140b0b3f8SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 20db9299fSJens Axboe /* 30db9299fSJens Axboe * Copyright (C) 2007 Jens Axboe <jens.axboe@oracle.com> 40db9299fSJens Axboe * 50db9299fSJens Axboe * Scatterlist handling helpers. 60db9299fSJens Axboe */ 78bc3bcc9SPaul Gortmaker #include <linux/export.h> 85a0e3ad6STejun Heo #include <linux/slab.h> 90db9299fSJens Axboe #include <linux/scatterlist.h> 10b1adaf65SFUJITA Tomonori #include <linux/highmem.h> 11b94de9bbSChris Wilson #include <linux/kmemleak.h> 12f5f82cd1SDavid Howells #include <linux/bvec.h> 13f5f82cd1SDavid Howells #include <linux/uio.h> 140db9299fSJens Axboe 150db9299fSJens Axboe /** 160db9299fSJens Axboe * sg_next - return the next scatterlist entry in a list 170db9299fSJens Axboe * @sg: The current sg entry 180db9299fSJens Axboe * 190db9299fSJens Axboe * Description: 200db9299fSJens Axboe * Usually the next entry will be @sg@ + 1, but if this sg element is part 210db9299fSJens Axboe * of a chained scatterlist, it could jump to the start of a new 220db9299fSJens Axboe * scatterlist array. 230db9299fSJens Axboe * 240db9299fSJens Axboe **/ 250db9299fSJens Axboe struct scatterlist *sg_next(struct scatterlist *sg) 260db9299fSJens Axboe { 270db9299fSJens Axboe if (sg_is_last(sg)) 280db9299fSJens Axboe return NULL; 290db9299fSJens Axboe 300db9299fSJens Axboe sg++; 310db9299fSJens Axboe if (unlikely(sg_is_chain(sg))) 320db9299fSJens Axboe sg = sg_chain_ptr(sg); 330db9299fSJens Axboe 340db9299fSJens Axboe return sg; 350db9299fSJens Axboe } 360db9299fSJens Axboe EXPORT_SYMBOL(sg_next); 370db9299fSJens Axboe 380db9299fSJens Axboe /** 392e484610SMaxim Levitsky * sg_nents - return total count of entries in scatterlist 402e484610SMaxim Levitsky * @sg: The scatterlist 412e484610SMaxim Levitsky * 422e484610SMaxim Levitsky * Description: 439dbbc3b9SZhen Lei * Allows to know how many entries are in sg, taking into account 442e484610SMaxim Levitsky * chaining as well 452e484610SMaxim Levitsky * 462e484610SMaxim Levitsky **/ 472e484610SMaxim Levitsky int sg_nents(struct scatterlist *sg) 482e484610SMaxim Levitsky { 49232f1b51SMaxim Levitsky int nents; 50232f1b51SMaxim Levitsky for (nents = 0; sg; sg = sg_next(sg)) 512e484610SMaxim Levitsky nents++; 522e484610SMaxim Levitsky return nents; 532e484610SMaxim Levitsky } 542e484610SMaxim Levitsky EXPORT_SYMBOL(sg_nents); 552e484610SMaxim Levitsky 56cfaed10dSTom Lendacky /** 57cfaed10dSTom Lendacky * sg_nents_for_len - return total count of entries in scatterlist 58cfaed10dSTom Lendacky * needed to satisfy the supplied length 59cfaed10dSTom Lendacky * @sg: The scatterlist 60cfaed10dSTom Lendacky * @len: The total required length 61cfaed10dSTom Lendacky * 62cfaed10dSTom Lendacky * Description: 63cfaed10dSTom Lendacky * Determines the number of entries in sg that are required to meet 649dbbc3b9SZhen Lei * the supplied length, taking into account chaining as well 65cfaed10dSTom Lendacky * 66cfaed10dSTom Lendacky * Returns: 67cfaed10dSTom Lendacky * the number of sg entries needed, negative error on failure 68cfaed10dSTom Lendacky * 69cfaed10dSTom Lendacky **/ 70cfaed10dSTom Lendacky int sg_nents_for_len(struct scatterlist *sg, u64 len) 71cfaed10dSTom Lendacky { 72cfaed10dSTom Lendacky int nents; 73cfaed10dSTom Lendacky u64 total; 74cfaed10dSTom Lendacky 75cfaed10dSTom Lendacky if (!len) 76cfaed10dSTom Lendacky return 0; 77cfaed10dSTom Lendacky 78cfaed10dSTom Lendacky for (nents = 0, total = 0; sg; sg = sg_next(sg)) { 79cfaed10dSTom Lendacky nents++; 80cfaed10dSTom Lendacky total += sg->length; 81cfaed10dSTom Lendacky if (total >= len) 82cfaed10dSTom Lendacky return nents; 83cfaed10dSTom Lendacky } 84cfaed10dSTom Lendacky 85cfaed10dSTom Lendacky return -EINVAL; 86cfaed10dSTom Lendacky } 87cfaed10dSTom Lendacky EXPORT_SYMBOL(sg_nents_for_len); 882e484610SMaxim Levitsky 892e484610SMaxim Levitsky /** 900db9299fSJens Axboe * sg_last - return the last scatterlist entry in a list 910db9299fSJens Axboe * @sgl: First entry in the scatterlist 920db9299fSJens Axboe * @nents: Number of entries in the scatterlist 930db9299fSJens Axboe * 940db9299fSJens Axboe * Description: 950db9299fSJens Axboe * Should only be used casually, it (currently) scans the entire list 960db9299fSJens Axboe * to get the last entry. 970db9299fSJens Axboe * 980db9299fSJens Axboe * Note that the @sgl@ pointer passed in need not be the first one, 990db9299fSJens Axboe * the important bit is that @nents@ denotes the number of entries that 1000db9299fSJens Axboe * exist from @sgl@. 1010db9299fSJens Axboe * 1020db9299fSJens Axboe **/ 1030db9299fSJens Axboe struct scatterlist *sg_last(struct scatterlist *sgl, unsigned int nents) 1040db9299fSJens Axboe { 1050db9299fSJens Axboe struct scatterlist *sg, *ret = NULL; 1060db9299fSJens Axboe unsigned int i; 1070db9299fSJens Axboe 1080db9299fSJens Axboe for_each_sg(sgl, sg, nents, i) 1090db9299fSJens Axboe ret = sg; 1100db9299fSJens Axboe 1110db9299fSJens Axboe BUG_ON(!sg_is_last(ret)); 1120db9299fSJens Axboe return ret; 1130db9299fSJens Axboe } 1140db9299fSJens Axboe EXPORT_SYMBOL(sg_last); 1150db9299fSJens Axboe 1160db9299fSJens Axboe /** 1170db9299fSJens Axboe * sg_init_table - Initialize SG table 1180db9299fSJens Axboe * @sgl: The SG table 1190db9299fSJens Axboe * @nents: Number of entries in table 1200db9299fSJens Axboe * 1210db9299fSJens Axboe * Notes: 1220db9299fSJens Axboe * If this is part of a chained sg table, sg_mark_end() should be 1230db9299fSJens Axboe * used only on the last table part. 1240db9299fSJens Axboe * 1250db9299fSJens Axboe **/ 1260db9299fSJens Axboe void sg_init_table(struct scatterlist *sgl, unsigned int nents) 1270db9299fSJens Axboe { 1280db9299fSJens Axboe memset(sgl, 0, sizeof(*sgl) * nents); 129f3851786SPrashant Bhole sg_init_marker(sgl, nents); 1300db9299fSJens Axboe } 1310db9299fSJens Axboe EXPORT_SYMBOL(sg_init_table); 1320db9299fSJens Axboe 1330db9299fSJens Axboe /** 1340db9299fSJens Axboe * sg_init_one - Initialize a single entry sg list 1350db9299fSJens Axboe * @sg: SG entry 1360db9299fSJens Axboe * @buf: Virtual address for IO 1370db9299fSJens Axboe * @buflen: IO length 1380db9299fSJens Axboe * 1390db9299fSJens Axboe **/ 1400db9299fSJens Axboe void sg_init_one(struct scatterlist *sg, const void *buf, unsigned int buflen) 1410db9299fSJens Axboe { 1420db9299fSJens Axboe sg_init_table(sg, 1); 1430db9299fSJens Axboe sg_set_buf(sg, buf, buflen); 1440db9299fSJens Axboe } 1450db9299fSJens Axboe EXPORT_SYMBOL(sg_init_one); 1460db9299fSJens Axboe 1470db9299fSJens Axboe /* 1480db9299fSJens Axboe * The default behaviour of sg_alloc_table() is to use these kmalloc/kfree 1490db9299fSJens Axboe * helpers. 1500db9299fSJens Axboe */ 1510db9299fSJens Axboe static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask) 1520db9299fSJens Axboe { 153b94de9bbSChris Wilson if (nents == SG_MAX_SINGLE_ALLOC) { 154b94de9bbSChris Wilson /* 155b94de9bbSChris Wilson * Kmemleak doesn't track page allocations as they are not 156b94de9bbSChris Wilson * commonly used (in a raw form) for kernel data structures. 157b94de9bbSChris Wilson * As we chain together a list of pages and then a normal 158b94de9bbSChris Wilson * kmalloc (tracked by kmemleak), in order to for that last 159b94de9bbSChris Wilson * allocation not to become decoupled (and thus a 160b94de9bbSChris Wilson * false-positive) we need to inform kmemleak of all the 161b94de9bbSChris Wilson * intermediate allocations. 162b94de9bbSChris Wilson */ 163b94de9bbSChris Wilson void *ptr = (void *) __get_free_page(gfp_mask); 164b94de9bbSChris Wilson kmemleak_alloc(ptr, PAGE_SIZE, 1, gfp_mask); 165b94de9bbSChris Wilson return ptr; 166b94de9bbSChris Wilson } else 1676da2ec56SKees Cook return kmalloc_array(nents, sizeof(struct scatterlist), 1686da2ec56SKees Cook gfp_mask); 1690db9299fSJens Axboe } 1700db9299fSJens Axboe 1710db9299fSJens Axboe static void sg_kfree(struct scatterlist *sg, unsigned int nents) 1720db9299fSJens Axboe { 173b94de9bbSChris Wilson if (nents == SG_MAX_SINGLE_ALLOC) { 174b94de9bbSChris Wilson kmemleak_free(sg); 1750db9299fSJens Axboe free_page((unsigned long) sg); 176b94de9bbSChris Wilson } else 1770db9299fSJens Axboe kfree(sg); 1780db9299fSJens Axboe } 1790db9299fSJens Axboe 1800db9299fSJens Axboe /** 1810db9299fSJens Axboe * __sg_free_table - Free a previously mapped sg table 1820db9299fSJens Axboe * @table: The sg table header to use 1837cedb1f1SJames Bottomley * @max_ents: The maximum number of entries per single scatterlist 1844635873cSMing Lei * @nents_first_chunk: Number of entries int the (preallocated) first 1854635873cSMing Lei * scatterlist chunk, 0 means no such preallocated first chunk 1860db9299fSJens Axboe * @free_fn: Free function 1873e302dbcSMaor Gottlieb * @num_ents: Number of entries in the table 1880db9299fSJens Axboe * 1890db9299fSJens Axboe * Description: 1907cedb1f1SJames Bottomley * Free an sg table previously allocated and setup with 1917cedb1f1SJames Bottomley * __sg_alloc_table(). The @max_ents value must be identical to 1927cedb1f1SJames Bottomley * that previously used with __sg_alloc_table(). 1930db9299fSJens Axboe * 1940db9299fSJens Axboe **/ 1957cedb1f1SJames Bottomley void __sg_free_table(struct sg_table *table, unsigned int max_ents, 1963e302dbcSMaor Gottlieb unsigned int nents_first_chunk, sg_free_fn *free_fn, 1973e302dbcSMaor Gottlieb unsigned int num_ents) 1980db9299fSJens Axboe { 1990db9299fSJens Axboe struct scatterlist *sgl, *next; 2004635873cSMing Lei unsigned curr_max_ents = nents_first_chunk ?: max_ents; 2010db9299fSJens Axboe 2020db9299fSJens Axboe if (unlikely(!table->sgl)) 2030db9299fSJens Axboe return; 2040db9299fSJens Axboe 2050db9299fSJens Axboe sgl = table->sgl; 2063e302dbcSMaor Gottlieb while (num_ents) { 2073e302dbcSMaor Gottlieb unsigned int alloc_size = num_ents; 2080db9299fSJens Axboe unsigned int sg_size; 2090db9299fSJens Axboe 2100db9299fSJens Axboe /* 2117cedb1f1SJames Bottomley * If we have more than max_ents segments left, 2120db9299fSJens Axboe * then assign 'next' to the sg table after the current one. 2130db9299fSJens Axboe * sg_size is then one less than alloc size, since the last 2140db9299fSJens Axboe * element is the chain pointer. 2150db9299fSJens Axboe */ 2164635873cSMing Lei if (alloc_size > curr_max_ents) { 2174635873cSMing Lei next = sg_chain_ptr(&sgl[curr_max_ents - 1]); 2184635873cSMing Lei alloc_size = curr_max_ents; 2190db9299fSJens Axboe sg_size = alloc_size - 1; 2200db9299fSJens Axboe } else { 2210db9299fSJens Axboe sg_size = alloc_size; 2220db9299fSJens Axboe next = NULL; 2230db9299fSJens Axboe } 2240db9299fSJens Axboe 2253e302dbcSMaor Gottlieb num_ents -= sg_size; 2264635873cSMing Lei if (nents_first_chunk) 2274635873cSMing Lei nents_first_chunk = 0; 228c21e59d8STony Battersby else 229c21e59d8STony Battersby free_fn(sgl, alloc_size); 2300db9299fSJens Axboe sgl = next; 2314635873cSMing Lei curr_max_ents = max_ents; 2320db9299fSJens Axboe } 2330db9299fSJens Axboe 2340db9299fSJens Axboe table->sgl = NULL; 2350db9299fSJens Axboe } 2360db9299fSJens Axboe EXPORT_SYMBOL(__sg_free_table); 2370db9299fSJens Axboe 2380db9299fSJens Axboe /** 2393e302dbcSMaor Gottlieb * sg_free_append_table - Free a previously allocated append sg table. 2403e302dbcSMaor Gottlieb * @table: The mapped sg append table header 2413e302dbcSMaor Gottlieb * 2423e302dbcSMaor Gottlieb **/ 2433e302dbcSMaor Gottlieb void sg_free_append_table(struct sg_append_table *table) 2443e302dbcSMaor Gottlieb { 2456d529ea8Swuchi __sg_free_table(&table->sgt, SG_MAX_SINGLE_ALLOC, 0, sg_kfree, 2463e302dbcSMaor Gottlieb table->total_nents); 2473e302dbcSMaor Gottlieb } 2483e302dbcSMaor Gottlieb EXPORT_SYMBOL(sg_free_append_table); 2493e302dbcSMaor Gottlieb 2503e302dbcSMaor Gottlieb 2513e302dbcSMaor Gottlieb /** 2520db9299fSJens Axboe * sg_free_table - Free a previously allocated sg table 2530db9299fSJens Axboe * @table: The mapped sg table header 2540db9299fSJens Axboe * 2550db9299fSJens Axboe **/ 2560db9299fSJens Axboe void sg_free_table(struct sg_table *table) 2570db9299fSJens Axboe { 2586d529ea8Swuchi __sg_free_table(table, SG_MAX_SINGLE_ALLOC, 0, sg_kfree, 2593e302dbcSMaor Gottlieb table->orig_nents); 2600db9299fSJens Axboe } 2610db9299fSJens Axboe EXPORT_SYMBOL(sg_free_table); 2620db9299fSJens Axboe 2630db9299fSJens Axboe /** 2640db9299fSJens Axboe * __sg_alloc_table - Allocate and initialize an sg table with given allocator 2650db9299fSJens Axboe * @table: The sg table header to use 2660db9299fSJens Axboe * @nents: Number of entries in sg list 2677cedb1f1SJames Bottomley * @max_ents: The maximum number of entries the allocator returns per call 268*c80da1fbSRandy Dunlap * @first_chunk: first SGL if preallocated (may be %NULL) 269*c80da1fbSRandy Dunlap * @nents_first_chunk: Number of entries in the (preallocated) first 2704635873cSMing Lei * scatterlist chunk, 0 means no such preallocated chunk provided by user 2710db9299fSJens Axboe * @gfp_mask: GFP allocation mask 2720db9299fSJens Axboe * @alloc_fn: Allocator to use 2730db9299fSJens Axboe * 2747cedb1f1SJames Bottomley * Description: 2757cedb1f1SJames Bottomley * This function returns a @table @nents long. The allocator is 2767cedb1f1SJames Bottomley * defined to return scatterlist chunks of maximum size @max_ents. 2777cedb1f1SJames Bottomley * Thus if @nents is bigger than @max_ents, the scatterlists will be 2787cedb1f1SJames Bottomley * chained in units of @max_ents. 2797cedb1f1SJames Bottomley * 2800db9299fSJens Axboe * Notes: 2810db9299fSJens Axboe * If this function returns non-0 (eg failure), the caller must call 2820db9299fSJens Axboe * __sg_free_table() to cleanup any leftover allocations. 2830db9299fSJens Axboe * 2840db9299fSJens Axboe **/ 2857cedb1f1SJames Bottomley int __sg_alloc_table(struct sg_table *table, unsigned int nents, 286c53c6d6aSChristoph Hellwig unsigned int max_ents, struct scatterlist *first_chunk, 2874635873cSMing Lei unsigned int nents_first_chunk, gfp_t gfp_mask, 2884635873cSMing Lei sg_alloc_fn *alloc_fn) 2890db9299fSJens Axboe { 2900db9299fSJens Axboe struct scatterlist *sg, *prv; 2910db9299fSJens Axboe unsigned int left; 2924635873cSMing Lei unsigned curr_max_ents = nents_first_chunk ?: max_ents; 2934635873cSMing Lei unsigned prv_max_ents; 2940db9299fSJens Axboe 29527daabd9SDan Carpenter memset(table, 0, sizeof(*table)); 29627daabd9SDan Carpenter 29727daabd9SDan Carpenter if (nents == 0) 29827daabd9SDan Carpenter return -EINVAL; 2997c703e54SChristoph Hellwig #ifdef CONFIG_ARCH_NO_SG_CHAIN 3006fd59a83SNick Bowler if (WARN_ON_ONCE(nents > max_ents)) 3016fd59a83SNick Bowler return -EINVAL; 3020db9299fSJens Axboe #endif 3030db9299fSJens Axboe 3040db9299fSJens Axboe left = nents; 3050db9299fSJens Axboe prv = NULL; 3060db9299fSJens Axboe do { 3070db9299fSJens Axboe unsigned int sg_size, alloc_size = left; 3080db9299fSJens Axboe 3094635873cSMing Lei if (alloc_size > curr_max_ents) { 3104635873cSMing Lei alloc_size = curr_max_ents; 3110db9299fSJens Axboe sg_size = alloc_size - 1; 3120db9299fSJens Axboe } else 3130db9299fSJens Axboe sg_size = alloc_size; 3140db9299fSJens Axboe 3150db9299fSJens Axboe left -= sg_size; 3160db9299fSJens Axboe 317c53c6d6aSChristoph Hellwig if (first_chunk) { 318c53c6d6aSChristoph Hellwig sg = first_chunk; 319c53c6d6aSChristoph Hellwig first_chunk = NULL; 320c53c6d6aSChristoph Hellwig } else { 3210db9299fSJens Axboe sg = alloc_fn(alloc_size, gfp_mask); 322c53c6d6aSChristoph Hellwig } 323edce6820SJeffrey Carlyle if (unlikely(!sg)) { 324edce6820SJeffrey Carlyle /* 325edce6820SJeffrey Carlyle * Adjust entry count to reflect that the last 326edce6820SJeffrey Carlyle * entry of the previous table won't be used for 327edce6820SJeffrey Carlyle * linkage. Without this, sg_kfree() may get 328edce6820SJeffrey Carlyle * confused. 329edce6820SJeffrey Carlyle */ 330edce6820SJeffrey Carlyle if (prv) 331edce6820SJeffrey Carlyle table->nents = ++table->orig_nents; 332edce6820SJeffrey Carlyle 3330db9299fSJens Axboe return -ENOMEM; 334edce6820SJeffrey Carlyle } 3350db9299fSJens Axboe 3360db9299fSJens Axboe sg_init_table(sg, alloc_size); 3370db9299fSJens Axboe table->nents = table->orig_nents += sg_size; 3380db9299fSJens Axboe 3390db9299fSJens Axboe /* 3400db9299fSJens Axboe * If this is the first mapping, assign the sg table header. 3410db9299fSJens Axboe * If this is not the first mapping, chain previous part. 3420db9299fSJens Axboe */ 3430db9299fSJens Axboe if (prv) 3444635873cSMing Lei sg_chain(prv, prv_max_ents, sg); 3450db9299fSJens Axboe else 3460db9299fSJens Axboe table->sgl = sg; 3470db9299fSJens Axboe 3480db9299fSJens Axboe /* 3490db9299fSJens Axboe * If no more entries after this one, mark the end 3500db9299fSJens Axboe */ 3510db9299fSJens Axboe if (!left) 3520db9299fSJens Axboe sg_mark_end(&sg[sg_size - 1]); 3530db9299fSJens Axboe 3540db9299fSJens Axboe prv = sg; 3554635873cSMing Lei prv_max_ents = curr_max_ents; 3564635873cSMing Lei curr_max_ents = max_ents; 3570db9299fSJens Axboe } while (left); 3580db9299fSJens Axboe 3590db9299fSJens Axboe return 0; 3600db9299fSJens Axboe } 3610db9299fSJens Axboe EXPORT_SYMBOL(__sg_alloc_table); 3620db9299fSJens Axboe 3630db9299fSJens Axboe /** 3640db9299fSJens Axboe * sg_alloc_table - Allocate and initialize an sg table 3650db9299fSJens Axboe * @table: The sg table header to use 3660db9299fSJens Axboe * @nents: Number of entries in sg list 3670db9299fSJens Axboe * @gfp_mask: GFP allocation mask 3680db9299fSJens Axboe * 3690db9299fSJens Axboe * Description: 3700db9299fSJens Axboe * Allocate and initialize an sg table. If @nents@ is larger than 3710db9299fSJens Axboe * SG_MAX_SINGLE_ALLOC a chained sg table will be setup. 3720db9299fSJens Axboe * 3730db9299fSJens Axboe **/ 3740db9299fSJens Axboe int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask) 3750db9299fSJens Axboe { 3760db9299fSJens Axboe int ret; 3770db9299fSJens Axboe 3787cedb1f1SJames Bottomley ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC, 3794635873cSMing Lei NULL, 0, gfp_mask, sg_kmalloc); 3800db9299fSJens Axboe if (unlikely(ret)) 3813e302dbcSMaor Gottlieb sg_free_table(table); 3820db9299fSJens Axboe return ret; 3830db9299fSJens Axboe } 3840db9299fSJens Axboe EXPORT_SYMBOL(sg_alloc_table); 385b1adaf65SFUJITA Tomonori 3863e302dbcSMaor Gottlieb static struct scatterlist *get_next_sg(struct sg_append_table *table, 38707da1223SMaor Gottlieb struct scatterlist *cur, 38807da1223SMaor Gottlieb unsigned long needed_sges, 38907da1223SMaor Gottlieb gfp_t gfp_mask) 39007da1223SMaor Gottlieb { 39107da1223SMaor Gottlieb struct scatterlist *new_sg, *next_sg; 39207da1223SMaor Gottlieb unsigned int alloc_size; 39307da1223SMaor Gottlieb 39407da1223SMaor Gottlieb if (cur) { 39507da1223SMaor Gottlieb next_sg = sg_next(cur); 39607da1223SMaor Gottlieb /* Check if last entry should be keeped for chainning */ 39707da1223SMaor Gottlieb if (!sg_is_last(next_sg) || needed_sges == 1) 39807da1223SMaor Gottlieb return next_sg; 39907da1223SMaor Gottlieb } 40007da1223SMaor Gottlieb 40107da1223SMaor Gottlieb alloc_size = min_t(unsigned long, needed_sges, SG_MAX_SINGLE_ALLOC); 40207da1223SMaor Gottlieb new_sg = sg_kmalloc(alloc_size, gfp_mask); 40307da1223SMaor Gottlieb if (!new_sg) 40407da1223SMaor Gottlieb return ERR_PTR(-ENOMEM); 40507da1223SMaor Gottlieb sg_init_table(new_sg, alloc_size); 40607da1223SMaor Gottlieb if (cur) { 4073e302dbcSMaor Gottlieb table->total_nents += alloc_size - 1; 40807da1223SMaor Gottlieb __sg_chain(next_sg, new_sg); 40907da1223SMaor Gottlieb } else { 4103e302dbcSMaor Gottlieb table->sgt.sgl = new_sg; 4113e302dbcSMaor Gottlieb table->total_nents = alloc_size; 41207da1223SMaor Gottlieb } 41307da1223SMaor Gottlieb return new_sg; 41407da1223SMaor Gottlieb } 41507da1223SMaor Gottlieb 4161567b49dSLogan Gunthorpe static bool pages_are_mergeable(struct page *a, struct page *b) 4171567b49dSLogan Gunthorpe { 4181567b49dSLogan Gunthorpe if (page_to_pfn(a) != page_to_pfn(b) + 1) 4191567b49dSLogan Gunthorpe return false; 4201567b49dSLogan Gunthorpe if (!zone_device_pages_have_same_pgmap(a, b)) 4211567b49dSLogan Gunthorpe return false; 4221567b49dSLogan Gunthorpe return true; 4231567b49dSLogan Gunthorpe } 4241567b49dSLogan Gunthorpe 425b1adaf65SFUJITA Tomonori /** 4263e302dbcSMaor Gottlieb * sg_alloc_append_table_from_pages - Allocate and initialize an append sg 4273e302dbcSMaor Gottlieb * table from an array of pages 4283e302dbcSMaor Gottlieb * @sgt_append: The sg append table to use 429efc42bc9STomasz Stanislawski * @pages: Pointer to an array of page pointers 430efc42bc9STomasz Stanislawski * @n_pages: Number of pages in the pages array 431efc42bc9STomasz Stanislawski * @offset: Offset from start of the first page to the start of a buffer 432efc42bc9STomasz Stanislawski * @size: Number of valid bytes in the buffer (after offset) 4339a40401cSJason Gunthorpe * @max_segment: Maximum size of a scatterlist element in bytes 43407da1223SMaor Gottlieb * @left_pages: Left pages caller have to set after this call 435efc42bc9STomasz Stanislawski * @gfp_mask: GFP allocation mask 436efc42bc9STomasz Stanislawski * 437efc42bc9STomasz Stanislawski * Description: 4383e302dbcSMaor Gottlieb * In the first call it allocate and initialize an sg table from a list of 4393e302dbcSMaor Gottlieb * pages, else reuse the scatterlist from sgt_append. Contiguous ranges of 4403e302dbcSMaor Gottlieb * the pages are squashed into a single scatterlist entry up to the maximum 4413e302dbcSMaor Gottlieb * size specified in @max_segment. A user may provide an offset at a start 4423e302dbcSMaor Gottlieb * and a size of valid data in a buffer specified by the page array. The 4433e302dbcSMaor Gottlieb * returned sg table is released by sg_free_append_table 444efc42bc9STomasz Stanislawski * 445efc42bc9STomasz Stanislawski * Returns: 4463e302dbcSMaor Gottlieb * 0 on success, negative error on failure 44707da1223SMaor Gottlieb * 44807da1223SMaor Gottlieb * Notes: 44907da1223SMaor Gottlieb * If this function returns non-0 (eg failure), the caller must call 4503e302dbcSMaor Gottlieb * sg_free_append_table() to cleanup any leftover allocations. 4513e302dbcSMaor Gottlieb * 4523e302dbcSMaor Gottlieb * In the fist call, sgt_append must by initialized. 453efc42bc9STomasz Stanislawski */ 4543e302dbcSMaor Gottlieb int sg_alloc_append_table_from_pages(struct sg_append_table *sgt_append, 45507da1223SMaor Gottlieb struct page **pages, unsigned int n_pages, unsigned int offset, 45689d8589cSTvrtko Ursulin unsigned long size, unsigned int max_segment, 4573e302dbcSMaor Gottlieb unsigned int left_pages, gfp_t gfp_mask) 458efc42bc9STomasz Stanislawski { 45907da1223SMaor Gottlieb unsigned int chunks, cur_page, seg_len, i, prv_len = 0; 46007da1223SMaor Gottlieb unsigned int added_nents = 0; 4613e302dbcSMaor Gottlieb struct scatterlist *s = sgt_append->prv; 4621567b49dSLogan Gunthorpe struct page *last_pg; 463efc42bc9STomasz Stanislawski 4649a40401cSJason Gunthorpe /* 4659a40401cSJason Gunthorpe * The algorithm below requires max_segment to be aligned to PAGE_SIZE 4669a40401cSJason Gunthorpe * otherwise it can overshoot. 4679a40401cSJason Gunthorpe */ 4689a40401cSJason Gunthorpe max_segment = ALIGN_DOWN(max_segment, PAGE_SIZE); 4699a40401cSJason Gunthorpe if (WARN_ON(max_segment < PAGE_SIZE)) 4703e302dbcSMaor Gottlieb return -EINVAL; 47107da1223SMaor Gottlieb 4723e302dbcSMaor Gottlieb if (IS_ENABLED(CONFIG_ARCH_NO_SG_CHAIN) && sgt_append->prv) 4733e302dbcSMaor Gottlieb return -EOPNOTSUPP; 47407da1223SMaor Gottlieb 4753e302dbcSMaor Gottlieb if (sgt_append->prv) { 4760f097f08SYishai Hadas unsigned long next_pfn = (page_to_phys(sg_page(sgt_append->prv)) + 4770f097f08SYishai Hadas sgt_append->prv->offset + sgt_append->prv->length) / PAGE_SIZE; 4780f097f08SYishai Hadas 47907da1223SMaor Gottlieb if (WARN_ON(offset)) 4803e302dbcSMaor Gottlieb return -EINVAL; 48107da1223SMaor Gottlieb 48207da1223SMaor Gottlieb /* Merge contiguous pages into the last SG */ 4833e302dbcSMaor Gottlieb prv_len = sgt_append->prv->length; 4840f097f08SYishai Hadas if (page_to_pfn(pages[0]) == next_pfn) { 4850f097f08SYishai Hadas last_pg = pfn_to_page(next_pfn - 1); 486e95d50d7SYishai Hadas while (n_pages && pages_are_mergeable(pages[0], last_pg)) { 4873e302dbcSMaor Gottlieb if (sgt_append->prv->length + PAGE_SIZE > max_segment) 48807da1223SMaor Gottlieb break; 4893e302dbcSMaor Gottlieb sgt_append->prv->length += PAGE_SIZE; 4901567b49dSLogan Gunthorpe last_pg = pages[0]; 49107da1223SMaor Gottlieb pages++; 49207da1223SMaor Gottlieb n_pages--; 49307da1223SMaor Gottlieb } 49407da1223SMaor Gottlieb if (!n_pages) 49507da1223SMaor Gottlieb goto out; 49607da1223SMaor Gottlieb } 4970f097f08SYishai Hadas } 49889d8589cSTvrtko Ursulin 499efc42bc9STomasz Stanislawski /* compute number of contiguous chunks */ 500efc42bc9STomasz Stanislawski chunks = 1; 501c125906bSTvrtko Ursulin seg_len = 0; 502c125906bSTvrtko Ursulin for (i = 1; i < n_pages; i++) { 503c125906bSTvrtko Ursulin seg_len += PAGE_SIZE; 504c125906bSTvrtko Ursulin if (seg_len >= max_segment || 5051567b49dSLogan Gunthorpe !pages_are_mergeable(pages[i], pages[i - 1])) { 506c125906bSTvrtko Ursulin chunks++; 507c125906bSTvrtko Ursulin seg_len = 0; 508c125906bSTvrtko Ursulin } 509c125906bSTvrtko Ursulin } 510efc42bc9STomasz Stanislawski 511efc42bc9STomasz Stanislawski /* merging chunks and putting them into the scatterlist */ 512efc42bc9STomasz Stanislawski cur_page = 0; 51307da1223SMaor Gottlieb for (i = 0; i < chunks; i++) { 514c125906bSTvrtko Ursulin unsigned int j, chunk_size; 515efc42bc9STomasz Stanislawski 516efc42bc9STomasz Stanislawski /* look for the end of the current chunk */ 517c125906bSTvrtko Ursulin seg_len = 0; 518c125906bSTvrtko Ursulin for (j = cur_page + 1; j < n_pages; j++) { 519c125906bSTvrtko Ursulin seg_len += PAGE_SIZE; 520c125906bSTvrtko Ursulin if (seg_len >= max_segment || 5211567b49dSLogan Gunthorpe !pages_are_mergeable(pages[j], pages[j - 1])) 522efc42bc9STomasz Stanislawski break; 523c125906bSTvrtko Ursulin } 524efc42bc9STomasz Stanislawski 52507da1223SMaor Gottlieb /* Pass how many chunks might be left */ 5263e302dbcSMaor Gottlieb s = get_next_sg(sgt_append, s, chunks - i + left_pages, 5273e302dbcSMaor Gottlieb gfp_mask); 52807da1223SMaor Gottlieb if (IS_ERR(s)) { 52907da1223SMaor Gottlieb /* 53007da1223SMaor Gottlieb * Adjust entry length to be as before function was 53107da1223SMaor Gottlieb * called. 53207da1223SMaor Gottlieb */ 5333e302dbcSMaor Gottlieb if (sgt_append->prv) 5343e302dbcSMaor Gottlieb sgt_append->prv->length = prv_len; 5353e302dbcSMaor Gottlieb return PTR_ERR(s); 53607da1223SMaor Gottlieb } 537efc42bc9STomasz Stanislawski chunk_size = ((j - cur_page) << PAGE_SHIFT) - offset; 538c125906bSTvrtko Ursulin sg_set_page(s, pages[cur_page], 539c125906bSTvrtko Ursulin min_t(unsigned long, size, chunk_size), offset); 54007da1223SMaor Gottlieb added_nents++; 541efc42bc9STomasz Stanislawski size -= chunk_size; 542efc42bc9STomasz Stanislawski offset = 0; 543efc42bc9STomasz Stanislawski cur_page = j; 544efc42bc9STomasz Stanislawski } 5453e302dbcSMaor Gottlieb sgt_append->sgt.nents += added_nents; 5463e302dbcSMaor Gottlieb sgt_append->sgt.orig_nents = sgt_append->sgt.nents; 5473e302dbcSMaor Gottlieb sgt_append->prv = s; 54807da1223SMaor Gottlieb out: 54907da1223SMaor Gottlieb if (!left_pages) 55007da1223SMaor Gottlieb sg_mark_end(s); 5513e302dbcSMaor Gottlieb return 0; 552efc42bc9STomasz Stanislawski } 55390e7a6deSMaor Gottlieb EXPORT_SYMBOL(sg_alloc_append_table_from_pages); 55489d8589cSTvrtko Ursulin 55589d8589cSTvrtko Ursulin /** 55690e7a6deSMaor Gottlieb * sg_alloc_table_from_pages_segment - Allocate and initialize an sg table from 55790e7a6deSMaor Gottlieb * an array of pages and given maximum 55890e7a6deSMaor Gottlieb * segment. 55989d8589cSTvrtko Ursulin * @sgt: The sg table header to use 56089d8589cSTvrtko Ursulin * @pages: Pointer to an array of page pointers 56189d8589cSTvrtko Ursulin * @n_pages: Number of pages in the pages array 56289d8589cSTvrtko Ursulin * @offset: Offset from start of the first page to the start of a buffer 56389d8589cSTvrtko Ursulin * @size: Number of valid bytes in the buffer (after offset) 56490e7a6deSMaor Gottlieb * @max_segment: Maximum size of a scatterlist element in bytes 56589d8589cSTvrtko Ursulin * @gfp_mask: GFP allocation mask 56689d8589cSTvrtko Ursulin * 56789d8589cSTvrtko Ursulin * Description: 56889d8589cSTvrtko Ursulin * Allocate and initialize an sg table from a list of pages. Contiguous 56990e7a6deSMaor Gottlieb * ranges of the pages are squashed into a single scatterlist node up to the 57090e7a6deSMaor Gottlieb * maximum size specified in @max_segment. A user may provide an offset at a 57190e7a6deSMaor Gottlieb * start and a size of valid data in a buffer specified by the page array. 57290e7a6deSMaor Gottlieb * 57390e7a6deSMaor Gottlieb * The returned sg table is released by sg_free_table. 57489d8589cSTvrtko Ursulin * 57589d8589cSTvrtko Ursulin * Returns: 57689d8589cSTvrtko Ursulin * 0 on success, negative error on failure 57789d8589cSTvrtko Ursulin */ 57890e7a6deSMaor Gottlieb int sg_alloc_table_from_pages_segment(struct sg_table *sgt, struct page **pages, 57989d8589cSTvrtko Ursulin unsigned int n_pages, unsigned int offset, 58090e7a6deSMaor Gottlieb unsigned long size, unsigned int max_segment, 58190e7a6deSMaor Gottlieb gfp_t gfp_mask) 58289d8589cSTvrtko Ursulin { 5833e302dbcSMaor Gottlieb struct sg_append_table append = {}; 5843e302dbcSMaor Gottlieb int err; 5853e302dbcSMaor Gottlieb 5863e302dbcSMaor Gottlieb err = sg_alloc_append_table_from_pages(&append, pages, n_pages, offset, 5873e302dbcSMaor Gottlieb size, max_segment, 0, gfp_mask); 5883e302dbcSMaor Gottlieb if (err) { 5893e302dbcSMaor Gottlieb sg_free_append_table(&append); 5903e302dbcSMaor Gottlieb return err; 5913e302dbcSMaor Gottlieb } 5923e302dbcSMaor Gottlieb memcpy(sgt, &append.sgt, sizeof(*sgt)); 5933e302dbcSMaor Gottlieb WARN_ON(append.total_nents != sgt->orig_nents); 5943e302dbcSMaor Gottlieb return 0; 59589d8589cSTvrtko Ursulin } 59690e7a6deSMaor Gottlieb EXPORT_SYMBOL(sg_alloc_table_from_pages_segment); 597efc42bc9STomasz Stanislawski 598e80a0af4SBart Van Assche #ifdef CONFIG_SGL_ALLOC 599e80a0af4SBart Van Assche 600e80a0af4SBart Van Assche /** 601e80a0af4SBart Van Assche * sgl_alloc_order - allocate a scatterlist and its pages 602e80a0af4SBart Van Assche * @length: Length in bytes of the scatterlist. Must be at least one 603e80a0af4SBart Van Assche * @order: Second argument for alloc_pages() 604e80a0af4SBart Van Assche * @chainable: Whether or not to allocate an extra element in the scatterlist 605e80a0af4SBart Van Assche * for scatterlist chaining purposes 606e80a0af4SBart Van Assche * @gfp: Memory allocation flags 607e80a0af4SBart Van Assche * @nent_p: [out] Number of entries in the scatterlist that have pages 608e80a0af4SBart Van Assche * 609e80a0af4SBart Van Assche * Returns: A pointer to an initialized scatterlist or %NULL upon failure. 610e80a0af4SBart Van Assche */ 611e80a0af4SBart Van Assche struct scatterlist *sgl_alloc_order(unsigned long long length, 612e80a0af4SBart Van Assche unsigned int order, bool chainable, 613e80a0af4SBart Van Assche gfp_t gfp, unsigned int *nent_p) 614e80a0af4SBart Van Assche { 615e80a0af4SBart Van Assche struct scatterlist *sgl, *sg; 616e80a0af4SBart Van Assche struct page *page; 617e80a0af4SBart Van Assche unsigned int nent, nalloc; 618e80a0af4SBart Van Assche u32 elem_len; 619e80a0af4SBart Van Assche 620e80a0af4SBart Van Assche nent = round_up(length, PAGE_SIZE << order) >> (PAGE_SHIFT + order); 621e80a0af4SBart Van Assche /* Check for integer overflow */ 622e80a0af4SBart Van Assche if (length > (nent << (PAGE_SHIFT + order))) 623e80a0af4SBart Van Assche return NULL; 624e80a0af4SBart Van Assche nalloc = nent; 625e80a0af4SBart Van Assche if (chainable) { 626e80a0af4SBart Van Assche /* Check for integer overflow */ 627e80a0af4SBart Van Assche if (nalloc + 1 < nalloc) 628e80a0af4SBart Van Assche return NULL; 629e80a0af4SBart Van Assche nalloc++; 630e80a0af4SBart Van Assche } 631e80a0af4SBart Van Assche sgl = kmalloc_array(nalloc, sizeof(struct scatterlist), 6326ed9b92eSChristophe JAILLET gfp & ~GFP_DMA); 633e80a0af4SBart Van Assche if (!sgl) 634e80a0af4SBart Van Assche return NULL; 635e80a0af4SBart Van Assche 6368c7a8d1cSBart Van Assche sg_init_table(sgl, nalloc); 637e80a0af4SBart Van Assche sg = sgl; 638e80a0af4SBart Van Assche while (length) { 639e80a0af4SBart Van Assche elem_len = min_t(u64, length, PAGE_SIZE << order); 640e80a0af4SBart Van Assche page = alloc_pages(gfp, order); 641e80a0af4SBart Van Assche if (!page) { 642b2a182a4SDouglas Gilbert sgl_free_order(sgl, order); 643e80a0af4SBart Van Assche return NULL; 644e80a0af4SBart Van Assche } 645e80a0af4SBart Van Assche 646e80a0af4SBart Van Assche sg_set_page(sg, page, elem_len, 0); 647e80a0af4SBart Van Assche length -= elem_len; 648e80a0af4SBart Van Assche sg = sg_next(sg); 649e80a0af4SBart Van Assche } 6508c7a8d1cSBart Van Assche WARN_ONCE(length, "length = %lld\n", length); 651e80a0af4SBart Van Assche if (nent_p) 652e80a0af4SBart Van Assche *nent_p = nent; 653e80a0af4SBart Van Assche return sgl; 654e80a0af4SBart Van Assche } 655e80a0af4SBart Van Assche EXPORT_SYMBOL(sgl_alloc_order); 656e80a0af4SBart Van Assche 657e80a0af4SBart Van Assche /** 658e80a0af4SBart Van Assche * sgl_alloc - allocate a scatterlist and its pages 659e80a0af4SBart Van Assche * @length: Length in bytes of the scatterlist 660e80a0af4SBart Van Assche * @gfp: Memory allocation flags 661e80a0af4SBart Van Assche * @nent_p: [out] Number of entries in the scatterlist 662e80a0af4SBart Van Assche * 663e80a0af4SBart Van Assche * Returns: A pointer to an initialized scatterlist or %NULL upon failure. 664e80a0af4SBart Van Assche */ 665e80a0af4SBart Van Assche struct scatterlist *sgl_alloc(unsigned long long length, gfp_t gfp, 666e80a0af4SBart Van Assche unsigned int *nent_p) 667e80a0af4SBart Van Assche { 668e80a0af4SBart Van Assche return sgl_alloc_order(length, 0, false, gfp, nent_p); 669e80a0af4SBart Van Assche } 670e80a0af4SBart Van Assche EXPORT_SYMBOL(sgl_alloc); 671e80a0af4SBart Van Assche 672e80a0af4SBart Van Assche /** 6738c7a8d1cSBart Van Assche * sgl_free_n_order - free a scatterlist and its pages 6748c7a8d1cSBart Van Assche * @sgl: Scatterlist with one or more elements 6758c7a8d1cSBart Van Assche * @nents: Maximum number of elements to free 6768c7a8d1cSBart Van Assche * @order: Second argument for __free_pages() 6778c7a8d1cSBart Van Assche * 6788c7a8d1cSBart Van Assche * Notes: 6798c7a8d1cSBart Van Assche * - If several scatterlists have been chained and each chain element is 6808c7a8d1cSBart Van Assche * freed separately then it's essential to set nents correctly to avoid that a 6818c7a8d1cSBart Van Assche * page would get freed twice. 6828c7a8d1cSBart Van Assche * - All pages in a chained scatterlist can be freed at once by setting @nents 6838c7a8d1cSBart Van Assche * to a high number. 6848c7a8d1cSBart Van Assche */ 6858c7a8d1cSBart Van Assche void sgl_free_n_order(struct scatterlist *sgl, int nents, int order) 6868c7a8d1cSBart Van Assche { 6878c7a8d1cSBart Van Assche struct scatterlist *sg; 6888c7a8d1cSBart Van Assche struct page *page; 6898c7a8d1cSBart Van Assche int i; 6908c7a8d1cSBart Van Assche 6918c7a8d1cSBart Van Assche for_each_sg(sgl, sg, nents, i) { 6928c7a8d1cSBart Van Assche if (!sg) 6938c7a8d1cSBart Van Assche break; 6948c7a8d1cSBart Van Assche page = sg_page(sg); 6958c7a8d1cSBart Van Assche if (page) 6968c7a8d1cSBart Van Assche __free_pages(page, order); 6978c7a8d1cSBart Van Assche } 6988c7a8d1cSBart Van Assche kfree(sgl); 6998c7a8d1cSBart Van Assche } 7008c7a8d1cSBart Van Assche EXPORT_SYMBOL(sgl_free_n_order); 7018c7a8d1cSBart Van Assche 7028c7a8d1cSBart Van Assche /** 703e80a0af4SBart Van Assche * sgl_free_order - free a scatterlist and its pages 704e80a0af4SBart Van Assche * @sgl: Scatterlist with one or more elements 705e80a0af4SBart Van Assche * @order: Second argument for __free_pages() 706e80a0af4SBart Van Assche */ 707e80a0af4SBart Van Assche void sgl_free_order(struct scatterlist *sgl, int order) 708e80a0af4SBart Van Assche { 7098c7a8d1cSBart Van Assche sgl_free_n_order(sgl, INT_MAX, order); 710e80a0af4SBart Van Assche } 711e80a0af4SBart Van Assche EXPORT_SYMBOL(sgl_free_order); 712e80a0af4SBart Van Assche 713e80a0af4SBart Van Assche /** 714e80a0af4SBart Van Assche * sgl_free - free a scatterlist and its pages 715e80a0af4SBart Van Assche * @sgl: Scatterlist with one or more elements 716e80a0af4SBart Van Assche */ 717e80a0af4SBart Van Assche void sgl_free(struct scatterlist *sgl) 718e80a0af4SBart Van Assche { 719e80a0af4SBart Van Assche sgl_free_order(sgl, 0); 720e80a0af4SBart Van Assche } 721e80a0af4SBart Van Assche EXPORT_SYMBOL(sgl_free); 722e80a0af4SBart Van Assche 723e80a0af4SBart Van Assche #endif /* CONFIG_SGL_ALLOC */ 724e80a0af4SBart Van Assche 725a321e91bSImre Deak void __sg_page_iter_start(struct sg_page_iter *piter, 726a321e91bSImre Deak struct scatterlist *sglist, unsigned int nents, 727a321e91bSImre Deak unsigned long pgoffset) 728a321e91bSImre Deak { 729a321e91bSImre Deak piter->__pg_advance = 0; 730a321e91bSImre Deak piter->__nents = nents; 731a321e91bSImre Deak 732a321e91bSImre Deak piter->sg = sglist; 733a321e91bSImre Deak piter->sg_pgoffset = pgoffset; 734a321e91bSImre Deak } 735a321e91bSImre Deak EXPORT_SYMBOL(__sg_page_iter_start); 736a321e91bSImre Deak 737a321e91bSImre Deak static int sg_page_count(struct scatterlist *sg) 738a321e91bSImre Deak { 739a321e91bSImre Deak return PAGE_ALIGN(sg->offset + sg->length) >> PAGE_SHIFT; 740a321e91bSImre Deak } 741a321e91bSImre Deak 742a321e91bSImre Deak bool __sg_page_iter_next(struct sg_page_iter *piter) 743a321e91bSImre Deak { 744a321e91bSImre Deak if (!piter->__nents || !piter->sg) 745a321e91bSImre Deak return false; 746a321e91bSImre Deak 747a321e91bSImre Deak piter->sg_pgoffset += piter->__pg_advance; 748a321e91bSImre Deak piter->__pg_advance = 1; 749a321e91bSImre Deak 750a321e91bSImre Deak while (piter->sg_pgoffset >= sg_page_count(piter->sg)) { 751a321e91bSImre Deak piter->sg_pgoffset -= sg_page_count(piter->sg); 752a321e91bSImre Deak piter->sg = sg_next(piter->sg); 753a321e91bSImre Deak if (!--piter->__nents || !piter->sg) 754a321e91bSImre Deak return false; 755a321e91bSImre Deak } 756a321e91bSImre Deak 757a321e91bSImre Deak return true; 758a321e91bSImre Deak } 759a321e91bSImre Deak EXPORT_SYMBOL(__sg_page_iter_next); 760a321e91bSImre Deak 761d901b276SJason Gunthorpe static int sg_dma_page_count(struct scatterlist *sg) 762d901b276SJason Gunthorpe { 763d901b276SJason Gunthorpe return PAGE_ALIGN(sg->offset + sg_dma_len(sg)) >> PAGE_SHIFT; 764d901b276SJason Gunthorpe } 765d901b276SJason Gunthorpe 766d901b276SJason Gunthorpe bool __sg_page_iter_dma_next(struct sg_dma_page_iter *dma_iter) 767d901b276SJason Gunthorpe { 768d901b276SJason Gunthorpe struct sg_page_iter *piter = &dma_iter->base; 769d901b276SJason Gunthorpe 770d901b276SJason Gunthorpe if (!piter->__nents || !piter->sg) 771d901b276SJason Gunthorpe return false; 772d901b276SJason Gunthorpe 773d901b276SJason Gunthorpe piter->sg_pgoffset += piter->__pg_advance; 774d901b276SJason Gunthorpe piter->__pg_advance = 1; 775d901b276SJason Gunthorpe 776d901b276SJason Gunthorpe while (piter->sg_pgoffset >= sg_dma_page_count(piter->sg)) { 777d901b276SJason Gunthorpe piter->sg_pgoffset -= sg_dma_page_count(piter->sg); 778d901b276SJason Gunthorpe piter->sg = sg_next(piter->sg); 779d901b276SJason Gunthorpe if (!--piter->__nents || !piter->sg) 780d901b276SJason Gunthorpe return false; 781d901b276SJason Gunthorpe } 782d901b276SJason Gunthorpe 783d901b276SJason Gunthorpe return true; 784d901b276SJason Gunthorpe } 785d901b276SJason Gunthorpe EXPORT_SYMBOL(__sg_page_iter_dma_next); 786d901b276SJason Gunthorpe 787efc42bc9STomasz Stanislawski /** 788137d3edbSTejun Heo * sg_miter_start - start mapping iteration over a sg list 789137d3edbSTejun Heo * @miter: sg mapping iter to be started 790137d3edbSTejun Heo * @sgl: sg list to iterate over 791137d3edbSTejun Heo * @nents: number of sg entries 792*c80da1fbSRandy Dunlap * @flags: sg iterator flags 793137d3edbSTejun Heo * 794137d3edbSTejun Heo * Description: 795137d3edbSTejun Heo * Starts mapping iterator @miter. 796137d3edbSTejun Heo * 797137d3edbSTejun Heo * Context: 798137d3edbSTejun Heo * Don't care. 799137d3edbSTejun Heo */ 800137d3edbSTejun Heo void sg_miter_start(struct sg_mapping_iter *miter, struct scatterlist *sgl, 801137d3edbSTejun Heo unsigned int nents, unsigned int flags) 802137d3edbSTejun Heo { 803137d3edbSTejun Heo memset(miter, 0, sizeof(struct sg_mapping_iter)); 804137d3edbSTejun Heo 8054225fc85SImre Deak __sg_page_iter_start(&miter->piter, sgl, nents, 0); 8066de7e356SSebastian Andrzej Siewior WARN_ON(!(flags & (SG_MITER_TO_SG | SG_MITER_FROM_SG))); 807137d3edbSTejun Heo miter->__flags = flags; 808137d3edbSTejun Heo } 809137d3edbSTejun Heo EXPORT_SYMBOL(sg_miter_start); 810137d3edbSTejun Heo 81111052004SAkinobu Mita static bool sg_miter_get_next_page(struct sg_mapping_iter *miter) 81211052004SAkinobu Mita { 81311052004SAkinobu Mita if (!miter->__remaining) { 81411052004SAkinobu Mita struct scatterlist *sg; 81511052004SAkinobu Mita 81611052004SAkinobu Mita if (!__sg_page_iter_next(&miter->piter)) 81711052004SAkinobu Mita return false; 81811052004SAkinobu Mita 81911052004SAkinobu Mita sg = miter->piter.sg; 82011052004SAkinobu Mita 821aeb87246SChristophe Leroy miter->__offset = miter->piter.sg_pgoffset ? 0 : sg->offset; 822aeb87246SChristophe Leroy miter->piter.sg_pgoffset += miter->__offset >> PAGE_SHIFT; 823aeb87246SChristophe Leroy miter->__offset &= PAGE_SIZE - 1; 82411052004SAkinobu Mita miter->__remaining = sg->offset + sg->length - 825aeb87246SChristophe Leroy (miter->piter.sg_pgoffset << PAGE_SHIFT) - 826aeb87246SChristophe Leroy miter->__offset; 82711052004SAkinobu Mita miter->__remaining = min_t(unsigned long, miter->__remaining, 82811052004SAkinobu Mita PAGE_SIZE - miter->__offset); 82911052004SAkinobu Mita } 83011052004SAkinobu Mita 83111052004SAkinobu Mita return true; 83211052004SAkinobu Mita } 83311052004SAkinobu Mita 834137d3edbSTejun Heo /** 835df642ceaSAkinobu Mita * sg_miter_skip - reposition mapping iterator 836df642ceaSAkinobu Mita * @miter: sg mapping iter to be skipped 837df642ceaSAkinobu Mita * @offset: number of bytes to plus the current location 838df642ceaSAkinobu Mita * 839df642ceaSAkinobu Mita * Description: 840df642ceaSAkinobu Mita * Sets the offset of @miter to its current location plus @offset bytes. 841df642ceaSAkinobu Mita * If mapping iterator @miter has been proceeded by sg_miter_next(), this 842df642ceaSAkinobu Mita * stops @miter. 843df642ceaSAkinobu Mita * 844df642ceaSAkinobu Mita * Context: 845723aca20SThomas Gleixner * Don't care. 846df642ceaSAkinobu Mita * 847df642ceaSAkinobu Mita * Returns: 848df642ceaSAkinobu Mita * true if @miter contains the valid mapping. false if end of sg 849df642ceaSAkinobu Mita * list is reached. 850df642ceaSAkinobu Mita */ 8510d6077f8SMing Lei bool sg_miter_skip(struct sg_mapping_iter *miter, off_t offset) 852df642ceaSAkinobu Mita { 853df642ceaSAkinobu Mita sg_miter_stop(miter); 854df642ceaSAkinobu Mita 855df642ceaSAkinobu Mita while (offset) { 856df642ceaSAkinobu Mita off_t consumed; 857df642ceaSAkinobu Mita 858df642ceaSAkinobu Mita if (!sg_miter_get_next_page(miter)) 859df642ceaSAkinobu Mita return false; 860df642ceaSAkinobu Mita 861df642ceaSAkinobu Mita consumed = min_t(off_t, offset, miter->__remaining); 862df642ceaSAkinobu Mita miter->__offset += consumed; 863df642ceaSAkinobu Mita miter->__remaining -= consumed; 864df642ceaSAkinobu Mita offset -= consumed; 865df642ceaSAkinobu Mita } 866df642ceaSAkinobu Mita 867df642ceaSAkinobu Mita return true; 868df642ceaSAkinobu Mita } 8690d6077f8SMing Lei EXPORT_SYMBOL(sg_miter_skip); 870df642ceaSAkinobu Mita 871df642ceaSAkinobu Mita /** 872137d3edbSTejun Heo * sg_miter_next - proceed mapping iterator to the next mapping 873137d3edbSTejun Heo * @miter: sg mapping iter to proceed 874137d3edbSTejun Heo * 875137d3edbSTejun Heo * Description: 8768290e2d2STejun Heo * Proceeds @miter to the next mapping. @miter should have been started 8778290e2d2STejun Heo * using sg_miter_start(). On successful return, @miter->page, 8788290e2d2STejun Heo * @miter->addr and @miter->length point to the current mapping. 879137d3edbSTejun Heo * 880137d3edbSTejun Heo * Context: 881723aca20SThomas Gleixner * May sleep if !SG_MITER_ATOMIC. 882137d3edbSTejun Heo * 883137d3edbSTejun Heo * Returns: 884137d3edbSTejun Heo * true if @miter contains the next mapping. false if end of sg 885137d3edbSTejun Heo * list is reached. 886137d3edbSTejun Heo */ 887137d3edbSTejun Heo bool sg_miter_next(struct sg_mapping_iter *miter) 888137d3edbSTejun Heo { 889137d3edbSTejun Heo sg_miter_stop(miter); 890137d3edbSTejun Heo 8914225fc85SImre Deak /* 8924225fc85SImre Deak * Get to the next page if necessary. 8934225fc85SImre Deak * __remaining, __offset is adjusted by sg_miter_stop 8944225fc85SImre Deak */ 89511052004SAkinobu Mita if (!sg_miter_get_next_page(miter)) 89623c560a9STejun Heo return false; 8974225fc85SImre Deak 8982db76d7cSImre Deak miter->page = sg_page_iter_page(&miter->piter); 8994225fc85SImre Deak miter->consumed = miter->length = miter->__remaining; 900137d3edbSTejun Heo 901137d3edbSTejun Heo if (miter->__flags & SG_MITER_ATOMIC) 9024225fc85SImre Deak miter->addr = kmap_atomic(miter->page) + miter->__offset; 903137d3edbSTejun Heo else 9044225fc85SImre Deak miter->addr = kmap(miter->page) + miter->__offset; 905137d3edbSTejun Heo 906137d3edbSTejun Heo return true; 907137d3edbSTejun Heo } 908137d3edbSTejun Heo EXPORT_SYMBOL(sg_miter_next); 909137d3edbSTejun Heo 910137d3edbSTejun Heo /** 911137d3edbSTejun Heo * sg_miter_stop - stop mapping iteration 912137d3edbSTejun Heo * @miter: sg mapping iter to be stopped 913137d3edbSTejun Heo * 914137d3edbSTejun Heo * Description: 915137d3edbSTejun Heo * Stops mapping iterator @miter. @miter should have been started 9164ba6a2b2SMasahiro Yamada * using sg_miter_start(). A stopped iteration can be resumed by 9174ba6a2b2SMasahiro Yamada * calling sg_miter_next() on it. This is useful when resources (kmap) 9184ba6a2b2SMasahiro Yamada * need to be released during iteration. 919137d3edbSTejun Heo * 920137d3edbSTejun Heo * Context: 921723aca20SThomas Gleixner * Don't care otherwise. 922137d3edbSTejun Heo */ 923137d3edbSTejun Heo void sg_miter_stop(struct sg_mapping_iter *miter) 924137d3edbSTejun Heo { 925137d3edbSTejun Heo WARN_ON(miter->consumed > miter->length); 926137d3edbSTejun Heo 927137d3edbSTejun Heo /* drop resources from the last iteration */ 928137d3edbSTejun Heo if (miter->addr) { 929137d3edbSTejun Heo miter->__offset += miter->consumed; 9304225fc85SImre Deak miter->__remaining -= miter->consumed; 931137d3edbSTejun Heo 9320e84f5dbSChristoph Hellwig if (miter->__flags & SG_MITER_TO_SG) 9330e84f5dbSChristoph Hellwig flush_dcache_page(miter->page); 9346de7e356SSebastian Andrzej Siewior 935137d3edbSTejun Heo if (miter->__flags & SG_MITER_ATOMIC) { 936723aca20SThomas Gleixner WARN_ON_ONCE(!pagefault_disabled()); 937c3eede8eSCong Wang kunmap_atomic(miter->addr); 938137d3edbSTejun Heo } else 939f652c521SArjan van de Ven kunmap(miter->page); 940137d3edbSTejun Heo 941137d3edbSTejun Heo miter->page = NULL; 942137d3edbSTejun Heo miter->addr = NULL; 943137d3edbSTejun Heo miter->length = 0; 944137d3edbSTejun Heo miter->consumed = 0; 945137d3edbSTejun Heo } 946137d3edbSTejun Heo } 947137d3edbSTejun Heo EXPORT_SYMBOL(sg_miter_stop); 948137d3edbSTejun Heo 949137d3edbSTejun Heo /** 950b1adaf65SFUJITA Tomonori * sg_copy_buffer - Copy data between a linear buffer and an SG list 951b1adaf65SFUJITA Tomonori * @sgl: The SG list 952b1adaf65SFUJITA Tomonori * @nents: Number of SG entries 953b1adaf65SFUJITA Tomonori * @buf: Where to copy from 954b1adaf65SFUJITA Tomonori * @buflen: The number of bytes to copy 955df642ceaSAkinobu Mita * @skip: Number of bytes to skip before copying 956df642ceaSAkinobu Mita * @to_buffer: transfer direction (true == from an sg list to a 9576e853185SGeert Uytterhoeven * buffer, false == from a buffer to an sg list) 958b1adaf65SFUJITA Tomonori * 959b1adaf65SFUJITA Tomonori * Returns the number of copied bytes. 960b1adaf65SFUJITA Tomonori * 961b1adaf65SFUJITA Tomonori **/ 962386ecb12SDave Gordon size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents, void *buf, 963386ecb12SDave Gordon size_t buflen, off_t skip, bool to_buffer) 964b1adaf65SFUJITA Tomonori { 965137d3edbSTejun Heo unsigned int offset = 0; 966137d3edbSTejun Heo struct sg_mapping_iter miter; 9676de7e356SSebastian Andrzej Siewior unsigned int sg_flags = SG_MITER_ATOMIC; 968b1adaf65SFUJITA Tomonori 9696de7e356SSebastian Andrzej Siewior if (to_buffer) 9706de7e356SSebastian Andrzej Siewior sg_flags |= SG_MITER_FROM_SG; 9716de7e356SSebastian Andrzej Siewior else 9726de7e356SSebastian Andrzej Siewior sg_flags |= SG_MITER_TO_SG; 9736de7e356SSebastian Andrzej Siewior 9746de7e356SSebastian Andrzej Siewior sg_miter_start(&miter, sgl, nents, sg_flags); 975b1adaf65SFUJITA Tomonori 976df642ceaSAkinobu Mita if (!sg_miter_skip(&miter, skip)) 9771f41be7dSDavid Disseldorp return 0; 978df642ceaSAkinobu Mita 9791d5210efSGilad Ben-Yossef while ((offset < buflen) && sg_miter_next(&miter)) { 980137d3edbSTejun Heo unsigned int len; 981b1adaf65SFUJITA Tomonori 982137d3edbSTejun Heo len = min(miter.length, buflen - offset); 983b1adaf65SFUJITA Tomonori 984b1adaf65SFUJITA Tomonori if (to_buffer) 985137d3edbSTejun Heo memcpy(buf + offset, miter.addr, len); 9866de7e356SSebastian Andrzej Siewior else 987137d3edbSTejun Heo memcpy(miter.addr, buf + offset, len); 988b1adaf65SFUJITA Tomonori 989137d3edbSTejun Heo offset += len; 990b1adaf65SFUJITA Tomonori } 991b1adaf65SFUJITA Tomonori 992137d3edbSTejun Heo sg_miter_stop(&miter); 993b1adaf65SFUJITA Tomonori 994137d3edbSTejun Heo return offset; 995b1adaf65SFUJITA Tomonori } 996386ecb12SDave Gordon EXPORT_SYMBOL(sg_copy_buffer); 997b1adaf65SFUJITA Tomonori 998b1adaf65SFUJITA Tomonori /** 999b1adaf65SFUJITA Tomonori * sg_copy_from_buffer - Copy from a linear buffer to an SG list 1000b1adaf65SFUJITA Tomonori * @sgl: The SG list 1001b1adaf65SFUJITA Tomonori * @nents: Number of SG entries 1002b1adaf65SFUJITA Tomonori * @buf: Where to copy from 1003b1adaf65SFUJITA Tomonori * @buflen: The number of bytes to copy 1004b1adaf65SFUJITA Tomonori * 1005b1adaf65SFUJITA Tomonori * Returns the number of copied bytes. 1006b1adaf65SFUJITA Tomonori * 1007b1adaf65SFUJITA Tomonori **/ 1008b1adaf65SFUJITA Tomonori size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents, 10092a1bf8f9SDave Gordon const void *buf, size_t buflen) 1010b1adaf65SFUJITA Tomonori { 10112a1bf8f9SDave Gordon return sg_copy_buffer(sgl, nents, (void *)buf, buflen, 0, false); 1012b1adaf65SFUJITA Tomonori } 1013b1adaf65SFUJITA Tomonori EXPORT_SYMBOL(sg_copy_from_buffer); 1014b1adaf65SFUJITA Tomonori 1015b1adaf65SFUJITA Tomonori /** 1016b1adaf65SFUJITA Tomonori * sg_copy_to_buffer - Copy from an SG list to a linear buffer 1017b1adaf65SFUJITA Tomonori * @sgl: The SG list 1018b1adaf65SFUJITA Tomonori * @nents: Number of SG entries 1019b1adaf65SFUJITA Tomonori * @buf: Where to copy to 1020b1adaf65SFUJITA Tomonori * @buflen: The number of bytes to copy 1021b1adaf65SFUJITA Tomonori * 1022b1adaf65SFUJITA Tomonori * Returns the number of copied bytes. 1023b1adaf65SFUJITA Tomonori * 1024b1adaf65SFUJITA Tomonori **/ 1025b1adaf65SFUJITA Tomonori size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents, 1026b1adaf65SFUJITA Tomonori void *buf, size_t buflen) 1027b1adaf65SFUJITA Tomonori { 1028df642ceaSAkinobu Mita return sg_copy_buffer(sgl, nents, buf, buflen, 0, true); 1029b1adaf65SFUJITA Tomonori } 1030b1adaf65SFUJITA Tomonori EXPORT_SYMBOL(sg_copy_to_buffer); 1031df642ceaSAkinobu Mita 1032df642ceaSAkinobu Mita /** 1033df642ceaSAkinobu Mita * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list 1034df642ceaSAkinobu Mita * @sgl: The SG list 1035df642ceaSAkinobu Mita * @nents: Number of SG entries 1036df642ceaSAkinobu Mita * @buf: Where to copy from 1037df642ceaSAkinobu Mita * @buflen: The number of bytes to copy 10384dc7daf8SDave Gordon * @skip: Number of bytes to skip before copying 1039df642ceaSAkinobu Mita * 1040df642ceaSAkinobu Mita * Returns the number of copied bytes. 1041df642ceaSAkinobu Mita * 1042df642ceaSAkinobu Mita **/ 1043df642ceaSAkinobu Mita size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents, 10442a1bf8f9SDave Gordon const void *buf, size_t buflen, off_t skip) 1045df642ceaSAkinobu Mita { 10462a1bf8f9SDave Gordon return sg_copy_buffer(sgl, nents, (void *)buf, buflen, skip, false); 1047df642ceaSAkinobu Mita } 1048df642ceaSAkinobu Mita EXPORT_SYMBOL(sg_pcopy_from_buffer); 1049df642ceaSAkinobu Mita 1050df642ceaSAkinobu Mita /** 1051df642ceaSAkinobu Mita * sg_pcopy_to_buffer - Copy from an SG list to a linear buffer 1052df642ceaSAkinobu Mita * @sgl: The SG list 1053df642ceaSAkinobu Mita * @nents: Number of SG entries 1054df642ceaSAkinobu Mita * @buf: Where to copy to 1055df642ceaSAkinobu Mita * @buflen: The number of bytes to copy 10564dc7daf8SDave Gordon * @skip: Number of bytes to skip before copying 1057df642ceaSAkinobu Mita * 1058df642ceaSAkinobu Mita * Returns the number of copied bytes. 1059df642ceaSAkinobu Mita * 1060df642ceaSAkinobu Mita **/ 1061df642ceaSAkinobu Mita size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents, 1062df642ceaSAkinobu Mita void *buf, size_t buflen, off_t skip) 1063df642ceaSAkinobu Mita { 1064df642ceaSAkinobu Mita return sg_copy_buffer(sgl, nents, buf, buflen, skip, true); 1065df642ceaSAkinobu Mita } 1066df642ceaSAkinobu Mita EXPORT_SYMBOL(sg_pcopy_to_buffer); 10670945e569SJohannes Thumshirn 10680945e569SJohannes Thumshirn /** 10690945e569SJohannes Thumshirn * sg_zero_buffer - Zero-out a part of a SG list 10700945e569SJohannes Thumshirn * @sgl: The SG list 10710945e569SJohannes Thumshirn * @nents: Number of SG entries 10720945e569SJohannes Thumshirn * @buflen: The number of bytes to zero out 10730945e569SJohannes Thumshirn * @skip: Number of bytes to skip before zeroing 10740945e569SJohannes Thumshirn * 10750945e569SJohannes Thumshirn * Returns the number of bytes zeroed. 10760945e569SJohannes Thumshirn **/ 10770945e569SJohannes Thumshirn size_t sg_zero_buffer(struct scatterlist *sgl, unsigned int nents, 10780945e569SJohannes Thumshirn size_t buflen, off_t skip) 10790945e569SJohannes Thumshirn { 10800945e569SJohannes Thumshirn unsigned int offset = 0; 10810945e569SJohannes Thumshirn struct sg_mapping_iter miter; 10820945e569SJohannes Thumshirn unsigned int sg_flags = SG_MITER_ATOMIC | SG_MITER_TO_SG; 10830945e569SJohannes Thumshirn 10840945e569SJohannes Thumshirn sg_miter_start(&miter, sgl, nents, sg_flags); 10850945e569SJohannes Thumshirn 10860945e569SJohannes Thumshirn if (!sg_miter_skip(&miter, skip)) 10870945e569SJohannes Thumshirn return false; 10880945e569SJohannes Thumshirn 10890945e569SJohannes Thumshirn while (offset < buflen && sg_miter_next(&miter)) { 10900945e569SJohannes Thumshirn unsigned int len; 10910945e569SJohannes Thumshirn 10920945e569SJohannes Thumshirn len = min(miter.length, buflen - offset); 10930945e569SJohannes Thumshirn memset(miter.addr, 0, len); 10940945e569SJohannes Thumshirn 10950945e569SJohannes Thumshirn offset += len; 10960945e569SJohannes Thumshirn } 10970945e569SJohannes Thumshirn 10980945e569SJohannes Thumshirn sg_miter_stop(&miter); 10990945e569SJohannes Thumshirn return offset; 11000945e569SJohannes Thumshirn } 11010945e569SJohannes Thumshirn EXPORT_SYMBOL(sg_zero_buffer); 1102f5f82cd1SDavid Howells 1103f5f82cd1SDavid Howells /* 1104f5f82cd1SDavid Howells * Extract and pin a list of up to sg_max pages from UBUF- or IOVEC-class 1105f5f82cd1SDavid Howells * iterators, and add them to the scatterlist. 1106f5f82cd1SDavid Howells */ 1107f5f82cd1SDavid Howells static ssize_t extract_user_to_sg(struct iov_iter *iter, 1108f5f82cd1SDavid Howells ssize_t maxsize, 1109f5f82cd1SDavid Howells struct sg_table *sgtable, 1110f5f82cd1SDavid Howells unsigned int sg_max, 1111f5f82cd1SDavid Howells iov_iter_extraction_t extraction_flags) 1112f5f82cd1SDavid Howells { 1113f5f82cd1SDavid Howells struct scatterlist *sg = sgtable->sgl + sgtable->nents; 1114f5f82cd1SDavid Howells struct page **pages; 1115f5f82cd1SDavid Howells unsigned int npages; 1116f5f82cd1SDavid Howells ssize_t ret = 0, res; 1117f5f82cd1SDavid Howells size_t len, off; 1118f5f82cd1SDavid Howells 1119f5f82cd1SDavid Howells /* We decant the page list into the tail of the scatterlist */ 1120f5f82cd1SDavid Howells pages = (void *)sgtable->sgl + 1121f5f82cd1SDavid Howells array_size(sg_max, sizeof(struct scatterlist)); 1122f5f82cd1SDavid Howells pages -= sg_max; 1123f5f82cd1SDavid Howells 1124f5f82cd1SDavid Howells do { 1125f5f82cd1SDavid Howells res = iov_iter_extract_pages(iter, &pages, maxsize, sg_max, 1126f5f82cd1SDavid Howells extraction_flags, &off); 1127f5f82cd1SDavid Howells if (res < 0) 1128f5f82cd1SDavid Howells goto failed; 1129f5f82cd1SDavid Howells 1130f5f82cd1SDavid Howells len = res; 1131f5f82cd1SDavid Howells maxsize -= len; 1132f5f82cd1SDavid Howells ret += len; 1133f5f82cd1SDavid Howells npages = DIV_ROUND_UP(off + len, PAGE_SIZE); 1134f5f82cd1SDavid Howells sg_max -= npages; 1135f5f82cd1SDavid Howells 1136f5f82cd1SDavid Howells for (; npages > 0; npages--) { 1137f5f82cd1SDavid Howells struct page *page = *pages; 1138f5f82cd1SDavid Howells size_t seg = min_t(size_t, PAGE_SIZE - off, len); 1139f5f82cd1SDavid Howells 1140f5f82cd1SDavid Howells *pages++ = NULL; 1141f5f82cd1SDavid Howells sg_set_page(sg, page, seg, off); 1142f5f82cd1SDavid Howells sgtable->nents++; 1143f5f82cd1SDavid Howells sg++; 1144f5f82cd1SDavid Howells len -= seg; 1145f5f82cd1SDavid Howells off = 0; 1146f5f82cd1SDavid Howells } 1147f5f82cd1SDavid Howells } while (maxsize > 0 && sg_max > 0); 1148f5f82cd1SDavid Howells 1149f5f82cd1SDavid Howells return ret; 1150f5f82cd1SDavid Howells 1151f5f82cd1SDavid Howells failed: 1152f5f82cd1SDavid Howells while (sgtable->nents > sgtable->orig_nents) 1153f443fd5aSDavid Howells unpin_user_page(sg_page(&sgtable->sgl[--sgtable->nents])); 1154f5f82cd1SDavid Howells return res; 1155f5f82cd1SDavid Howells } 1156f5f82cd1SDavid Howells 1157f5f82cd1SDavid Howells /* 1158f5f82cd1SDavid Howells * Extract up to sg_max pages from a BVEC-type iterator and add them to the 1159f5f82cd1SDavid Howells * scatterlist. The pages are not pinned. 1160f5f82cd1SDavid Howells */ 1161f5f82cd1SDavid Howells static ssize_t extract_bvec_to_sg(struct iov_iter *iter, 1162f5f82cd1SDavid Howells ssize_t maxsize, 1163f5f82cd1SDavid Howells struct sg_table *sgtable, 1164f5f82cd1SDavid Howells unsigned int sg_max, 1165f5f82cd1SDavid Howells iov_iter_extraction_t extraction_flags) 1166f5f82cd1SDavid Howells { 1167f5f82cd1SDavid Howells const struct bio_vec *bv = iter->bvec; 1168f5f82cd1SDavid Howells struct scatterlist *sg = sgtable->sgl + sgtable->nents; 1169f5f82cd1SDavid Howells unsigned long start = iter->iov_offset; 1170f5f82cd1SDavid Howells unsigned int i; 1171f5f82cd1SDavid Howells ssize_t ret = 0; 1172f5f82cd1SDavid Howells 1173f5f82cd1SDavid Howells for (i = 0; i < iter->nr_segs; i++) { 1174f5f82cd1SDavid Howells size_t off, len; 1175f5f82cd1SDavid Howells 1176f5f82cd1SDavid Howells len = bv[i].bv_len; 1177f5f82cd1SDavid Howells if (start >= len) { 1178f5f82cd1SDavid Howells start -= len; 1179f5f82cd1SDavid Howells continue; 1180f5f82cd1SDavid Howells } 1181f5f82cd1SDavid Howells 1182f5f82cd1SDavid Howells len = min_t(size_t, maxsize, len - start); 1183f5f82cd1SDavid Howells off = bv[i].bv_offset + start; 1184f5f82cd1SDavid Howells 1185f5f82cd1SDavid Howells sg_set_page(sg, bv[i].bv_page, len, off); 1186f5f82cd1SDavid Howells sgtable->nents++; 1187f5f82cd1SDavid Howells sg++; 1188f5f82cd1SDavid Howells sg_max--; 1189f5f82cd1SDavid Howells 1190f5f82cd1SDavid Howells ret += len; 1191f5f82cd1SDavid Howells maxsize -= len; 1192f5f82cd1SDavid Howells if (maxsize <= 0 || sg_max == 0) 1193f5f82cd1SDavid Howells break; 1194f5f82cd1SDavid Howells start = 0; 1195f5f82cd1SDavid Howells } 1196f5f82cd1SDavid Howells 1197f5f82cd1SDavid Howells if (ret > 0) 1198f5f82cd1SDavid Howells iov_iter_advance(iter, ret); 1199f5f82cd1SDavid Howells return ret; 1200f5f82cd1SDavid Howells } 1201f5f82cd1SDavid Howells 1202f5f82cd1SDavid Howells /* 1203f5f82cd1SDavid Howells * Extract up to sg_max pages from a KVEC-type iterator and add them to the 1204f5f82cd1SDavid Howells * scatterlist. This can deal with vmalloc'd buffers as well as kmalloc'd or 1205f5f82cd1SDavid Howells * static buffers. The pages are not pinned. 1206f5f82cd1SDavid Howells */ 1207f5f82cd1SDavid Howells static ssize_t extract_kvec_to_sg(struct iov_iter *iter, 1208f5f82cd1SDavid Howells ssize_t maxsize, 1209f5f82cd1SDavid Howells struct sg_table *sgtable, 1210f5f82cd1SDavid Howells unsigned int sg_max, 1211f5f82cd1SDavid Howells iov_iter_extraction_t extraction_flags) 1212f5f82cd1SDavid Howells { 1213f5f82cd1SDavid Howells const struct kvec *kv = iter->kvec; 1214f5f82cd1SDavid Howells struct scatterlist *sg = sgtable->sgl + sgtable->nents; 1215f5f82cd1SDavid Howells unsigned long start = iter->iov_offset; 1216f5f82cd1SDavid Howells unsigned int i; 1217f5f82cd1SDavid Howells ssize_t ret = 0; 1218f5f82cd1SDavid Howells 1219f5f82cd1SDavid Howells for (i = 0; i < iter->nr_segs; i++) { 1220f5f82cd1SDavid Howells struct page *page; 1221f5f82cd1SDavid Howells unsigned long kaddr; 1222f5f82cd1SDavid Howells size_t off, len, seg; 1223f5f82cd1SDavid Howells 1224f5f82cd1SDavid Howells len = kv[i].iov_len; 1225f5f82cd1SDavid Howells if (start >= len) { 1226f5f82cd1SDavid Howells start -= len; 1227f5f82cd1SDavid Howells continue; 1228f5f82cd1SDavid Howells } 1229f5f82cd1SDavid Howells 1230f5f82cd1SDavid Howells kaddr = (unsigned long)kv[i].iov_base + start; 1231f5f82cd1SDavid Howells off = kaddr & ~PAGE_MASK; 1232f5f82cd1SDavid Howells len = min_t(size_t, maxsize, len - start); 1233f5f82cd1SDavid Howells kaddr &= PAGE_MASK; 1234f5f82cd1SDavid Howells 1235f5f82cd1SDavid Howells maxsize -= len; 1236f5f82cd1SDavid Howells ret += len; 1237f5f82cd1SDavid Howells do { 1238f5f82cd1SDavid Howells seg = min_t(size_t, len, PAGE_SIZE - off); 1239f5f82cd1SDavid Howells if (is_vmalloc_or_module_addr((void *)kaddr)) 1240f5f82cd1SDavid Howells page = vmalloc_to_page((void *)kaddr); 1241f5f82cd1SDavid Howells else 12427b82e904SLinus Torvalds page = virt_to_page((void *)kaddr); 1243f5f82cd1SDavid Howells 1244f5f82cd1SDavid Howells sg_set_page(sg, page, len, off); 1245f5f82cd1SDavid Howells sgtable->nents++; 1246f5f82cd1SDavid Howells sg++; 1247f5f82cd1SDavid Howells sg_max--; 1248f5f82cd1SDavid Howells 1249f5f82cd1SDavid Howells len -= seg; 1250f5f82cd1SDavid Howells kaddr += PAGE_SIZE; 1251f5f82cd1SDavid Howells off = 0; 1252f5f82cd1SDavid Howells } while (len > 0 && sg_max > 0); 1253f5f82cd1SDavid Howells 1254f5f82cd1SDavid Howells if (maxsize <= 0 || sg_max == 0) 1255f5f82cd1SDavid Howells break; 1256f5f82cd1SDavid Howells start = 0; 1257f5f82cd1SDavid Howells } 1258f5f82cd1SDavid Howells 1259f5f82cd1SDavid Howells if (ret > 0) 1260f5f82cd1SDavid Howells iov_iter_advance(iter, ret); 1261f5f82cd1SDavid Howells return ret; 1262f5f82cd1SDavid Howells } 1263f5f82cd1SDavid Howells 1264f5f82cd1SDavid Howells /* 1265f5f82cd1SDavid Howells * Extract up to sg_max folios from an XARRAY-type iterator and add them to 1266f5f82cd1SDavid Howells * the scatterlist. The pages are not pinned. 1267f5f82cd1SDavid Howells */ 1268f5f82cd1SDavid Howells static ssize_t extract_xarray_to_sg(struct iov_iter *iter, 1269f5f82cd1SDavid Howells ssize_t maxsize, 1270f5f82cd1SDavid Howells struct sg_table *sgtable, 1271f5f82cd1SDavid Howells unsigned int sg_max, 1272f5f82cd1SDavid Howells iov_iter_extraction_t extraction_flags) 1273f5f82cd1SDavid Howells { 1274f5f82cd1SDavid Howells struct scatterlist *sg = sgtable->sgl + sgtable->nents; 1275f5f82cd1SDavid Howells struct xarray *xa = iter->xarray; 1276f5f82cd1SDavid Howells struct folio *folio; 1277f5f82cd1SDavid Howells loff_t start = iter->xarray_start + iter->iov_offset; 1278f5f82cd1SDavid Howells pgoff_t index = start / PAGE_SIZE; 1279f5f82cd1SDavid Howells ssize_t ret = 0; 1280f5f82cd1SDavid Howells size_t offset, len; 1281f5f82cd1SDavid Howells XA_STATE(xas, xa, index); 1282f5f82cd1SDavid Howells 1283f5f82cd1SDavid Howells rcu_read_lock(); 1284f5f82cd1SDavid Howells 1285f5f82cd1SDavid Howells xas_for_each(&xas, folio, ULONG_MAX) { 1286f5f82cd1SDavid Howells if (xas_retry(&xas, folio)) 1287f5f82cd1SDavid Howells continue; 1288f5f82cd1SDavid Howells if (WARN_ON(xa_is_value(folio))) 1289f5f82cd1SDavid Howells break; 1290f5f82cd1SDavid Howells if (WARN_ON(folio_test_hugetlb(folio))) 1291f5f82cd1SDavid Howells break; 1292f5f82cd1SDavid Howells 1293f5f82cd1SDavid Howells offset = offset_in_folio(folio, start); 1294f5f82cd1SDavid Howells len = min_t(size_t, maxsize, folio_size(folio) - offset); 1295f5f82cd1SDavid Howells 1296f5f82cd1SDavid Howells sg_set_page(sg, folio_page(folio, 0), len, offset); 1297f5f82cd1SDavid Howells sgtable->nents++; 1298f5f82cd1SDavid Howells sg++; 1299f5f82cd1SDavid Howells sg_max--; 1300f5f82cd1SDavid Howells 1301f5f82cd1SDavid Howells maxsize -= len; 1302f5f82cd1SDavid Howells ret += len; 1303f5f82cd1SDavid Howells if (maxsize <= 0 || sg_max == 0) 1304f5f82cd1SDavid Howells break; 1305f5f82cd1SDavid Howells } 1306f5f82cd1SDavid Howells 1307f5f82cd1SDavid Howells rcu_read_unlock(); 1308f5f82cd1SDavid Howells if (ret > 0) 1309f5f82cd1SDavid Howells iov_iter_advance(iter, ret); 1310f5f82cd1SDavid Howells return ret; 1311f5f82cd1SDavid Howells } 1312f5f82cd1SDavid Howells 1313f5f82cd1SDavid Howells /** 1314f5f82cd1SDavid Howells * extract_iter_to_sg - Extract pages from an iterator and add to an sglist 1315f5f82cd1SDavid Howells * @iter: The iterator to extract from 1316f5f82cd1SDavid Howells * @maxsize: The amount of iterator to copy 1317f5f82cd1SDavid Howells * @sgtable: The scatterlist table to fill in 1318f5f82cd1SDavid Howells * @sg_max: Maximum number of elements in @sgtable that may be filled 1319f5f82cd1SDavid Howells * @extraction_flags: Flags to qualify the request 1320f5f82cd1SDavid Howells * 1321f5f82cd1SDavid Howells * Extract the page fragments from the given amount of the source iterator and 1322f5f82cd1SDavid Howells * add them to a scatterlist that refers to all of those bits, to a maximum 1323f5f82cd1SDavid Howells * addition of @sg_max elements. 1324f5f82cd1SDavid Howells * 1325f5f82cd1SDavid Howells * The pages referred to by UBUF- and IOVEC-type iterators are extracted and 1326f5f82cd1SDavid Howells * pinned; BVEC-, KVEC- and XARRAY-type are extracted but aren't pinned; PIPE- 1327f5f82cd1SDavid Howells * and DISCARD-type are not supported. 1328f5f82cd1SDavid Howells * 1329f5f82cd1SDavid Howells * No end mark is placed on the scatterlist; that's left to the caller. 1330f5f82cd1SDavid Howells * 1331f5f82cd1SDavid Howells * @extraction_flags can have ITER_ALLOW_P2PDMA set to request peer-to-peer DMA 1332f5f82cd1SDavid Howells * be allowed on the pages extracted. 1333f5f82cd1SDavid Howells * 1334f5f82cd1SDavid Howells * If successful, @sgtable->nents is updated to include the number of elements 1335f5f82cd1SDavid Howells * added and the number of bytes added is returned. @sgtable->orig_nents is 1336f5f82cd1SDavid Howells * left unaltered. 1337f5f82cd1SDavid Howells * 1338f5f82cd1SDavid Howells * The iov_iter_extract_mode() function should be used to query how cleanup 1339f5f82cd1SDavid Howells * should be performed. 1340f5f82cd1SDavid Howells */ 1341f5f82cd1SDavid Howells ssize_t extract_iter_to_sg(struct iov_iter *iter, size_t maxsize, 1342f5f82cd1SDavid Howells struct sg_table *sgtable, unsigned int sg_max, 1343f5f82cd1SDavid Howells iov_iter_extraction_t extraction_flags) 1344f5f82cd1SDavid Howells { 1345f5f82cd1SDavid Howells if (maxsize == 0) 1346f5f82cd1SDavid Howells return 0; 1347f5f82cd1SDavid Howells 1348f5f82cd1SDavid Howells switch (iov_iter_type(iter)) { 1349f5f82cd1SDavid Howells case ITER_UBUF: 1350f5f82cd1SDavid Howells case ITER_IOVEC: 1351f5f82cd1SDavid Howells return extract_user_to_sg(iter, maxsize, sgtable, sg_max, 1352f5f82cd1SDavid Howells extraction_flags); 1353f5f82cd1SDavid Howells case ITER_BVEC: 1354f5f82cd1SDavid Howells return extract_bvec_to_sg(iter, maxsize, sgtable, sg_max, 1355f5f82cd1SDavid Howells extraction_flags); 1356f5f82cd1SDavid Howells case ITER_KVEC: 1357f5f82cd1SDavid Howells return extract_kvec_to_sg(iter, maxsize, sgtable, sg_max, 1358f5f82cd1SDavid Howells extraction_flags); 1359f5f82cd1SDavid Howells case ITER_XARRAY: 1360f5f82cd1SDavid Howells return extract_xarray_to_sg(iter, maxsize, sgtable, sg_max, 1361f5f82cd1SDavid Howells extraction_flags); 1362f5f82cd1SDavid Howells default: 1363f5f82cd1SDavid Howells pr_err("%s(%u) unsupported\n", __func__, iov_iter_type(iter)); 1364f5f82cd1SDavid Howells WARN_ON_ONCE(1); 1365f5f82cd1SDavid Howells return -EIO; 1366f5f82cd1SDavid Howells } 1367f5f82cd1SDavid Howells } 1368f5f82cd1SDavid Howells EXPORT_SYMBOL_GPL(extract_iter_to_sg); 1369