1184299Snwhitehorn/****************************************************************************** 2184299Snwhitehorn * gntdev.c 3184299Snwhitehorn * 4184299Snwhitehorn * Device for accessing (in user-space) pages that have been granted by other 5184299Snwhitehorn * domains. 6184299Snwhitehorn * 7184299Snwhitehorn * Copyright (c) 2006-2007, D G Murray. 8184299Snwhitehorn * (c) 2009 Gerd Hoffmann <kraxel@redhat.com> 9184299Snwhitehorn * (c) 2018 Oleksandr Andrushchenko, EPAM Systems Inc. 10184299Snwhitehorn * 11184299Snwhitehorn * This program is distributed in the hope that it will be useful, 12184299Snwhitehorn * but WITHOUT ANY WARRANTY; without even the implied warranty of 13184299Snwhitehorn * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 14184299Snwhitehorn * GNU General Public License for more details. 15184299Snwhitehorn * 16184299Snwhitehorn * You should have received a copy of the GNU General Public License 17184299Snwhitehorn * along with this program; if not, write to the Free Software 18184299Snwhitehorn * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 19184299Snwhitehorn */ 20184299Snwhitehorn 21184299Snwhitehorn#undef DEBUG 22184299Snwhitehorn 23184299Snwhitehorn#define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt 24184299Snwhitehorn 25184299Snwhitehorn#include <linux/dma-mapping.h> 26184299Snwhitehorn#include <linux/module.h> 27184299Snwhitehorn#include <linux/kernel.h> 28184299Snwhitehorn#include <linux/init.h> 29184299Snwhitehorn#include <linux/miscdevice.h> 30184299Snwhitehorn#include <linux/fs.h> 31184299Snwhitehorn#include <linux/uaccess.h> 32184299Snwhitehorn#include <linux/sched.h> 33184299Snwhitehorn#include <linux/sched/mm.h> 34184299Snwhitehorn#include <linux/spinlock.h> 35184299Snwhitehorn#include <linux/slab.h> 36184299Snwhitehorn#include <linux/highmem.h> 37184299Snwhitehorn#include <linux/refcount.h> 38205506Snwhitehorn#include <linux/workqueue.h> 39212054Snwhitehorn 40185727Snwhitehorn#include <xen/xen.h> 41184299Snwhitehorn#include <xen/grant_table.h> 42184299Snwhitehorn#include <xen/balloon.h> 43184299Snwhitehorn#include <xen/gntdev.h> 44185782Snwhitehorn#include <xen/events.h> 45184299Snwhitehorn#include <xen/page.h> 46184299Snwhitehorn#include <asm/xen/hypervisor.h> 47184299Snwhitehorn#include <asm/xen/hypercall.h> 48184299Snwhitehorn 49184299Snwhitehorn#include "gntdev-common.h" 50184299Snwhitehorn#ifdef CONFIG_XEN_GNTDEV_DMABUF 51184299Snwhitehorn#include "gntdev-dmabuf.h" 52184299Snwhitehorn#endif 53184299Snwhitehorn 54184299SnwhitehornMODULE_LICENSE("GPL"); 55184299SnwhitehornMODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, " 56184299Snwhitehorn "Gerd Hoffmann <kraxel@redhat.com>"); 57184299SnwhitehornMODULE_DESCRIPTION("User-space granted page access driver"); 58184299Snwhitehorn 59205506Snwhitehornstatic unsigned int limit = 64*1024; 60184299Snwhitehornmodule_param(limit, uint, 0644); 61184299SnwhitehornMODULE_PARM_DESC(limit, 62184299Snwhitehorn "Maximum number of grants that may be mapped by one mapping request"); 63184299Snwhitehorn 64205506Snwhitehorn/* True in PV mode, false otherwise */ 65184299Snwhitehornstatic int use_ptemod; 66184299Snwhitehorn 67184299Snwhitehornstatic void unmap_grant_pages(struct gntdev_grant_map *map, 68184299Snwhitehorn int offset, int pages); 69184299Snwhitehorn 70205506Snwhitehornstatic struct miscdevice gntdev_miscdev; 71205506Snwhitehorn 72205506Snwhitehorn/* ------------------------------------------------------------------ */ 73205506Snwhitehorn 74205506Snwhitehornbool gntdev_test_page_count(unsigned int count) 75205506Snwhitehorn{ 76205506Snwhitehorn return !count || count > limit; 77205506Snwhitehorn} 78205506Snwhitehorn 79205506Snwhitehornstatic void gntdev_print_maps(struct gntdev_priv *priv, 80185727Snwhitehorn char *text, int text_index) 81185754Snwhitehorn{ 82185727Snwhitehorn#ifdef DEBUG 83194027Savg struct gntdev_grant_map *map; 84185754Snwhitehorn 85212054Snwhitehorn pr_debug("%s: maps list (priv %p)\n", __func__, priv); 86212054Snwhitehorn list_for_each_entry(map, &priv->maps, next) 87212054Snwhitehorn pr_debug(" index %2d, count %2d %s\n", 88212054Snwhitehorn map->index, map->count, 89212054Snwhitehorn map->index == text_index && text ? text : ""); 90185782Snwhitehorn#endif 91185727Snwhitehorn} 92193159Snwhitehorn 93185754Snwhitehornstatic void gntdev_free_map(struct gntdev_grant_map *map) 94185754Snwhitehorn{ 95185754Snwhitehorn if (map == NULL) 96184299Snwhitehorn return; 97185754Snwhitehorn 98185754Snwhitehorn#ifdef CONFIG_XEN_GRANT_DMA_ALLOC 99185754Snwhitehorn if (map->dma_vaddr) { 100185754Snwhitehorn struct gnttab_dma_alloc_args args; 101185754Snwhitehorn 102185754Snwhitehorn args.dev = map->dma_dev; 103185754Snwhitehorn args.coherent = !!(map->dma_flags & GNTDEV_DMA_FLAG_COHERENT); 104185754Snwhitehorn args.nr_pages = map->count; 105185754Snwhitehorn args.pages = map->pages; 106185754Snwhitehorn args.frames = map->frames; 107185754Snwhitehorn args.vaddr = map->dma_vaddr; 108185754Snwhitehorn args.dev_bus_addr = map->dma_bus_addr; 109185754Snwhitehorn 110185754Snwhitehorn gnttab_dma_free_pages(&args); 111185754Snwhitehorn } else 112184299Snwhitehorn#endif 113184299Snwhitehorn if (map->pages) 114184299Snwhitehorn gnttab_free_pages(map->count, map->pages); 115184299Snwhitehorn 116184299Snwhitehorn#ifdef CONFIG_XEN_GRANT_DMA_ALLOC 117184299Snwhitehorn kvfree(map->frames); 118184299Snwhitehorn#endif 119184299Snwhitehorn kvfree(map->pages); 120184299Snwhitehorn kvfree(map->grants); 121184299Snwhitehorn kvfree(map->map_ops); 122184299Snwhitehorn kvfree(map->unmap_ops); 123184299Snwhitehorn kvfree(map->kmap_ops); 124184299Snwhitehorn kvfree(map->kunmap_ops); 125184299Snwhitehorn kvfree(map->being_removed); 126184299Snwhitehorn kfree(map); 127184299Snwhitehorn} 128184299Snwhitehorn 129184299Snwhitehornstruct gntdev_grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count, 130205506Snwhitehorn int dma_flags) 131205506Snwhitehorn{ 132205506Snwhitehorn struct gntdev_grant_map *add; 133205506Snwhitehorn int i; 134184299Snwhitehorn 135184299Snwhitehorn add = kzalloc(sizeof(*add), GFP_KERNEL); 136184299Snwhitehorn if (NULL == add) 137184299Snwhitehorn return NULL; 138184299Snwhitehorn 139184299Snwhitehorn add->grants = kvmalloc_array(count, sizeof(add->grants[0]), 140184299Snwhitehorn GFP_KERNEL); 141184299Snwhitehorn add->map_ops = kvmalloc_array(count, sizeof(add->map_ops[0]), 142184299Snwhitehorn GFP_KERNEL); 143184299Snwhitehorn add->unmap_ops = kvmalloc_array(count, sizeof(add->unmap_ops[0]), 144184299Snwhitehorn GFP_KERNEL); 145184299Snwhitehorn add->pages = kvcalloc(count, sizeof(add->pages[0]), GFP_KERNEL); 146184299Snwhitehorn add->being_removed = 147184299Snwhitehorn kvcalloc(count, sizeof(add->being_removed[0]), GFP_KERNEL); 148184299Snwhitehorn if (NULL == add->grants || 149184299Snwhitehorn NULL == add->map_ops || 150184299Snwhitehorn NULL == add->unmap_ops || 151184299Snwhitehorn NULL == add->pages || 152184299Snwhitehorn NULL == add->being_removed) 153184299Snwhitehorn goto err; 154184299Snwhitehorn if (use_ptemod) { 155184299Snwhitehorn add->kmap_ops = kvmalloc_array(count, sizeof(add->kmap_ops[0]), 156184299Snwhitehorn GFP_KERNEL); 157184299Snwhitehorn add->kunmap_ops = kvmalloc_array(count, sizeof(add->kunmap_ops[0]), 158184299Snwhitehorn GFP_KERNEL); 159184299Snwhitehorn if (NULL == add->kmap_ops || NULL == add->kunmap_ops) 160184299Snwhitehorn goto err; 161184299Snwhitehorn } 162184299Snwhitehorn 163184299Snwhitehorn#ifdef CONFIG_XEN_GRANT_DMA_ALLOC 164184299Snwhitehorn add->dma_flags = dma_flags; 165184299Snwhitehorn 166184299Snwhitehorn /* 167184299Snwhitehorn * Check if this mapping is requested to be backed 168184299Snwhitehorn * by a DMA buffer. 169184299Snwhitehorn */ 170184299Snwhitehorn if (dma_flags & (GNTDEV_DMA_FLAG_WC | GNTDEV_DMA_FLAG_COHERENT)) { 171184299Snwhitehorn struct gnttab_dma_alloc_args args; 172184299Snwhitehorn 173184299Snwhitehorn add->frames = kvcalloc(count, sizeof(add->frames[0]), 174184299Snwhitehorn GFP_KERNEL); 175184299Snwhitehorn if (!add->frames) 176184299Snwhitehorn goto err; 177184299Snwhitehorn 178184299Snwhitehorn /* Remember the device, so we can free DMA memory. */ 179184299Snwhitehorn add->dma_dev = priv->dma_dev; 180184299Snwhitehorn 181184299Snwhitehorn args.dev = priv->dma_dev; 182184299Snwhitehorn args.coherent = !!(dma_flags & GNTDEV_DMA_FLAG_COHERENT); 183184299Snwhitehorn args.nr_pages = count; 184184299Snwhitehorn args.pages = add->pages; 185184299Snwhitehorn args.frames = add->frames; 186184299Snwhitehorn 187184299Snwhitehorn if (gnttab_dma_alloc_pages(&args)) 188184299Snwhitehorn goto err; 189184299Snwhitehorn 190184299Snwhitehorn add->dma_vaddr = args.vaddr; 191184299Snwhitehorn add->dma_bus_addr = args.dev_bus_addr; 192184299Snwhitehorn } else 193184299Snwhitehorn#endif 194184299Snwhitehorn if (gnttab_alloc_pages(count, add->pages)) 195184299Snwhitehorn goto err; 196184299Snwhitehorn 197184299Snwhitehorn for (i = 0; i < count; i++) { 198184299Snwhitehorn add->grants[i].domid = DOMID_INVALID; 199184299Snwhitehorn add->grants[i].ref = INVALID_GRANT_REF; 200184299Snwhitehorn add->map_ops[i].handle = INVALID_GRANT_HANDLE; 201184299Snwhitehorn add->unmap_ops[i].handle = INVALID_GRANT_HANDLE; 202184299Snwhitehorn if (use_ptemod) { 203184299Snwhitehorn add->kmap_ops[i].handle = INVALID_GRANT_HANDLE; 204184299Snwhitehorn add->kunmap_ops[i].handle = INVALID_GRANT_HANDLE; 205184299Snwhitehorn } 206184299Snwhitehorn } 207184299Snwhitehorn 208184299Snwhitehorn add->index = 0; 209184299Snwhitehorn add->count = count; 210184299Snwhitehorn refcount_set(&add->users, 1); 211184299Snwhitehorn 212184299Snwhitehorn return add; 213184299Snwhitehorn 214184299Snwhitehornerr: 215184299Snwhitehorn gntdev_free_map(add); 216184299Snwhitehorn return NULL; 217184299Snwhitehorn} 218184299Snwhitehorn 219184299Snwhitehornvoid gntdev_add_map(struct gntdev_priv *priv, struct gntdev_grant_map *add) 220184299Snwhitehorn{ 221184299Snwhitehorn struct gntdev_grant_map *map; 222184299Snwhitehorn 223184299Snwhitehorn list_for_each_entry(map, &priv->maps, next) { 224184299Snwhitehorn if (add->index + add->count < map->index) { 225184299Snwhitehorn list_add_tail(&add->next, &map->next); 226184299Snwhitehorn goto done; 227184299Snwhitehorn } 228184299Snwhitehorn add->index = map->index + map->count; 229184299Snwhitehorn } 230184299Snwhitehorn list_add_tail(&add->next, &priv->maps); 231184299Snwhitehorn 232184299Snwhitehorndone: 233184299Snwhitehorn gntdev_print_maps(priv, "[new]", add->index); 234184299Snwhitehorn} 235184299Snwhitehorn 236184299Snwhitehornstatic struct gntdev_grant_map *gntdev_find_map_index(struct gntdev_priv *priv, 237184299Snwhitehorn int index, int count) 238184299Snwhitehorn{ 239184299Snwhitehorn struct gntdev_grant_map *map; 240184299Snwhitehorn 241184299Snwhitehorn list_for_each_entry(map, &priv->maps, next) { 242184299Snwhitehorn if (map->index != index) 243184299Snwhitehorn continue; 244184299Snwhitehorn if (count && map->count != count) 245184299Snwhitehorn continue; 246184299Snwhitehorn return map; 247184299Snwhitehorn } 248184299Snwhitehorn return NULL; 249184299Snwhitehorn} 250184299Snwhitehorn 251184299Snwhitehornvoid gntdev_put_map(struct gntdev_priv *priv, struct gntdev_grant_map *map) 252184299Snwhitehorn{ 253184299Snwhitehorn if (!map) 254184299Snwhitehorn return; 255184299Snwhitehorn 256184299Snwhitehorn if (!refcount_dec_and_test(&map->users)) 257184299Snwhitehorn return; 258184299Snwhitehorn 259184299Snwhitehorn if (map->pages && !use_ptemod) { 260184299Snwhitehorn /* 261184299Snwhitehorn * Increment the reference count. This ensures that the 262184299Snwhitehorn * subsequent call to unmap_grant_pages() will not wind up 263184299Snwhitehorn * re-entering itself. It *can* wind up calling 264184299Snwhitehorn * gntdev_put_map() recursively, but such calls will be with a 265184299Snwhitehorn * reference count greater than 1, so they will return before 266184299Snwhitehorn * this code is reached. The recursion depth is thus limited to 267184299Snwhitehorn * 1. Do NOT use refcount_inc() here, as it will detect that 268184299Snwhitehorn * the reference count is zero and WARN(). 269184299Snwhitehorn */ 270184299Snwhitehorn refcount_set(&map->users, 1); 271184299Snwhitehorn 272184299Snwhitehorn /* 273184299Snwhitehorn * Unmap the grants. This may or may not be asynchronous, so it 274184299Snwhitehorn * is possible that the reference count is 1 on return, but it 275184299Snwhitehorn * could also be greater than 1. 276184299Snwhitehorn */ 277184299Snwhitehorn unmap_grant_pages(map, 0, map->count); 278184299Snwhitehorn 279184299Snwhitehorn /* Check if the memory now needs to be freed */ 280184299Snwhitehorn if (!refcount_dec_and_test(&map->users)) 281184299Snwhitehorn return; 282184299Snwhitehorn 283184299Snwhitehorn /* 284184299Snwhitehorn * All pages have been returned to the hypervisor, so free the 285184299Snwhitehorn * map. 286184299Snwhitehorn */ 287184299Snwhitehorn } 288184299Snwhitehorn 289184299Snwhitehorn if (use_ptemod && map->notifier_init) 290184299Snwhitehorn mmu_interval_notifier_remove(&map->notifier); 291184299Snwhitehorn 292184299Snwhitehorn if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) { 293184299Snwhitehorn notify_remote_via_evtchn(map->notify.event); 294184299Snwhitehorn evtchn_put(map->notify.event); 295184299Snwhitehorn } 296184299Snwhitehorn gntdev_free_map(map); 297184299Snwhitehorn} 298184299Snwhitehorn 299184299Snwhitehorn/* ------------------------------------------------------------------ */ 300184299Snwhitehorn 301184299Snwhitehornstatic int find_grant_ptes(pte_t *pte, unsigned long addr, void *data) 302184299Snwhitehorn{ 303184299Snwhitehorn struct gntdev_grant_map *map = data; 304184299Snwhitehorn unsigned int pgnr = (addr - map->pages_vm_start) >> PAGE_SHIFT; 305184299Snwhitehorn int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte | 306184299Snwhitehorn (1 << _GNTMAP_guest_avail0); 307184299Snwhitehorn u64 pte_maddr; 308184299Snwhitehorn 309184299Snwhitehorn BUG_ON(pgnr >= map->count); 310184299Snwhitehorn pte_maddr = arbitrary_virt_to_machine(pte).maddr; 311184299Snwhitehorn 312184299Snwhitehorn gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags, 313184299Snwhitehorn map->grants[pgnr].ref, 314184299Snwhitehorn map->grants[pgnr].domid); 315184299Snwhitehorn gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags, 316184299Snwhitehorn INVALID_GRANT_HANDLE); 317184299Snwhitehorn return 0; 318184299Snwhitehorn} 319184299Snwhitehorn 320184299Snwhitehornint gntdev_map_grant_pages(struct gntdev_grant_map *map) 321184299Snwhitehorn{ 322184299Snwhitehorn size_t alloced = 0; 323184299Snwhitehorn int i, err = 0; 324185754Snwhitehorn 325184299Snwhitehorn if (!use_ptemod) { 326184299Snwhitehorn /* Note: it could already be mapped */ 327184299Snwhitehorn if (map->map_ops[0].handle != INVALID_GRANT_HANDLE) 328184299Snwhitehorn return 0; 329185727Snwhitehorn for (i = 0; i < map->count; i++) { 330185727Snwhitehorn unsigned long addr = (unsigned long) 331184299Snwhitehorn pfn_to_kaddr(page_to_pfn(map->pages[i])); 332184299Snwhitehorn gnttab_set_map_op(&map->map_ops[i], addr, map->flags, 333184299Snwhitehorn map->grants[i].ref, 334184299Snwhitehorn map->grants[i].domid); 335184299Snwhitehorn gnttab_set_unmap_op(&map->unmap_ops[i], addr, 336184299Snwhitehorn map->flags, INVALID_GRANT_HANDLE); 337184299Snwhitehorn } 338184299Snwhitehorn } else { 339184299Snwhitehorn /* 340184299Snwhitehorn * Setup the map_ops corresponding to the pte entries pointing 341184299Snwhitehorn * to the kernel linear addresses of the struct pages. 342184299Snwhitehorn * These ptes are completely different from the user ptes dealt 343184299Snwhitehorn * with find_grant_ptes. 344184299Snwhitehorn * Note that GNTMAP_device_map isn't needed here: The 345184299Snwhitehorn * dev_bus_addr output field gets consumed only from ->map_ops, 346184299Snwhitehorn * and by not requesting it when mapping we also avoid needing 347184299Snwhitehorn * to mirror dev_bus_addr into ->unmap_ops (and holding an extra 348184299Snwhitehorn * reference to the page in the hypervisor). 349184299Snwhitehorn */ 350184299Snwhitehorn unsigned int flags = (map->flags & ~GNTMAP_device_map) | 351184299Snwhitehorn GNTMAP_host_map; 352184299Snwhitehorn 353184299Snwhitehorn for (i = 0; i < map->count; i++) { 354184299Snwhitehorn unsigned long address = (unsigned long) 355184299Snwhitehorn pfn_to_kaddr(page_to_pfn(map->pages[i])); 356184299Snwhitehorn BUG_ON(PageHighMem(map->pages[i])); 357184299Snwhitehorn 358185782Snwhitehorn gnttab_set_map_op(&map->kmap_ops[i], address, flags, 359185782Snwhitehorn map->grants[i].ref, 360185782Snwhitehorn map->grants[i].domid); 361184299Snwhitehorn gnttab_set_unmap_op(&map->kunmap_ops[i], address, 362184299Snwhitehorn flags, INVALID_GRANT_HANDLE); 363184299Snwhitehorn } 364184299Snwhitehorn } 365184299Snwhitehorn 366184299Snwhitehorn pr_debug("map %d+%d\n", map->index, map->count); 367184299Snwhitehorn err = gnttab_map_refs(map->map_ops, map->kmap_ops, map->pages, 368184299Snwhitehorn map->count); 369184299Snwhitehorn 370184299Snwhitehorn for (i = 0; i < map->count; i++) { 371184299Snwhitehorn if (map->map_ops[i].status == GNTST_okay) { 372184299Snwhitehorn map->unmap_ops[i].handle = map->map_ops[i].handle; 373184299Snwhitehorn alloced++; 374184299Snwhitehorn } else if (!err) 375184299Snwhitehorn err = -EINVAL; 376184299Snwhitehorn 377184299Snwhitehorn if (map->flags & GNTMAP_device_map) 378184299Snwhitehorn map->unmap_ops[i].dev_bus_addr = map->map_ops[i].dev_bus_addr; 379184299Snwhitehorn 380184299Snwhitehorn if (use_ptemod) { 381184299Snwhitehorn if (map->kmap_ops[i].status == GNTST_okay) { 382184299Snwhitehorn alloced++; 383184299Snwhitehorn map->kunmap_ops[i].handle = map->kmap_ops[i].handle; 384184299Snwhitehorn } else if (!err) 385184299Snwhitehorn err = -EINVAL; 386184299Snwhitehorn } 387184299Snwhitehorn } 388184299Snwhitehorn atomic_add(alloced, &map->live_grants); 389185754Snwhitehorn return err; 390185754Snwhitehorn} 391185754Snwhitehorn 392185754Snwhitehornstatic void __unmap_grant_pages_done(int result, 393185754Snwhitehorn struct gntab_unmap_queue_data *data) 394185754Snwhitehorn{ 395185754Snwhitehorn unsigned int i; 396185754Snwhitehorn struct gntdev_grant_map *map = data->data; 397185754Snwhitehorn unsigned int offset = data->unmap_ops - map->unmap_ops; 398185754Snwhitehorn int successful_unmaps = 0; 399185754Snwhitehorn int live_grants; 400185754Snwhitehorn 401184299Snwhitehorn for (i = 0; i < data->count; i++) { 402184299Snwhitehorn if (map->unmap_ops[offset + i].status == GNTST_okay && 403185727Snwhitehorn map->unmap_ops[offset + i].handle != INVALID_GRANT_HANDLE) 404185727Snwhitehorn successful_unmaps++; 405185727Snwhitehorn 406185727Snwhitehorn WARN_ON(map->unmap_ops[offset + i].status != GNTST_okay && 407185727Snwhitehorn map->unmap_ops[offset + i].handle != INVALID_GRANT_HANDLE); 408185727Snwhitehorn pr_debug("unmap handle=%d st=%d\n", 409185727Snwhitehorn map->unmap_ops[offset+i].handle, 410185727Snwhitehorn map->unmap_ops[offset+i].status); 411185754Snwhitehorn map->unmap_ops[offset+i].handle = INVALID_GRANT_HANDLE; 412185754Snwhitehorn if (use_ptemod) { 413185727Snwhitehorn if (map->kunmap_ops[offset + i].status == GNTST_okay && 414185754Snwhitehorn map->kunmap_ops[offset + i].handle != INVALID_GRANT_HANDLE) 415185754Snwhitehorn successful_unmaps++; 416185754Snwhitehorn 417185754Snwhitehorn WARN_ON(map->kunmap_ops[offset + i].status != GNTST_okay && 418193159Snwhitehorn map->kunmap_ops[offset + i].handle != INVALID_GRANT_HANDLE); 419193159Snwhitehorn pr_debug("kunmap handle=%u st=%d\n", 420193159Snwhitehorn map->kunmap_ops[offset+i].handle, 421193159Snwhitehorn map->kunmap_ops[offset+i].status); 422185754Snwhitehorn map->kunmap_ops[offset+i].handle = INVALID_GRANT_HANDLE; 423185754Snwhitehorn } 424185754Snwhitehorn } 425185754Snwhitehorn 426185754Snwhitehorn /* 427185754Snwhitehorn * Decrease the live-grant counter. This must happen after the loop to 428185754Snwhitehorn * prevent premature reuse of the grants by gnttab_mmap(). 429185754Snwhitehorn */ 430185754Snwhitehorn live_grants = atomic_sub_return(successful_unmaps, &map->live_grants); 431185754Snwhitehorn if (WARN_ON(live_grants < 0)) 432185754Snwhitehorn pr_err("%s: live_grants became negative (%d) after unmapping %d pages!\n", 433185754Snwhitehorn __func__, live_grants, successful_unmaps); 434185754Snwhitehorn 435185754Snwhitehorn /* Release reference taken by __unmap_grant_pages */ 436185754Snwhitehorn gntdev_put_map(NULL, map); 437185754Snwhitehorn} 438185754Snwhitehorn 439185754Snwhitehornstatic void __unmap_grant_pages(struct gntdev_grant_map *map, int offset, 440185754Snwhitehorn int pages) 441185754Snwhitehorn{ 442185754Snwhitehorn if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) { 443185754Snwhitehorn int pgno = (map->notify.addr >> PAGE_SHIFT); 444185754Snwhitehorn 445185754Snwhitehorn if (pgno >= offset && pgno < offset + pages) { 446185754Snwhitehorn /* No need for kmap, pages are in lowmem */ 447185754Snwhitehorn uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno])); 448185754Snwhitehorn 449185754Snwhitehorn tmp[map->notify.addr & (PAGE_SIZE-1)] = 0; 450185754Snwhitehorn map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE; 451185754Snwhitehorn } 452185754Snwhitehorn } 453185754Snwhitehorn 454185754Snwhitehorn map->unmap_data.unmap_ops = map->unmap_ops + offset; 455185754Snwhitehorn map->unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL; 456185754Snwhitehorn map->unmap_data.pages = map->pages + offset; 457185754Snwhitehorn map->unmap_data.count = pages; 458185754Snwhitehorn map->unmap_data.done = __unmap_grant_pages_done; 459185754Snwhitehorn map->unmap_data.data = map; 460185754Snwhitehorn refcount_inc(&map->users); /* to keep map alive during async call below */ 461185754Snwhitehorn 462185754Snwhitehorn gnttab_unmap_refs_async(&map->unmap_data); 463185754Snwhitehorn} 464185754Snwhitehorn 465185754Snwhitehornstatic void unmap_grant_pages(struct gntdev_grant_map *map, int offset, 466185754Snwhitehorn int pages) 467185754Snwhitehorn{ 468185754Snwhitehorn int range; 469185754Snwhitehorn 470185754Snwhitehorn if (atomic_read(&map->live_grants) == 0) 471185782Snwhitehorn return; /* Nothing to do */ 472185782Snwhitehorn 473185782Snwhitehorn pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages); 474185782Snwhitehorn 475185782Snwhitehorn /* It is possible the requested range will have a "hole" where we 476185782Snwhitehorn * already unmapped some of the grants. Only unmap valid ranges. 477205506Snwhitehorn */ 478205506Snwhitehorn while (pages) { 479205506Snwhitehorn while (pages && map->being_removed[offset]) { 480205506Snwhitehorn offset++; 481205506Snwhitehorn pages--; 482205506Snwhitehorn } 483212054Snwhitehorn range = 0; 484212054Snwhitehorn while (range < pages) { 485212054Snwhitehorn if (map->being_removed[offset + range]) 486212054Snwhitehorn break; 487212054Snwhitehorn map->being_removed[offset + range] = true; 488212054Snwhitehorn range++; 489184299Snwhitehorn } 490184299Snwhitehorn if (range) 491184299Snwhitehorn __unmap_grant_pages(map, offset, range); 492184299Snwhitehorn offset += range; 493184299Snwhitehorn pages -= range; 494184299Snwhitehorn } 495184299Snwhitehorn} 496184299Snwhitehorn 497184299Snwhitehorn/* ------------------------------------------------------------------ */ 498184299Snwhitehorn 499185782Snwhitehornstatic void gntdev_vma_open(struct vm_area_struct *vma) 500185782Snwhitehorn{ 501185782Snwhitehorn struct gntdev_grant_map *map = vma->vm_private_data; 502184299Snwhitehorn 503184299Snwhitehorn pr_debug("gntdev_vma_open %p\n", vma); 504184299Snwhitehorn refcount_inc(&map->users); 505184299Snwhitehorn} 506184299Snwhitehorn 507184299Snwhitehornstatic void gntdev_vma_close(struct vm_area_struct *vma) 508184299Snwhitehorn{ 509184299Snwhitehorn struct gntdev_grant_map *map = vma->vm_private_data; 510184299Snwhitehorn struct file *file = vma->vm_file; 511184299Snwhitehorn struct gntdev_priv *priv = file->private_data; 512184299Snwhitehorn 513184299Snwhitehorn pr_debug("gntdev_vma_close %p\n", vma); 514184299Snwhitehorn 515184299Snwhitehorn vma->vm_private_data = NULL; 516184299Snwhitehorn gntdev_put_map(priv, map); 517184299Snwhitehorn} 518184299Snwhitehorn 519184299Snwhitehornstatic struct page *gntdev_vma_find_special_page(struct vm_area_struct *vma, 520184299Snwhitehorn unsigned long addr) 521184299Snwhitehorn{ 522184299Snwhitehorn struct gntdev_grant_map *map = vma->vm_private_data; 523184299Snwhitehorn 524184299Snwhitehorn return map->pages[(addr - map->pages_vm_start) >> PAGE_SHIFT]; 525184299Snwhitehorn} 526184299Snwhitehorn 527184299Snwhitehornstatic const struct vm_operations_struct gntdev_vmops = { 528184299Snwhitehorn .open = gntdev_vma_open, 529184299Snwhitehorn .close = gntdev_vma_close, 530184299Snwhitehorn .find_special_page = gntdev_vma_find_special_page, 531184299Snwhitehorn}; 532184299Snwhitehorn 533184299Snwhitehorn/* ------------------------------------------------------------------ */ 534184299Snwhitehorn 535184299Snwhitehornstatic bool gntdev_invalidate(struct mmu_interval_notifier *mn, 536184299Snwhitehorn const struct mmu_notifier_range *range, 537184299Snwhitehorn unsigned long cur_seq) 538184299Snwhitehorn{ 539184299Snwhitehorn struct gntdev_grant_map *map = 540184299Snwhitehorn container_of(mn, struct gntdev_grant_map, notifier); 541184299Snwhitehorn unsigned long mstart, mend; 542184299Snwhitehorn unsigned long map_start, map_end; 543184299Snwhitehorn 544184299Snwhitehorn if (!mmu_notifier_range_blockable(range)) 545184299Snwhitehorn return false; 546184299Snwhitehorn 547184299Snwhitehorn map_start = map->pages_vm_start; 548184299Snwhitehorn map_end = map->pages_vm_start + (map->count << PAGE_SHIFT); 549184299Snwhitehorn 550184299Snwhitehorn /* 551184299Snwhitehorn * If the VMA is split or otherwise changed the notifier is not 552184299Snwhitehorn * updated, but we don't want to process VA's outside the modified 553184299Snwhitehorn * VMA. FIXME: It would be much more understandable to just prevent 554184299Snwhitehorn * modifying the VMA in the first place. 555184299Snwhitehorn */ 556184299Snwhitehorn if (map_start >= range->end || map_end <= range->start) 557184299Snwhitehorn return true; 558184299Snwhitehorn 559184299Snwhitehorn mstart = max(range->start, map_start); 560184299Snwhitehorn mend = min(range->end, map_end); 561184299Snwhitehorn pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n", 562184299Snwhitehorn map->index, map->count, map_start, map_end, 563184299Snwhitehorn range->start, range->end, mstart, mend); 564184299Snwhitehorn unmap_grant_pages(map, (mstart - map_start) >> PAGE_SHIFT, 565184299Snwhitehorn (mend - mstart) >> PAGE_SHIFT); 566184299Snwhitehorn 567184299Snwhitehorn return true; 568184299Snwhitehorn} 569184299Snwhitehorn 570184299Snwhitehornstatic const struct mmu_interval_notifier_ops gntdev_mmu_ops = { 571184299Snwhitehorn .invalidate = gntdev_invalidate, 572184299Snwhitehorn}; 573184299Snwhitehorn 574184299Snwhitehorn/* ------------------------------------------------------------------ */ 575184299Snwhitehorn 576184299Snwhitehornstatic int gntdev_open(struct inode *inode, struct file *flip) 577184299Snwhitehorn{ 578184299Snwhitehorn struct gntdev_priv *priv; 579184299Snwhitehorn 580184299Snwhitehorn priv = kzalloc(sizeof(*priv), GFP_KERNEL); 581184299Snwhitehorn if (!priv) 582184299Snwhitehorn return -ENOMEM; 583184299Snwhitehorn 584184299Snwhitehorn INIT_LIST_HEAD(&priv->maps); 585184299Snwhitehorn mutex_init(&priv->lock); 586184299Snwhitehorn 587184299Snwhitehorn#ifdef CONFIG_XEN_GNTDEV_DMABUF 588184299Snwhitehorn priv->dmabuf_priv = gntdev_dmabuf_init(flip); 589184299Snwhitehorn if (IS_ERR(priv->dmabuf_priv)) { 590184299Snwhitehorn int ret = PTR_ERR(priv->dmabuf_priv); 591184299Snwhitehorn 592184299Snwhitehorn kfree(priv); 593184299Snwhitehorn return ret; 594184299Snwhitehorn } 595184299Snwhitehorn#endif 596184299Snwhitehorn 597184299Snwhitehorn flip->private_data = priv; 598184299Snwhitehorn#ifdef CONFIG_XEN_GRANT_DMA_ALLOC 599184299Snwhitehorn priv->dma_dev = gntdev_miscdev.this_device; 600184299Snwhitehorn dma_coerce_mask_and_coherent(priv->dma_dev, DMA_BIT_MASK(64)); 601184299Snwhitehorn#endif 602184299Snwhitehorn pr_debug("priv %p\n", priv); 603184299Snwhitehorn 604184299Snwhitehorn return 0; 605184299Snwhitehorn} 606184299Snwhitehorn 607184299Snwhitehornstatic int gntdev_release(struct inode *inode, struct file *flip) 608184299Snwhitehorn{ 609184299Snwhitehorn struct gntdev_priv *priv = flip->private_data; 610184299Snwhitehorn struct gntdev_grant_map *map; 611194027Savg 612184299Snwhitehorn pr_debug("priv %p\n", priv); 613184299Snwhitehorn 614184299Snwhitehorn mutex_lock(&priv->lock); 615194027Savg while (!list_empty(&priv->maps)) { 616184299Snwhitehorn map = list_entry(priv->maps.next, 617184299Snwhitehorn struct gntdev_grant_map, next); 618184299Snwhitehorn list_del(&map->next); 619184299Snwhitehorn gntdev_put_map(NULL /* already removed */, map); 620184299Snwhitehorn } 621184299Snwhitehorn mutex_unlock(&priv->lock); 622184299Snwhitehorn 623184299Snwhitehorn#ifdef CONFIG_XEN_GNTDEV_DMABUF 624184299Snwhitehorn gntdev_dmabuf_fini(priv->dmabuf_priv); 625184299Snwhitehorn#endif 626184299Snwhitehorn 627184299Snwhitehorn kfree(priv); 628184299Snwhitehorn return 0; 629184299Snwhitehorn} 630184299Snwhitehorn 631184299Snwhitehornstatic long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv, 632184299Snwhitehorn struct ioctl_gntdev_map_grant_ref __user *u) 633184299Snwhitehorn{ 634184299Snwhitehorn struct ioctl_gntdev_map_grant_ref op; 635184299Snwhitehorn struct gntdev_grant_map *map; 636184299Snwhitehorn int err; 637184299Snwhitehorn 638184299Snwhitehorn if (copy_from_user(&op, u, sizeof(op)) != 0) 639184299Snwhitehorn return -EFAULT; 640184299Snwhitehorn pr_debug("priv %p, add %d\n", priv, op.count); 641184299Snwhitehorn if (unlikely(gntdev_test_page_count(op.count))) 642184299Snwhitehorn return -EINVAL; 643184299Snwhitehorn 644184299Snwhitehorn err = -ENOMEM; 645184299Snwhitehorn map = gntdev_alloc_map(priv, op.count, 0 /* This is not a dma-buf. */); 646184299Snwhitehorn if (!map) 647184299Snwhitehorn return err; 648184299Snwhitehorn 649184299Snwhitehorn if (copy_from_user(map->grants, &u->refs, 650184299Snwhitehorn sizeof(map->grants[0]) * op.count) != 0) { 651184299Snwhitehorn gntdev_put_map(NULL, map); 652184299Snwhitehorn return -EFAULT; 653184299Snwhitehorn } 654184299Snwhitehorn 655184299Snwhitehorn mutex_lock(&priv->lock); 656184299Snwhitehorn gntdev_add_map(priv, map); 657184299Snwhitehorn op.index = map->index << PAGE_SHIFT; 658184299Snwhitehorn mutex_unlock(&priv->lock); 659184299Snwhitehorn 660184299Snwhitehorn if (copy_to_user(u, &op, sizeof(op)) != 0) 661184299Snwhitehorn return -EFAULT; 662184299Snwhitehorn 663184299Snwhitehorn return 0; 664184299Snwhitehorn} 665184299Snwhitehorn 666184299Snwhitehornstatic long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv, 667184299Snwhitehorn struct ioctl_gntdev_unmap_grant_ref __user *u) 668184299Snwhitehorn{ 669184299Snwhitehorn struct ioctl_gntdev_unmap_grant_ref op; 670184299Snwhitehorn struct gntdev_grant_map *map; 671184299Snwhitehorn int err = -ENOENT; 672184299Snwhitehorn 673184299Snwhitehorn if (copy_from_user(&op, u, sizeof(op)) != 0) 674184299Snwhitehorn return -EFAULT; 675184299Snwhitehorn pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count); 676184299Snwhitehorn 677184299Snwhitehorn mutex_lock(&priv->lock); 678184299Snwhitehorn map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count); 679184299Snwhitehorn if (map) { 680184299Snwhitehorn list_del(&map->next); 681184299Snwhitehorn err = 0; 682184299Snwhitehorn } 683184299Snwhitehorn mutex_unlock(&priv->lock); 684184299Snwhitehorn if (map) 685184299Snwhitehorn gntdev_put_map(priv, map); 686184299Snwhitehorn return err; 687184299Snwhitehorn} 688184299Snwhitehorn 689184299Snwhitehornstatic long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv, 690184299Snwhitehorn struct ioctl_gntdev_get_offset_for_vaddr __user *u) 691184299Snwhitehorn{ 692184299Snwhitehorn struct ioctl_gntdev_get_offset_for_vaddr op; 693184299Snwhitehorn struct vm_area_struct *vma; 694184299Snwhitehorn struct gntdev_grant_map *map; 695184299Snwhitehorn int rv = -EINVAL; 696184299Snwhitehorn 697184299Snwhitehorn if (copy_from_user(&op, u, sizeof(op)) != 0) 698184299Snwhitehorn return -EFAULT; 699184299Snwhitehorn pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr); 700184299Snwhitehorn 701184299Snwhitehorn mmap_read_lock(current->mm); 702184299Snwhitehorn vma = find_vma(current->mm, op.vaddr); 703184299Snwhitehorn if (!vma || vma->vm_ops != &gntdev_vmops) 704184299Snwhitehorn goto out_unlock; 705184299Snwhitehorn 706184299Snwhitehorn map = vma->vm_private_data; 707184299Snwhitehorn if (!map) 708184299Snwhitehorn goto out_unlock; 709184299Snwhitehorn 710184299Snwhitehorn op.offset = map->index << PAGE_SHIFT; 711184299Snwhitehorn op.count = map->count; 712184299Snwhitehorn rv = 0; 713184299Snwhitehorn 714184299Snwhitehorn out_unlock: 715184299Snwhitehorn mmap_read_unlock(current->mm); 716184299Snwhitehorn 717184299Snwhitehorn if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0) 718184299Snwhitehorn return -EFAULT; 719184299Snwhitehorn return rv; 720184299Snwhitehorn} 721184299Snwhitehorn 722184299Snwhitehornstatic long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u) 723184299Snwhitehorn{ 724184299Snwhitehorn struct ioctl_gntdev_unmap_notify op; 725184299Snwhitehorn struct gntdev_grant_map *map; 726184299Snwhitehorn int rc; 727184299Snwhitehorn int out_flags; 728184299Snwhitehorn evtchn_port_t out_event; 729184299Snwhitehorn 730184299Snwhitehorn if (copy_from_user(&op, u, sizeof(op))) 731184299Snwhitehorn return -EFAULT; 732184299Snwhitehorn 733184299Snwhitehorn if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT)) 734184299Snwhitehorn return -EINVAL; 735184299Snwhitehorn 736184299Snwhitehorn /* We need to grab a reference to the event channel we are going to use 737184299Snwhitehorn * to send the notify before releasing the reference we may already have 738184299Snwhitehorn * (if someone has called this ioctl twice). This is required so that 739184299Snwhitehorn * it is possible to change the clear_byte part of the notification 740184299Snwhitehorn * without disturbing the event channel part, which may now be the last 741184299Snwhitehorn * reference to that event channel. 742184299Snwhitehorn */ 743184299Snwhitehorn if (op.action & UNMAP_NOTIFY_SEND_EVENT) { 744184299Snwhitehorn if (evtchn_get(op.event_channel_port)) 745184299Snwhitehorn return -EINVAL; 746184299Snwhitehorn } 747184299Snwhitehorn 748184299Snwhitehorn out_flags = op.action; 749184299Snwhitehorn out_event = op.event_channel_port; 750184299Snwhitehorn 751184299Snwhitehorn mutex_lock(&priv->lock); 752184299Snwhitehorn 753184299Snwhitehorn list_for_each_entry(map, &priv->maps, next) { 754184299Snwhitehorn uint64_t begin = map->index << PAGE_SHIFT; 755184299Snwhitehorn uint64_t end = (map->index + map->count) << PAGE_SHIFT; 756184299Snwhitehorn if (op.index >= begin && op.index < end) 757184299Snwhitehorn goto found; 758184299Snwhitehorn } 759184299Snwhitehorn rc = -ENOENT; 760184299Snwhitehorn goto unlock_out; 761184299Snwhitehorn 762184299Snwhitehorn found: 763184299Snwhitehorn if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) && 764185727Snwhitehorn (map->flags & GNTMAP_readonly)) { 765185782Snwhitehorn rc = -EINVAL; 766212054Snwhitehorn goto unlock_out; 767212054Snwhitehorn } 768212054Snwhitehorn 769212054Snwhitehorn out_flags = map->notify.flags; 770212054Snwhitehorn out_event = map->notify.event; 771212054Snwhitehorn 772212054Snwhitehorn map->notify.flags = op.action; 773212054Snwhitehorn map->notify.addr = op.index - (map->index << PAGE_SHIFT); 774212054Snwhitehorn map->notify.event = op.event_channel_port; 775212054Snwhitehorn 776212054Snwhitehorn rc = 0; 777212054Snwhitehorn 778212054Snwhitehorn unlock_out: 779212054Snwhitehorn mutex_unlock(&priv->lock); 780185782Snwhitehorn 781185782Snwhitehorn /* Drop the reference to the event channel we did not save in the map */ 782185782Snwhitehorn if (out_flags & UNMAP_NOTIFY_SEND_EVENT) 783185782Snwhitehorn evtchn_put(out_event); 784185782Snwhitehorn 785185782Snwhitehorn return rc; 786185782Snwhitehorn} 787185782Snwhitehorn 788185782Snwhitehorn#define GNTDEV_COPY_BATCH 16 789185782Snwhitehorn 790185782Snwhitehornstruct gntdev_copy_batch { 791185782Snwhitehorn struct gnttab_copy ops[GNTDEV_COPY_BATCH]; 792185727Snwhitehorn struct page *pages[GNTDEV_COPY_BATCH]; 793185727Snwhitehorn s16 __user *status[GNTDEV_COPY_BATCH]; 794185727Snwhitehorn unsigned int nr_ops; 795185727Snwhitehorn unsigned int nr_pages; 796185727Snwhitehorn bool writeable; 797185727Snwhitehorn}; 798185727Snwhitehorn 799185727Snwhitehornstatic int gntdev_get_page(struct gntdev_copy_batch *batch, void __user *virt, 800185727Snwhitehorn unsigned long *gfn) 801185727Snwhitehorn{ 802185727Snwhitehorn unsigned long addr = (unsigned long)virt; 803185727Snwhitehorn struct page *page; 804185727Snwhitehorn unsigned long xen_pfn; 805185727Snwhitehorn int ret; 806185727Snwhitehorn 807185727Snwhitehorn ret = pin_user_pages_fast(addr, 1, batch->writeable ? FOLL_WRITE : 0, &page); 808185727Snwhitehorn if (ret < 0) 809185727Snwhitehorn return ret; 810185727Snwhitehorn 811185727Snwhitehorn batch->pages[batch->nr_pages++] = page; 812185727Snwhitehorn 813185727Snwhitehorn xen_pfn = page_to_xen_pfn(page) + XEN_PFN_DOWN(addr & ~PAGE_MASK); 814185727Snwhitehorn *gfn = pfn_to_gfn(xen_pfn); 815185727Snwhitehorn 816185727Snwhitehorn return 0; 817185727Snwhitehorn} 818185727Snwhitehorn 819185727Snwhitehornstatic void gntdev_put_pages(struct gntdev_copy_batch *batch) 820185727Snwhitehorn{ 821185727Snwhitehorn unpin_user_pages_dirty_lock(batch->pages, batch->nr_pages, batch->writeable); 822185727Snwhitehorn batch->nr_pages = 0; 823185727Snwhitehorn batch->writeable = false; 824185727Snwhitehorn} 825185727Snwhitehorn 826185727Snwhitehornstatic int gntdev_copy(struct gntdev_copy_batch *batch) 827185727Snwhitehorn{ 828185727Snwhitehorn unsigned int i; 829185727Snwhitehorn 830185727Snwhitehorn gnttab_batch_copy(batch->ops, batch->nr_ops); 831185727Snwhitehorn gntdev_put_pages(batch); 832185727Snwhitehorn 833185727Snwhitehorn /* 834185754Snwhitehorn * For each completed op, update the status if the op failed 835185754Snwhitehorn * and all previous ops for the segment were successful. 836185754Snwhitehorn */ 837185754Snwhitehorn for (i = 0; i < batch->nr_ops; i++) { 838185754Snwhitehorn s16 status = batch->ops[i].status; 839185754Snwhitehorn s16 old_status; 840185754Snwhitehorn 841185754Snwhitehorn if (status == GNTST_okay) 842185754Snwhitehorn continue; 843185754Snwhitehorn 844185754Snwhitehorn if (__get_user(old_status, batch->status[i])) 845185754Snwhitehorn return -EFAULT; 846185754Snwhitehorn 847185754Snwhitehorn if (old_status != GNTST_okay) 848185754Snwhitehorn continue; 849185754Snwhitehorn 850185754Snwhitehorn if (__put_user(status, batch->status[i])) 851185754Snwhitehorn return -EFAULT; 852185754Snwhitehorn } 853185754Snwhitehorn 854185754Snwhitehorn batch->nr_ops = 0; 855185754Snwhitehorn return 0; 856185754Snwhitehorn} 857185754Snwhitehorn 858185754Snwhitehornstatic int gntdev_grant_copy_seg(struct gntdev_copy_batch *batch, 859185754Snwhitehorn struct gntdev_grant_copy_segment *seg, 860185754Snwhitehorn s16 __user *status) 861185754Snwhitehorn{ 862185754Snwhitehorn uint16_t copied = 0; 863185754Snwhitehorn 864185754Snwhitehorn /* 865185754Snwhitehorn * Disallow local -> local copies since there is only space in 866185754Snwhitehorn * batch->pages for one page per-op and this would be a very 867185754Snwhitehorn * expensive memcpy(). 868185754Snwhitehorn */ 869185754Snwhitehorn if (!(seg->flags & (GNTCOPY_source_gref | GNTCOPY_dest_gref))) 870185754Snwhitehorn return -EINVAL; 871185754Snwhitehorn 872185754Snwhitehorn /* Can't cross page if source/dest is a grant ref. */ 873185754Snwhitehorn if (seg->flags & GNTCOPY_source_gref) { 874185754Snwhitehorn if (seg->source.foreign.offset + seg->len > XEN_PAGE_SIZE) 875185754Snwhitehorn return -EINVAL; 876185754Snwhitehorn } 877185754Snwhitehorn if (seg->flags & GNTCOPY_dest_gref) { 878185754Snwhitehorn if (seg->dest.foreign.offset + seg->len > XEN_PAGE_SIZE) 879185754Snwhitehorn return -EINVAL; 880185754Snwhitehorn } 881185754Snwhitehorn 882185754Snwhitehorn if (put_user(GNTST_okay, status)) 883185754Snwhitehorn return -EFAULT; 884185754Snwhitehorn 885185754Snwhitehorn while (copied < seg->len) { 886185754Snwhitehorn struct gnttab_copy *op; 887185754Snwhitehorn void __user *virt; 888185754Snwhitehorn size_t len, off; 889185754Snwhitehorn unsigned long gfn; 890185754Snwhitehorn int ret; 891185754Snwhitehorn 892185754Snwhitehorn if (batch->nr_ops >= GNTDEV_COPY_BATCH) { 893185754Snwhitehorn ret = gntdev_copy(batch); 894185754Snwhitehorn if (ret < 0) 895185754Snwhitehorn return ret; 896185754Snwhitehorn } 897185754Snwhitehorn 898185754Snwhitehorn len = seg->len - copied; 899193159Snwhitehorn 900193159Snwhitehorn op = &batch->ops[batch->nr_ops]; 901193159Snwhitehorn op->flags = 0; 902193159Snwhitehorn 903193159Snwhitehorn if (seg->flags & GNTCOPY_source_gref) { 904193159Snwhitehorn op->source.u.ref = seg->source.foreign.ref; 905193159Snwhitehorn op->source.domid = seg->source.foreign.domid; 906193159Snwhitehorn op->source.offset = seg->source.foreign.offset + copied; 907193159Snwhitehorn op->flags |= GNTCOPY_source_gref; 908193159Snwhitehorn } else { 909193159Snwhitehorn virt = seg->source.virt + copied; 910193159Snwhitehorn off = (unsigned long)virt & ~XEN_PAGE_MASK; 911193159Snwhitehorn len = min(len, (size_t)XEN_PAGE_SIZE - off); 912193159Snwhitehorn batch->writeable = false; 913193159Snwhitehorn 914193159Snwhitehorn ret = gntdev_get_page(batch, virt, &gfn); 915193159Snwhitehorn if (ret < 0) 916193159Snwhitehorn return ret; 917193159Snwhitehorn 918193159Snwhitehorn op->source.u.gmfn = gfn; 919193159Snwhitehorn op->source.domid = DOMID_SELF; 920185754Snwhitehorn op->source.offset = off; 921185754Snwhitehorn } 922185754Snwhitehorn 923185754Snwhitehorn if (seg->flags & GNTCOPY_dest_gref) { 924185754Snwhitehorn op->dest.u.ref = seg->dest.foreign.ref; 925185754Snwhitehorn op->dest.domid = seg->dest.foreign.domid; 926185754Snwhitehorn op->dest.offset = seg->dest.foreign.offset + copied; 927185754Snwhitehorn op->flags |= GNTCOPY_dest_gref; 928185754Snwhitehorn } else { 929185754Snwhitehorn virt = seg->dest.virt + copied; 930185754Snwhitehorn off = (unsigned long)virt & ~XEN_PAGE_MASK; 931185754Snwhitehorn len = min(len, (size_t)XEN_PAGE_SIZE - off); 932185754Snwhitehorn batch->writeable = true; 933185754Snwhitehorn 934185754Snwhitehorn ret = gntdev_get_page(batch, virt, &gfn); 935185754Snwhitehorn if (ret < 0) 936185754Snwhitehorn return ret; 937185754Snwhitehorn 938185754Snwhitehorn op->dest.u.gmfn = gfn; 939185754Snwhitehorn op->dest.domid = DOMID_SELF; 940185754Snwhitehorn op->dest.offset = off; 941185754Snwhitehorn } 942185754Snwhitehorn 943185754Snwhitehorn op->len = len; 944185754Snwhitehorn copied += len; 945185754Snwhitehorn 946185754Snwhitehorn batch->status[batch->nr_ops] = status; 947185754Snwhitehorn batch->nr_ops++; 948185754Snwhitehorn } 949185754Snwhitehorn 950185754Snwhitehorn return 0; 951185754Snwhitehorn} 952185754Snwhitehorn 953185754Snwhitehornstatic long gntdev_ioctl_grant_copy(struct gntdev_priv *priv, void __user *u) 954185754Snwhitehorn{ 955185754Snwhitehorn struct ioctl_gntdev_grant_copy copy; 956185754Snwhitehorn struct gntdev_copy_batch batch; 957185754Snwhitehorn unsigned int i; 958185754Snwhitehorn int ret = 0; 959185754Snwhitehorn 960185754Snwhitehorn if (copy_from_user(©, u, sizeof(copy))) 961185754Snwhitehorn return -EFAULT; 962185754Snwhitehorn 963185754Snwhitehorn batch.nr_ops = 0; 964185754Snwhitehorn batch.nr_pages = 0; 965185754Snwhitehorn 966185754Snwhitehorn for (i = 0; i < copy.count; i++) { 967185754Snwhitehorn struct gntdev_grant_copy_segment seg; 968185754Snwhitehorn 969185754Snwhitehorn if (copy_from_user(&seg, ©.segments[i], sizeof(seg))) { 970185754Snwhitehorn ret = -EFAULT; 971185754Snwhitehorn goto out; 972185754Snwhitehorn } 973185754Snwhitehorn 974185754Snwhitehorn ret = gntdev_grant_copy_seg(&batch, &seg, ©.segments[i].status); 975185754Snwhitehorn if (ret < 0) 976205506Snwhitehorn goto out; 977205506Snwhitehorn 978205506Snwhitehorn cond_resched(); 979205506Snwhitehorn } 980205506Snwhitehorn if (batch.nr_ops) 981205506Snwhitehorn ret = gntdev_copy(&batch); 982205506Snwhitehorn return ret; 983205506Snwhitehorn 984205506Snwhitehorn out: 985205506Snwhitehorn gntdev_put_pages(&batch); 986205506Snwhitehorn return ret; 987205506Snwhitehorn} 988205506Snwhitehorn 989205506Snwhitehornstatic long gntdev_ioctl(struct file *flip, 990205506Snwhitehorn unsigned int cmd, unsigned long arg) 991205506Snwhitehorn{ 992205506Snwhitehorn struct gntdev_priv *priv = flip->private_data; 993205506Snwhitehorn void __user *ptr = (void __user *)arg; 994205506Snwhitehorn 995205506Snwhitehorn switch (cmd) { 996205506Snwhitehorn case IOCTL_GNTDEV_MAP_GRANT_REF: 997205506Snwhitehorn return gntdev_ioctl_map_grant_ref(priv, ptr); 998205506Snwhitehorn 999205506Snwhitehorn case IOCTL_GNTDEV_UNMAP_GRANT_REF: 1000205506Snwhitehorn return gntdev_ioctl_unmap_grant_ref(priv, ptr); 1001205506Snwhitehorn 1002205506Snwhitehorn case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR: 1003205506Snwhitehorn return gntdev_ioctl_get_offset_for_vaddr(priv, ptr); 1004205506Snwhitehorn 1005205506Snwhitehorn case IOCTL_GNTDEV_SET_UNMAP_NOTIFY: 1006205506Snwhitehorn return gntdev_ioctl_notify(priv, ptr); 1007205506Snwhitehorn 1008205506Snwhitehorn case IOCTL_GNTDEV_GRANT_COPY: 1009205506Snwhitehorn return gntdev_ioctl_grant_copy(priv, ptr); 1010205506Snwhitehorn 1011#ifdef CONFIG_XEN_GNTDEV_DMABUF 1012 case IOCTL_GNTDEV_DMABUF_EXP_FROM_REFS: 1013 return gntdev_ioctl_dmabuf_exp_from_refs(priv, use_ptemod, ptr); 1014 1015 case IOCTL_GNTDEV_DMABUF_EXP_WAIT_RELEASED: 1016 return gntdev_ioctl_dmabuf_exp_wait_released(priv, ptr); 1017 1018 case IOCTL_GNTDEV_DMABUF_IMP_TO_REFS: 1019 return gntdev_ioctl_dmabuf_imp_to_refs(priv, ptr); 1020 1021 case IOCTL_GNTDEV_DMABUF_IMP_RELEASE: 1022 return gntdev_ioctl_dmabuf_imp_release(priv, ptr); 1023#endif 1024 1025 default: 1026 pr_debug("priv %p, unknown cmd %x\n", priv, cmd); 1027 return -ENOIOCTLCMD; 1028 } 1029 1030 return 0; 1031} 1032 1033static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) 1034{ 1035 struct gntdev_priv *priv = flip->private_data; 1036 int index = vma->vm_pgoff; 1037 int count = vma_pages(vma); 1038 struct gntdev_grant_map *map; 1039 int err = -EINVAL; 1040 1041 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) 1042 return -EINVAL; 1043 1044 pr_debug("map %d+%d at %lx (pgoff %lx)\n", 1045 index, count, vma->vm_start, vma->vm_pgoff); 1046 1047 mutex_lock(&priv->lock); 1048 map = gntdev_find_map_index(priv, index, count); 1049 if (!map) 1050 goto unlock_out; 1051 if (!atomic_add_unless(&map->in_use, 1, 1)) 1052 goto unlock_out; 1053 1054 refcount_inc(&map->users); 1055 1056 vma->vm_ops = &gntdev_vmops; 1057 1058 vm_flags_set(vma, VM_DONTEXPAND | VM_DONTDUMP | VM_MIXEDMAP); 1059 1060 if (use_ptemod) 1061 vm_flags_set(vma, VM_DONTCOPY); 1062 1063 vma->vm_private_data = map; 1064 if (map->flags) { 1065 if ((vma->vm_flags & VM_WRITE) && 1066 (map->flags & GNTMAP_readonly)) 1067 goto out_unlock_put; 1068 } else { 1069 map->flags = GNTMAP_host_map; 1070 if (!(vma->vm_flags & VM_WRITE)) 1071 map->flags |= GNTMAP_readonly; 1072 } 1073 1074 map->pages_vm_start = vma->vm_start; 1075 1076 if (use_ptemod) { 1077 err = mmu_interval_notifier_insert_locked( 1078 &map->notifier, vma->vm_mm, vma->vm_start, 1079 vma->vm_end - vma->vm_start, &gntdev_mmu_ops); 1080 if (err) 1081 goto out_unlock_put; 1082 1083 map->notifier_init = true; 1084 } 1085 mutex_unlock(&priv->lock); 1086 1087 if (use_ptemod) { 1088 /* 1089 * gntdev takes the address of the PTE in find_grant_ptes() and 1090 * passes it to the hypervisor in gntdev_map_grant_pages(). The 1091 * purpose of the notifier is to prevent the hypervisor pointer 1092 * to the PTE from going stale. 1093 * 1094 * Since this vma's mappings can't be touched without the 1095 * mmap_lock, and we are holding it now, there is no need for 1096 * the notifier_range locking pattern. 1097 */ 1098 mmu_interval_read_begin(&map->notifier); 1099 1100 err = apply_to_page_range(vma->vm_mm, vma->vm_start, 1101 vma->vm_end - vma->vm_start, 1102 find_grant_ptes, map); 1103 if (err) { 1104 pr_warn("find_grant_ptes() failure.\n"); 1105 goto out_put_map; 1106 } 1107 } 1108 1109 err = gntdev_map_grant_pages(map); 1110 if (err) 1111 goto out_put_map; 1112 1113 if (!use_ptemod) { 1114 err = vm_map_pages_zero(vma, map->pages, map->count); 1115 if (err) 1116 goto out_put_map; 1117 } 1118 1119 return 0; 1120 1121unlock_out: 1122 mutex_unlock(&priv->lock); 1123 return err; 1124 1125out_unlock_put: 1126 mutex_unlock(&priv->lock); 1127out_put_map: 1128 if (use_ptemod) 1129 unmap_grant_pages(map, 0, map->count); 1130 gntdev_put_map(priv, map); 1131 return err; 1132} 1133 1134static const struct file_operations gntdev_fops = { 1135 .owner = THIS_MODULE, 1136 .open = gntdev_open, 1137 .release = gntdev_release, 1138 .mmap = gntdev_mmap, 1139 .unlocked_ioctl = gntdev_ioctl 1140}; 1141 1142static struct miscdevice gntdev_miscdev = { 1143 .minor = MISC_DYNAMIC_MINOR, 1144 .name = "xen/gntdev", 1145 .fops = &gntdev_fops, 1146}; 1147 1148/* ------------------------------------------------------------------ */ 1149 1150static int __init gntdev_init(void) 1151{ 1152 int err; 1153 1154 if (!xen_domain()) 1155 return -ENODEV; 1156 1157 use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap); 1158 1159 err = misc_register(&gntdev_miscdev); 1160 if (err != 0) { 1161 pr_err("Could not register gntdev device\n"); 1162 return err; 1163 } 1164 return 0; 1165} 1166 1167static void __exit gntdev_exit(void) 1168{ 1169 misc_deregister(&gntdev_miscdev); 1170} 1171 1172module_init(gntdev_init); 1173module_exit(gntdev_exit); 1174 1175/* ------------------------------------------------------------------ */ 1176