]>
Commit | Line | Data |
---|---|---|
e53d678d MM |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | ||
22 | /* | |
23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
24 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. | |
25 | * Copyright (c) 2015 by Chunwei Chen. All rights reserved. | |
26 | * Copyright 2017 Nexenta Systems, Inc. | |
27 | */ | |
28 | ||
29 | /* Portions Copyright 2007 Jeremy Teo */ | |
30 | /* Portions Copyright 2010 Robert Milkowski */ | |
31 | ||
32 | #include <sys/types.h> | |
33 | #include <sys/param.h> | |
34 | #include <sys/time.h> | |
35 | #include <sys/sysmacros.h> | |
36 | #include <sys/vfs.h> | |
c0801bf3 | 37 | #include <sys/uio_impl.h> |
e53d678d MM |
38 | #include <sys/file.h> |
39 | #include <sys/stat.h> | |
40 | #include <sys/kmem.h> | |
41 | #include <sys/cmn_err.h> | |
42 | #include <sys/errno.h> | |
43 | #include <sys/zfs_dir.h> | |
44 | #include <sys/zfs_acl.h> | |
45 | #include <sys/zfs_ioctl.h> | |
46 | #include <sys/fs/zfs.h> | |
47 | #include <sys/dmu.h> | |
48 | #include <sys/dmu_objset.h> | |
49 | #include <sys/spa.h> | |
50 | #include <sys/txg.h> | |
51 | #include <sys/dbuf.h> | |
52 | #include <sys/policy.h> | |
53 | #include <sys/zfs_vnops.h> | |
54 | #include <sys/zfs_quota.h> | |
ab8c935e CS |
55 | #include <sys/zfs_vfsops.h> |
56 | #include <sys/zfs_znode.h> | |
e53d678d MM |
57 | |
58 | ||
59 | static ulong_t zfs_fsync_sync_cnt = 4; | |
60 | ||
61 | int | |
62 | zfs_fsync(znode_t *zp, int syncflag, cred_t *cr) | |
63 | { | |
64 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
65 | ||
66 | (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); | |
67 | ||
68 | if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { | |
69 | ZFS_ENTER(zfsvfs); | |
70 | ZFS_VERIFY_ZP(zp); | |
71 | zil_commit(zfsvfs->z_log, zp->z_id); | |
72 | ZFS_EXIT(zfsvfs); | |
73 | } | |
74 | tsd_set(zfs_fsyncer_key, NULL); | |
75 | ||
76 | return (0); | |
77 | } | |
78 | ||
8583540c MM |
79 | |
80 | #if defined(SEEK_HOLE) && defined(SEEK_DATA) | |
81 | /* | |
82 | * Lseek support for finding holes (cmd == SEEK_HOLE) and | |
83 | * data (cmd == SEEK_DATA). "off" is an in/out parameter. | |
84 | */ | |
85 | static int | |
86 | zfs_holey_common(znode_t *zp, ulong_t cmd, loff_t *off) | |
87 | { | |
88 | uint64_t noff = (uint64_t)*off; /* new offset */ | |
89 | uint64_t file_sz; | |
90 | int error; | |
91 | boolean_t hole; | |
92 | ||
93 | file_sz = zp->z_size; | |
94 | if (noff >= file_sz) { | |
95 | return (SET_ERROR(ENXIO)); | |
96 | } | |
97 | ||
98 | if (cmd == F_SEEK_HOLE) | |
99 | hole = B_TRUE; | |
100 | else | |
101 | hole = B_FALSE; | |
102 | ||
103 | error = dmu_offset_next(ZTOZSB(zp)->z_os, zp->z_id, hole, &noff); | |
104 | ||
105 | if (error == ESRCH) | |
106 | return (SET_ERROR(ENXIO)); | |
107 | ||
108 | /* file was dirty, so fall back to using generic logic */ | |
109 | if (error == EBUSY) { | |
110 | if (hole) | |
111 | *off = file_sz; | |
112 | ||
113 | return (0); | |
114 | } | |
115 | ||
116 | /* | |
117 | * We could find a hole that begins after the logical end-of-file, | |
118 | * because dmu_offset_next() only works on whole blocks. If the | |
119 | * EOF falls mid-block, then indicate that the "virtual hole" | |
120 | * at the end of the file begins at the logical EOF, rather than | |
121 | * at the end of the last block. | |
122 | */ | |
123 | if (noff > file_sz) { | |
124 | ASSERT(hole); | |
125 | noff = file_sz; | |
126 | } | |
127 | ||
128 | if (noff < *off) | |
129 | return (error); | |
130 | *off = noff; | |
131 | return (error); | |
132 | } | |
133 | ||
134 | int | |
135 | zfs_holey(znode_t *zp, ulong_t cmd, loff_t *off) | |
136 | { | |
137 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
138 | int error; | |
139 | ||
140 | ZFS_ENTER(zfsvfs); | |
141 | ZFS_VERIFY_ZP(zp); | |
142 | ||
143 | error = zfs_holey_common(zp, cmd, off); | |
144 | ||
145 | ZFS_EXIT(zfsvfs); | |
146 | return (error); | |
147 | } | |
148 | #endif /* SEEK_HOLE && SEEK_DATA */ | |
149 | ||
150 | /*ARGSUSED*/ | |
151 | int | |
152 | zfs_access(znode_t *zp, int mode, int flag, cred_t *cr) | |
153 | { | |
154 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
155 | int error; | |
156 | ||
157 | ZFS_ENTER(zfsvfs); | |
158 | ZFS_VERIFY_ZP(zp); | |
159 | ||
160 | if (flag & V_ACE_MASK) | |
161 | error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); | |
162 | else | |
163 | error = zfs_zaccess_rwx(zp, mode, flag, cr); | |
164 | ||
165 | ZFS_EXIT(zfsvfs); | |
166 | return (error); | |
167 | } | |
168 | ||
e53d678d MM |
169 | static unsigned long zfs_vnops_read_chunk_size = 1024 * 1024; /* Tunable */ |
170 | ||
171 | /* | |
172 | * Read bytes from specified file into supplied buffer. | |
173 | * | |
174 | * IN: zp - inode of file to be read from. | |
175 | * uio - structure supplying read location, range info, | |
176 | * and return buffer. | |
177 | * ioflag - O_SYNC flags; used to provide FRSYNC semantics. | |
178 | * O_DIRECT flag; used to bypass page cache. | |
179 | * cr - credentials of caller. | |
180 | * | |
181 | * OUT: uio - updated offset and range, buffer filled. | |
182 | * | |
183 | * RETURN: 0 on success, error code on failure. | |
184 | * | |
185 | * Side Effects: | |
186 | * inode - atime updated if byte count > 0 | |
187 | */ | |
188 | /* ARGSUSED */ | |
189 | int | |
d0cd9a5c | 190 | zfs_read(struct znode *zp, zfs_uio_t *uio, int ioflag, cred_t *cr) |
e53d678d MM |
191 | { |
192 | int error = 0; | |
193 | boolean_t frsync = B_FALSE; | |
194 | ||
195 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
196 | ZFS_ENTER(zfsvfs); | |
197 | ZFS_VERIFY_ZP(zp); | |
198 | ||
199 | if (zp->z_pflags & ZFS_AV_QUARANTINED) { | |
200 | ZFS_EXIT(zfsvfs); | |
201 | return (SET_ERROR(EACCES)); | |
202 | } | |
203 | ||
204 | /* We don't copy out anything useful for directories. */ | |
205 | if (Z_ISDIR(ZTOTYPE(zp))) { | |
206 | ZFS_EXIT(zfsvfs); | |
207 | return (SET_ERROR(EISDIR)); | |
208 | } | |
209 | ||
210 | /* | |
211 | * Validate file offset | |
212 | */ | |
d0cd9a5c | 213 | if (zfs_uio_offset(uio) < (offset_t)0) { |
e53d678d MM |
214 | ZFS_EXIT(zfsvfs); |
215 | return (SET_ERROR(EINVAL)); | |
216 | } | |
217 | ||
218 | /* | |
219 | * Fasttrack empty reads | |
220 | */ | |
d0cd9a5c | 221 | if (zfs_uio_resid(uio) == 0) { |
e53d678d MM |
222 | ZFS_EXIT(zfsvfs); |
223 | return (0); | |
224 | } | |
225 | ||
226 | #ifdef FRSYNC | |
227 | /* | |
228 | * If we're in FRSYNC mode, sync out this znode before reading it. | |
229 | * Only do this for non-snapshots. | |
230 | * | |
231 | * Some platforms do not support FRSYNC and instead map it | |
232 | * to O_SYNC, which results in unnecessary calls to zil_commit. We | |
233 | * only honor FRSYNC requests on platforms which support it. | |
234 | */ | |
235 | frsync = !!(ioflag & FRSYNC); | |
236 | #endif | |
237 | if (zfsvfs->z_log && | |
238 | (frsync || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS)) | |
239 | zil_commit(zfsvfs->z_log, zp->z_id); | |
240 | ||
241 | /* | |
242 | * Lock the range against changes. | |
243 | */ | |
244 | zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock, | |
d0cd9a5c | 245 | zfs_uio_offset(uio), zfs_uio_resid(uio), RL_READER); |
e53d678d MM |
246 | |
247 | /* | |
248 | * If we are reading past end-of-file we can skip | |
249 | * to the end; but we might still need to set atime. | |
250 | */ | |
d0cd9a5c | 251 | if (zfs_uio_offset(uio) >= zp->z_size) { |
e53d678d MM |
252 | error = 0; |
253 | goto out; | |
254 | } | |
255 | ||
d0cd9a5c | 256 | ASSERT(zfs_uio_offset(uio) < zp->z_size); |
05679465 | 257 | #if defined(__linux__) |
59eab109 | 258 | ssize_t start_offset = zfs_uio_offset(uio); |
05679465 | 259 | #endif |
d0cd9a5c | 260 | ssize_t n = MIN(zfs_uio_resid(uio), zp->z_size - zfs_uio_offset(uio)); |
e53d678d MM |
261 | ssize_t start_resid = n; |
262 | ||
263 | while (n > 0) { | |
264 | ssize_t nbytes = MIN(n, zfs_vnops_read_chunk_size - | |
d0cd9a5c | 265 | P2PHASE(zfs_uio_offset(uio), zfs_vnops_read_chunk_size)); |
e53d678d | 266 | #ifdef UIO_NOCOPY |
d0cd9a5c | 267 | if (zfs_uio_segflg(uio) == UIO_NOCOPY) |
e53d678d MM |
268 | error = mappedread_sf(zp, nbytes, uio); |
269 | else | |
270 | #endif | |
271 | if (zn_has_cached_data(zp) && !(ioflag & O_DIRECT)) { | |
272 | error = mappedread(zp, nbytes, uio); | |
273 | } else { | |
274 | error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), | |
275 | uio, nbytes); | |
276 | } | |
277 | ||
278 | if (error) { | |
279 | /* convert checksum errors into IO errors */ | |
280 | if (error == ECKSUM) | |
281 | error = SET_ERROR(EIO); | |
05679465 RE |
282 | |
283 | #if defined(__linux__) | |
59eab109 RE |
284 | /* |
285 | * if we actually read some bytes, bubbling EFAULT | |
05679465 RE |
286 | * up to become EAGAIN isn't what we want here... |
287 | * | |
288 | * ...on Linux, at least. On FBSD, doing this breaks. | |
59eab109 RE |
289 | */ |
290 | if (error == EFAULT && | |
291 | (zfs_uio_offset(uio) - start_offset) != 0) | |
292 | error = 0; | |
05679465 | 293 | #endif |
e53d678d MM |
294 | break; |
295 | } | |
296 | ||
297 | n -= nbytes; | |
298 | } | |
299 | ||
300 | int64_t nread = start_resid - n; | |
301 | dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, nread); | |
302 | task_io_account_read(nread); | |
303 | out: | |
304 | zfs_rangelock_exit(lr); | |
305 | ||
306 | ZFS_ACCESSTIME_STAMP(zfsvfs, zp); | |
307 | ZFS_EXIT(zfsvfs); | |
308 | return (error); | |
309 | } | |
310 | ||
311 | /* | |
312 | * Write the bytes to a file. | |
313 | * | |
314 | * IN: zp - znode of file to be written to. | |
315 | * uio - structure supplying write location, range info, | |
316 | * and data buffer. | |
317 | * ioflag - O_APPEND flag set if in append mode. | |
318 | * O_DIRECT flag; used to bypass page cache. | |
319 | * cr - credentials of caller. | |
320 | * | |
321 | * OUT: uio - updated offset and range. | |
322 | * | |
323 | * RETURN: 0 if success | |
324 | * error code if failure | |
325 | * | |
326 | * Timestamps: | |
327 | * ip - ctime|mtime updated if byte count > 0 | |
328 | */ | |
329 | ||
330 | /* ARGSUSED */ | |
331 | int | |
d0cd9a5c | 332 | zfs_write(znode_t *zp, zfs_uio_t *uio, int ioflag, cred_t *cr) |
e53d678d MM |
333 | { |
334 | int error = 0; | |
d0cd9a5c | 335 | ssize_t start_resid = zfs_uio_resid(uio); |
e53d678d MM |
336 | |
337 | /* | |
338 | * Fasttrack empty write | |
339 | */ | |
340 | ssize_t n = start_resid; | |
341 | if (n == 0) | |
342 | return (0); | |
343 | ||
e53d678d MM |
344 | zfsvfs_t *zfsvfs = ZTOZSB(zp); |
345 | ZFS_ENTER(zfsvfs); | |
346 | ZFS_VERIFY_ZP(zp); | |
347 | ||
348 | sa_bulk_attr_t bulk[4]; | |
349 | int count = 0; | |
350 | uint64_t mtime[2], ctime[2]; | |
351 | SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); | |
352 | SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); | |
353 | SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, | |
354 | &zp->z_size, 8); | |
355 | SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, | |
356 | &zp->z_pflags, 8); | |
357 | ||
358 | /* | |
359 | * Callers might not be able to detect properly that we are read-only, | |
360 | * so check it explicitly here. | |
361 | */ | |
362 | if (zfs_is_readonly(zfsvfs)) { | |
363 | ZFS_EXIT(zfsvfs); | |
364 | return (SET_ERROR(EROFS)); | |
365 | } | |
366 | ||
367 | /* | |
4b2e2082 RM |
368 | * If immutable or not appending then return EPERM. |
369 | * Intentionally allow ZFS_READONLY through here. | |
370 | * See zfs_zaccess_common() | |
e53d678d | 371 | */ |
4b2e2082 | 372 | if ((zp->z_pflags & ZFS_IMMUTABLE) || |
e53d678d | 373 | ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & O_APPEND) && |
d0cd9a5c | 374 | (zfs_uio_offset(uio) < zp->z_size))) { |
e53d678d MM |
375 | ZFS_EXIT(zfsvfs); |
376 | return (SET_ERROR(EPERM)); | |
377 | } | |
378 | ||
379 | /* | |
380 | * Validate file offset | |
381 | */ | |
d0cd9a5c | 382 | offset_t woff = ioflag & O_APPEND ? zp->z_size : zfs_uio_offset(uio); |
e53d678d MM |
383 | if (woff < 0) { |
384 | ZFS_EXIT(zfsvfs); | |
385 | return (SET_ERROR(EINVAL)); | |
386 | } | |
387 | ||
9a764716 | 388 | const uint64_t max_blksz = zfsvfs->z_max_blksz; |
e53d678d MM |
389 | |
390 | /* | |
391 | * Pre-fault the pages to ensure slow (eg NFS) pages | |
392 | * don't hold up txg. | |
393 | * Skip this if uio contains loaned arc_buf. | |
394 | */ | |
d0cd9a5c | 395 | if (zfs_uio_prefaultpages(MIN(n, max_blksz), uio)) { |
e53d678d MM |
396 | ZFS_EXIT(zfsvfs); |
397 | return (SET_ERROR(EFAULT)); | |
398 | } | |
399 | ||
400 | /* | |
401 | * If in append mode, set the io offset pointer to eof. | |
402 | */ | |
403 | zfs_locked_range_t *lr; | |
404 | if (ioflag & O_APPEND) { | |
405 | /* | |
406 | * Obtain an appending range lock to guarantee file append | |
407 | * semantics. We reset the write offset once we have the lock. | |
408 | */ | |
409 | lr = zfs_rangelock_enter(&zp->z_rangelock, 0, n, RL_APPEND); | |
410 | woff = lr->lr_offset; | |
411 | if (lr->lr_length == UINT64_MAX) { | |
412 | /* | |
413 | * We overlocked the file because this write will cause | |
414 | * the file block size to increase. | |
415 | * Note that zp_size cannot change with this lock held. | |
416 | */ | |
417 | woff = zp->z_size; | |
418 | } | |
d0cd9a5c | 419 | zfs_uio_setoffset(uio, woff); |
e53d678d MM |
420 | } else { |
421 | /* | |
422 | * Note that if the file block size will change as a result of | |
423 | * this write, then this range lock will lock the entire file | |
424 | * so that we can re-write the block safely. | |
425 | */ | |
426 | lr = zfs_rangelock_enter(&zp->z_rangelock, woff, n, RL_WRITER); | |
427 | } | |
428 | ||
d0cd9a5c | 429 | if (zn_rlimit_fsize(zp, uio)) { |
e53d678d MM |
430 | zfs_rangelock_exit(lr); |
431 | ZFS_EXIT(zfsvfs); | |
7e3617de | 432 | return (SET_ERROR(EFBIG)); |
e53d678d MM |
433 | } |
434 | ||
d1dd72a2 RM |
435 | const rlim64_t limit = MAXOFFSET_T; |
436 | ||
e53d678d MM |
437 | if (woff >= limit) { |
438 | zfs_rangelock_exit(lr); | |
439 | ZFS_EXIT(zfsvfs); | |
440 | return (SET_ERROR(EFBIG)); | |
441 | } | |
442 | ||
d1dd72a2 | 443 | if (n > limit - woff) |
e53d678d MM |
444 | n = limit - woff; |
445 | ||
446 | uint64_t end_size = MAX(zp->z_size, woff + n); | |
447 | zilog_t *zilog = zfsvfs->z_log; | |
448 | ||
eec6646e RM |
449 | const uint64_t uid = KUID_TO_SUID(ZTOUID(zp)); |
450 | const uint64_t gid = KGID_TO_SGID(ZTOGID(zp)); | |
451 | const uint64_t projid = zp->z_projid; | |
452 | ||
e53d678d MM |
453 | /* |
454 | * Write the file in reasonable size chunks. Each chunk is written | |
455 | * in a separate transaction; this keeps the intent log records small | |
456 | * and allows us to do more fine-grained space accounting. | |
457 | */ | |
458 | while (n > 0) { | |
d0cd9a5c | 459 | woff = zfs_uio_offset(uio); |
e53d678d | 460 | |
eec6646e RM |
461 | if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, uid) || |
462 | zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, gid) || | |
463 | (projid != ZFS_DEFAULT_PROJID && | |
e53d678d | 464 | zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, |
eec6646e | 465 | projid))) { |
e53d678d MM |
466 | error = SET_ERROR(EDQUOT); |
467 | break; | |
468 | } | |
469 | ||
470 | arc_buf_t *abuf = NULL; | |
471 | if (n >= max_blksz && woff >= zp->z_size && | |
472 | P2PHASE(woff, max_blksz) == 0 && | |
473 | zp->z_blksz == max_blksz) { | |
474 | /* | |
475 | * This write covers a full block. "Borrow" a buffer | |
476 | * from the dmu so that we can fill it before we enter | |
477 | * a transaction. This avoids the possibility of | |
478 | * holding up the transaction if the data copy hangs | |
479 | * up on a pagefault (e.g., from an NFS server mapping). | |
480 | */ | |
481 | size_t cbytes; | |
482 | ||
483 | abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), | |
484 | max_blksz); | |
485 | ASSERT(abuf != NULL); | |
486 | ASSERT(arc_buf_size(abuf) == max_blksz); | |
d0cd9a5c | 487 | if ((error = zfs_uiocopy(abuf->b_data, max_blksz, |
e53d678d MM |
488 | UIO_WRITE, uio, &cbytes))) { |
489 | dmu_return_arcbuf(abuf); | |
490 | break; | |
491 | } | |
1c2358c1 | 492 | ASSERT3S(cbytes, ==, max_blksz); |
e53d678d MM |
493 | } |
494 | ||
495 | /* | |
496 | * Start a transaction. | |
497 | */ | |
498 | dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); | |
499 | dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); | |
500 | dmu_buf_impl_t *db = (dmu_buf_impl_t *)sa_get_db(zp->z_sa_hdl); | |
501 | DB_DNODE_ENTER(db); | |
502 | dmu_tx_hold_write_by_dnode(tx, DB_DNODE(db), woff, | |
503 | MIN(n, max_blksz)); | |
504 | DB_DNODE_EXIT(db); | |
505 | zfs_sa_upgrade_txholds(tx, zp); | |
506 | error = dmu_tx_assign(tx, TXG_WAIT); | |
507 | if (error) { | |
508 | dmu_tx_abort(tx); | |
509 | if (abuf != NULL) | |
510 | dmu_return_arcbuf(abuf); | |
511 | break; | |
512 | } | |
513 | ||
514 | /* | |
515 | * If rangelock_enter() over-locked we grow the blocksize | |
516 | * and then reduce the lock range. This will only happen | |
517 | * on the first iteration since rangelock_reduce() will | |
518 | * shrink down lr_length to the appropriate size. | |
519 | */ | |
520 | if (lr->lr_length == UINT64_MAX) { | |
521 | uint64_t new_blksz; | |
522 | ||
523 | if (zp->z_blksz > max_blksz) { | |
524 | /* | |
525 | * File's blocksize is already larger than the | |
526 | * "recordsize" property. Only let it grow to | |
527 | * the next power of 2. | |
528 | */ | |
529 | ASSERT(!ISP2(zp->z_blksz)); | |
530 | new_blksz = MIN(end_size, | |
531 | 1 << highbit64(zp->z_blksz)); | |
532 | } else { | |
533 | new_blksz = MIN(end_size, max_blksz); | |
534 | } | |
535 | zfs_grow_blocksize(zp, new_blksz, tx); | |
536 | zfs_rangelock_reduce(lr, woff, n); | |
537 | } | |
538 | ||
539 | /* | |
540 | * XXX - should we really limit each write to z_max_blksz? | |
541 | * Perhaps we should use SPA_MAXBLOCKSIZE chunks? | |
542 | */ | |
85703f61 RM |
543 | const ssize_t nbytes = |
544 | MIN(n, max_blksz - P2PHASE(woff, max_blksz)); | |
e53d678d MM |
545 | |
546 | ssize_t tx_bytes; | |
547 | if (abuf == NULL) { | |
d0cd9a5c BA |
548 | tx_bytes = zfs_uio_resid(uio); |
549 | zfs_uio_fault_disable(uio, B_TRUE); | |
e53d678d MM |
550 | error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), |
551 | uio, nbytes, tx); | |
d0cd9a5c | 552 | zfs_uio_fault_disable(uio, B_FALSE); |
e53d678d MM |
553 | #ifdef __linux__ |
554 | if (error == EFAULT) { | |
555 | dmu_tx_commit(tx); | |
556 | /* | |
557 | * Account for partial writes before | |
558 | * continuing the loop. | |
559 | * Update needs to occur before the next | |
d0cd9a5c | 560 | * zfs_uio_prefaultpages, or prefaultpages may |
e53d678d MM |
561 | * error, and we may break the loop early. |
562 | */ | |
d0cd9a5c BA |
563 | if (tx_bytes != zfs_uio_resid(uio)) |
564 | n -= tx_bytes - zfs_uio_resid(uio); | |
565 | if (zfs_uio_prefaultpages(MIN(n, max_blksz), | |
566 | uio)) { | |
e53d678d MM |
567 | break; |
568 | } | |
569 | continue; | |
570 | } | |
571 | #endif | |
572 | if (error != 0) { | |
573 | dmu_tx_commit(tx); | |
574 | break; | |
575 | } | |
d0cd9a5c | 576 | tx_bytes -= zfs_uio_resid(uio); |
e53d678d | 577 | } else { |
85703f61 RM |
578 | /* Implied by abuf != NULL: */ |
579 | ASSERT3S(n, >=, max_blksz); | |
85703f61 | 580 | ASSERT0(P2PHASE(woff, max_blksz)); |
e53d678d | 581 | /* |
85703f61 RM |
582 | * We can simplify nbytes to MIN(n, max_blksz) since |
583 | * P2PHASE(woff, max_blksz) is 0, and knowing | |
584 | * n >= max_blksz lets us simplify further: | |
e53d678d | 585 | */ |
85703f61 | 586 | ASSERT3S(nbytes, ==, max_blksz); |
e53d678d | 587 | /* |
85703f61 RM |
588 | * Thus, we're writing a full block at a block-aligned |
589 | * offset and extending the file past EOF. | |
590 | * | |
591 | * dmu_assign_arcbuf_by_dbuf() will directly assign the | |
592 | * arc buffer to a dbuf. | |
e53d678d | 593 | */ |
85703f61 RM |
594 | error = dmu_assign_arcbuf_by_dbuf( |
595 | sa_get_db(zp->z_sa_hdl), woff, abuf, tx); | |
596 | if (error != 0) { | |
597 | dmu_return_arcbuf(abuf); | |
598 | dmu_tx_commit(tx); | |
599 | break; | |
e53d678d | 600 | } |
d0cd9a5c BA |
601 | ASSERT3S(nbytes, <=, zfs_uio_resid(uio)); |
602 | zfs_uioskip(uio, nbytes); | |
85703f61 | 603 | tx_bytes = nbytes; |
e53d678d MM |
604 | } |
605 | if (tx_bytes && zn_has_cached_data(zp) && | |
606 | !(ioflag & O_DIRECT)) { | |
8a9634e2 | 607 | update_pages(zp, woff, tx_bytes, zfsvfs->z_os); |
e53d678d MM |
608 | } |
609 | ||
610 | /* | |
611 | * If we made no progress, we're done. If we made even | |
612 | * partial progress, update the znode and ZIL accordingly. | |
613 | */ | |
614 | if (tx_bytes == 0) { | |
615 | (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), | |
616 | (void *)&zp->z_size, sizeof (uint64_t), tx); | |
617 | dmu_tx_commit(tx); | |
618 | ASSERT(error != 0); | |
619 | break; | |
620 | } | |
621 | ||
622 | /* | |
623 | * Clear Set-UID/Set-GID bits on successful write if not | |
624 | * privileged and at least one of the execute bits is set. | |
625 | * | |
626 | * It would be nice to do this after all writes have | |
627 | * been done, but that would still expose the ISUID/ISGID | |
628 | * to another app after the partial write is committed. | |
629 | * | |
630 | * Note: we don't call zfs_fuid_map_id() here because | |
631 | * user 0 is not an ephemeral uid. | |
632 | */ | |
633 | mutex_enter(&zp->z_acl_lock); | |
e53d678d MM |
634 | if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | |
635 | (S_IXUSR >> 6))) != 0 && | |
636 | (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && | |
637 | secpolicy_vnode_setid_retain(zp, cr, | |
638 | ((zp->z_mode & S_ISUID) != 0 && uid == 0)) != 0) { | |
639 | uint64_t newmode; | |
640 | zp->z_mode &= ~(S_ISUID | S_ISGID); | |
f8ce8aed | 641 | newmode = zp->z_mode; |
e53d678d MM |
642 | (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), |
643 | (void *)&newmode, sizeof (uint64_t), tx); | |
644 | } | |
645 | mutex_exit(&zp->z_acl_lock); | |
646 | ||
647 | zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime); | |
648 | ||
649 | /* | |
650 | * Update the file size (zp_size) if it has changed; | |
651 | * account for possible concurrent updates. | |
652 | */ | |
d0cd9a5c | 653 | while ((end_size = zp->z_size) < zfs_uio_offset(uio)) { |
e53d678d | 654 | (void) atomic_cas_64(&zp->z_size, end_size, |
d0cd9a5c | 655 | zfs_uio_offset(uio)); |
e53d678d MM |
656 | ASSERT(error == 0); |
657 | } | |
658 | /* | |
659 | * If we are replaying and eof is non zero then force | |
660 | * the file size to the specified eof. Note, there's no | |
661 | * concurrency during replay. | |
662 | */ | |
663 | if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) | |
664 | zp->z_size = zfsvfs->z_replay_eof; | |
665 | ||
666 | error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); | |
667 | ||
668 | zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag, | |
669 | NULL, NULL); | |
670 | dmu_tx_commit(tx); | |
671 | ||
672 | if (error != 0) | |
673 | break; | |
1c2358c1 | 674 | ASSERT3S(tx_bytes, ==, nbytes); |
e53d678d MM |
675 | n -= nbytes; |
676 | ||
677 | if (n > 0) { | |
d0cd9a5c | 678 | if (zfs_uio_prefaultpages(MIN(n, max_blksz), uio)) { |
7e3617de | 679 | error = SET_ERROR(EFAULT); |
e53d678d MM |
680 | break; |
681 | } | |
682 | } | |
683 | } | |
684 | ||
fc273894 | 685 | zfs_znode_update_vfs(zp); |
e53d678d MM |
686 | zfs_rangelock_exit(lr); |
687 | ||
688 | /* | |
7e3617de RM |
689 | * If we're in replay mode, or we made no progress, or the |
690 | * uio data is inaccessible return an error. Otherwise, it's | |
691 | * at least a partial write, so it's successful. | |
e53d678d | 692 | */ |
d0cd9a5c | 693 | if (zfsvfs->z_replay || zfs_uio_resid(uio) == start_resid || |
7e3617de | 694 | error == EFAULT) { |
e53d678d MM |
695 | ZFS_EXIT(zfsvfs); |
696 | return (error); | |
697 | } | |
698 | ||
699 | if (ioflag & (O_SYNC | O_DSYNC) || | |
700 | zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) | |
701 | zil_commit(zilog, zp->z_id); | |
702 | ||
d0cd9a5c | 703 | const int64_t nwritten = start_resid - zfs_uio_resid(uio); |
e53d678d MM |
704 | dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, nwritten); |
705 | task_io_account_write(nwritten); | |
706 | ||
707 | ZFS_EXIT(zfsvfs); | |
708 | return (0); | |
709 | } | |
710 | ||
711 | /*ARGSUSED*/ | |
712 | int | |
713 | zfs_getsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr) | |
714 | { | |
715 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
716 | int error; | |
717 | boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; | |
718 | ||
719 | ZFS_ENTER(zfsvfs); | |
720 | ZFS_VERIFY_ZP(zp); | |
721 | error = zfs_getacl(zp, vsecp, skipaclchk, cr); | |
722 | ZFS_EXIT(zfsvfs); | |
723 | ||
724 | return (error); | |
725 | } | |
726 | ||
727 | /*ARGSUSED*/ | |
728 | int | |
729 | zfs_setsecattr(znode_t *zp, vsecattr_t *vsecp, int flag, cred_t *cr) | |
730 | { | |
731 | zfsvfs_t *zfsvfs = ZTOZSB(zp); | |
732 | int error; | |
733 | boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; | |
734 | zilog_t *zilog = zfsvfs->z_log; | |
735 | ||
736 | ZFS_ENTER(zfsvfs); | |
737 | ZFS_VERIFY_ZP(zp); | |
738 | ||
739 | error = zfs_setacl(zp, vsecp, skipaclchk, cr); | |
740 | ||
741 | if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) | |
742 | zil_commit(zilog, 0); | |
743 | ||
744 | ZFS_EXIT(zfsvfs); | |
745 | return (error); | |
746 | } | |
747 | ||
ab8c935e CS |
748 | #ifdef ZFS_DEBUG |
749 | static int zil_fault_io = 0; | |
750 | #endif | |
751 | ||
752 | static void zfs_get_done(zgd_t *zgd, int error); | |
753 | ||
754 | /* | |
755 | * Get data to generate a TX_WRITE intent log record. | |
756 | */ | |
757 | int | |
296a4a36 CC |
758 | zfs_get_data(void *arg, uint64_t gen, lr_write_t *lr, char *buf, |
759 | struct lwb *lwb, zio_t *zio) | |
ab8c935e CS |
760 | { |
761 | zfsvfs_t *zfsvfs = arg; | |
762 | objset_t *os = zfsvfs->z_os; | |
763 | znode_t *zp; | |
764 | uint64_t object = lr->lr_foid; | |
765 | uint64_t offset = lr->lr_offset; | |
766 | uint64_t size = lr->lr_length; | |
767 | dmu_buf_t *db; | |
768 | zgd_t *zgd; | |
769 | int error = 0; | |
296a4a36 | 770 | uint64_t zp_gen; |
ab8c935e CS |
771 | |
772 | ASSERT3P(lwb, !=, NULL); | |
773 | ASSERT3P(zio, !=, NULL); | |
774 | ASSERT3U(size, !=, 0); | |
775 | ||
776 | /* | |
777 | * Nothing to do if the file has been removed | |
778 | */ | |
779 | if (zfs_zget(zfsvfs, object, &zp) != 0) | |
780 | return (SET_ERROR(ENOENT)); | |
781 | if (zp->z_unlinked) { | |
782 | /* | |
783 | * Release the vnode asynchronously as we currently have the | |
784 | * txg stopped from syncing. | |
785 | */ | |
786 | zfs_zrele_async(zp); | |
787 | return (SET_ERROR(ENOENT)); | |
788 | } | |
296a4a36 CC |
789 | /* check if generation number matches */ |
790 | if (sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, | |
791 | sizeof (zp_gen)) != 0) { | |
792 | zfs_zrele_async(zp); | |
793 | return (SET_ERROR(EIO)); | |
794 | } | |
795 | if (zp_gen != gen) { | |
796 | zfs_zrele_async(zp); | |
797 | return (SET_ERROR(ENOENT)); | |
798 | } | |
ab8c935e CS |
799 | |
800 | zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); | |
801 | zgd->zgd_lwb = lwb; | |
802 | zgd->zgd_private = zp; | |
803 | ||
804 | /* | |
805 | * Write records come in two flavors: immediate and indirect. | |
806 | * For small writes it's cheaper to store the data with the | |
807 | * log record (immediate); for large writes it's cheaper to | |
808 | * sync the data and get a pointer to it (indirect) so that | |
809 | * we don't have to write the data twice. | |
810 | */ | |
811 | if (buf != NULL) { /* immediate write */ | |
812 | zgd->zgd_lr = zfs_rangelock_enter(&zp->z_rangelock, | |
813 | offset, size, RL_READER); | |
814 | /* test for truncation needs to be done while range locked */ | |
815 | if (offset >= zp->z_size) { | |
816 | error = SET_ERROR(ENOENT); | |
817 | } else { | |
818 | error = dmu_read(os, object, offset, size, buf, | |
819 | DMU_READ_NO_PREFETCH); | |
820 | } | |
821 | ASSERT(error == 0 || error == ENOENT); | |
822 | } else { /* indirect write */ | |
823 | /* | |
824 | * Have to lock the whole block to ensure when it's | |
825 | * written out and its checksum is being calculated | |
826 | * that no one can change the data. We need to re-check | |
827 | * blocksize after we get the lock in case it's changed! | |
828 | */ | |
829 | for (;;) { | |
830 | uint64_t blkoff; | |
831 | size = zp->z_blksz; | |
832 | blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; | |
833 | offset -= blkoff; | |
834 | zgd->zgd_lr = zfs_rangelock_enter(&zp->z_rangelock, | |
835 | offset, size, RL_READER); | |
836 | if (zp->z_blksz == size) | |
837 | break; | |
838 | offset += blkoff; | |
839 | zfs_rangelock_exit(zgd->zgd_lr); | |
840 | } | |
841 | /* test for truncation needs to be done while range locked */ | |
842 | if (lr->lr_offset >= zp->z_size) | |
843 | error = SET_ERROR(ENOENT); | |
844 | #ifdef ZFS_DEBUG | |
845 | if (zil_fault_io) { | |
846 | error = SET_ERROR(EIO); | |
847 | zil_fault_io = 0; | |
848 | } | |
849 | #endif | |
850 | if (error == 0) | |
851 | error = dmu_buf_hold(os, object, offset, zgd, &db, | |
852 | DMU_READ_NO_PREFETCH); | |
853 | ||
854 | if (error == 0) { | |
855 | blkptr_t *bp = &lr->lr_blkptr; | |
856 | ||
857 | zgd->zgd_db = db; | |
858 | zgd->zgd_bp = bp; | |
859 | ||
860 | ASSERT(db->db_offset == offset); | |
861 | ASSERT(db->db_size == size); | |
862 | ||
863 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
864 | zfs_get_done, zgd); | |
865 | ASSERT(error || lr->lr_length <= size); | |
866 | ||
867 | /* | |
868 | * On success, we need to wait for the write I/O | |
869 | * initiated by dmu_sync() to complete before we can | |
870 | * release this dbuf. We will finish everything up | |
871 | * in the zfs_get_done() callback. | |
872 | */ | |
873 | if (error == 0) | |
874 | return (0); | |
875 | ||
876 | if (error == EALREADY) { | |
877 | lr->lr_common.lrc_txtype = TX_WRITE2; | |
878 | /* | |
879 | * TX_WRITE2 relies on the data previously | |
880 | * written by the TX_WRITE that caused | |
881 | * EALREADY. We zero out the BP because | |
882 | * it is the old, currently-on-disk BP. | |
883 | */ | |
884 | zgd->zgd_bp = NULL; | |
885 | BP_ZERO(bp); | |
886 | error = 0; | |
887 | } | |
888 | } | |
889 | } | |
890 | ||
891 | zfs_get_done(zgd, error); | |
892 | ||
893 | return (error); | |
894 | } | |
895 | ||
896 | ||
897 | /* ARGSUSED */ | |
898 | static void | |
899 | zfs_get_done(zgd_t *zgd, int error) | |
900 | { | |
901 | znode_t *zp = zgd->zgd_private; | |
902 | ||
903 | if (zgd->zgd_db) | |
904 | dmu_buf_rele(zgd->zgd_db, zgd); | |
905 | ||
906 | zfs_rangelock_exit(zgd->zgd_lr); | |
907 | ||
908 | /* | |
909 | * Release the vnode asynchronously as we currently have the | |
910 | * txg stopped from syncing. | |
911 | */ | |
912 | zfs_zrele_async(zp); | |
913 | ||
914 | kmem_free(zgd, sizeof (zgd_t)); | |
915 | } | |
916 | ||
8583540c | 917 | EXPORT_SYMBOL(zfs_access); |
e53d678d | 918 | EXPORT_SYMBOL(zfs_fsync); |
8583540c | 919 | EXPORT_SYMBOL(zfs_holey); |
e53d678d MM |
920 | EXPORT_SYMBOL(zfs_read); |
921 | EXPORT_SYMBOL(zfs_write); | |
922 | EXPORT_SYMBOL(zfs_getsecattr); | |
923 | EXPORT_SYMBOL(zfs_setsecattr); | |
924 | ||
925 | ZFS_MODULE_PARAM(zfs_vnops, zfs_vnops_, read_chunk_size, ULONG, ZMOD_RW, | |
926 | "Bytes to read per chunk"); |