks - conf->max_degraded; 8304536bf9bSArtur Paszkiewicz strip_sectors = (data_size >> 9) / data_disks; 8314536bf9bSArtur Paszkiewicz } 8324536bf9bSArtur Paszkiewicz r_sector_last = r_sector_first + 8334536bf9bSArtur Paszkiewicz (data_disks - 1) * conf->chunk_sectors + 8344536bf9bSArtur Paszkiewicz strip_sectors; 8354536bf9bSArtur Paszkiewicz } else { 8364536bf9bSArtur Paszkiewicz data_disks = conf->raid_disks - conf->max_degraded; 8374536bf9bSArtur Paszkiewicz strip_sectors = conf->chunk_sectors; 8384536bf9bSArtur Paszkiewicz r_sector_last = r_sector_first + (data_size >> 9); 8394536bf9bSArtur Paszkiewicz } 8404536bf9bSArtur Paszkiewicz 8414536bf9bSArtur Paszkiewicz pr_debug("%s: array sector first: %llu last: %llu\n", __func__, 8424536bf9bSArtur Paszkiewicz (unsigned long long)r_sector_first, 8434536bf9bSArtur Paszkiewicz (unsigned long long)r_sector_last); 8444536bf9bSArtur Paszkiewicz 8454536bf9bSArtur Paszkiewicz /* if start and end is 4k aligned, use a 4k block */ 8464536bf9bSArtur Paszkiewicz if (block_size == 512 && 8474536bf9bSArtur Paszkiewicz (r_sector_first & (STRIPE_SECTORS - 1)) == 0 && 8484536bf9bSArtur Paszkiewicz (r_sector_last & (STRIPE_SECTORS - 1)) == 0) 8494536bf9bSArtur Paszkiewicz block_size = STRIPE_SIZE; 8504536bf9bSArtur Paszkiewicz 8514536bf9bSArtur Paszkiewicz /* iterate through blocks in strip */ 8524536bf9bSArtur Paszkiewicz for (i = 0; i < strip_sectors; i += (block_size >> 9)) { 8534536bf9bSArtur Paszkiewicz bool update_parity = false; 8544536bf9bSArtur Paszkiewicz sector_t parity_sector; 8554536bf9bSArtur Paszkiewicz struct md_rdev *parity_rdev; 8564536bf9bSArtur Paszkiewicz struct stripe_head sh; 8574536bf9bSArtur Paszkiewicz int disk; 8584536bf9bSArtur Paszkiewicz int indent = 0; 8594536bf9bSArtur Paszkiewicz 8604536bf9bSArtur Paszkiewicz pr_debug("%s:%*s iter %d start\n", __func__, indent, "", i); 8614536bf9bSArtur Paszkiewicz indent += 2; 8624536bf9bSArtur Paszkiewicz 8634536bf9bSArtur Paszkiewicz memset(page_address(page1), 0, PAGE_SIZE); 8644536bf9bSArtur Paszkiewicz 8654536bf9bSArtur Paszkiewicz /* iterate through data member disks */ 8664536bf9bSArtur Paszkiewicz for (disk = 0; disk < data_disks; disk++) { 8674536bf9bSArtur Paszkiewicz int dd_idx; 8684536bf9bSArtur Paszkiewicz struct md_rdev *rdev; 8694536bf9bSArtur Paszkiewicz sector_t sector; 8704536bf9bSArtur Paszkiewicz sector_t r_sector = r_sector_first + i + 8714536bf9bSArtur Paszkiewicz (disk * conf->chunk_sectors); 8724536bf9bSArtur Paszkiewicz 8734536bf9bSArtur Paszkiewicz pr_debug("%s:%*s data member disk %d start\n", 8744536bf9bSArtur Paszkiewicz __func__, indent, "", disk); 8754536bf9bSArtur Paszkiewicz indent += 2; 8764536bf9bSArtur Paszkiewicz 8774536bf9bSArtur Paszkiewicz if (r_sector >= r_sector_last) { 8784536bf9bSArtur Paszkiewicz pr_debug("%s:%*s array sector %llu doesn't need parity update\n", 8794536bf9bSArtur Paszkiewicz __func__, indent, "", 8804536bf9bSArtur Paszkiewicz (unsigned long long)r_sector); 8814536bf9bSArtur Paszkiewicz indent -= 2; 8824536bf9bSArtur Paszkiewicz continue; 8834536bf9bSArtur Paszkiewicz } 8844536bf9bSArtur Paszkiewicz 8854536bf9bSArtur Paszkiewicz update_parity = true; 8864536bf9bSArtur Paszkiewicz 8874536bf9bSArtur Paszkiewicz /* map raid sector to member disk */ 8884536bf9bSArtur Paszkiewicz sector = raid5_compute_sector(conf, r_sector, 0, 8894536bf9bSArtur Paszkiewicz &dd_idx, NULL); 8904536bf9bSArtur Paszkiewicz pr_debug("%s:%*s processing array sector %llu => data member disk %d, sector %llu\n", 8914536bf9bSArtur Paszkiewicz __func__, indent, "", 8924536bf9bSArtur Paszkiewicz (unsigned long long)r_sector, dd_idx, 8934536bf9bSArtur Paszkiewicz (unsigned long long)sector); 8944536bf9bSArtur Paszkiewicz 8954536bf9bSArtur Paszkiewicz rdev = conf->disks[dd_idx].rdev; 89607719ff7SArtur Paszkiewicz if (!rdev || (!test_bit(In_sync, &rdev->flags) && 89707719ff7SArtur Paszkiewicz sector >= rdev->recovery_offset)) { 8984536bf9bSArtur Paszkiewicz pr_debug("%s:%*s data member disk %d missing\n", 8994536bf9bSArtur Paszkiewicz __func__, indent, "", dd_idx); 9004536bf9bSArtur Paszkiewicz update_parity = false; 9014536bf9bSArtur Paszkiewicz break; 9024536bf9bSArtur Paszkiewicz } 9034536bf9bSArtur Paszkiewicz 9044536bf9bSArtur Paszkiewicz pr_debug("%s:%*s reading data member disk %s sector %llu\n", 9054536bf9bSArtur Paszkiewicz __func__, indent, "", bdevname(rdev->bdev, b), 9064536bf9bSArtur Paszkiewicz (unsigned long long)sector); 9074536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, sector, block_size, page2, 9084536bf9bSArtur Paszkiewicz REQ_OP_READ, 0, false)) { 9094536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 9104536bf9bSArtur Paszkiewicz pr_debug("%s:%*s read failed!\n", __func__, 9114536bf9bSArtur Paszkiewicz indent, ""); 9124536bf9bSArtur Paszkiewicz ret = -EIO; 9134536bf9bSArtur Paszkiewicz goto out; 9144536bf9bSArtur Paszkiewicz } 9154536bf9bSArtur Paszkiewicz 9164536bf9bSArtur Paszkiewicz ppl_xor(block_size, page1, page2); 9174536bf9bSArtur Paszkiewicz 9184536bf9bSArtur Paszkiewicz indent -= 2; 9194536bf9bSArtur Paszkiewicz } 9204536bf9bSArtur Paszkiewicz 9214536bf9bSArtur Paszkiewicz if (!update_parity) 9224536bf9bSArtur Paszkiewicz continue; 9234536bf9bSArtur Paszkiewicz 9244536bf9bSArtur Paszkiewicz if (pp_size > 0) { 9254536bf9bSArtur Paszkiewicz pr_debug("%s:%*s reading pp disk sector %llu\n", 9264536bf9bSArtur Paszkiewicz __func__, indent, "", 9274536bf9bSArtur Paszkiewicz (unsigned long long)(ppl_sector + i)); 9284536bf9bSArtur Paszkiewicz if (!sync_page_io(log->rdev, 9294536bf9bSArtur Paszkiewicz ppl_sector - log->rdev->data_offset + i, 9304536bf9bSArtur Paszkiewicz block_size, page2, REQ_OP_READ, 0, 9314536bf9bSArtur Paszkiewicz false)) { 9324536bf9bSArtur Paszkiewicz pr_debug("%s:%*s read failed!\n", __func__, 9334536bf9bSArtur Paszkiewicz indent, ""); 9344536bf9bSArtur Paszkiewicz md_error(mddev, log->rdev); 9354536bf9bSArtur Paszkiewicz ret = -EIO; 9364536bf9bSArtur Paszkiewicz goto out; 9374536bf9bSArtur Paszkiewicz } 9384536bf9bSArtur Paszkiewicz 9394536bf9bSArtur Paszkiewicz ppl_xor(block_size, page1, page2); 9404536bf9bSArtur Paszkiewicz } 9414536bf9bSArtur Paszkiewicz 9424536bf9bSArtur Paszkiewicz /* map raid sector to parity disk */ 9434536bf9bSArtur Paszkiewicz parity_sector = raid5_compute_sector(conf, r_sector_first + i, 9444536bf9bSArtur Paszkiewicz 0, &disk, &sh); 9454536bf9bSArtur Paszkiewicz BUG_ON(sh.pd_idx != le32_to_cpu(e->parity_disk)); 9464536bf9bSArtur Paszkiewicz parity_rdev = conf->disks[sh.pd_idx].rdev; 9474536bf9bSArtur Paszkiewicz 9484536bf9bSArtur Paszkiewicz BUG_ON(parity_rdev->bdev->bd_dev != log->rdev->bdev->bd_dev); 9494536bf9bSArtur Paszkiewicz pr_debug("%s:%*s write parity at sector %llu, disk %s\n", 9504536bf9bSArtur Paszkiewicz __func__, indent, "", 9514536bf9bSArtur Paszkiewicz (unsigned long long)parity_sector, 9524536bf9bSArtur Paszkiewicz bdevname(parity_rdev->bdev, b)); 9534536bf9bSArtur Paszkiewicz if (!sync_page_io(parity_rdev, parity_sector, block_size, 9544536bf9bSArtur Paszkiewicz page1, REQ_OP_WRITE, 0, false)) { 9554536bf9bSArtur Paszkiewicz pr_debug("%s:%*s parity write error!\n", __func__, 9564536bf9bSArtur Paszkiewicz indent, ""); 9574536bf9bSArtur Paszkiewicz md_error(mddev, parity_rdev); 9584536bf9bSArtur Paszkiewicz ret = -EIO; 9594536bf9bSArtur Paszkiewicz goto out; 9604536bf9bSArtur Paszkiewicz } 9614536bf9bSArtur Paszkiewicz } 9624536bf9bSArtur Paszkiewicz out: 9634536bf9bSArtur Paszkiewicz if (page1) 9644536bf9bSArtur Paszkiewicz __free_page(page1); 9654536bf9bSArtur Paszkiewicz if (page2) 9664536bf9bSArtur Paszkiewicz __free_page(page2); 9674536bf9bSArtur Paszkiewicz return ret; 9684536bf9bSArtur Paszkiewicz } 9694536bf9bSArtur Paszkiewicz 970675dc2ccSPawel Baldysiak static int ppl_recover(struct ppl_log *log, struct ppl_header *pplhdr, 971675dc2ccSPawel Baldysiak sector_t offset) 9724536bf9bSArtur Paszkiewicz { 9734536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 9744536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 9754536bf9bSArtur Paszkiewicz struct mddev *mddev = rdev->mddev; 976675dc2ccSPawel Baldysiak sector_t ppl_sector = rdev->ppl.sector + offset + 977675dc2ccSPawel Baldysiak (PPL_HEADER_SIZE >> 9); 9784536bf9bSArtur Paszkiewicz struct page *page; 9794536bf9bSArtur Paszkiewicz int i; 9804536bf9bSArtur Paszkiewicz int ret = 0; 9814536bf9bSArtur Paszkiewicz 9824536bf9bSArtur Paszkiewicz page = alloc_page(GFP_KERNEL); 9834536bf9bSArtur Paszkiewicz if (!page) 9844536bf9bSArtur Paszkiewicz return -ENOMEM; 9854536bf9bSArtur Paszkiewicz 9864536bf9bSArtur Paszkiewicz /* iterate through all PPL entries saved */ 9874536bf9bSArtur Paszkiewicz for (i = 0; i < le32_to_cpu(pplhdr->entries_count); i++) { 9884536bf9bSArtur Paszkiewicz struct ppl_header_entry *e = &pplhdr->entries[i]; 9894536bf9bSArtur Paszkiewicz u32 pp_size = le32_to_cpu(e->pp_size); 9904536bf9bSArtur Paszkiewicz sector_t sector = ppl_sector; 9914536bf9bSArtur Paszkiewicz int ppl_entry_sectors = pp_size >> 9; 9924536bf9bSArtur Paszkiewicz u32 crc, crc_stored; 9934536bf9bSArtur Paszkiewicz 9944536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d entry: %d ppl_sector: %llu pp_size: %u\n", 9954536bf9bSArtur Paszkiewicz __func__, rdev->raid_disk, i, 9964536bf9bSArtur Paszkiewicz (unsigned long long)ppl_sector, pp_size); 9974536bf9bSArtur Paszkiewicz 9984536bf9bSArtur Paszkiewicz crc = ~0; 9994536bf9bSArtur Paszkiewicz crc_stored = le32_to_cpu(e->checksum); 10004536bf9bSArtur Paszkiewicz 10014536bf9bSArtur Paszkiewicz /* read parial parity for this entry and calculate its checksum */ 10024536bf9bSArtur Paszkiewicz while (pp_size) { 10034536bf9bSArtur Paszkiewicz int s = pp_size > PAGE_SIZE ? PAGE_SIZE : pp_size; 10044536bf9bSArtur Paszkiewicz 10054536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, sector - rdev->data_offset, 10064536bf9bSArtur Paszkiewicz s, page, REQ_OP_READ, 0, false)) { 10074536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 10084536bf9bSArtur Paszkiewicz ret = -EIO; 10094536bf9bSArtur Paszkiewicz goto out; 10104536bf9bSArtur Paszkiewicz } 10114536bf9bSArtur Paszkiewicz 10124536bf9bSArtur Paszkiewicz crc = crc32c_le(crc, page_address(page), s); 10134536bf9bSArtur Paszkiewicz 10144536bf9bSArtur Paszkiewicz pp_size -= s; 10154536bf9bSArtur Paszkiewicz sector += s >> 9; 10164536bf9bSArtur Paszkiewicz } 10174536bf9bSArtur Paszkiewicz 10184536bf9bSArtur Paszkiewicz crc = ~crc; 10194536bf9bSArtur Paszkiewicz 10204536bf9bSArtur Paszkiewicz if (crc != crc_stored) { 10214536bf9bSArtur Paszkiewicz /* 10224536bf9bSArtur Paszkiewicz * Don't recover this entry if the checksum does not 10234536bf9bSArtur Paszkiewicz * match, but keep going and try to recover other 10244536bf9bSArtur Paszkiewicz * entries. 10254536bf9bSArtur Paszkiewicz */ 10264536bf9bSArtur Paszkiewicz pr_debug("%s: ppl entry crc does not match: stored: 0x%x calculated: 0x%x\n", 10274536bf9bSArtur Paszkiewicz __func__, crc_stored, crc); 10284536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count++; 10294536bf9bSArtur Paszkiewicz } else { 10304536bf9bSArtur Paszkiewicz ret = ppl_recover_entry(log, e, ppl_sector); 10314536bf9bSArtur Paszkiewicz if (ret) 10324536bf9bSArtur Paszkiewicz goto out; 10334536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries++; 10344536bf9bSArtur Paszkiewicz } 10354536bf9bSArtur Paszkiewicz 10364536bf9bSArtur Paszkiewicz ppl_sector += ppl_entry_sectors; 10374536bf9bSArtur Paszkiewicz } 10384536bf9bSArtur Paszkiewicz 10394536bf9bSArtur Paszkiewicz /* flush the disk cache after recovery if necessary */ 10404536bf9bSArtur Paszkiewicz ret = blkdev_issue_flush(rdev->bdev, GFP_KERNEL, NULL); 10414536bf9bSArtur Paszkiewicz out: 10424536bf9bSArtur Paszkiewicz __free_page(page); 10434536bf9bSArtur Paszkiewicz return ret; 10444536bf9bSArtur Paszkiewicz } 10454536bf9bSArtur Paszkiewicz 10464536bf9bSArtur Paszkiewicz static int ppl_write_empty_header(struct ppl_log *log) 10474536bf9bSArtur Paszkiewicz { 10484536bf9bSArtur Paszkiewicz struct page *page; 10494536bf9bSArtur Paszkiewicz struct ppl_header *pplhdr; 10504536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 10514536bf9bSArtur Paszkiewicz int ret = 0; 10524536bf9bSArtur Paszkiewicz 10534536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d ppl_sector: %llu\n", __func__, 10544536bf9bSArtur Paszkiewicz rdev->raid_disk, (unsigned long long)rdev->ppl.sector); 10554536bf9bSArtur Paszkiewicz 10564536bf9bSArtur Paszkiewicz page = alloc_page(GFP_NOIO | __GFP_ZERO); 10574536bf9bSArtur Paszkiewicz if (!page) 10584536bf9bSArtur Paszkiewicz return -ENOMEM; 10594536bf9bSArtur Paszkiewicz 10604536bf9bSArtur Paszkiewicz pplhdr = page_address(page); 1061675dc2ccSPawel Baldysiak /* zero out PPL space to avoid collision with old PPLs */ 1062675dc2ccSPawel Baldysiak blkdev_issue_zeroout(rdev->bdev, rdev->ppl.sector, 1063675dc2ccSPawel Baldysiak log->rdev->ppl.size, GFP_NOIO, 0); 10644536bf9bSArtur Paszkiewicz memset(pplhdr->reserved, 0xff, PPL_HDR_RESERVED); 10654536bf9bSArtur Paszkiewicz pplhdr->signature = cpu_to_le32(log->ppl_conf->signature); 10664536bf9bSArtur Paszkiewicz pplhdr->checksum = cpu_to_le32(~crc32c_le(~0, pplhdr, PAGE_SIZE)); 10674536bf9bSArtur Paszkiewicz 10684536bf9bSArtur Paszkiewicz if (!sync_page_io(rdev, rdev->ppl.sector - rdev->data_offset, 10695a8948f8SJan Kara PPL_HEADER_SIZE, page, REQ_OP_WRITE | REQ_SYNC | 10705a8948f8SJan Kara REQ_FUA, 0, false)) { 10714536bf9bSArtur Paszkiewicz md_error(rdev->mddev, rdev); 10724536bf9bSArtur Paszkiewicz ret = -EIO; 10734536bf9bSArtur Paszkiewicz } 10744536bf9bSArtur Paszkiewicz 10754536bf9bSArtur Paszkiewicz __free_page(page); 10764536bf9bSArtur Paszkiewicz return ret; 10774536bf9bSArtur Paszkiewicz } 10784536bf9bSArtur Paszkiewicz 10794536bf9bSArtur Paszkiewicz static int ppl_load_distributed(struct ppl_log *log) 10804536bf9bSArtur Paszkiewicz { 10814536bf9bSArtur Paszkiewicz struct ppl_conf *ppl_conf = log->ppl_conf; 10824536bf9bSArtur Paszkiewicz struct md_rdev *rdev = log->rdev; 10834536bf9bSArtur Paszkiewicz struct mddev *mddev = rdev->mddev; 1084675dc2ccSPawel Baldysiak struct page *page, *page2, *tmp; 1085675dc2ccSPawel Baldysiak struct ppl_header *pplhdr = NULL, *prev_pplhdr = NULL; 10864536bf9bSArtur Paszkiewicz u32 crc, crc_stored; 10874536bf9bSArtur Paszkiewicz u32 signature; 1088675dc2ccSPawel Baldysiak int ret = 0, i; 1089675dc2ccSPawel Baldysiak sector_t pplhdr_offset = 0, prev_pplhdr_offset = 0; 10904536bf9bSArtur Paszkiewicz 10914536bf9bSArtur Paszkiewicz pr_debug("%s: disk: %d\n", __func__, rdev->raid_disk); 1092675dc2ccSPawel Baldysiak /* read PPL headers, find the recent one */ 10934536bf9bSArtur Paszkiewicz page = alloc_page(GFP_KERNEL); 10944536bf9bSArtur Paszkiewicz if (!page) 10954536bf9bSArtur Paszkiewicz return -ENOMEM; 10964536bf9bSArtur Paszkiewicz 1097675dc2ccSPawel Baldysiak page2 = alloc_page(GFP_KERNEL); 1098675dc2ccSPawel Baldysiak if (!page2) { 1099675dc2ccSPawel Baldysiak __free_page(page); 1100675dc2ccSPawel Baldysiak return -ENOMEM; 1101675dc2ccSPawel Baldysiak } 1102675dc2ccSPawel Baldysiak 1103675dc2ccSPawel Baldysiak /* searching ppl area for latest ppl */ 1104675dc2ccSPawel Baldysiak while (pplhdr_offset < rdev->ppl.size - (PPL_HEADER_SIZE >> 9)) { 1105675dc2ccSPawel Baldysiak if (!sync_page_io(rdev, 1106675dc2ccSPawel Baldysiak rdev->ppl.sector - rdev->data_offset + 1107675dc2ccSPawel Baldysiak pplhdr_offset, PAGE_SIZE, page, REQ_OP_READ, 1108675dc2ccSPawel Baldysiak 0, false)) { 11094536bf9bSArtur Paszkiewicz md_error(mddev, rdev); 11104536bf9bSArtur Paszkiewicz ret = -EIO; 1111675dc2ccSPawel Baldysiak /* if not able to read - don't recover any PPL */ 1112675dc2ccSPawel Baldysiak pplhdr = NULL; 1113675dc2ccSPawel Baldysiak break; 11144536bf9bSArtur Paszkiewicz } 11154536bf9bSArtur Paszkiewicz pplhdr = page_address(page); 11164536bf9bSArtur Paszkiewicz 11174536bf9bSArtur Paszkiewicz /* check header validity */ 11184536bf9bSArtur Paszkiewicz crc_stored = le32_to_cpu(pplhdr->checksum); 11194536bf9bSArtur Paszkiewicz pplhdr->checksum = 0; 11204536bf9bSArtur Paszkiewicz crc = ~crc32c_le(~0, pplhdr, PAGE_SIZE); 11214536bf9bSArtur Paszkiewicz 11224536bf9bSArtur Paszkiewicz if (crc_stored != crc) { 1123675dc2ccSPawel Baldysiak pr_debug("%s: ppl header crc does not match: stored: 0x%x calculated: 0x%x (offset: %llu)\n", 1124675dc2ccSPawel Baldysiak __func__, crc_stored, crc, 1125675dc2ccSPawel Baldysiak (unsigned long long)pplhdr_offset); 1126675dc2ccSPawel Baldysiak pplhdr = prev_pplhdr; 1127675dc2ccSPawel Baldysiak pplhdr_offset = prev_pplhdr_offset; 1128675dc2ccSPawel Baldysiak break; 11294536bf9bSArtur Paszkiewicz } 11304536bf9bSArtur Paszkiewicz 11314536bf9bSArtur Paszkiewicz signature = le32_to_cpu(pplhdr->signature); 11324536bf9bSArtur Paszkiewicz 11334536bf9bSArtur Paszkiewicz if (mddev->external) { 11344536bf9bSArtur Paszkiewicz /* 11354536bf9bSArtur Paszkiewicz * For external metadata the header signature is set and 11364536bf9bSArtur Paszkiewicz * validated in userspace. 11374536bf9bSArtur Paszkiewicz */ 11384536bf9bSArtur Paszkiewicz ppl_conf->signature = signature; 11394536bf9bSArtur Paszkiewicz } else if (ppl_conf->signature != signature) { 1140675dc2ccSPawel Baldysiak pr_debug("%s: ppl header signature does not match: stored: 0x%x configured: 0x%x (offset: %llu)\n", 1141675dc2ccSPawel Baldysiak __func__, signature, ppl_conf->signature, 1142675dc2ccSPawel Baldysiak (unsigned long long)pplhdr_offset); 1143675dc2ccSPawel Baldysiak pplhdr = prev_pplhdr; 1144675dc2ccSPawel Baldysiak pplhdr_offset = prev_pplhdr_offset; 1145675dc2ccSPawel Baldysiak break; 11464536bf9bSArtur Paszkiewicz } 11474536bf9bSArtur Paszkiewicz 1148675dc2ccSPawel Baldysiak if (prev_pplhdr && le64_to_cpu(prev_pplhdr->generation) > 1149675dc2ccSPawel Baldysiak le64_to_cpu(pplhdr->generation)) { 1150675dc2ccSPawel Baldysiak /* previous was newest */ 1151675dc2ccSPawel Baldysiak pplhdr = prev_pplhdr; 1152675dc2ccSPawel Baldysiak pplhdr_offset = prev_pplhdr_offset; 1153675dc2ccSPawel Baldysiak break; 1154675dc2ccSPawel Baldysiak } 1155675dc2ccSPawel Baldysiak 1156675dc2ccSPawel Baldysiak prev_pplhdr_offset = pplhdr_offset; 1157675dc2ccSPawel Baldysiak prev_pplhdr = pplhdr; 1158675dc2ccSPawel Baldysiak 1159675dc2ccSPawel Baldysiak tmp = page; 1160675dc2ccSPawel Baldysiak page = page2; 1161675dc2ccSPawel Baldysiak page2 = tmp; 1162675dc2ccSPawel Baldysiak 1163675dc2ccSPawel Baldysiak /* calculate next potential ppl offset */ 1164675dc2ccSPawel Baldysiak for (i = 0; i < le32_to_cpu(pplhdr->entries_count); i++) 1165675dc2ccSPawel Baldysiak pplhdr_offset += 1166675dc2ccSPawel Baldysiak le32_to_cpu(pplhdr->entries[i].pp_size) >> 9; 1167675dc2ccSPawel Baldysiak pplhdr_offset += PPL_HEADER_SIZE >> 9; 1168675dc2ccSPawel Baldysiak } 1169675dc2ccSPawel Baldysiak 1170675dc2ccSPawel Baldysiak /* no valid ppl found */ 1171675dc2ccSPawel Baldysiak if (!pplhdr) 1172675dc2ccSPawel Baldysiak ppl_conf->mismatch_count++; 1173675dc2ccSPawel Baldysiak else 1174675dc2ccSPawel Baldysiak pr_debug("%s: latest PPL found at offset: %llu, with generation: %llu\n", 1175675dc2ccSPawel Baldysiak __func__, (unsigned long long)pplhdr_offset, 1176675dc2ccSPawel Baldysiak le64_to_cpu(pplhdr->generation)); 1177675dc2ccSPawel Baldysiak 11784536bf9bSArtur Paszkiewicz /* attempt to recover from log if we are starting a dirty array */ 1179675dc2ccSPawel Baldysiak if (pplhdr && !mddev->pers && mddev->recovery_cp != MaxSector) 1180675dc2ccSPawel Baldysiak ret = ppl_recover(log, pplhdr, pplhdr_offset); 1181675dc2ccSPawel Baldysiak 11824536bf9bSArtur Paszkiewicz /* write empty header if we are starting the array */ 11834536bf9bSArtur Paszkiewicz if (!ret && !mddev->pers) 11844536bf9bSArtur Paszkiewicz ret = ppl_write_empty_header(log); 11854536bf9bSArtur Paszkiewicz 11864536bf9bSArtur Paszkiewicz __free_page(page); 1187675dc2ccSPawel Baldysiak __free_page(page2); 11884536bf9bSArtur Paszkiewicz 11894536bf9bSArtur Paszkiewicz pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n", 11904536bf9bSArtur Paszkiewicz __func__, ret, ppl_conf->mismatch_count, 11914536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries); 11924536bf9bSArtur Paszkiewicz return ret; 11934536bf9bSArtur Paszkiewicz } 11944536bf9bSArtur Paszkiewicz 11954536bf9bSArtur Paszkiewicz static int ppl_load(struct ppl_conf *ppl_conf) 11964536bf9bSArtur Paszkiewicz { 11974536bf9bSArtur Paszkiewicz int ret = 0; 11984536bf9bSArtur Paszkiewicz u32 signature = 0; 11994536bf9bSArtur Paszkiewicz bool signature_set = false; 12004536bf9bSArtur Paszkiewicz int i; 12014536bf9bSArtur Paszkiewicz 12024536bf9bSArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 12034536bf9bSArtur Paszkiewicz struct ppl_log *log = &ppl_conf->child_logs[i]; 12044536bf9bSArtur Paszkiewicz 12054536bf9bSArtur Paszkiewicz /* skip missing drive */ 12064536bf9bSArtur Paszkiewicz if (!log->rdev) 12074536bf9bSArtur Paszkiewicz continue; 12084536bf9bSArtur Paszkiewicz 12094536bf9bSArtur Paszkiewicz ret = ppl_load_distributed(log); 12104536bf9bSArtur Paszkiewicz if (ret) 12114536bf9bSArtur Paszkiewicz break; 12124536bf9bSArtur Paszkiewicz 12134536bf9bSArtur Paszkiewicz /* 12144536bf9bSArtur Paszkiewicz * For external metadata we can't check if the signature is 12154536bf9bSArtur Paszkiewicz * correct on a single drive, but we can check if it is the same 12164536bf9bSArtur Paszkiewicz * on all drives. 12174536bf9bSArtur Paszkiewicz */ 12184536bf9bSArtur Paszkiewicz if (ppl_conf->mddev->external) { 12194536bf9bSArtur Paszkiewicz if (!signature_set) { 12204536bf9bSArtur Paszkiewicz signature = ppl_conf->signature; 12214536bf9bSArtur Paszkiewicz signature_set = true; 12224536bf9bSArtur Paszkiewicz } else if (signature != ppl_conf->signature) { 12234536bf9bSArtur Paszkiewicz pr_warn("md/raid:%s: PPL header signature does not match on all member drives\n", 12244536bf9bSArtur Paszkiewicz mdname(ppl_conf->mddev)); 12254536bf9bSArtur Paszkiewicz ret = -EINVAL; 12264536bf9bSArtur Paszkiewicz break; 12274536bf9bSArtur Paszkiewicz } 12284536bf9bSArtur Paszkiewicz } 12294536bf9bSArtur Paszkiewicz } 12304536bf9bSArtur Paszkiewicz 12314536bf9bSArtur Paszkiewicz pr_debug("%s: return: %d mismatch_count: %d recovered_entries: %d\n", 12324536bf9bSArtur Paszkiewicz __func__, ret, ppl_conf->mismatch_count, 12334536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries); 12344536bf9bSArtur Paszkiewicz return ret; 12354536bf9bSArtur Paszkiewicz } 12364536bf9bSArtur Paszkiewicz 12373418d036SArtur Paszkiewicz static void __ppl_exit_log(struct ppl_conf *ppl_conf) 12383418d036SArtur Paszkiewicz { 12393418d036SArtur Paszkiewicz clear_bit(MD_HAS_PPL, &ppl_conf->mddev->flags); 1240ddc08823SPawel Baldysiak clear_bit(MD_HAS_MULTIPLE_PPLS, &ppl_conf->mddev->flags); 12413418d036SArtur Paszkiewicz 12423418d036SArtur Paszkiewicz kfree(ppl_conf->child_logs); 12433418d036SArtur Paszkiewicz 1244afeee514SKent Overstreet bioset_exit(&ppl_conf->bs); 1245afeee514SKent Overstreet bioset_exit(&ppl_conf->flush_bs); 1246afeee514SKent Overstreet mempool_exit(&ppl_conf->io_pool); 12473418d036SArtur Paszkiewicz kmem_cache_destroy(ppl_conf->io_kc); 12483418d036SArtur Paszkiewicz 12493418d036SArtur Paszkiewicz kfree(ppl_conf); 12503418d036SArtur Paszkiewicz } 12513418d036SArtur Paszkiewicz 12523418d036SArtur Paszkiewicz void ppl_exit_log(struct r5conf *conf) 12533418d036SArtur Paszkiewicz { 12543418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 12553418d036SArtur Paszkiewicz 12563418d036SArtur Paszkiewicz if (ppl_conf) { 12573418d036SArtur Paszkiewicz __ppl_exit_log(ppl_conf); 12583418d036SArtur Paszkiewicz conf->log_private = NULL; 12593418d036SArtur Paszkiewicz } 12603418d036SArtur Paszkiewicz } 12613418d036SArtur Paszkiewicz 12623418d036SArtur Paszkiewicz static int ppl_validate_rdev(struct md_rdev *rdev) 12633418d036SArtur Paszkiewicz { 12643418d036SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 12653418d036SArtur Paszkiewicz int ppl_data_sectors; 12663418d036SArtur Paszkiewicz int ppl_size_new; 12673418d036SArtur Paszkiewicz 12683418d036SArtur Paszkiewicz /* 12693418d036SArtur Paszkiewicz * The configured PPL size must be enough to store 12703418d036SArtur Paszkiewicz * the header and (at the very least) partial parity 12713418d036SArtur Paszkiewicz * for one stripe. Round it down to ensure the data 12723418d036SArtur Paszkiewicz * space is cleanly divisible by stripe size. 12733418d036SArtur Paszkiewicz */ 12743418d036SArtur Paszkiewicz ppl_data_sectors = rdev->ppl.size - (PPL_HEADER_SIZE >> 9); 12753418d036SArtur Paszkiewicz 12763418d036SArtur Paszkiewicz if (ppl_data_sectors > 0) 12773418d036SArtur Paszkiewicz ppl_data_sectors = rounddown(ppl_data_sectors, STRIPE_SECTORS); 12783418d036SArtur Paszkiewicz 12793418d036SArtur Paszkiewicz if (ppl_data_sectors <= 0) { 12803418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space too small on %s\n", 12813418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 12823418d036SArtur Paszkiewicz return -ENOSPC; 12833418d036SArtur Paszkiewicz } 12843418d036SArtur Paszkiewicz 12853418d036SArtur Paszkiewicz ppl_size_new = ppl_data_sectors + (PPL_HEADER_SIZE >> 9); 12863418d036SArtur Paszkiewicz 12873418d036SArtur Paszkiewicz if ((rdev->ppl.sector < rdev->data_offset && 12883418d036SArtur Paszkiewicz rdev->ppl.sector + ppl_size_new > rdev->data_offset) || 12893418d036SArtur Paszkiewicz (rdev->ppl.sector >= rdev->data_offset && 12903418d036SArtur Paszkiewicz rdev->data_offset + rdev->sectors > rdev->ppl.sector)) { 12913418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space overlaps with data on %s\n", 12923418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 12933418d036SArtur Paszkiewicz return -EINVAL; 12943418d036SArtur Paszkiewicz } 12953418d036SArtur Paszkiewicz 12963418d036SArtur Paszkiewicz if (!rdev->mddev->external && 12973418d036SArtur Paszkiewicz ((rdev->ppl.offset > 0 && rdev->ppl.offset < (rdev->sb_size >> 9)) || 12983418d036SArtur Paszkiewicz (rdev->ppl.offset <= 0 && rdev->ppl.offset + ppl_size_new > 0))) { 12993418d036SArtur Paszkiewicz pr_warn("md/raid:%s: PPL space overlaps with superblock on %s\n", 13003418d036SArtur Paszkiewicz mdname(rdev->mddev), bdevname(rdev->bdev, b)); 13013418d036SArtur Paszkiewicz return -EINVAL; 13023418d036SArtur Paszkiewicz } 13033418d036SArtur Paszkiewicz 13043418d036SArtur Paszkiewicz rdev->ppl.size = ppl_size_new; 13053418d036SArtur Paszkiewicz 13063418d036SArtur Paszkiewicz return 0; 13073418d036SArtur Paszkiewicz } 13083418d036SArtur Paszkiewicz 1309ddc08823SPawel Baldysiak static void ppl_init_child_log(struct ppl_log *log, struct md_rdev *rdev) 1310ddc08823SPawel Baldysiak { 13111532d9e8STomasz Majchrzak struct request_queue *q; 13121532d9e8STomasz Majchrzak 1313ddc08823SPawel Baldysiak if ((rdev->ppl.size << 9) >= (PPL_SPACE_SIZE + 1314ddc08823SPawel Baldysiak PPL_HEADER_SIZE) * 2) { 1315ddc08823SPawel Baldysiak log->use_multippl = true; 1316ddc08823SPawel Baldysiak set_bit(MD_HAS_MULTIPLE_PPLS, 1317ddc08823SPawel Baldysiak &log->ppl_conf->mddev->flags); 1318ddc08823SPawel Baldysiak log->entry_space = PPL_SPACE_SIZE; 1319ddc08823SPawel Baldysiak } else { 1320ddc08823SPawel Baldysiak log->use_multippl = false; 1321ddc08823SPawel Baldysiak log->entry_space = (log->rdev->ppl.size << 9) - 1322ddc08823SPawel Baldysiak PPL_HEADER_SIZE; 1323ddc08823SPawel Baldysiak } 1324ddc08823SPawel Baldysiak log->next_io_sector = rdev->ppl.sector; 13251532d9e8STomasz Majchrzak 13261532d9e8STomasz Majchrzak q = bdev_get_queue(rdev->bdev); 13271532d9e8STomasz Majchrzak if (test_bit(QUEUE_FLAG_WC, &q->queue_flags)) 13281532d9e8STomasz Majchrzak log->wb_cache_on = true; 1329ddc08823SPawel Baldysiak } 1330ddc08823SPawel Baldysiak 13313418d036SArtur Paszkiewicz int ppl_init_log(struct r5conf *conf) 13323418d036SArtur Paszkiewicz { 13333418d036SArtur Paszkiewicz struct ppl_conf *ppl_conf; 13343418d036SArtur Paszkiewicz struct mddev *mddev = conf->mddev; 13353418d036SArtur Paszkiewicz int ret = 0; 13361532d9e8STomasz Majchrzak int max_disks; 13373418d036SArtur Paszkiewicz int i; 13383418d036SArtur Paszkiewicz 13393418d036SArtur Paszkiewicz pr_debug("md/raid:%s: enabling distributed Partial Parity Log\n", 13403418d036SArtur Paszkiewicz mdname(conf->mddev)); 13413418d036SArtur Paszkiewicz 13423418d036SArtur Paszkiewicz if (PAGE_SIZE != 4096) 13433418d036SArtur Paszkiewicz return -EINVAL; 13443418d036SArtur Paszkiewicz 13453418d036SArtur Paszkiewicz if (mddev->level != 5) { 13463418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with raid level %d\n", 13473418d036SArtur Paszkiewicz mdname(mddev), mddev->level); 13483418d036SArtur Paszkiewicz return -EINVAL; 13493418d036SArtur Paszkiewicz } 13503418d036SArtur Paszkiewicz 13513418d036SArtur Paszkiewicz if (mddev->bitmap_info.file || mddev->bitmap_info.offset) { 13523418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with bitmap\n", 13533418d036SArtur Paszkiewicz mdname(mddev)); 13543418d036SArtur Paszkiewicz return -EINVAL; 13553418d036SArtur Paszkiewicz } 13563418d036SArtur Paszkiewicz 13573418d036SArtur Paszkiewicz if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { 13583418d036SArtur Paszkiewicz pr_warn("md/raid:%s PPL is not compatible with journal\n", 13593418d036SArtur Paszkiewicz mdname(mddev)); 13603418d036SArtur Paszkiewicz return -EINVAL; 13613418d036SArtur Paszkiewicz } 13623418d036SArtur Paszkiewicz 13631532d9e8STomasz Majchrzak max_disks = FIELD_SIZEOF(struct ppl_log, disk_flush_bitmap) * 13641532d9e8STomasz Majchrzak BITS_PER_BYTE; 13651532d9e8STomasz Majchrzak if (conf->raid_disks > max_disks) { 13661532d9e8STomasz Majchrzak pr_warn("md/raid:%s PPL doesn't support over %d disks in the array\n", 13671532d9e8STomasz Majchrzak mdname(mddev), max_disks); 13681532d9e8STomasz Majchrzak return -EINVAL; 13691532d9e8STomasz Majchrzak } 13701532d9e8STomasz Majchrzak 13713418d036SArtur Paszkiewicz ppl_conf = kzalloc(sizeof(struct ppl_conf), GFP_KERNEL); 13723418d036SArtur Paszkiewicz if (!ppl_conf) 13733418d036SArtur Paszkiewicz return -ENOMEM; 13743418d036SArtur Paszkiewicz 13753418d036SArtur Paszkiewicz ppl_conf->mddev = mddev; 13763418d036SArtur Paszkiewicz 13773418d036SArtur Paszkiewicz ppl_conf->io_kc = KMEM_CACHE(ppl_io_unit, 0); 13783418d036SArtur Paszkiewicz if (!ppl_conf->io_kc) { 1379fcd403afSArtur Paszkiewicz ret = -ENOMEM; 13803418d036SArtur Paszkiewicz goto err; 13813418d036SArtur Paszkiewicz } 13823418d036SArtur Paszkiewicz 1383afeee514SKent Overstreet ret = mempool_init(&ppl_conf->io_pool, conf->raid_disks, ppl_io_pool_alloc, 1384fcd403afSArtur Paszkiewicz ppl_io_pool_free, ppl_conf->io_kc); 1385afeee514SKent Overstreet if (ret) 13863418d036SArtur Paszkiewicz goto err; 13873418d036SArtur Paszkiewicz 1388afeee514SKent Overstreet ret = bioset_init(&ppl_conf->bs, conf->raid_disks, 0, BIOSET_NEED_BVECS); 1389afeee514SKent Overstreet if (ret) 13903418d036SArtur Paszkiewicz goto err; 13913418d036SArtur Paszkiewicz 1392afeee514SKent Overstreet ret = bioset_init(&ppl_conf->flush_bs, conf->raid_disks, 0, 0); 1393afeee514SKent Overstreet if (ret) 13941532d9e8STomasz Majchrzak goto err; 13951532d9e8STomasz Majchrzak 13963418d036SArtur Paszkiewicz ppl_conf->count = conf->raid_disks; 13973418d036SArtur Paszkiewicz ppl_conf->child_logs = kcalloc(ppl_conf->count, sizeof(struct ppl_log), 13983418d036SArtur Paszkiewicz GFP_KERNEL); 13993418d036SArtur Paszkiewicz if (!ppl_conf->child_logs) { 14003418d036SArtur Paszkiewicz ret = -ENOMEM; 14013418d036SArtur Paszkiewicz goto err; 14023418d036SArtur Paszkiewicz } 14033418d036SArtur Paszkiewicz 14043418d036SArtur Paszkiewicz atomic64_set(&ppl_conf->seq, 0); 140594568f64SArtur Paszkiewicz INIT_LIST_HEAD(&ppl_conf->no_mem_stripes); 140694568f64SArtur Paszkiewicz spin_lock_init(&ppl_conf->no_mem_stripes_lock); 14070815ef3cSEugene Syromiatnikov ppl_conf->write_hint = RWH_WRITE_LIFE_NOT_SET; 14083418d036SArtur Paszkiewicz 14093418d036SArtur Paszkiewicz if (!mddev->external) { 14103418d036SArtur Paszkiewicz ppl_conf->signature = ~crc32c_le(~0, mddev->uuid, sizeof(mddev->uuid)); 14113418d036SArtur Paszkiewicz ppl_conf->block_size = 512; 14123418d036SArtur Paszkiewicz } else { 14133418d036SArtur Paszkiewicz ppl_conf->block_size = queue_logical_block_size(mddev->queue); 14143418d036SArtur Paszkiewicz } 14153418d036SArtur Paszkiewicz 14163418d036SArtur Paszkiewicz for (i = 0; i < ppl_conf->count; i++) { 14173418d036SArtur Paszkiewicz struct ppl_log *log = &ppl_conf->child_logs[i]; 14183418d036SArtur Paszkiewicz struct md_rdev *rdev = conf->disks[i].rdev; 14193418d036SArtur Paszkiewicz 14203418d036SArtur Paszkiewicz mutex_init(&log->io_mutex); 14213418d036SArtur Paszkiewicz spin_lock_init(&log->io_list_lock); 14223418d036SArtur Paszkiewicz INIT_LIST_HEAD(&log->io_list); 14233418d036SArtur Paszkiewicz 14243418d036SArtur Paszkiewicz log->ppl_conf = ppl_conf; 14253418d036SArtur Paszkiewicz log->rdev = rdev; 14263418d036SArtur Paszkiewicz 14273418d036SArtur Paszkiewicz if (rdev) { 14283418d036SArtur Paszkiewicz ret = ppl_validate_rdev(rdev); 14293418d036SArtur Paszkiewicz if (ret) 14303418d036SArtur Paszkiewicz goto err; 14313418d036SArtur Paszkiewicz 1432ddc08823SPawel Baldysiak ppl_init_child_log(log, rdev); 14333418d036SArtur Paszkiewicz } 14343418d036SArtur Paszkiewicz } 14353418d036SArtur Paszkiewicz 14364536bf9bSArtur Paszkiewicz /* load and possibly recover the logs from the member disks */ 14374536bf9bSArtur Paszkiewicz ret = ppl_load(ppl_conf); 14384536bf9bSArtur Paszkiewicz 14394536bf9bSArtur Paszkiewicz if (ret) { 14404536bf9bSArtur Paszkiewicz goto err; 1441611426e2SArtur Paszkiewicz } else if (!mddev->pers && mddev->recovery_cp == 0 && 14424536bf9bSArtur Paszkiewicz ppl_conf->recovered_entries > 0 && 14434536bf9bSArtur Paszkiewicz ppl_conf->mismatch_count == 0) { 14444536bf9bSArtur Paszkiewicz /* 14454536bf9bSArtur Paszkiewicz * If we are starting a dirty array and the recovery succeeds 14464536bf9bSArtur Paszkiewicz * without any issues, set the array as clean. 14474536bf9bSArtur Paszkiewicz */ 14484536bf9bSArtur Paszkiewicz mddev->recovery_cp = MaxSector; 14494536bf9bSArtur Paszkiewicz set_bit(MD_SB_CHANGE_CLEAN, &mddev->sb_flags); 1450ba903a3eSArtur Paszkiewicz } else if (mddev->pers && ppl_conf->mismatch_count > 0) { 1451ba903a3eSArtur Paszkiewicz /* no mismatch allowed when enabling PPL for a running array */ 1452ba903a3eSArtur Paszkiewicz ret = -EINVAL; 1453ba903a3eSArtur Paszkiewicz goto err; 14544536bf9bSArtur Paszkiewicz } 14554536bf9bSArtur Paszkiewicz 14563418d036SArtur Paszkiewicz conf->log_private = ppl_conf; 1457845b9e22SArtur Paszkiewicz set_bit(MD_HAS_PPL, &ppl_conf->mddev->flags); 14583418d036SArtur Paszkiewicz 14593418d036SArtur Paszkiewicz return 0; 14603418d036SArtur Paszkiewicz err: 14613418d036SArtur Paszkiewicz __ppl_exit_log(ppl_conf); 14623418d036SArtur Paszkiewicz return ret; 14633418d036SArtur Paszkiewicz } 14646358c239SArtur Paszkiewicz 14656358c239SArtur Paszkiewicz int ppl_modify_log(struct r5conf *conf, struct md_rdev *rdev, bool add) 14666358c239SArtur Paszkiewicz { 14676358c239SArtur Paszkiewicz struct ppl_conf *ppl_conf = conf->log_private; 14686358c239SArtur Paszkiewicz struct ppl_log *log; 14696358c239SArtur Paszkiewicz int ret = 0; 14706358c239SArtur Paszkiewicz char b[BDEVNAME_SIZE]; 14716358c239SArtur Paszkiewicz 14726358c239SArtur Paszkiewicz if (!rdev) 14736358c239SArtur Paszkiewicz return -EINVAL; 14746358c239SArtur Paszkiewicz 14756358c239SArtur Paszkiewicz pr_debug("%s: disk: %d operation: %s dev: %s\n", 14766358c239SArtur Paszkiewicz __func__, rdev->raid_disk, add ? "add" : "remove", 14776358c239SArtur Paszkiewicz bdevname(rdev->bdev, b)); 14786358c239SArtur Paszkiewicz 14796358c239SArtur Paszkiewicz if (rdev->raid_disk < 0) 14806358c239SArtur Paszkiewicz return 0; 14816358c239SArtur Paszkiewicz 14826358c239SArtur Paszkiewicz if (rdev->raid_disk >= ppl_conf->count) 14836358c239SArtur Paszkiewicz return -ENODEV; 14846358c239SArtur Paszkiewicz 14856358c239SArtur Paszkiewicz log = &ppl_conf->child_logs[rdev->raid_disk]; 14866358c239SArtur Paszkiewicz 14876358c239SArtur Paszkiewicz mutex_lock(&log->io_mutex); 14886358c239SArtur Paszkiewicz if (add) { 14896358c239SArtur Paszkiewicz ret = ppl_validate_rdev(rdev); 14906358c239SArtur Paszkiewicz if (!ret) { 14916358c239SArtur Paszkiewicz log->rdev = rdev; 14926358c239SArtur Paszkiewicz ret = ppl_write_empty_header(log); 1493ddc08823SPawel Baldysiak ppl_init_child_log(log, rdev); 14946358c239SArtur Paszkiewicz } 14956358c239SArtur Paszkiewicz } else { 14966358c239SArtur Paszkiewicz log->rdev = NULL; 14976358c239SArtur Paszkiewicz } 14986358c239SArtur Paszkiewicz mutex_unlock(&log->io_mutex); 14996358c239SArtur Paszkiewicz 15006358c239SArtur Paszkiewicz return ret; 15016358c239SArtur Paszkiewicz } 1502a596d086SMariusz Dabrowski 1503a596d086SMariusz Dabrowski static ssize_t 1504a596d086SMariusz Dabrowski ppl_write_hint_show(struct mddev *mddev, char *buf) 1505a596d086SMariusz Dabrowski { 1506a596d086SMariusz Dabrowski size_t ret = 0; 1507a596d086SMariusz Dabrowski struct r5conf *conf; 1508a596d086SMariusz Dabrowski struct ppl_conf *ppl_conf = NULL; 1509a596d086SMariusz Dabrowski 1510a596d086SMariusz Dabrowski spin_lock(&mddev->lock); 1511a596d086SMariusz Dabrowski conf = mddev->private; 1512a596d086SMariusz Dabrowski if (conf && raid5_has_ppl(conf)) 1513a596d086SMariusz Dabrowski ppl_conf = conf->log_private; 1514a596d086SMariusz Dabrowski ret = sprintf(buf, "%d\n", ppl_conf ? ppl_conf->write_hint : 0); 1515a596d086SMariusz Dabrowski spin_unlock(&mddev->lock); 1516a596d086SMariusz Dabrowski 1517a596d086SMariusz Dabrowski return ret; 1518a596d086SMariusz Dabrowski } 1519a596d086SMariusz Dabrowski 1520a596d086SMariusz Dabrowski static ssize_t 1521a596d086SMariusz Dabrowski ppl_write_hint_store(struct mddev *mddev, const char *page, size_t len) 1522a596d086SMariusz Dabrowski { 1523a596d086SMariusz Dabrowski struct r5conf *conf; 1524a596d086SMariusz Dabrowski struct ppl_conf *ppl_conf; 1525a596d086SMariusz Dabrowski int err = 0; 1526a596d086SMariusz Dabrowski unsigned short new; 1527a596d086SMariusz Dabrowski 1528a596d086SMariusz Dabrowski if (len >= PAGE_SIZE) 1529a596d086SMariusz Dabrowski return -EINVAL; 1530a596d086SMariusz Dabrowski if (kstrtou16(page, 10, &new)) 1531a596d086SMariusz Dabrowski return -EINVAL; 1532a596d086SMariusz Dabrowski 1533a596d086SMariusz Dabrowski err = mddev_lock(mddev); 1534a596d086SMariusz Dabrowski if (err) 1535a596d086SMariusz Dabrowski return err; 1536a596d086SMariusz Dabrowski 1537a596d086SMariusz Dabrowski conf = mddev->private; 1538a596d086SMariusz Dabrowski if (!conf) { 1539a596d086SMariusz Dabrowski err = -ENODEV; 1540a596d086SMariusz Dabrowski } else if (raid5_has_ppl(conf)) { 1541a596d086SMariusz Dabrowski ppl_conf = conf->log_private; 1542a596d086SMariusz Dabrowski if (!ppl_conf) 1543a596d086SMariusz Dabrowski err = -EINVAL; 1544a596d086SMariusz Dabrowski else 1545a596d086SMariusz Dabrowski ppl_conf->write_hint = new; 1546a596d086SMariusz Dabrowski } else { 1547a596d086SMariusz Dabrowski err = -EINVAL; 1548a596d086SMariusz Dabrowski } 1549a596d086SMariusz Dabrowski 1550a596d086SMariusz Dabrowski mddev_unlock(mddev); 1551a596d086SMariusz Dabrowski 1552a596d086SMariusz Dabrowski return err ?: len; 1553a596d086SMariusz Dabrowski } 1554a596d086SMariusz Dabrowski 1555a596d086SMariusz Dabrowski struct md_sysfs_entry 1556a596d086SMariusz Dabrowski ppl_write_hint = __ATTR(ppl_write_hint, S_IRUGO | S_IWUSR, 1557a596d086SMariusz Dabrowski ppl_write_hint_show, 1558a596d086SMariusz Dabrowski ppl_write_hint_store); 1559