mm: brk: downgrade mmap_sem to read when shrinking
Hello Yang Shi,
The patch ca761a3ea456: "mm: brk: downgrade mmap_sem to read when shrinking" from Oct 4, 2018, leads to the following static checker warning:
mm/mmap.c:252 __do_sys_brk() warn: unsigned 'retval' is never less than zero.
mm/mmap.c
223 /*
224 * Check against rlimit here. If this check is done later after the test
225 * of oldbrk with newbrk then it can escape the test and let the data
226 * segment grow beyond its set limit the in case where the limit is
227 * not page aligned -Ram Gupta
228 */
229 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk,
230 mm->end_data, mm->start_data))
231 goto out;
232
233 newbrk = PAGE_ALIGN(brk);
234 oldbrk = PAGE_ALIGN(mm->brk);
235 if (oldbrk == newbrk) {
236 mm->brk = brk;
237 goto success;
238 }
239
240 /*
241 * Always allow shrinking brk.
242 * __do_munmap() may downgrade mmap_sem to read.
243 */
244 if (brk <= mm->brk) {
245 /*
246 * mm->brk must to be protected by write mmap_sem so update it
247 * before downgrading mmap_sem. When __do_munmap() fails,
248 * mm->brk will be restored from origbrk.
249 */
250 mm->brk = brk;
251 retval = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true);
252 if (retval < 0) {
^^^^^^^^^^Impossible.
253 mm->brk = origbrk; 254 goto out; 255 } else if (retval == 1) 256 downgraded = true; 257 goto success; 258 } 259
See also:mm/mremap.c:571 __do_sys_mremap() warn: unsigned 'ret' is never less than zero.
regards,
dan carpenter
Hi Dan,
Thanks for reporting this bug. Colin Ian King reported the same issue a couple of days ago, please see https://www.mail-archive.com/[email protected]/msg1780112.html.
And, the fixes have been merged into Andrew's -mm tree.
Regards,
Yang
Quoted 56 lines · Dan Carpenter
Hello Yang Shi,
The patch
ca761a3ea456: "mm: brk: downgrademmap_semto read when shrinking" from Oct 4, 2018, leads to the following static checker warning:mm/mmap.c:252 __do_sys_brk() warn: unsigned 'retval' is never less than zero.
mm/mmap.c223 /* 224 * Check against rlimit here. If this check is done later after the test 225 * of oldbrk with newbrk then it can escape the test and let the data 226 * segment grow beyond its set limit the in case where the limit is 227 * not page aligned -Ram Gupta 228 */ 229 if (check_data_rlimit(rlimit(RLIMIT_DATA), brk, mm->start_brk, 230 mm->end_data, mm->start_data)) 231 goto out; 232 233 newbrk = PAGE_ALIGN(brk); 234 oldbrk = PAGE_ALIGN(mm->brk); 235 if (oldbrk == newbrk) { 236 mm->brk = brk; 237 goto success; 238 } 239 240 /* 241 * Always allow shrinking brk. 242 * __do_munmap() may downgrade mmap_sem to read. 243 */ 244 if (brk <= mm->brk) { 245 /* 246 * mm->brk must to be protected by write mmap_sem so update it 247 * before downgrading mmap_sem. When __do_munmap() fails, 248 * mm->brk will be restored from origbrk. 249 */ 250 mm->brk = brk; 251 retval = __do_munmap(mm, newbrk, oldbrk-newbrk, &uf, true); 252 if (retval < 0) { ^^^^^^^^^^Impossible.
253 mm->brk = origbrk; 254 goto out; 255 } else if (retval == 1) 256 downgraded = true; 257 goto success; 258 } 259See also:
mm/mremap.c:571__do_sys_mremap()warn: unsigned 'ret' is never less than zero.regards,
dan carpenter