14db6bfe0SAlasdair G Kergon /* 24db6bfe0SAlasdair G Kergon * Copyright (C) 2001-2002 Sistina Software (UK) Limited. 34db6bfe0SAlasdair G Kergon * Copyright (C) 2006-2008 Red Hat GmbH 44db6bfe0SAlasdair G Kergon * 54db6bfe0SAlasdair G Kergon * This file is released under the GPL. 64db6bfe0SAlasdair G Kergon */ 74db6bfe0SAlasdair G Kergon 84db6bfe0SAlasdair G Kergon #include "dm-exception-store.h" 94db6bfe0SAlasdair G Kergon 10b0d3cc01SMike Snitzer #include <linux/ctype.h> 114db6bfe0SAlasdair G Kergon #include <linux/mm.h> 124db6bfe0SAlasdair G Kergon #include <linux/pagemap.h> 134db6bfe0SAlasdair G Kergon #include <linux/vmalloc.h> 14daaa5f7cSPaul Gortmaker #include <linux/export.h> 154db6bfe0SAlasdair G Kergon #include <linux/slab.h> 164db6bfe0SAlasdair G Kergon #include <linux/dm-io.h> 17afa53df8SMikulas Patocka #include <linux/dm-bufio.h> 184db6bfe0SAlasdair G Kergon 194db6bfe0SAlasdair G Kergon #define DM_MSG_PREFIX "persistent snapshot" 20ad6bf88aSMikulas Patocka #define DM_CHUNK_SIZE_DEFAULT_SECTORS 32U /* 16KB */ 214db6bfe0SAlasdair G Kergon 2255b082e6SMikulas Patocka #define DM_PREFETCH_CHUNKS 12 2355b082e6SMikulas Patocka 244db6bfe0SAlasdair G Kergon /*----------------------------------------------------------------- 254db6bfe0SAlasdair G Kergon * Persistent snapshots, by persistent we mean that the snapshot 264db6bfe0SAlasdair G Kergon * will survive a reboot. 274db6bfe0SAlasdair G Kergon *---------------------------------------------------------------*/ 284db6bfe0SAlasdair G Kergon 294db6bfe0SAlasdair G Kergon /* 304db6bfe0SAlasdair G Kergon * We need to store a record of which parts of the origin have 314db6bfe0SAlasdair G Kergon * been copied to the snapshot device. The snapshot code 324db6bfe0SAlasdair G Kergon * requires that we copy exception chunks to chunk aligned areas 334db6bfe0SAlasdair G Kergon * of the COW store. It makes sense therefore, to store the 344db6bfe0SAlasdair G Kergon * metadata in chunk size blocks. 354db6bfe0SAlasdair G Kergon * 364db6bfe0SAlasdair G Kergon * There is no backward or forward compatibility implemented, 374db6bfe0SAlasdair G Kergon * snapshots with different disk versions than the kernel will 384db6bfe0SAlasdair G Kergon * not be usable. It is expected that "lvcreate" will blank out 394db6bfe0SAlasdair G Kergon * the start of a fresh COW device before calling the snapshot 404db6bfe0SAlasdair G Kergon * constructor. 414db6bfe0SAlasdair G Kergon * 424db6bfe0SAlasdair G Kergon * The first chunk of the COW device just contains the header. 434db6bfe0SAlasdair G Kergon * After this there is a chunk filled with exception metadata, 444db6bfe0SAlasdair G Kergon * followed by as many exception chunks as can fit in the 454db6bfe0SAlasdair G Kergon * metadata areas. 464db6bfe0SAlasdair G Kergon * 474db6bfe0SAlasdair G Kergon * All on disk structures are in little-endian format. The end 484db6bfe0SAlasdair G Kergon * of the exceptions info is indicated by an exception with a 494db6bfe0SAlasdair G Kergon * new_chunk of 0, which is invalid since it would point to the 504db6bfe0SAlasdair G Kergon * header chunk. 514db6bfe0SAlasdair G Kergon */ 524db6bfe0SAlasdair G Kergon 534db6bfe0SAlasdair G Kergon /* 544db6bfe0SAlasdair G Kergon * Magic for persistent snapshots: "SnAp" - Feeble isn't it. 554db6bfe0SAlasdair G Kergon */ 564db6bfe0SAlasdair G Kergon #define SNAP_MAGIC 0x70416e53 574db6bfe0SAlasdair G Kergon 584db6bfe0SAlasdair G Kergon /* 594db6bfe0SAlasdair G Kergon * The on-disk version of the metadata. 604db6bfe0SAlasdair G Kergon */ 614db6bfe0SAlasdair G Kergon #define SNAPSHOT_DISK_VERSION 1 624db6bfe0SAlasdair G Kergon 634454a621SMikulas Patocka #define NUM_SNAPSHOT_HDR_CHUNKS 1 644454a621SMikulas Patocka 654db6bfe0SAlasdair G Kergon struct disk_header { 66283a8328SAlasdair G Kergon __le32 magic; 674db6bfe0SAlasdair G Kergon 684db6bfe0SAlasdair G Kergon /* 694db6bfe0SAlasdair G Kergon * Is this snapshot valid. There is no way of recovering 704db6bfe0SAlasdair G Kergon * an invalid snapshot. 714db6bfe0SAlasdair G Kergon */ 72283a8328SAlasdair G Kergon __le32 valid; 734db6bfe0SAlasdair G Kergon 744db6bfe0SAlasdair G Kergon /* 754db6bfe0SAlasdair G Kergon * Simple, incrementing version. no backward 764db6bfe0SAlasdair G Kergon * compatibility. 774db6bfe0SAlasdair G Kergon */ 78283a8328SAlasdair G Kergon __le32 version; 794db6bfe0SAlasdair G Kergon 804db6bfe0SAlasdair G Kergon /* In sectors */ 81283a8328SAlasdair G Kergon __le32 chunk_size; 82283a8328SAlasdair G Kergon } __packed; 834db6bfe0SAlasdair G Kergon 844db6bfe0SAlasdair G Kergon struct disk_exception { 85283a8328SAlasdair G Kergon __le64 old_chunk; 86283a8328SAlasdair G Kergon __le64 new_chunk; 87283a8328SAlasdair G Kergon } __packed; 88283a8328SAlasdair G Kergon 89283a8328SAlasdair G Kergon struct core_exception { 904db6bfe0SAlasdair G Kergon uint64_t old_chunk; 914db6bfe0SAlasdair G Kergon uint64_t new_chunk; 924db6bfe0SAlasdair G Kergon }; 934db6bfe0SAlasdair G Kergon 944db6bfe0SAlasdair G Kergon struct commit_callback { 954db6bfe0SAlasdair G Kergon void (*callback)(void *, int success); 964db6bfe0SAlasdair G Kergon void *context; 974db6bfe0SAlasdair G Kergon }; 984db6bfe0SAlasdair G Kergon 994db6bfe0SAlasdair G Kergon /* 1004db6bfe0SAlasdair G Kergon * The top level structure for a persistent exception store. 1014db6bfe0SAlasdair G Kergon */ 1024db6bfe0SAlasdair G Kergon struct pstore { 10371fab00aSJonathan Brassow struct dm_exception_store *store; 1044db6bfe0SAlasdair G Kergon int version; 1054db6bfe0SAlasdair G Kergon int valid; 1064db6bfe0SAlasdair G Kergon uint32_t exceptions_per_area; 1074db6bfe0SAlasdair G Kergon 1084db6bfe0SAlasdair G Kergon /* 1094db6bfe0SAlasdair G Kergon * Now that we have an asynchronous kcopyd there is no 1104db6bfe0SAlasdair G Kergon * need for large chunk sizes, so it wont hurt to have a 1114db6bfe0SAlasdair G Kergon * whole chunks worth of metadata in memory at once. 1124db6bfe0SAlasdair G Kergon */ 1134db6bfe0SAlasdair G Kergon void *area; 1144db6bfe0SAlasdair G Kergon 1154db6bfe0SAlasdair G Kergon /* 1164db6bfe0SAlasdair G Kergon * An area of zeros used to clear the next area. 1174db6bfe0SAlasdair G Kergon */ 1184db6bfe0SAlasdair G Kergon void *zero_area; 1194db6bfe0SAlasdair G Kergon 1204db6bfe0SAlasdair G Kergon /* 12161578dcdSMikulas Patocka * An area used for header. The header can be written 12261578dcdSMikulas Patocka * concurrently with metadata (when invalidating the snapshot), 12361578dcdSMikulas Patocka * so it needs a separate buffer. 12461578dcdSMikulas Patocka */ 12561578dcdSMikulas Patocka void *header_area; 12661578dcdSMikulas Patocka 12761578dcdSMikulas Patocka /* 1284db6bfe0SAlasdair G Kergon * Used to keep track of which metadata area the data in 1294db6bfe0SAlasdair G Kergon * 'chunk' refers to. 1304db6bfe0SAlasdair G Kergon */ 1314db6bfe0SAlasdair G Kergon chunk_t current_area; 1324db6bfe0SAlasdair G Kergon 1334db6bfe0SAlasdair G Kergon /* 1344db6bfe0SAlasdair G Kergon * The next free chunk for an exception. 1354454a621SMikulas Patocka * 1364454a621SMikulas Patocka * When creating exceptions, all the chunks here and above are 1374454a621SMikulas Patocka * free. It holds the next chunk to be allocated. On rare 1384454a621SMikulas Patocka * occasions (e.g. after a system crash) holes can be left in 1394454a621SMikulas Patocka * the exception store because chunks can be committed out of 1404454a621SMikulas Patocka * order. 1414454a621SMikulas Patocka * 1424454a621SMikulas Patocka * When merging exceptions, it does not necessarily mean all the 1434454a621SMikulas Patocka * chunks here and above are free. It holds the value it would 1444454a621SMikulas Patocka * have held if all chunks had been committed in order of 1454454a621SMikulas Patocka * allocation. Consequently the value may occasionally be 1464454a621SMikulas Patocka * slightly too low, but since it's only used for 'status' and 1474454a621SMikulas Patocka * it can never reach its minimum value too early this doesn't 1484454a621SMikulas Patocka * matter. 1494db6bfe0SAlasdair G Kergon */ 1504454a621SMikulas Patocka 1514db6bfe0SAlasdair G Kergon chunk_t next_free; 1524db6bfe0SAlasdair G Kergon 1534db6bfe0SAlasdair G Kergon /* 1544db6bfe0SAlasdair G Kergon * The index of next free exception in the current 1554db6bfe0SAlasdair G Kergon * metadata area. 1564db6bfe0SAlasdair G Kergon */ 1574db6bfe0SAlasdair G Kergon uint32_t current_committed; 1584db6bfe0SAlasdair G Kergon 1594db6bfe0SAlasdair G Kergon atomic_t pending_count; 1604db6bfe0SAlasdair G Kergon uint32_t callback_count; 1614db6bfe0SAlasdair G Kergon struct commit_callback *callbacks; 1624db6bfe0SAlasdair G Kergon struct dm_io_client *io_client; 1634db6bfe0SAlasdair G Kergon 1644db6bfe0SAlasdair G Kergon struct workqueue_struct *metadata_wq; 1654db6bfe0SAlasdair G Kergon }; 1664db6bfe0SAlasdair G Kergon 1674db6bfe0SAlasdair G Kergon static int alloc_area(struct pstore *ps) 1684db6bfe0SAlasdair G Kergon { 1694db6bfe0SAlasdair G Kergon int r = -ENOMEM; 1704db6bfe0SAlasdair G Kergon size_t len; 1714db6bfe0SAlasdair G Kergon 17271fab00aSJonathan Brassow len = ps->store->chunk_size << SECTOR_SHIFT; 1734db6bfe0SAlasdair G Kergon 1744db6bfe0SAlasdair G Kergon /* 1754db6bfe0SAlasdair G Kergon * Allocate the chunk_size block of memory that will hold 1764db6bfe0SAlasdair G Kergon * a single metadata area. 1774db6bfe0SAlasdair G Kergon */ 1784db6bfe0SAlasdair G Kergon ps->area = vmalloc(len); 1794db6bfe0SAlasdair G Kergon if (!ps->area) 18061578dcdSMikulas Patocka goto err_area; 1814db6bfe0SAlasdair G Kergon 182e29e65aaSJoe Perches ps->zero_area = vzalloc(len); 18361578dcdSMikulas Patocka if (!ps->zero_area) 18461578dcdSMikulas Patocka goto err_zero_area; 1854db6bfe0SAlasdair G Kergon 18661578dcdSMikulas Patocka ps->header_area = vmalloc(len); 18761578dcdSMikulas Patocka if (!ps->header_area) 18861578dcdSMikulas Patocka goto err_header_area; 18961578dcdSMikulas Patocka 1904db6bfe0SAlasdair G Kergon return 0; 19161578dcdSMikulas Patocka 19261578dcdSMikulas Patocka err_header_area: 19361578dcdSMikulas Patocka vfree(ps->zero_area); 19461578dcdSMikulas Patocka 19561578dcdSMikulas Patocka err_zero_area: 19661578dcdSMikulas Patocka vfree(ps->area); 19761578dcdSMikulas Patocka 19861578dcdSMikulas Patocka err_area: 19961578dcdSMikulas Patocka return r; 2004db6bfe0SAlasdair G Kergon } 2014db6bfe0SAlasdair G Kergon 2024db6bfe0SAlasdair G Kergon static void free_area(struct pstore *ps) 2034db6bfe0SAlasdair G Kergon { 2044db6bfe0SAlasdair G Kergon vfree(ps->area); 2054db6bfe0SAlasdair G Kergon ps->area = NULL; 2064db6bfe0SAlasdair G Kergon vfree(ps->zero_area); 2074db6bfe0SAlasdair G Kergon ps->zero_area = NULL; 20861578dcdSMikulas Patocka vfree(ps->header_area); 20961578dcdSMikulas Patocka ps->header_area = NULL; 2104db6bfe0SAlasdair G Kergon } 2114db6bfe0SAlasdair G Kergon 2124db6bfe0SAlasdair G Kergon struct mdata_req { 2134db6bfe0SAlasdair G Kergon struct dm_io_region *where; 2144db6bfe0SAlasdair G Kergon struct dm_io_request *io_req; 2154db6bfe0SAlasdair G Kergon struct work_struct work; 2164db6bfe0SAlasdair G Kergon int result; 2174db6bfe0SAlasdair G Kergon }; 2184db6bfe0SAlasdair G Kergon 2194db6bfe0SAlasdair G Kergon static void do_metadata(struct work_struct *work) 2204db6bfe0SAlasdair G Kergon { 2214db6bfe0SAlasdair G Kergon struct mdata_req *req = container_of(work, struct mdata_req, work); 2224db6bfe0SAlasdair G Kergon 2234db6bfe0SAlasdair G Kergon req->result = dm_io(req->io_req, 1, req->where, NULL); 2244db6bfe0SAlasdair G Kergon } 2254db6bfe0SAlasdair G Kergon 2264db6bfe0SAlasdair G Kergon /* 2274db6bfe0SAlasdair G Kergon * Read or write a chunk aligned and sized block of data from a device. 2284db6bfe0SAlasdair G Kergon */ 229e6047149SMike Christie static int chunk_io(struct pstore *ps, void *area, chunk_t chunk, int op, 230e6047149SMike Christie int op_flags, int metadata) 2314db6bfe0SAlasdair G Kergon { 2324db6bfe0SAlasdair G Kergon struct dm_io_region where = { 233fc56f6fbSMike Snitzer .bdev = dm_snap_cow(ps->store->snap)->bdev, 23471fab00aSJonathan Brassow .sector = ps->store->chunk_size * chunk, 23571fab00aSJonathan Brassow .count = ps->store->chunk_size, 2364db6bfe0SAlasdair G Kergon }; 2374db6bfe0SAlasdair G Kergon struct dm_io_request io_req = { 238e6047149SMike Christie .bi_op = op, 239e6047149SMike Christie .bi_op_flags = op_flags, 2404db6bfe0SAlasdair G Kergon .mem.type = DM_IO_VMA, 24102d2fd31SMikulas Patocka .mem.ptr.vma = area, 2424db6bfe0SAlasdair G Kergon .client = ps->io_client, 2434db6bfe0SAlasdair G Kergon .notify.fn = NULL, 2444db6bfe0SAlasdair G Kergon }; 2454db6bfe0SAlasdair G Kergon struct mdata_req req; 2464db6bfe0SAlasdair G Kergon 2474db6bfe0SAlasdair G Kergon if (!metadata) 2484db6bfe0SAlasdair G Kergon return dm_io(&io_req, 1, &where, NULL); 2494db6bfe0SAlasdair G Kergon 2504db6bfe0SAlasdair G Kergon req.where = &where; 2514db6bfe0SAlasdair G Kergon req.io_req = &io_req; 2524db6bfe0SAlasdair G Kergon 2534db6bfe0SAlasdair G Kergon /* 2544db6bfe0SAlasdair G Kergon * Issue the synchronous I/O from a different thread 255ed00aabdSChristoph Hellwig * to avoid submit_bio_noacct recursion. 2564db6bfe0SAlasdair G Kergon */ 257ca1cab37SAndrew Morton INIT_WORK_ONSTACK(&req.work, do_metadata); 2584db6bfe0SAlasdair G Kergon queue_work(ps->metadata_wq, &req.work); 2595ea330a7SMikulas Patocka flush_workqueue(ps->metadata_wq); 260c1a64160SChuansheng Liu destroy_work_on_stack(&req.work); 2614db6bfe0SAlasdair G Kergon 2624db6bfe0SAlasdair G Kergon return req.result; 2634db6bfe0SAlasdair G Kergon } 2644db6bfe0SAlasdair G Kergon 2654db6bfe0SAlasdair G Kergon /* 2664db6bfe0SAlasdair G Kergon * Convert a metadata area index to a chunk index. 2674db6bfe0SAlasdair G Kergon */ 2684db6bfe0SAlasdair G Kergon static chunk_t area_location(struct pstore *ps, chunk_t area) 2694db6bfe0SAlasdair G Kergon { 27087c961cbSTomohiro Kusumi return NUM_SNAPSHOT_HDR_CHUNKS + ((ps->exceptions_per_area + 1) * area); 2714db6bfe0SAlasdair G Kergon } 2724db6bfe0SAlasdair G Kergon 273e9c6a182SMikulas Patocka static void skip_metadata(struct pstore *ps) 274e9c6a182SMikulas Patocka { 275e9c6a182SMikulas Patocka uint32_t stride = ps->exceptions_per_area + 1; 276e9c6a182SMikulas Patocka chunk_t next_free = ps->next_free; 277e9c6a182SMikulas Patocka if (sector_div(next_free, stride) == NUM_SNAPSHOT_HDR_CHUNKS) 278e9c6a182SMikulas Patocka ps->next_free++; 279e9c6a182SMikulas Patocka } 280e9c6a182SMikulas Patocka 2814db6bfe0SAlasdair G Kergon /* 2824db6bfe0SAlasdair G Kergon * Read or write a metadata area. Remembering to skip the first 2834db6bfe0SAlasdair G Kergon * chunk which holds the header. 2844db6bfe0SAlasdair G Kergon */ 285e6047149SMike Christie static int area_io(struct pstore *ps, int op, int op_flags) 2864db6bfe0SAlasdair G Kergon { 2877d837c0dSQinglang Miao chunk_t chunk = area_location(ps, ps->current_area); 2884db6bfe0SAlasdair G Kergon 2897d837c0dSQinglang Miao return chunk_io(ps, ps->area, chunk, op, op_flags, 0); 2904db6bfe0SAlasdair G Kergon } 2914db6bfe0SAlasdair G Kergon 2924db6bfe0SAlasdair G Kergon static void zero_memory_area(struct pstore *ps) 2934db6bfe0SAlasdair G Kergon { 29471fab00aSJonathan Brassow memset(ps->area, 0, ps->store->chunk_size << SECTOR_SHIFT); 2954db6bfe0SAlasdair G Kergon } 2964db6bfe0SAlasdair G Kergon 2974db6bfe0SAlasdair G Kergon static int zero_disk_area(struct pstore *ps, chunk_t area) 2984db6bfe0SAlasdair G Kergon { 299e6047149SMike Christie return chunk_io(ps, ps->zero_area, area_location(ps, area), 300e6047149SMike Christie REQ_OP_WRITE, 0, 0); 3014db6bfe0SAlasdair G Kergon } 3024db6bfe0SAlasdair G Kergon 3034db6bfe0SAlasdair G Kergon static int read_header(struct pstore *ps, int *new_snapshot) 3044db6bfe0SAlasdair G Kergon { 3054db6bfe0SAlasdair G Kergon int r; 3064db6bfe0SAlasdair G Kergon struct disk_header *dh; 307df96eee6SMikulas Patocka unsigned chunk_size; 3084db6bfe0SAlasdair G Kergon int chunk_size_supplied = 1; 309ae0b7448SMikulas Patocka char *chunk_err; 3104db6bfe0SAlasdair G Kergon 3114db6bfe0SAlasdair G Kergon /* 312df96eee6SMikulas Patocka * Use default chunk size (or logical_block_size, if larger) 313df96eee6SMikulas Patocka * if none supplied 3144db6bfe0SAlasdair G Kergon */ 31571fab00aSJonathan Brassow if (!ps->store->chunk_size) { 31671fab00aSJonathan Brassow ps->store->chunk_size = max(DM_CHUNK_SIZE_DEFAULT_SECTORS, 317fc56f6fbSMike Snitzer bdev_logical_block_size(dm_snap_cow(ps->store->snap)-> 318fc56f6fbSMike Snitzer bdev) >> 9); 31971fab00aSJonathan Brassow ps->store->chunk_mask = ps->store->chunk_size - 1; 320a3d939aeSMikulas Patocka ps->store->chunk_shift = __ffs(ps->store->chunk_size); 3214db6bfe0SAlasdair G Kergon chunk_size_supplied = 0; 3224db6bfe0SAlasdair G Kergon } 3234db6bfe0SAlasdair G Kergon 324bda8efecSMikulas Patocka ps->io_client = dm_io_client_create(); 3254db6bfe0SAlasdair G Kergon if (IS_ERR(ps->io_client)) 3264db6bfe0SAlasdair G Kergon return PTR_ERR(ps->io_client); 3274db6bfe0SAlasdair G Kergon 3284db6bfe0SAlasdair G Kergon r = alloc_area(ps); 3294db6bfe0SAlasdair G Kergon if (r) 3304db6bfe0SAlasdair G Kergon return r; 3314db6bfe0SAlasdair G Kergon 332e6047149SMike Christie r = chunk_io(ps, ps->header_area, 0, REQ_OP_READ, 0, 1); 3334db6bfe0SAlasdair G Kergon if (r) 3344db6bfe0SAlasdair G Kergon goto bad; 3354db6bfe0SAlasdair G Kergon 33661578dcdSMikulas Patocka dh = ps->header_area; 3374db6bfe0SAlasdair G Kergon 3384db6bfe0SAlasdair G Kergon if (le32_to_cpu(dh->magic) == 0) { 3394db6bfe0SAlasdair G Kergon *new_snapshot = 1; 3404db6bfe0SAlasdair G Kergon return 0; 3414db6bfe0SAlasdair G Kergon } 3424db6bfe0SAlasdair G Kergon 3434db6bfe0SAlasdair G Kergon if (le32_to_cpu(dh->magic) != SNAP_MAGIC) { 3444db6bfe0SAlasdair G Kergon DMWARN("Invalid or corrupt snapshot"); 3454db6bfe0SAlasdair G Kergon r = -ENXIO; 3464db6bfe0SAlasdair G Kergon goto bad; 3474db6bfe0SAlasdair G Kergon } 3484db6bfe0SAlasdair G Kergon 3494db6bfe0SAlasdair G Kergon *new_snapshot = 0; 3504db6bfe0SAlasdair G Kergon ps->valid = le32_to_cpu(dh->valid); 3514db6bfe0SAlasdair G Kergon ps->version = le32_to_cpu(dh->version); 3524db6bfe0SAlasdair G Kergon chunk_size = le32_to_cpu(dh->chunk_size); 3534db6bfe0SAlasdair G Kergon 354ae0b7448SMikulas Patocka if (ps->store->chunk_size == chunk_size) 3554db6bfe0SAlasdair G Kergon return 0; 3564db6bfe0SAlasdair G Kergon 357ae0b7448SMikulas Patocka if (chunk_size_supplied) 358df96eee6SMikulas Patocka DMWARN("chunk size %u in device metadata overrides " 359df96eee6SMikulas Patocka "table chunk size of %u.", 360df96eee6SMikulas Patocka chunk_size, ps->store->chunk_size); 3614db6bfe0SAlasdair G Kergon 3624db6bfe0SAlasdair G Kergon /* We had a bogus chunk_size. Fix stuff up. */ 3634db6bfe0SAlasdair G Kergon free_area(ps); 3644db6bfe0SAlasdair G Kergon 365ae0b7448SMikulas Patocka r = dm_exception_store_set_chunk_size(ps->store, chunk_size, 366ae0b7448SMikulas Patocka &chunk_err); 367ae0b7448SMikulas Patocka if (r) { 368df96eee6SMikulas Patocka DMERR("invalid on-disk chunk size %u: %s.", 369df96eee6SMikulas Patocka chunk_size, chunk_err); 370ae0b7448SMikulas Patocka return r; 371ae0b7448SMikulas Patocka } 3724db6bfe0SAlasdair G Kergon 3734db6bfe0SAlasdair G Kergon r = alloc_area(ps); 3744db6bfe0SAlasdair G Kergon return r; 3754db6bfe0SAlasdair G Kergon 3764db6bfe0SAlasdair G Kergon bad: 3774db6bfe0SAlasdair G Kergon free_area(ps); 3784db6bfe0SAlasdair G Kergon return r; 3794db6bfe0SAlasdair G Kergon } 3804db6bfe0SAlasdair G Kergon 3814db6bfe0SAlasdair G Kergon static int write_header(struct pstore *ps) 3824db6bfe0SAlasdair G Kergon { 3834db6bfe0SAlasdair G Kergon struct disk_header *dh; 3844db6bfe0SAlasdair G Kergon 38561578dcdSMikulas Patocka memset(ps->header_area, 0, ps->store->chunk_size << SECTOR_SHIFT); 3864db6bfe0SAlasdair G Kergon 38761578dcdSMikulas Patocka dh = ps->header_area; 3884db6bfe0SAlasdair G Kergon dh->magic = cpu_to_le32(SNAP_MAGIC); 3894db6bfe0SAlasdair G Kergon dh->valid = cpu_to_le32(ps->valid); 3904db6bfe0SAlasdair G Kergon dh->version = cpu_to_le32(ps->version); 39171fab00aSJonathan Brassow dh->chunk_size = cpu_to_le32(ps->store->chunk_size); 3924db6bfe0SAlasdair G Kergon 393e6047149SMike Christie return chunk_io(ps, ps->header_area, 0, REQ_OP_WRITE, 0, 1); 3944db6bfe0SAlasdair G Kergon } 3954db6bfe0SAlasdair G Kergon 3964db6bfe0SAlasdair G Kergon /* 3974db6bfe0SAlasdair G Kergon * Access functions for the disk exceptions, these do the endian conversions. 3984db6bfe0SAlasdair G Kergon */ 3992cadabd5SMikulas Patocka static struct disk_exception *get_exception(struct pstore *ps, void *ps_area, 4002cadabd5SMikulas Patocka uint32_t index) 4014db6bfe0SAlasdair G Kergon { 4024db6bfe0SAlasdair G Kergon BUG_ON(index >= ps->exceptions_per_area); 4034db6bfe0SAlasdair G Kergon 4042cadabd5SMikulas Patocka return ((struct disk_exception *) ps_area) + index; 4054db6bfe0SAlasdair G Kergon } 4064db6bfe0SAlasdair G Kergon 4072cadabd5SMikulas Patocka static void read_exception(struct pstore *ps, void *ps_area, 408283a8328SAlasdair G Kergon uint32_t index, struct core_exception *result) 4094db6bfe0SAlasdair G Kergon { 4102cadabd5SMikulas Patocka struct disk_exception *de = get_exception(ps, ps_area, index); 4114db6bfe0SAlasdair G Kergon 4124db6bfe0SAlasdair G Kergon /* copy it */ 413283a8328SAlasdair G Kergon result->old_chunk = le64_to_cpu(de->old_chunk); 414283a8328SAlasdair G Kergon result->new_chunk = le64_to_cpu(de->new_chunk); 4154db6bfe0SAlasdair G Kergon } 4164db6bfe0SAlasdair G Kergon 4174db6bfe0SAlasdair G Kergon static void write_exception(struct pstore *ps, 418283a8328SAlasdair G Kergon uint32_t index, struct core_exception *e) 4194db6bfe0SAlasdair G Kergon { 4202cadabd5SMikulas Patocka struct disk_exception *de = get_exception(ps, ps->area, index); 4214db6bfe0SAlasdair G Kergon 4224db6bfe0SAlasdair G Kergon /* copy it */ 423283a8328SAlasdair G Kergon de->old_chunk = cpu_to_le64(e->old_chunk); 424283a8328SAlasdair G Kergon de->new_chunk = cpu_to_le64(e->new_chunk); 4254db6bfe0SAlasdair G Kergon } 4264db6bfe0SAlasdair G Kergon 4274454a621SMikulas Patocka static void clear_exception(struct pstore *ps, uint32_t index) 4284454a621SMikulas Patocka { 4292cadabd5SMikulas Patocka struct disk_exception *de = get_exception(ps, ps->area, index); 4304454a621SMikulas Patocka 4314454a621SMikulas Patocka /* clear it */ 432283a8328SAlasdair G Kergon de->old_chunk = 0; 433283a8328SAlasdair G Kergon de->new_chunk = 0; 4344454a621SMikulas Patocka } 4354454a621SMikulas Patocka 4364db6bfe0SAlasdair G Kergon /* 4374db6bfe0SAlasdair G Kergon * Registers the exceptions that are present in the current area. 4384db6bfe0SAlasdair G Kergon * 'full' is filled in to indicate if the area has been 4394db6bfe0SAlasdair G Kergon * filled. 4404db6bfe0SAlasdair G Kergon */ 4412cadabd5SMikulas Patocka static int insert_exceptions(struct pstore *ps, void *ps_area, 442a159c1acSJonathan Brassow int (*callback)(void *callback_context, 443a159c1acSJonathan Brassow chunk_t old, chunk_t new), 444a159c1acSJonathan Brassow void *callback_context, 445a159c1acSJonathan Brassow int *full) 4464db6bfe0SAlasdair G Kergon { 4474db6bfe0SAlasdair G Kergon int r; 4484db6bfe0SAlasdair G Kergon unsigned int i; 449283a8328SAlasdair G Kergon struct core_exception e; 4504db6bfe0SAlasdair G Kergon 4514db6bfe0SAlasdair G Kergon /* presume the area is full */ 4524db6bfe0SAlasdair G Kergon *full = 1; 4534db6bfe0SAlasdair G Kergon 4544db6bfe0SAlasdair G Kergon for (i = 0; i < ps->exceptions_per_area; i++) { 4552cadabd5SMikulas Patocka read_exception(ps, ps_area, i, &e); 4564db6bfe0SAlasdair G Kergon 4574db6bfe0SAlasdair G Kergon /* 4584db6bfe0SAlasdair G Kergon * If the new_chunk is pointing at the start of 4594db6bfe0SAlasdair G Kergon * the COW device, where the first metadata area 4604db6bfe0SAlasdair G Kergon * is we know that we've hit the end of the 4614db6bfe0SAlasdair G Kergon * exceptions. Therefore the area is not full. 4624db6bfe0SAlasdair G Kergon */ 463283a8328SAlasdair G Kergon if (e.new_chunk == 0LL) { 4644db6bfe0SAlasdair G Kergon ps->current_committed = i; 4654db6bfe0SAlasdair G Kergon *full = 0; 4664db6bfe0SAlasdair G Kergon break; 4674db6bfe0SAlasdair G Kergon } 4684db6bfe0SAlasdair G Kergon 4694db6bfe0SAlasdair G Kergon /* 4704db6bfe0SAlasdair G Kergon * Keep track of the start of the free chunks. 4714db6bfe0SAlasdair G Kergon */ 472283a8328SAlasdair G Kergon if (ps->next_free <= e.new_chunk) 473283a8328SAlasdair G Kergon ps->next_free = e.new_chunk + 1; 4744db6bfe0SAlasdair G Kergon 4754db6bfe0SAlasdair G Kergon /* 4764db6bfe0SAlasdair G Kergon * Otherwise we add the exception to the snapshot. 4774db6bfe0SAlasdair G Kergon */ 478283a8328SAlasdair G Kergon r = callback(callback_context, e.old_chunk, e.new_chunk); 4794db6bfe0SAlasdair G Kergon if (r) 4804db6bfe0SAlasdair G Kergon return r; 4814db6bfe0SAlasdair G Kergon } 4824db6bfe0SAlasdair G Kergon 4834db6bfe0SAlasdair G Kergon return 0; 4844db6bfe0SAlasdair G Kergon } 4854db6bfe0SAlasdair G Kergon 486a159c1acSJonathan Brassow static int read_exceptions(struct pstore *ps, 487a159c1acSJonathan Brassow int (*callback)(void *callback_context, chunk_t old, 488a159c1acSJonathan Brassow chunk_t new), 489a159c1acSJonathan Brassow void *callback_context) 4904db6bfe0SAlasdair G Kergon { 4914db6bfe0SAlasdair G Kergon int r, full = 1; 49255494bf2SMikulas Patocka struct dm_bufio_client *client; 49355b082e6SMikulas Patocka chunk_t prefetch_area = 0; 49455494bf2SMikulas Patocka 49555494bf2SMikulas Patocka client = dm_bufio_client_create(dm_snap_cow(ps->store->snap)->bdev, 49655494bf2SMikulas Patocka ps->store->chunk_size << SECTOR_SHIFT, 497*0fcb100dSNathan Huckleberry 1, 0, NULL, NULL, 0); 49855494bf2SMikulas Patocka 49955494bf2SMikulas Patocka if (IS_ERR(client)) 50055494bf2SMikulas Patocka return PTR_ERR(client); 5014db6bfe0SAlasdair G Kergon 5024db6bfe0SAlasdair G Kergon /* 50355b082e6SMikulas Patocka * Setup for one current buffer + desired readahead buffers. 50455b082e6SMikulas Patocka */ 50555b082e6SMikulas Patocka dm_bufio_set_minimum_buffers(client, 1 + DM_PREFETCH_CHUNKS); 50655b082e6SMikulas Patocka 50755b082e6SMikulas Patocka /* 5084db6bfe0SAlasdair G Kergon * Keeping reading chunks and inserting exceptions until 5094db6bfe0SAlasdair G Kergon * we find a partially full area. 5104db6bfe0SAlasdair G Kergon */ 5114db6bfe0SAlasdair G Kergon for (ps->current_area = 0; full; ps->current_area++) { 51255494bf2SMikulas Patocka struct dm_buffer *bp; 51355494bf2SMikulas Patocka void *area; 51455b082e6SMikulas Patocka chunk_t chunk; 51555b082e6SMikulas Patocka 51655b082e6SMikulas Patocka if (unlikely(prefetch_area < ps->current_area)) 51755b082e6SMikulas Patocka prefetch_area = ps->current_area; 51855b082e6SMikulas Patocka 51955b082e6SMikulas Patocka if (DM_PREFETCH_CHUNKS) do { 52055b082e6SMikulas Patocka chunk_t pf_chunk = area_location(ps, prefetch_area); 52155b082e6SMikulas Patocka if (unlikely(pf_chunk >= dm_bufio_get_device_size(client))) 52255b082e6SMikulas Patocka break; 52355b082e6SMikulas Patocka dm_bufio_prefetch(client, pf_chunk, 1); 52455b082e6SMikulas Patocka prefetch_area++; 52555b082e6SMikulas Patocka if (unlikely(!prefetch_area)) 52655b082e6SMikulas Patocka break; 52755b082e6SMikulas Patocka } while (prefetch_area <= ps->current_area + DM_PREFETCH_CHUNKS); 52855b082e6SMikulas Patocka 52955b082e6SMikulas Patocka chunk = area_location(ps, ps->current_area); 5304db6bfe0SAlasdair G Kergon 53155494bf2SMikulas Patocka area = dm_bufio_read(client, chunk, &bp); 532fc0a4461Sviresh kumar if (IS_ERR(area)) { 53355494bf2SMikulas Patocka r = PTR_ERR(area); 53455494bf2SMikulas Patocka goto ret_destroy_bufio; 53555494bf2SMikulas Patocka } 53655494bf2SMikulas Patocka 53755494bf2SMikulas Patocka r = insert_exceptions(ps, area, callback, callback_context, 5382cadabd5SMikulas Patocka &full); 53955494bf2SMikulas Patocka 5402c945820SMikulas Patocka if (!full) 5412c945820SMikulas Patocka memcpy(ps->area, area, ps->store->chunk_size << SECTOR_SHIFT); 5422c945820SMikulas Patocka 54355494bf2SMikulas Patocka dm_bufio_release(bp); 54455494bf2SMikulas Patocka 54555494bf2SMikulas Patocka dm_bufio_forget(client, chunk); 54655494bf2SMikulas Patocka 54755494bf2SMikulas Patocka if (unlikely(r)) 54855494bf2SMikulas Patocka goto ret_destroy_bufio; 5494db6bfe0SAlasdair G Kergon } 5504db6bfe0SAlasdair G Kergon 5514db6bfe0SAlasdair G Kergon ps->current_area--; 5524db6bfe0SAlasdair G Kergon 553e9c6a182SMikulas Patocka skip_metadata(ps); 554e9c6a182SMikulas Patocka 55555494bf2SMikulas Patocka r = 0; 55655494bf2SMikulas Patocka 55755494bf2SMikulas Patocka ret_destroy_bufio: 55855494bf2SMikulas Patocka dm_bufio_client_destroy(client); 55955494bf2SMikulas Patocka 56055494bf2SMikulas Patocka return r; 5614db6bfe0SAlasdair G Kergon } 5624db6bfe0SAlasdair G Kergon 5634db6bfe0SAlasdair G Kergon static struct pstore *get_info(struct dm_exception_store *store) 5644db6bfe0SAlasdair G Kergon { 5654db6bfe0SAlasdair G Kergon return (struct pstore *) store->context; 5664db6bfe0SAlasdair G Kergon } 5674db6bfe0SAlasdair G Kergon 568985903bbSMike Snitzer static void persistent_usage(struct dm_exception_store *store, 569985903bbSMike Snitzer sector_t *total_sectors, 570985903bbSMike Snitzer sector_t *sectors_allocated, 571985903bbSMike Snitzer sector_t *metadata_sectors) 5724db6bfe0SAlasdair G Kergon { 573985903bbSMike Snitzer struct pstore *ps = get_info(store); 574985903bbSMike Snitzer 575985903bbSMike Snitzer *sectors_allocated = ps->next_free * store->chunk_size; 576fc56f6fbSMike Snitzer *total_sectors = get_dev_size(dm_snap_cow(store->snap)->bdev); 577985903bbSMike Snitzer 578985903bbSMike Snitzer /* 579985903bbSMike Snitzer * First chunk is the fixed header. 580985903bbSMike Snitzer * Then there are (ps->current_area + 1) metadata chunks, each one 581985903bbSMike Snitzer * separated from the next by ps->exceptions_per_area data chunks. 582985903bbSMike Snitzer */ 5834454a621SMikulas Patocka *metadata_sectors = (ps->current_area + 1 + NUM_SNAPSHOT_HDR_CHUNKS) * 5844454a621SMikulas Patocka store->chunk_size; 5854db6bfe0SAlasdair G Kergon } 5864db6bfe0SAlasdair G Kergon 587493df71cSJonathan Brassow static void persistent_dtr(struct dm_exception_store *store) 5884db6bfe0SAlasdair G Kergon { 5894db6bfe0SAlasdair G Kergon struct pstore *ps = get_info(store); 5904db6bfe0SAlasdair G Kergon 5914db6bfe0SAlasdair G Kergon destroy_workqueue(ps->metadata_wq); 592a32079ceSJonathan Brassow 593a32079ceSJonathan Brassow /* Created in read_header */ 594a32079ceSJonathan Brassow if (ps->io_client) 5954db6bfe0SAlasdair G Kergon dm_io_client_destroy(ps->io_client); 5964db6bfe0SAlasdair G Kergon free_area(ps); 597a32079ceSJonathan Brassow 598a32079ceSJonathan Brassow /* Allocated in persistent_read_metadata */ 5997a35693aSMatthew Wilcox (Oracle) kvfree(ps->callbacks); 600a32079ceSJonathan Brassow 6014db6bfe0SAlasdair G Kergon kfree(ps); 6024db6bfe0SAlasdair G Kergon } 6034db6bfe0SAlasdair G Kergon 604a159c1acSJonathan Brassow static int persistent_read_metadata(struct dm_exception_store *store, 605a159c1acSJonathan Brassow int (*callback)(void *callback_context, 606a159c1acSJonathan Brassow chunk_t old, chunk_t new), 607a159c1acSJonathan Brassow void *callback_context) 6084db6bfe0SAlasdair G Kergon { 6093f649ab7SKees Cook int r, new_snapshot; 6104db6bfe0SAlasdair G Kergon struct pstore *ps = get_info(store); 6114db6bfe0SAlasdair G Kergon 6124db6bfe0SAlasdair G Kergon /* 6134db6bfe0SAlasdair G Kergon * Read the snapshot header. 6144db6bfe0SAlasdair G Kergon */ 6154db6bfe0SAlasdair G Kergon r = read_header(ps, &new_snapshot); 6164db6bfe0SAlasdair G Kergon if (r) 6174db6bfe0SAlasdair G Kergon return r; 6184db6bfe0SAlasdair G Kergon 6194db6bfe0SAlasdair G Kergon /* 6204db6bfe0SAlasdair G Kergon * Now we know correct chunk_size, complete the initialisation. 6214db6bfe0SAlasdair G Kergon */ 62271fab00aSJonathan Brassow ps->exceptions_per_area = (ps->store->chunk_size << SECTOR_SHIFT) / 62371fab00aSJonathan Brassow sizeof(struct disk_exception); 6247a35693aSMatthew Wilcox (Oracle) ps->callbacks = kvcalloc(ps->exceptions_per_area, 6257a35693aSMatthew Wilcox (Oracle) sizeof(*ps->callbacks), GFP_KERNEL); 6264db6bfe0SAlasdair G Kergon if (!ps->callbacks) 6274db6bfe0SAlasdair G Kergon return -ENOMEM; 6284db6bfe0SAlasdair G Kergon 6294db6bfe0SAlasdair G Kergon /* 6304db6bfe0SAlasdair G Kergon * Do we need to setup a new snapshot ? 6314db6bfe0SAlasdair G Kergon */ 6324db6bfe0SAlasdair G Kergon if (new_snapshot) { 6334db6bfe0SAlasdair G Kergon r = write_header(ps); 6344db6bfe0SAlasdair G Kergon if (r) { 6354db6bfe0SAlasdair G Kergon DMWARN("write_header failed"); 6364db6bfe0SAlasdair G Kergon return r; 6374db6bfe0SAlasdair G Kergon } 6384db6bfe0SAlasdair G Kergon 6394db6bfe0SAlasdair G Kergon ps->current_area = 0; 6404db6bfe0SAlasdair G Kergon zero_memory_area(ps); 6414db6bfe0SAlasdair G Kergon r = zero_disk_area(ps, 0); 642f5acc834SJon Brassow if (r) 6434db6bfe0SAlasdair G Kergon DMWARN("zero_disk_area(0) failed"); 6444db6bfe0SAlasdair G Kergon return r; 6454db6bfe0SAlasdair G Kergon } 6464db6bfe0SAlasdair G Kergon /* 6474db6bfe0SAlasdair G Kergon * Sanity checks. 6484db6bfe0SAlasdair G Kergon */ 6494db6bfe0SAlasdair G Kergon if (ps->version != SNAPSHOT_DISK_VERSION) { 6504db6bfe0SAlasdair G Kergon DMWARN("unable to handle snapshot disk version %d", 6514db6bfe0SAlasdair G Kergon ps->version); 6524db6bfe0SAlasdair G Kergon return -EINVAL; 6534db6bfe0SAlasdair G Kergon } 6544db6bfe0SAlasdair G Kergon 6554db6bfe0SAlasdair G Kergon /* 6564db6bfe0SAlasdair G Kergon * Metadata are valid, but snapshot is invalidated 6574db6bfe0SAlasdair G Kergon */ 6584db6bfe0SAlasdair G Kergon if (!ps->valid) 6594db6bfe0SAlasdair G Kergon return 1; 6604db6bfe0SAlasdair G Kergon 6614db6bfe0SAlasdair G Kergon /* 6624db6bfe0SAlasdair G Kergon * Read the metadata. 6634db6bfe0SAlasdair G Kergon */ 664a159c1acSJonathan Brassow r = read_exceptions(ps, callback, callback_context); 6654db6bfe0SAlasdair G Kergon 666f5acc834SJon Brassow return r; 6674db6bfe0SAlasdair G Kergon } 6684db6bfe0SAlasdair G Kergon 669a159c1acSJonathan Brassow static int persistent_prepare_exception(struct dm_exception_store *store, 6701d4989c8SJon Brassow struct dm_exception *e) 6714db6bfe0SAlasdair G Kergon { 6724db6bfe0SAlasdair G Kergon struct pstore *ps = get_info(store); 673fc56f6fbSMike Snitzer sector_t size = get_dev_size(dm_snap_cow(store->snap)->bdev); 6744db6bfe0SAlasdair G Kergon 6754db6bfe0SAlasdair G Kergon /* Is there enough room ? */ 676d0216849SJonathan Brassow if (size < ((ps->next_free + 1) * store->chunk_size)) 6774db6bfe0SAlasdair G Kergon return -ENOSPC; 6784db6bfe0SAlasdair G Kergon 6794db6bfe0SAlasdair G Kergon e->new_chunk = ps->next_free; 6804db6bfe0SAlasdair G Kergon 6814db6bfe0SAlasdair G Kergon /* 6824db6bfe0SAlasdair G Kergon * Move onto the next free pending, making sure to take 6834db6bfe0SAlasdair G Kergon * into account the location of the metadata chunks. 6844db6bfe0SAlasdair G Kergon */ 6854db6bfe0SAlasdair G Kergon ps->next_free++; 686e9c6a182SMikulas Patocka skip_metadata(ps); 6874db6bfe0SAlasdair G Kergon 6884db6bfe0SAlasdair G Kergon atomic_inc(&ps->pending_count); 6894db6bfe0SAlasdair G Kergon return 0; 6904db6bfe0SAlasdair G Kergon } 6914db6bfe0SAlasdair G Kergon 692a159c1acSJonathan Brassow static void persistent_commit_exception(struct dm_exception_store *store, 693385277bfSMikulas Patocka struct dm_exception *e, int valid, 6944db6bfe0SAlasdair G Kergon void (*callback) (void *, int success), 6954db6bfe0SAlasdair G Kergon void *callback_context) 6964db6bfe0SAlasdair G Kergon { 6974db6bfe0SAlasdair G Kergon unsigned int i; 6984db6bfe0SAlasdair G Kergon struct pstore *ps = get_info(store); 699283a8328SAlasdair G Kergon struct core_exception ce; 7004db6bfe0SAlasdair G Kergon struct commit_callback *cb; 7014db6bfe0SAlasdair G Kergon 702385277bfSMikulas Patocka if (!valid) 703385277bfSMikulas Patocka ps->valid = 0; 704385277bfSMikulas Patocka 705283a8328SAlasdair G Kergon ce.old_chunk = e->old_chunk; 706283a8328SAlasdair G Kergon ce.new_chunk = e->new_chunk; 707283a8328SAlasdair G Kergon write_exception(ps, ps->current_committed++, &ce); 7084db6bfe0SAlasdair G Kergon 7094db6bfe0SAlasdair G Kergon /* 7104db6bfe0SAlasdair G Kergon * Add the callback to the back of the array. This code 7114db6bfe0SAlasdair G Kergon * is the only place where the callback array is 7124db6bfe0SAlasdair G Kergon * manipulated, and we know that it will never be called 7134db6bfe0SAlasdair G Kergon * multiple times concurrently. 7144db6bfe0SAlasdair G Kergon */ 7154db6bfe0SAlasdair G Kergon cb = ps->callbacks + ps->callback_count++; 7164db6bfe0SAlasdair G Kergon cb->callback = callback; 7174db6bfe0SAlasdair G Kergon cb->context = callback_context; 7184db6bfe0SAlasdair G Kergon 7194db6bfe0SAlasdair G Kergon /* 7204db6bfe0SAlasdair G Kergon * If there are exceptions in flight and we have not yet 7214db6bfe0SAlasdair G Kergon * filled this metadata area there's nothing more to do. 7224db6bfe0SAlasdair G Kergon */ 7234db6bfe0SAlasdair G Kergon if (!atomic_dec_and_test(&ps->pending_count) && 7244db6bfe0SAlasdair G Kergon (ps->current_committed != ps->exceptions_per_area)) 7254db6bfe0SAlasdair G Kergon return; 7264db6bfe0SAlasdair G Kergon 7274db6bfe0SAlasdair G Kergon /* 7284db6bfe0SAlasdair G Kergon * If we completely filled the current area, then wipe the next one. 7294db6bfe0SAlasdair G Kergon */ 7304db6bfe0SAlasdair G Kergon if ((ps->current_committed == ps->exceptions_per_area) && 7314db6bfe0SAlasdair G Kergon zero_disk_area(ps, ps->current_area + 1)) 7324db6bfe0SAlasdair G Kergon ps->valid = 0; 7334db6bfe0SAlasdair G Kergon 7344db6bfe0SAlasdair G Kergon /* 7354db6bfe0SAlasdair G Kergon * Commit exceptions to disk. 7364db6bfe0SAlasdair G Kergon */ 737ff0361b3SJan Kara if (ps->valid && area_io(ps, REQ_OP_WRITE, 738ff0361b3SJan Kara REQ_PREFLUSH | REQ_FUA | REQ_SYNC)) 7394db6bfe0SAlasdair G Kergon ps->valid = 0; 7404db6bfe0SAlasdair G Kergon 7414db6bfe0SAlasdair G Kergon /* 7424db6bfe0SAlasdair G Kergon * Advance to the next area if this one is full. 7434db6bfe0SAlasdair G Kergon */ 7444db6bfe0SAlasdair G Kergon if (ps->current_committed == ps->exceptions_per_area) { 7454db6bfe0SAlasdair G Kergon ps->current_committed = 0; 7464db6bfe0SAlasdair G Kergon ps->current_area++; 7474db6bfe0SAlasdair G Kergon zero_memory_area(ps); 7484db6bfe0SAlasdair G Kergon } 7494db6bfe0SAlasdair G Kergon 7504db6bfe0SAlasdair G Kergon for (i = 0; i < ps->callback_count; i++) { 7514db6bfe0SAlasdair G Kergon cb = ps->callbacks + i; 7524db6bfe0SAlasdair G Kergon cb->callback(cb->context, ps->valid); 7534db6bfe0SAlasdair G Kergon } 7544db6bfe0SAlasdair G Kergon 7554db6bfe0SAlasdair G Kergon ps->callback_count = 0; 7564db6bfe0SAlasdair G Kergon } 7574db6bfe0SAlasdair G Kergon 7584454a621SMikulas Patocka static int persistent_prepare_merge(struct dm_exception_store *store, 7594454a621SMikulas Patocka chunk_t *last_old_chunk, 7604454a621SMikulas Patocka chunk_t *last_new_chunk) 7614454a621SMikulas Patocka { 7624454a621SMikulas Patocka struct pstore *ps = get_info(store); 763283a8328SAlasdair G Kergon struct core_exception ce; 7644454a621SMikulas Patocka int nr_consecutive; 7654454a621SMikulas Patocka int r; 7664454a621SMikulas Patocka 7674454a621SMikulas Patocka /* 7684454a621SMikulas Patocka * When current area is empty, move back to preceding area. 7694454a621SMikulas Patocka */ 7704454a621SMikulas Patocka if (!ps->current_committed) { 7714454a621SMikulas Patocka /* 7724454a621SMikulas Patocka * Have we finished? 7734454a621SMikulas Patocka */ 7744454a621SMikulas Patocka if (!ps->current_area) 7754454a621SMikulas Patocka return 0; 7764454a621SMikulas Patocka 7774454a621SMikulas Patocka ps->current_area--; 778e6047149SMike Christie r = area_io(ps, REQ_OP_READ, 0); 7794454a621SMikulas Patocka if (r < 0) 7804454a621SMikulas Patocka return r; 7814454a621SMikulas Patocka ps->current_committed = ps->exceptions_per_area; 7824454a621SMikulas Patocka } 7834454a621SMikulas Patocka 7842cadabd5SMikulas Patocka read_exception(ps, ps->area, ps->current_committed - 1, &ce); 785283a8328SAlasdair G Kergon *last_old_chunk = ce.old_chunk; 786283a8328SAlasdair G Kergon *last_new_chunk = ce.new_chunk; 7874454a621SMikulas Patocka 7884454a621SMikulas Patocka /* 7894454a621SMikulas Patocka * Find number of consecutive chunks within the current area, 7904454a621SMikulas Patocka * working backwards. 7914454a621SMikulas Patocka */ 7924454a621SMikulas Patocka for (nr_consecutive = 1; nr_consecutive < ps->current_committed; 7934454a621SMikulas Patocka nr_consecutive++) { 7942cadabd5SMikulas Patocka read_exception(ps, ps->area, 7952cadabd5SMikulas Patocka ps->current_committed - 1 - nr_consecutive, &ce); 796283a8328SAlasdair G Kergon if (ce.old_chunk != *last_old_chunk - nr_consecutive || 797283a8328SAlasdair G Kergon ce.new_chunk != *last_new_chunk - nr_consecutive) 7984454a621SMikulas Patocka break; 7994454a621SMikulas Patocka } 8004454a621SMikulas Patocka 8014454a621SMikulas Patocka return nr_consecutive; 8024454a621SMikulas Patocka } 8034454a621SMikulas Patocka 8044454a621SMikulas Patocka static int persistent_commit_merge(struct dm_exception_store *store, 8054454a621SMikulas Patocka int nr_merged) 8064454a621SMikulas Patocka { 8074454a621SMikulas Patocka int r, i; 8084454a621SMikulas Patocka struct pstore *ps = get_info(store); 8094454a621SMikulas Patocka 8104454a621SMikulas Patocka BUG_ON(nr_merged > ps->current_committed); 8114454a621SMikulas Patocka 8124454a621SMikulas Patocka for (i = 0; i < nr_merged; i++) 8134454a621SMikulas Patocka clear_exception(ps, ps->current_committed - 1 - i); 8144454a621SMikulas Patocka 81570fd7614SChristoph Hellwig r = area_io(ps, REQ_OP_WRITE, REQ_PREFLUSH | REQ_FUA); 8164454a621SMikulas Patocka if (r < 0) 8174454a621SMikulas Patocka return r; 8184454a621SMikulas Patocka 8194454a621SMikulas Patocka ps->current_committed -= nr_merged; 8204454a621SMikulas Patocka 8214454a621SMikulas Patocka /* 8224454a621SMikulas Patocka * At this stage, only persistent_usage() uses ps->next_free, so 8234454a621SMikulas Patocka * we make no attempt to keep ps->next_free strictly accurate 8244454a621SMikulas Patocka * as exceptions may have been committed out-of-order originally. 8254454a621SMikulas Patocka * Once a snapshot has become merging, we set it to the value it 8264454a621SMikulas Patocka * would have held had all the exceptions been committed in order. 8274454a621SMikulas Patocka * 8284454a621SMikulas Patocka * ps->current_area does not get reduced by prepare_merge() until 8294454a621SMikulas Patocka * after commit_merge() has removed the nr_merged previous exceptions. 8304454a621SMikulas Patocka */ 83187c961cbSTomohiro Kusumi ps->next_free = area_location(ps, ps->current_area) + 83287c961cbSTomohiro Kusumi ps->current_committed + 1; 8334454a621SMikulas Patocka 8344454a621SMikulas Patocka return 0; 8354454a621SMikulas Patocka } 8364454a621SMikulas Patocka 837a159c1acSJonathan Brassow static void persistent_drop_snapshot(struct dm_exception_store *store) 8384db6bfe0SAlasdair G Kergon { 8394db6bfe0SAlasdair G Kergon struct pstore *ps = get_info(store); 8404db6bfe0SAlasdair G Kergon 8414db6bfe0SAlasdair G Kergon ps->valid = 0; 8424db6bfe0SAlasdair G Kergon if (write_header(ps)) 8434db6bfe0SAlasdair G Kergon DMWARN("write header failed"); 8444db6bfe0SAlasdair G Kergon } 8454db6bfe0SAlasdair G Kergon 846b0d3cc01SMike Snitzer static int persistent_ctr(struct dm_exception_store *store, char *options) 8474db6bfe0SAlasdair G Kergon { 8484db6bfe0SAlasdair G Kergon struct pstore *ps; 849a2a678edSSudip Mukherjee int r; 8504db6bfe0SAlasdair G Kergon 8514db6bfe0SAlasdair G Kergon /* allocate the pstore */ 852a32079ceSJonathan Brassow ps = kzalloc(sizeof(*ps), GFP_KERNEL); 8534db6bfe0SAlasdair G Kergon if (!ps) 8544db6bfe0SAlasdair G Kergon return -ENOMEM; 8554db6bfe0SAlasdair G Kergon 85671fab00aSJonathan Brassow ps->store = store; 8574db6bfe0SAlasdair G Kergon ps->valid = 1; 8584db6bfe0SAlasdair G Kergon ps->version = SNAPSHOT_DISK_VERSION; 8594db6bfe0SAlasdair G Kergon ps->area = NULL; 86061578dcdSMikulas Patocka ps->zero_area = NULL; 86161578dcdSMikulas Patocka ps->header_area = NULL; 8624454a621SMikulas Patocka ps->next_free = NUM_SNAPSHOT_HDR_CHUNKS + 1; /* header and 1st area */ 8634db6bfe0SAlasdair G Kergon ps->current_committed = 0; 8644db6bfe0SAlasdair G Kergon 8654db6bfe0SAlasdair G Kergon ps->callback_count = 0; 8664db6bfe0SAlasdair G Kergon atomic_set(&ps->pending_count, 0); 8674db6bfe0SAlasdair G Kergon ps->callbacks = NULL; 8684db6bfe0SAlasdair G Kergon 869239c8dd5STejun Heo ps->metadata_wq = alloc_workqueue("ksnaphd", WQ_MEM_RECLAIM, 0); 8704db6bfe0SAlasdair G Kergon if (!ps->metadata_wq) { 8714db6bfe0SAlasdair G Kergon DMERR("couldn't start header metadata update thread"); 872a2a678edSSudip Mukherjee r = -ENOMEM; 873a2a678edSSudip Mukherjee goto err_workqueue; 8744db6bfe0SAlasdair G Kergon } 8754db6bfe0SAlasdair G Kergon 876b0d3cc01SMike Snitzer if (options) { 877b0d3cc01SMike Snitzer char overflow = toupper(options[0]); 878b0d3cc01SMike Snitzer if (overflow == 'O') 879b0d3cc01SMike Snitzer store->userspace_supports_overflow = true; 880b0d3cc01SMike Snitzer else { 881b0d3cc01SMike Snitzer DMERR("Unsupported persistent store option: %s", options); 882a2a678edSSudip Mukherjee r = -EINVAL; 883a2a678edSSudip Mukherjee goto err_options; 884b0d3cc01SMike Snitzer } 885b0d3cc01SMike Snitzer } 886b0d3cc01SMike Snitzer 8874db6bfe0SAlasdair G Kergon store->context = ps; 8884db6bfe0SAlasdair G Kergon 8894db6bfe0SAlasdair G Kergon return 0; 890a2a678edSSudip Mukherjee 891a2a678edSSudip Mukherjee err_options: 892a2a678edSSudip Mukherjee destroy_workqueue(ps->metadata_wq); 893a2a678edSSudip Mukherjee err_workqueue: 894a2a678edSSudip Mukherjee kfree(ps); 895a2a678edSSudip Mukherjee 896a2a678edSSudip Mukherjee return r; 8974db6bfe0SAlasdair G Kergon } 8984db6bfe0SAlasdair G Kergon 8991e302a92SJonathan Brassow static unsigned persistent_status(struct dm_exception_store *store, 900493df71cSJonathan Brassow status_type_t status, char *result, 9011e302a92SJonathan Brassow unsigned maxlen) 902493df71cSJonathan Brassow { 9031e302a92SJonathan Brassow unsigned sz = 0; 9041e302a92SJonathan Brassow 9051e302a92SJonathan Brassow switch (status) { 9061e302a92SJonathan Brassow case STATUSTYPE_INFO: 9071e302a92SJonathan Brassow break; 9081e302a92SJonathan Brassow case STATUSTYPE_TABLE: 909b0d3cc01SMike Snitzer DMEMIT(" %s %llu", store->userspace_supports_overflow ? "PO" : "P", 910b0d3cc01SMike Snitzer (unsigned long long)store->chunk_size); 9118ec45662STushar Sugandhi break; 9128ec45662STushar Sugandhi case STATUSTYPE_IMA: 9138ec45662STushar Sugandhi *result = '\0'; 9148ec45662STushar Sugandhi break; 9151e302a92SJonathan Brassow } 916493df71cSJonathan Brassow 917493df71cSJonathan Brassow return sz; 918493df71cSJonathan Brassow } 919493df71cSJonathan Brassow 920493df71cSJonathan Brassow static struct dm_exception_store_type _persistent_type = { 921493df71cSJonathan Brassow .name = "persistent", 922493df71cSJonathan Brassow .module = THIS_MODULE, 923493df71cSJonathan Brassow .ctr = persistent_ctr, 924493df71cSJonathan Brassow .dtr = persistent_dtr, 925493df71cSJonathan Brassow .read_metadata = persistent_read_metadata, 926493df71cSJonathan Brassow .prepare_exception = persistent_prepare_exception, 927493df71cSJonathan Brassow .commit_exception = persistent_commit_exception, 9284454a621SMikulas Patocka .prepare_merge = persistent_prepare_merge, 9294454a621SMikulas Patocka .commit_merge = persistent_commit_merge, 930493df71cSJonathan Brassow .drop_snapshot = persistent_drop_snapshot, 931985903bbSMike Snitzer .usage = persistent_usage, 932493df71cSJonathan Brassow .status = persistent_status, 933493df71cSJonathan Brassow }; 934493df71cSJonathan Brassow 935493df71cSJonathan Brassow static struct dm_exception_store_type _persistent_compat_type = { 936493df71cSJonathan Brassow .name = "P", 937493df71cSJonathan Brassow .module = THIS_MODULE, 938493df71cSJonathan Brassow .ctr = persistent_ctr, 939493df71cSJonathan Brassow .dtr = persistent_dtr, 940493df71cSJonathan Brassow .read_metadata = persistent_read_metadata, 941493df71cSJonathan Brassow .prepare_exception = persistent_prepare_exception, 942493df71cSJonathan Brassow .commit_exception = persistent_commit_exception, 9434454a621SMikulas Patocka .prepare_merge = persistent_prepare_merge, 9444454a621SMikulas Patocka .commit_merge = persistent_commit_merge, 945493df71cSJonathan Brassow .drop_snapshot = persistent_drop_snapshot, 946985903bbSMike Snitzer .usage = persistent_usage, 947493df71cSJonathan Brassow .status = persistent_status, 948493df71cSJonathan Brassow }; 949493df71cSJonathan Brassow 9504db6bfe0SAlasdair G Kergon int dm_persistent_snapshot_init(void) 9514db6bfe0SAlasdair G Kergon { 952493df71cSJonathan Brassow int r; 953493df71cSJonathan Brassow 954493df71cSJonathan Brassow r = dm_exception_store_type_register(&_persistent_type); 955493df71cSJonathan Brassow if (r) { 956493df71cSJonathan Brassow DMERR("Unable to register persistent exception store type"); 957493df71cSJonathan Brassow return r; 958493df71cSJonathan Brassow } 959493df71cSJonathan Brassow 960493df71cSJonathan Brassow r = dm_exception_store_type_register(&_persistent_compat_type); 961493df71cSJonathan Brassow if (r) { 962493df71cSJonathan Brassow DMERR("Unable to register old-style persistent exception " 963493df71cSJonathan Brassow "store type"); 964493df71cSJonathan Brassow dm_exception_store_type_unregister(&_persistent_type); 965493df71cSJonathan Brassow return r; 966493df71cSJonathan Brassow } 967493df71cSJonathan Brassow 968493df71cSJonathan Brassow return r; 9694db6bfe0SAlasdair G Kergon } 9704db6bfe0SAlasdair G Kergon 9714db6bfe0SAlasdair G Kergon void dm_persistent_snapshot_exit(void) 9724db6bfe0SAlasdair G Kergon { 973493df71cSJonathan Brassow dm_exception_store_type_unregister(&_persistent_type); 974493df71cSJonathan Brassow dm_exception_store_type_unregister(&_persistent_compat_type); 9754db6bfe0SAlasdair G Kergon } 976