xfs_trans_buf.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784
  1. /*
  2. * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
  3. * All Rights Reserved.
  4. *
  5. * This program is free software; you can redistribute it and/or
  6. * modify it under the terms of the GNU General Public License as
  7. * published by the Free Software Foundation.
  8. *
  9. * This program is distributed in the hope that it would be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License
  15. * along with this program; if not, write the Free Software Foundation,
  16. * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. #include "xfs.h"
  19. #include "xfs_fs.h"
  20. #include "xfs_types.h"
  21. #include "xfs_bit.h"
  22. #include "xfs_log.h"
  23. #include "xfs_trans.h"
  24. #include "xfs_sb.h"
  25. #include "xfs_ag.h"
  26. #include "xfs_mount.h"
  27. #include "xfs_bmap_btree.h"
  28. #include "xfs_alloc_btree.h"
  29. #include "xfs_ialloc_btree.h"
  30. #include "xfs_dinode.h"
  31. #include "xfs_inode.h"
  32. #include "xfs_buf_item.h"
  33. #include "xfs_trans_priv.h"
  34. #include "xfs_error.h"
  35. #include "xfs_rw.h"
  36. #include "xfs_trace.h"
  37. /*
  38. * Check to see if a buffer matching the given parameters is already
  39. * a part of the given transaction.
  40. */
  41. STATIC struct xfs_buf *
  42. xfs_trans_buf_item_match(
  43. struct xfs_trans *tp,
  44. struct xfs_buftarg *target,
  45. xfs_daddr_t blkno,
  46. int len)
  47. {
  48. struct xfs_log_item_desc *lidp;
  49. struct xfs_buf_log_item *blip;
  50. len = BBTOB(len);
  51. list_for_each_entry(lidp, &tp->t_items, lid_trans) {
  52. blip = (struct xfs_buf_log_item *)lidp->lid_item;
  53. if (blip->bli_item.li_type == XFS_LI_BUF &&
  54. blip->bli_buf->b_target == target &&
  55. XFS_BUF_ADDR(blip->bli_buf) == blkno &&
  56. BBTOB(blip->bli_buf->b_length) == len)
  57. return blip->bli_buf;
  58. }
  59. return NULL;
  60. }
  61. /*
  62. * Add the locked buffer to the transaction.
  63. *
  64. * The buffer must be locked, and it cannot be associated with any
  65. * transaction.
  66. *
  67. * If the buffer does not yet have a buf log item associated with it,
  68. * then allocate one for it. Then add the buf item to the transaction.
  69. */
  70. STATIC void
  71. _xfs_trans_bjoin(
  72. struct xfs_trans *tp,
  73. struct xfs_buf *bp,
  74. int reset_recur)
  75. {
  76. struct xfs_buf_log_item *bip;
  77. ASSERT(bp->b_transp == NULL);
  78. /*
  79. * The xfs_buf_log_item pointer is stored in b_fsprivate. If
  80. * it doesn't have one yet, then allocate one and initialize it.
  81. * The checks to see if one is there are in xfs_buf_item_init().
  82. */
  83. xfs_buf_item_init(bp, tp->t_mountp);
  84. bip = bp->b_fspriv;
  85. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  86. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  87. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  88. if (reset_recur)
  89. bip->bli_recur = 0;
  90. /*
  91. * Take a reference for this transaction on the buf item.
  92. */
  93. atomic_inc(&bip->bli_refcount);
  94. /*
  95. * Get a log_item_desc to point at the new item.
  96. */
  97. xfs_trans_add_item(tp, &bip->bli_item);
  98. /*
  99. * Initialize b_fsprivate2 so we can find it with incore_match()
  100. * in xfs_trans_get_buf() and friends above.
  101. */
  102. bp->b_transp = tp;
  103. }
  104. void
  105. xfs_trans_bjoin(
  106. struct xfs_trans *tp,
  107. struct xfs_buf *bp)
  108. {
  109. _xfs_trans_bjoin(tp, bp, 0);
  110. trace_xfs_trans_bjoin(bp->b_fspriv);
  111. }
  112. /*
  113. * Get and lock the buffer for the caller if it is not already
  114. * locked within the given transaction. If it is already locked
  115. * within the transaction, just increment its lock recursion count
  116. * and return a pointer to it.
  117. *
  118. * If the transaction pointer is NULL, make this just a normal
  119. * get_buf() call.
  120. */
  121. xfs_buf_t *
  122. xfs_trans_get_buf(xfs_trans_t *tp,
  123. xfs_buftarg_t *target_dev,
  124. xfs_daddr_t blkno,
  125. int len,
  126. uint flags)
  127. {
  128. xfs_buf_t *bp;
  129. xfs_buf_log_item_t *bip;
  130. if (flags == 0)
  131. flags = XBF_MAPPED;
  132. /*
  133. * Default to a normal get_buf() call if the tp is NULL.
  134. */
  135. if (tp == NULL)
  136. return xfs_buf_get(target_dev, blkno, len, flags);
  137. /*
  138. * If we find the buffer in the cache with this transaction
  139. * pointer in its b_fsprivate2 field, then we know we already
  140. * have it locked. In this case we just increment the lock
  141. * recursion count and return the buffer to the caller.
  142. */
  143. bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
  144. if (bp != NULL) {
  145. ASSERT(xfs_buf_islocked(bp));
  146. if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
  147. xfs_buf_stale(bp);
  148. XFS_BUF_DONE(bp);
  149. }
  150. ASSERT(bp->b_transp == tp);
  151. bip = bp->b_fspriv;
  152. ASSERT(bip != NULL);
  153. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  154. bip->bli_recur++;
  155. trace_xfs_trans_get_buf_recur(bip);
  156. return (bp);
  157. }
  158. bp = xfs_buf_get(target_dev, blkno, len, flags);
  159. if (bp == NULL) {
  160. return NULL;
  161. }
  162. ASSERT(!bp->b_error);
  163. _xfs_trans_bjoin(tp, bp, 1);
  164. trace_xfs_trans_get_buf(bp->b_fspriv);
  165. return (bp);
  166. }
  167. /*
  168. * Get and lock the superblock buffer of this file system for the
  169. * given transaction.
  170. *
  171. * We don't need to use incore_match() here, because the superblock
  172. * buffer is a private buffer which we keep a pointer to in the
  173. * mount structure.
  174. */
  175. xfs_buf_t *
  176. xfs_trans_getsb(xfs_trans_t *tp,
  177. struct xfs_mount *mp,
  178. int flags)
  179. {
  180. xfs_buf_t *bp;
  181. xfs_buf_log_item_t *bip;
  182. /*
  183. * Default to just trying to lock the superblock buffer
  184. * if tp is NULL.
  185. */
  186. if (tp == NULL) {
  187. return (xfs_getsb(mp, flags));
  188. }
  189. /*
  190. * If the superblock buffer already has this transaction
  191. * pointer in its b_fsprivate2 field, then we know we already
  192. * have it locked. In this case we just increment the lock
  193. * recursion count and return the buffer to the caller.
  194. */
  195. bp = mp->m_sb_bp;
  196. if (bp->b_transp == tp) {
  197. bip = bp->b_fspriv;
  198. ASSERT(bip != NULL);
  199. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  200. bip->bli_recur++;
  201. trace_xfs_trans_getsb_recur(bip);
  202. return (bp);
  203. }
  204. bp = xfs_getsb(mp, flags);
  205. if (bp == NULL)
  206. return NULL;
  207. _xfs_trans_bjoin(tp, bp, 1);
  208. trace_xfs_trans_getsb(bp->b_fspriv);
  209. return (bp);
  210. }
  211. #ifdef DEBUG
  212. xfs_buftarg_t *xfs_error_target;
  213. int xfs_do_error;
  214. int xfs_req_num;
  215. int xfs_error_mod = 33;
  216. #endif
  217. /*
  218. * Get and lock the buffer for the caller if it is not already
  219. * locked within the given transaction. If it has not yet been
  220. * read in, read it from disk. If it is already locked
  221. * within the transaction and already read in, just increment its
  222. * lock recursion count and return a pointer to it.
  223. *
  224. * If the transaction pointer is NULL, make this just a normal
  225. * read_buf() call.
  226. */
  227. int
  228. xfs_trans_read_buf(
  229. xfs_mount_t *mp,
  230. xfs_trans_t *tp,
  231. xfs_buftarg_t *target,
  232. xfs_daddr_t blkno,
  233. int len,
  234. uint flags,
  235. xfs_buf_t **bpp)
  236. {
  237. xfs_buf_t *bp;
  238. xfs_buf_log_item_t *bip;
  239. int error;
  240. *bpp = NULL;
  241. if (flags == 0)
  242. flags = XBF_MAPPED;
  243. /*
  244. * Default to a normal get_buf() call if the tp is NULL.
  245. */
  246. if (tp == NULL) {
  247. bp = xfs_buf_read(target, blkno, len, flags);
  248. if (!bp)
  249. return (flags & XBF_TRYLOCK) ?
  250. EAGAIN : XFS_ERROR(ENOMEM);
  251. if (bp->b_error) {
  252. error = bp->b_error;
  253. xfs_buf_ioerror_alert(bp, __func__);
  254. XFS_BUF_UNDONE(bp);
  255. xfs_buf_stale(bp);
  256. xfs_buf_relse(bp);
  257. return error;
  258. }
  259. #ifdef DEBUG
  260. if (xfs_do_error) {
  261. if (xfs_error_target == target) {
  262. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  263. xfs_buf_relse(bp);
  264. xfs_debug(mp, "Returning error!");
  265. return XFS_ERROR(EIO);
  266. }
  267. }
  268. }
  269. #endif
  270. if (XFS_FORCED_SHUTDOWN(mp))
  271. goto shutdown_abort;
  272. *bpp = bp;
  273. return 0;
  274. }
  275. /*
  276. * If we find the buffer in the cache with this transaction
  277. * pointer in its b_fsprivate2 field, then we know we already
  278. * have it locked. If it is already read in we just increment
  279. * the lock recursion count and return the buffer to the caller.
  280. * If the buffer is not yet read in, then we read it in, increment
  281. * the lock recursion count, and return it to the caller.
  282. */
  283. bp = xfs_trans_buf_item_match(tp, target, blkno, len);
  284. if (bp != NULL) {
  285. ASSERT(xfs_buf_islocked(bp));
  286. ASSERT(bp->b_transp == tp);
  287. ASSERT(bp->b_fspriv != NULL);
  288. ASSERT(!bp->b_error);
  289. if (!(XFS_BUF_ISDONE(bp))) {
  290. trace_xfs_trans_read_buf_io(bp, _RET_IP_);
  291. ASSERT(!XFS_BUF_ISASYNC(bp));
  292. XFS_BUF_READ(bp);
  293. xfsbdstrat(tp->t_mountp, bp);
  294. error = xfs_buf_iowait(bp);
  295. if (error) {
  296. xfs_buf_ioerror_alert(bp, __func__);
  297. xfs_buf_relse(bp);
  298. /*
  299. * We can gracefully recover from most read
  300. * errors. Ones we can't are those that happen
  301. * after the transaction's already dirty.
  302. */
  303. if (tp->t_flags & XFS_TRANS_DIRTY)
  304. xfs_force_shutdown(tp->t_mountp,
  305. SHUTDOWN_META_IO_ERROR);
  306. return error;
  307. }
  308. }
  309. /*
  310. * We never locked this buf ourselves, so we shouldn't
  311. * brelse it either. Just get out.
  312. */
  313. if (XFS_FORCED_SHUTDOWN(mp)) {
  314. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  315. *bpp = NULL;
  316. return XFS_ERROR(EIO);
  317. }
  318. bip = bp->b_fspriv;
  319. bip->bli_recur++;
  320. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  321. trace_xfs_trans_read_buf_recur(bip);
  322. *bpp = bp;
  323. return 0;
  324. }
  325. bp = xfs_buf_read(target, blkno, len, flags);
  326. if (bp == NULL) {
  327. *bpp = NULL;
  328. return (flags & XBF_TRYLOCK) ?
  329. 0 : XFS_ERROR(ENOMEM);
  330. }
  331. if (bp->b_error) {
  332. error = bp->b_error;
  333. xfs_buf_stale(bp);
  334. XFS_BUF_DONE(bp);
  335. xfs_buf_ioerror_alert(bp, __func__);
  336. if (tp->t_flags & XFS_TRANS_DIRTY)
  337. xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
  338. xfs_buf_relse(bp);
  339. return error;
  340. }
  341. #ifdef DEBUG
  342. if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
  343. if (xfs_error_target == target) {
  344. if (((xfs_req_num++) % xfs_error_mod) == 0) {
  345. xfs_force_shutdown(tp->t_mountp,
  346. SHUTDOWN_META_IO_ERROR);
  347. xfs_buf_relse(bp);
  348. xfs_debug(mp, "Returning trans error!");
  349. return XFS_ERROR(EIO);
  350. }
  351. }
  352. }
  353. #endif
  354. if (XFS_FORCED_SHUTDOWN(mp))
  355. goto shutdown_abort;
  356. _xfs_trans_bjoin(tp, bp, 1);
  357. trace_xfs_trans_read_buf(bp->b_fspriv);
  358. *bpp = bp;
  359. return 0;
  360. shutdown_abort:
  361. trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
  362. xfs_buf_relse(bp);
  363. *bpp = NULL;
  364. return XFS_ERROR(EIO);
  365. }
  366. /*
  367. * Release the buffer bp which was previously acquired with one of the
  368. * xfs_trans_... buffer allocation routines if the buffer has not
  369. * been modified within this transaction. If the buffer is modified
  370. * within this transaction, do decrement the recursion count but do
  371. * not release the buffer even if the count goes to 0. If the buffer is not
  372. * modified within the transaction, decrement the recursion count and
  373. * release the buffer if the recursion count goes to 0.
  374. *
  375. * If the buffer is to be released and it was not modified before
  376. * this transaction began, then free the buf_log_item associated with it.
  377. *
  378. * If the transaction pointer is NULL, make this just a normal
  379. * brelse() call.
  380. */
  381. void
  382. xfs_trans_brelse(xfs_trans_t *tp,
  383. xfs_buf_t *bp)
  384. {
  385. xfs_buf_log_item_t *bip;
  386. /*
  387. * Default to a normal brelse() call if the tp is NULL.
  388. */
  389. if (tp == NULL) {
  390. ASSERT(bp->b_transp == NULL);
  391. xfs_buf_relse(bp);
  392. return;
  393. }
  394. ASSERT(bp->b_transp == tp);
  395. bip = bp->b_fspriv;
  396. ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
  397. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  398. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  399. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  400. trace_xfs_trans_brelse(bip);
  401. /*
  402. * If the release is just for a recursive lock,
  403. * then decrement the count and return.
  404. */
  405. if (bip->bli_recur > 0) {
  406. bip->bli_recur--;
  407. return;
  408. }
  409. /*
  410. * If the buffer is dirty within this transaction, we can't
  411. * release it until we commit.
  412. */
  413. if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
  414. return;
  415. /*
  416. * If the buffer has been invalidated, then we can't release
  417. * it until the transaction commits to disk unless it is re-dirtied
  418. * as part of this transaction. This prevents us from pulling
  419. * the item from the AIL before we should.
  420. */
  421. if (bip->bli_flags & XFS_BLI_STALE)
  422. return;
  423. ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
  424. /*
  425. * Free up the log item descriptor tracking the released item.
  426. */
  427. xfs_trans_del_item(&bip->bli_item);
  428. /*
  429. * Clear the hold flag in the buf log item if it is set.
  430. * We wouldn't want the next user of the buffer to
  431. * get confused.
  432. */
  433. if (bip->bli_flags & XFS_BLI_HOLD) {
  434. bip->bli_flags &= ~XFS_BLI_HOLD;
  435. }
  436. /*
  437. * Drop our reference to the buf log item.
  438. */
  439. atomic_dec(&bip->bli_refcount);
  440. /*
  441. * If the buf item is not tracking data in the log, then
  442. * we must free it before releasing the buffer back to the
  443. * free pool. Before releasing the buffer to the free pool,
  444. * clear the transaction pointer in b_fsprivate2 to dissolve
  445. * its relation to this transaction.
  446. */
  447. if (!xfs_buf_item_dirty(bip)) {
  448. /***
  449. ASSERT(bp->b_pincount == 0);
  450. ***/
  451. ASSERT(atomic_read(&bip->bli_refcount) == 0);
  452. ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
  453. ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
  454. xfs_buf_item_relse(bp);
  455. }
  456. bp->b_transp = NULL;
  457. xfs_buf_relse(bp);
  458. }
  459. /*
  460. * Mark the buffer as not needing to be unlocked when the buf item's
  461. * IOP_UNLOCK() routine is called. The buffer must already be locked
  462. * and associated with the given transaction.
  463. */
  464. /* ARGSUSED */
  465. void
  466. xfs_trans_bhold(xfs_trans_t *tp,
  467. xfs_buf_t *bp)
  468. {
  469. xfs_buf_log_item_t *bip = bp->b_fspriv;
  470. ASSERT(bp->b_transp == tp);
  471. ASSERT(bip != NULL);
  472. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  473. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  474. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  475. bip->bli_flags |= XFS_BLI_HOLD;
  476. trace_xfs_trans_bhold(bip);
  477. }
  478. /*
  479. * Cancel the previous buffer hold request made on this buffer
  480. * for this transaction.
  481. */
  482. void
  483. xfs_trans_bhold_release(xfs_trans_t *tp,
  484. xfs_buf_t *bp)
  485. {
  486. xfs_buf_log_item_t *bip = bp->b_fspriv;
  487. ASSERT(bp->b_transp == tp);
  488. ASSERT(bip != NULL);
  489. ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
  490. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
  491. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  492. ASSERT(bip->bli_flags & XFS_BLI_HOLD);
  493. bip->bli_flags &= ~XFS_BLI_HOLD;
  494. trace_xfs_trans_bhold_release(bip);
  495. }
  496. /*
  497. * This is called to mark bytes first through last inclusive of the given
  498. * buffer as needing to be logged when the transaction is committed.
  499. * The buffer must already be associated with the given transaction.
  500. *
  501. * First and last are numbers relative to the beginning of this buffer,
  502. * so the first byte in the buffer is numbered 0 regardless of the
  503. * value of b_blkno.
  504. */
  505. void
  506. xfs_trans_log_buf(xfs_trans_t *tp,
  507. xfs_buf_t *bp,
  508. uint first,
  509. uint last)
  510. {
  511. xfs_buf_log_item_t *bip = bp->b_fspriv;
  512. ASSERT(bp->b_transp == tp);
  513. ASSERT(bip != NULL);
  514. ASSERT(first <= last && last < BBTOB(bp->b_length));
  515. ASSERT(bp->b_iodone == NULL ||
  516. bp->b_iodone == xfs_buf_iodone_callbacks);
  517. /*
  518. * Mark the buffer as needing to be written out eventually,
  519. * and set its iodone function to remove the buffer's buf log
  520. * item from the AIL and free it when the buffer is flushed
  521. * to disk. See xfs_buf_attach_iodone() for more details
  522. * on li_cb and xfs_buf_iodone_callbacks().
  523. * If we end up aborting this transaction, we trap this buffer
  524. * inside the b_bdstrat callback so that this won't get written to
  525. * disk.
  526. */
  527. XFS_BUF_DONE(bp);
  528. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  529. bp->b_iodone = xfs_buf_iodone_callbacks;
  530. bip->bli_item.li_cb = xfs_buf_iodone;
  531. trace_xfs_trans_log_buf(bip);
  532. /*
  533. * If we invalidated the buffer within this transaction, then
  534. * cancel the invalidation now that we're dirtying the buffer
  535. * again. There are no races with the code in xfs_buf_item_unpin(),
  536. * because we have a reference to the buffer this entire time.
  537. */
  538. if (bip->bli_flags & XFS_BLI_STALE) {
  539. bip->bli_flags &= ~XFS_BLI_STALE;
  540. ASSERT(XFS_BUF_ISSTALE(bp));
  541. XFS_BUF_UNSTALE(bp);
  542. bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
  543. }
  544. tp->t_flags |= XFS_TRANS_DIRTY;
  545. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  546. bip->bli_flags |= XFS_BLI_LOGGED;
  547. xfs_buf_item_log(bip, first, last);
  548. }
  549. /*
  550. * Invalidate a buffer that is being used within a transaction.
  551. *
  552. * Typically this is because the blocks in the buffer are being freed, so we
  553. * need to prevent it from being written out when we're done. Allowing it
  554. * to be written again might overwrite data in the free blocks if they are
  555. * reallocated to a file.
  556. *
  557. * We prevent the buffer from being written out by marking it stale. We can't
  558. * get rid of the buf log item at this point because the buffer may still be
  559. * pinned by another transaction. If that is the case, then we'll wait until
  560. * the buffer is committed to disk for the last time (we can tell by the ref
  561. * count) and free it in xfs_buf_item_unpin(). Until that happens we will
  562. * keep the buffer locked so that the buffer and buf log item are not reused.
  563. *
  564. * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
  565. * the buf item. This will be used at recovery time to determine that copies
  566. * of the buffer in the log before this should not be replayed.
  567. *
  568. * We mark the item descriptor and the transaction dirty so that we'll hold
  569. * the buffer until after the commit.
  570. *
  571. * Since we're invalidating the buffer, we also clear the state about which
  572. * parts of the buffer have been logged. We also clear the flag indicating
  573. * that this is an inode buffer since the data in the buffer will no longer
  574. * be valid.
  575. *
  576. * We set the stale bit in the buffer as well since we're getting rid of it.
  577. */
  578. void
  579. xfs_trans_binval(
  580. xfs_trans_t *tp,
  581. xfs_buf_t *bp)
  582. {
  583. xfs_buf_log_item_t *bip = bp->b_fspriv;
  584. ASSERT(bp->b_transp == tp);
  585. ASSERT(bip != NULL);
  586. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  587. trace_xfs_trans_binval(bip);
  588. if (bip->bli_flags & XFS_BLI_STALE) {
  589. /*
  590. * If the buffer is already invalidated, then
  591. * just return.
  592. */
  593. ASSERT(XFS_BUF_ISSTALE(bp));
  594. ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
  595. ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
  596. ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
  597. ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
  598. ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
  599. return;
  600. }
  601. xfs_buf_stale(bp);
  602. bip->bli_flags |= XFS_BLI_STALE;
  603. bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
  604. bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
  605. bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
  606. memset((char *)(bip->bli_format.blf_data_map), 0,
  607. (bip->bli_format.blf_map_size * sizeof(uint)));
  608. bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
  609. tp->t_flags |= XFS_TRANS_DIRTY;
  610. }
  611. /*
  612. * This call is used to indicate that the buffer contains on-disk inodes which
  613. * must be handled specially during recovery. They require special handling
  614. * because only the di_next_unlinked from the inodes in the buffer should be
  615. * recovered. The rest of the data in the buffer is logged via the inodes
  616. * themselves.
  617. *
  618. * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
  619. * transferred to the buffer's log format structure so that we'll know what to
  620. * do at recovery time.
  621. */
  622. void
  623. xfs_trans_inode_buf(
  624. xfs_trans_t *tp,
  625. xfs_buf_t *bp)
  626. {
  627. xfs_buf_log_item_t *bip = bp->b_fspriv;
  628. ASSERT(bp->b_transp == tp);
  629. ASSERT(bip != NULL);
  630. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  631. bip->bli_flags |= XFS_BLI_INODE_BUF;
  632. }
  633. /*
  634. * This call is used to indicate that the buffer is going to
  635. * be staled and was an inode buffer. This means it gets
  636. * special processing during unpin - where any inodes
  637. * associated with the buffer should be removed from ail.
  638. * There is also special processing during recovery,
  639. * any replay of the inodes in the buffer needs to be
  640. * prevented as the buffer may have been reused.
  641. */
  642. void
  643. xfs_trans_stale_inode_buf(
  644. xfs_trans_t *tp,
  645. xfs_buf_t *bp)
  646. {
  647. xfs_buf_log_item_t *bip = bp->b_fspriv;
  648. ASSERT(bp->b_transp == tp);
  649. ASSERT(bip != NULL);
  650. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  651. bip->bli_flags |= XFS_BLI_STALE_INODE;
  652. bip->bli_item.li_cb = xfs_buf_iodone;
  653. }
  654. /*
  655. * Mark the buffer as being one which contains newly allocated
  656. * inodes. We need to make sure that even if this buffer is
  657. * relogged as an 'inode buf' we still recover all of the inode
  658. * images in the face of a crash. This works in coordination with
  659. * xfs_buf_item_committed() to ensure that the buffer remains in the
  660. * AIL at its original location even after it has been relogged.
  661. */
  662. /* ARGSUSED */
  663. void
  664. xfs_trans_inode_alloc_buf(
  665. xfs_trans_t *tp,
  666. xfs_buf_t *bp)
  667. {
  668. xfs_buf_log_item_t *bip = bp->b_fspriv;
  669. ASSERT(bp->b_transp == tp);
  670. ASSERT(bip != NULL);
  671. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  672. bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
  673. }
  674. /*
  675. * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
  676. * dquots. However, unlike in inode buffer recovery, dquot buffers get
  677. * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
  678. * The only thing that makes dquot buffers different from regular
  679. * buffers is that we must not replay dquot bufs when recovering
  680. * if a _corresponding_ quotaoff has happened. We also have to distinguish
  681. * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
  682. * can be turned off independently.
  683. */
  684. /* ARGSUSED */
  685. void
  686. xfs_trans_dquot_buf(
  687. xfs_trans_t *tp,
  688. xfs_buf_t *bp,
  689. uint type)
  690. {
  691. xfs_buf_log_item_t *bip = bp->b_fspriv;
  692. ASSERT(bp->b_transp == tp);
  693. ASSERT(bip != NULL);
  694. ASSERT(type == XFS_BLF_UDQUOT_BUF ||
  695. type == XFS_BLF_PDQUOT_BUF ||
  696. type == XFS_BLF_GDQUOT_BUF);
  697. ASSERT(atomic_read(&bip->bli_refcount) > 0);
  698. bip->bli_format.blf_flags |= type;
  699. }